Theory and Explanation in Social Psychology 9781462518517, 9781462518487

Citation preview

ebook THE GUILFORD PRESS

Theory and Explanation in Social Psychology



Also Available Cognitive Consistency: A Fundamental Principle in Social Cognition Edited by Bertram Gawronski and Fritz Strack

Dual-Process Theories of the Social Mind Edited by Jeffrey W. Sherman, Bertram Gawronski, and Yaacov Trope

Handbook of Implicit Social Cognition: Measurement, Theory, and Applications Edited by Bertram Gawronski and B. Keith Payne



Theory and Explanation in Social Psychology edited by

Bertram Gawronski Galen V. Bodenhausen

The Guilford Press New York  London



© 2015 The Guilford Press A Division of Guilford Publications, Inc. 72 Spring Street, New York, NY 10012 www.guilford.com All rights reserved No part of this book may be reproduced, translated, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, microfilming, recording, or otherwise, without written permission from the publisher. Printed in the United States of America This book is printed on acid-free paper. Last digit is print number: 9 8 7 6 5 4 3 2 1 Library of Congress Cataloging-in-Publication Data Theory and explanation in social psychology / edited by Bertram Gawronski, Galen V. Bodenhausen.   pages cm   Includes bibliographical references and index.   ISBN 978-1-4625-1848-7 (hardback)   1.  Social psychology.  2.  Cognitive psychology.  I.  Gawronski, Bertram.  II.  Bodenhausen, Galen V. (Galen Von), 1961–   HM1033.T436 2014  302—dc23 2014040661



About the Editors

Bertram Gawronski, PhD, is Professor of Psychology at the University of Texas at Austin. His research investigates the mental underpinnings and behavioral consequences of spontaneous and deliberate evaluations of objects, individuals, groups, and social issues. Dr. Gawronski’s work has been recognized with the Theoretical Innovation Prize from the Society for Personality and Social Psychology, the Career Trajectory Award from the Society of Experimental Social Psychology, the Early Career Award from the International Social Cognition Network, the Early Researcher Award from the Ministry of Research and Innovation of Ontario, and the Charlotte and Karl Bühler Award from the German Psychological Society. He is an elected Fellow of the Association for Psychological Science, the Society of Experimental Social Psychology, the Society for Personality and Social Psychology, and the Midwestern Psychological Association, and a member of several editorial boards, including the Journal of Personality and Social Psychology, the Personality and Social Psychology Bulletin, the European Review of Social Psychology, and Social Psychological and Personality Science. Galen V. Bodenhausen, PhD, is Lawyer Taylor Professor of Psychology and Professor of Marketing at Northwestern University. Dr. Bodenhausen studies a wide variety of issues related to social cognition, such as the origins, nature, and consequences of social attitudes, including both explicit and implicit (or automatic) attitudes; the role of identity concerns in judgment and behavior; the influence of prejudice and stereotypes on perception, judgment, memory, and behavior; how moods and other kinds of emotional states influence judgment and preference; and the nature and consequences of social and consumer values. He is an elected Fellow of the Association for Psychological Science, the American Psychological Association, the Society for Personality and Social Psychology, and the Society for the Psychological Study of Social Issues, and a member of several editorial boards, including the Journal of Experimental Psychology: General, the Journal of Personality and Social Psychology, and Personality and Social Psychology Review. v

Contributors

Ximena B. Arriaga, PhD, Department of Psychological Sciences, Purdue University, West Lafayette, Indiana Asaf Beasley, BS, Department of Psychological and Brain Sciences, Indiana University, Bloomington, Bloomington, Indiana Jennifer S. Beer, PhD, Department of Psychology and Imaging Research Center, University of Texas at Austin, Austin, Texas Galen V. Bodenhausen, PhD, Department of Psychology, Northwestern University, Evanston, Illinois Tracy L. Caldwell, PhD, Department of Psychology, Dominican University, River Forest, Illinois Daniel Cervone, PhD, Department of Psychology, University of Illinois at Chicago, Chicago, Illinois Jan De Houwer, PhD, Department of Experimental Clinical and Health Psychology, Ghent University, Ghent, Belgium Roland Deutsch, PhD, Department of Psychology, Dresden University of Technology, Dresden, Germany David Dunning, PhD, Department of Psychology, Cornell University, Ithaca, New York Kimin Eom, MA, Department of Psychological and Brain Sciences, University of California, Santa Barbara, Santa Barbara, California Klaus Fiedler, PhD, Department of Psychology, Heidelberg University, Heidelberg, Germany Katrina Fincher, BS, Department of Psychology, University of Pennsylvania, Philadelphia, Pennsylvania vi

Contributors vii Margarida V. Garrido, PhD, Department of Social and Organizational Psychology, University Institute of Lisbon, Lisbon, Portugal Bertram Gawronski, PhD, Department of Psychology, University of Texas at Austin, Austin, Texas Wendy Johnson, PhD, Department of Psychology, University of Edinburgh, Edinburgh, Scotland, United Kingdom Yoshihisa Kashima, PhD, Melbourne School of Psychological Sciences, University of Melbourne, Parkville, Victoria, Australia Timothy Ketelaar, PhD, Department of Psychology, New Mexico State University, Las Cruces, New Mexico Heejung S. Kim, PhD, Department of Psychological and Brain Sciences, University of California, Santa Barbara, Santa Barbara, California Karl Christoph Klauer, PhD, Institute for Psychology, Albert Ludwigs University of Freiburg, Freiburg, Germany Florian L. Kutzner, PhD, Department of Psychology, Heidelberg University, Heidelberg, Germany Antony S. R. Manstead, PhD, School of Psychology, Cardiff University, Cardiff, Wales, United Kingdom Nicole D. Mayer, MA, Department of Psychology, University of Illinois at Chicago, Chicago, Illinois Agnes Moors, PhD, Department of Experimental Clinical and Health Psychology, Ghent University, Ghent, Belgium Brian Parkinson, PhD, Department of Experimental Psychology, University of Oxford, Oxford, United Kingdom Lars Penke, PhD, Georg Elias Müller Institute of Psychology, Georg August University Göttingen, Göttingen, Germany Harry T. Reis, PhD, Department of Clinical and Social Sciences in Psychology, University of Rochester, Rochester, New York Gün R. Semin, PhD, Director, William James Center for Research, ISPA–Instituto Universitário, Lisbon, Portugal Eliot R. Smith, PhD, Department of Psychological and Brain Sciences, Indiana University, Bloomington, Bloomington, Indiana Philip E. Tetlock, PhD, Department of Psychology, University of Pennsylvania, Philadelphia, Pennsylvania David Trafimow, PhD, Department of Psychology, New Mexico State University, Las Cruces, New Mexico

Preface

I

n a comment on the state of theory in social psychology, Kruglanski (2001) noted the surprising absence of training in the construction and evaluation of theories in our discipline. Most graduate programs include excellent training in research methods and data analysis, but there is hardly any training in the criteria for good social-psychological theorizing. Although the contents of classic and contemporary theories are very well covered by most curricula, metatheoretical principles for the construction and evaluation of theories are hardly ever explicitly addressed. As Kruglanski observed, this omission can be detrimental for scientific progress in social psychology, because it can hamper epistemologically sound and logically valid explanations of social phenomena. There are undoubtedly multiple factors that have contributed to the development of this state of affairs. Moreover, to the extent that earlier generations of scholars did not receive training in theory construction and theory evaluation, they are unlikely to provide such training to the next generations of social psychologists, thereby perpetuating the status quo. The goal of the current volume is to break this status quo by offering a comprehensive resource for the development of knowledge and skills bearing on the construction and evaluation of social-psychological theory. Prioritizing this oft-neglected aspect of the scholarly toolbox promises not only to strengthen the contributions of individual researchers but also, ultimately, to place the entire discipline on a firmer conceptual footing. Toward this end, we invited leading theorists to review the conceptual foundations of the field’s most influential theoretical approaches, scrutinizing the range and limits of theories in various areas of inquiry. The chapters describe basic principles of logical inference, illustrate common fallacies in viii

Preface ix

theoretical interpretations of empirical findings, and address recurring challenges facing theory development across the gamut of social-psychological topics. By outlining the philosophical foundations of scientific explanation, the volume illuminates the unique contributions of different levels of analysis within social psychology. The chapters of the first section review domain-independent principles of social-psychological theorizing, including general criteria for theory evaluation, the explanatory function of different levels of analysis, and the relation between causal and meaning-based explanation. The chapters of the following sections analyze the logical and metatheoretical foundations of social-psychological theories within various areas of inquiry. These areas are divided into five broader categories, including theories that explain social phenomena with reference to mental states (e.g., social-cognitive theories, emotion theories, motivation theories, personality theories, dual-process theories), theories addressing the biological underpinnings of social phenomena (e.g., cognitive neuroscience, genetics of social behavior, evolutionary theories), theories that emphasize pragmatic aspects of social situations (e.g., rational actor theories, social functionalism, situated and embodied theories), theories about the role of social contexts and social structures (e.g., interdependence theories, cultural theories), and formalized theories of social behavior (e.g., computer simulations, mathematical models, agentbased models). Important questions that are addressed in the chapters are the following: What phenomena do particular theories aim to explain and what are the theoretical assumptions that are proposed to explain these phenomena? Are there particular structural features of theories and explanations in a given area? What are their implications for the kind of data that would be required to confirm or disconfirm a theory or hypothesis? Are there unquestioned core assumptions or premises? Are there different levels of analysis? How do these levels support and build on each other? Are there trade-offs between explanatory breadth and predictive power? What characteristics should a theory have to avoid such trade-offs? Are there common logical fallacies in theoretical interpretations of empirical data? What can be done to avoid these fallacies? Are there justified concerns about lack of falsifiability? Are there concerns about theoretical parsimony? In general, what characteristics should a good social-psychological theory have? What are unique challenges for theories and explanations in a particular area?

x Preface

What can theories in a given area tell us? What are their limits? We hope that the current volume will be useful to both students and established researchers who are interested in scientific explanations of social behavior. Our aim is to help readers profit from an enhanced ability to evaluate theoretical claims critically and to develop more powerful explanatory accounts in their own scholarship. The book can be used as a primary or supplementary text for graduate courses in social psychology; individual chapters also can serve as supplements for specialized courses on particular topics. Reference Kruglanski, A. W. (2001). That “vision thing”: Theory construction in social and personality psychology at the edge of the new millennium. Journal of Personality and Social Psychology, 80, 871–875.

Contents

PART I.  Basics  1. Theory Evaluation

3

Bertram Gawronski and Galen V. Bodenhausen

 2. Levels of Analysis in Social Psychology

24

Jan De Houwer and Agnes Moors

 3. Causal and Meaning-Based Explanation

41

Yoshihisa Kashima

PART II.  Mental State Theories  4. Social-Cognitive Theories

65

Bertram Gawronski and Galen V. Bodenhausen

 5. Emotion Theories

84

Antony S. R. Manstead and Brian Parkinson

 6. Motivational Theories David Dunning xi

108

xii Contents

 7. Duality Models in Social Psychology

132

Roland Deutsch

 8. Personality Systems and the Coherence of Social Behavior

157

Daniel Cervone, Tracy L. Caldwell, and Nicole D. Mayer

PART III.  Biological Theories   9. Cognitive Neuroscience of Social Behavior

183

Jennifer S. Beer

10. Genetics of Social Behavior

205

Wendy Johnson and Lars Penke

11. Evolutionary Theories

224

Timothy Ketelaar

PART IV.  Pragmatic Theories 12. Rational Actor Theories

245

David Trafimow

13. Social Functionalism

266

Philip E. Tetlock and Katrina Fincher

14. Socially Situated Cognition

283

Gün R. Semin and Margarida V. Garrido

PART V. Social Theories 15. Interdependence Theory and Related Theories

305

Harry T. Reis and Ximena B. Arriaga

16. Cultural Psychological Theory Kimin Eom and Heejung S. Kim

328

Contents xiii

PART VI. Formal Theories 17. Computer Simulation

347

Klaus Fiedler and Florian L. Kutzner

18. Mathematical Modeling

371

Karl Christoph Klauer

19. Agent-Based Modeling

390

Eliot R. Smith and Asaf Beasley

Author Index

409



423

Subject Index



Part I Basics





1 Theory Evaluation Bertram Gawronski Galen V. Bodenhausen

F

ew aphorisms have been repeated more often by social psychologists than Kurt Lewin’s (1951) claim that there is nothing as practical as a good theory. Social psychologists strive to explain and predict social behavior not only to expand our understanding of human nature but also to identify leverage points for effective interventions that can remediate pressing social problems. As Lewin argued, the key to these endeavors is the formulation of good theories—but what constitutes a good theory? Psychological theories aim at identifying general principles that can be used to better understand why people behave the way they do (explanation) and to forecast how people will behave in particular situations (prediction). Hence, good theories should be consistent with empirical observations in that they can make sense of past observations of behavior and correctly predict future observations. These empirical criteria for deciding whether or not a particular idea constitutes a “good theory” seem intuitively straightforward, and social psychologists have been prolific in formulating an abundance of theories that are consistent with the empirical findings generated in a wide range of areas (Van Lange, Kruglanski, & Higgins, 2012). To document the goodness of their theories, researchers typically seek to obtain confirmatory evidence in that they present evidence that is consis 3

4 BASICS

tent with their preferred theoretical account. This confirmatory approach to theory evaluation has well-understood limitations (Greenwald, Pratkanis, Leippe, & Baumgardner, 1986), including the failure to consider potentially superior rival theoretical accounts for the same phenomenon. Consistency with empirical findings is a necessary but not a sufficient condition for establishing the goodness of a theory. Thus, a central question is how researchers can evaluate social-psychological theories over and above their mere consistency with empirical observations. Philosophers of science have identified several criteria that are suitable for the evaluation of scientific theories. The main goal of the current chapter is to review these criteria and to discuss their relevance for theorizing in social psychology.

Induction, Deduction, and the Logic of Falsification The question of theory evaluation played an important role in historical debates on how to distinguish between scientific and nonscientific statements. In the philosophy of science, this question is known as the demarcation problem (Popper, 1934). Resonating with the philosophical notion of positivism, a common answer to this question is that scientific theories should be based on empirical observations. To the extent that a statement is not based on empirical observations, it should be regarded as speculative rather than scientific. From this perspective, the scientific status of a given theory increases as a function of the number of empirical observations it is based on: the more empirical observations have been gathered to support a given theory, the higher is its scientific status. Although the positivist approach to theory evaluation resonates with lay conceptions of science as establishing irrevocable truths, it has been criticized for presupposing a logical principle of induction that could establish the truth of a general statement on the basis of individual observations— a principle that does not exist. As prominently outlined by Popper (1934), it is logically impossible to inductively verify the truth of a general statement on the basis of individual observations. For example, it is impossible to establish the truth of the general statement all swans are white on the basis of individual observations of white swans. After all, it is still possible that, despite painstaking counting of white swans, there is a black swan somewhere out there that was missed. Yet, classical logic does allow establishing the falsity of the general statement all swans are white through the observation of a single black swan.1 In general terms, Popper’s analysis suggests that, although it is impossible to inductively establish the truth of scientific theories (verification), it is possible to deductively establish their falsity (falsification). In classical logic, the two kinds of inferences have been depicted in the form of valid and invalid syllogisms. The invalid syllogism underlying inductive inferences of truth can be depicted as



Theory Evaluation 5

  T → O

[If theory T is true, then observation O should occur.]

  O

[Observation O occurs.]

  Therefore, T

[Therefore, theory T is true.]

As noted above, this syllogism is invalid because there is no logical principle that allows inductive inferences about the truth of a general statement on the basis of individual observations. Yet, it is possible to draw deductive inferences about the falsity of a general statement, as depicted in the valid syllogism of the modus tollens:   T → O

[If theory T is true, then observation O should occur.]

  ¬O

[Observation O does not occur.]

  Therefore, ¬T

[Therefore, theory T is not true.]

These insights have important implications for the evaluation of scientific theories. In contrast to the positivist answer to the demarcation problem, Popper (1934) argued that scientific theories should be distinguished from nonscientific theories on the basis of their falsifiability. According to Popper, the falsifiability of a theory increases not as a function of the observations that are implied by the theory, but as a function of the observations that are prohibited by the theory. To the extent that a theory does not prohibit any observation, it is consistent with any potential finding, and thus unfalsifiable. For example, a theory of attitude change implying that attitudes may either change or remain unaffected in response to a persuasive message would be consistent with any empirical outcome, and thus unfalsifiable. Yet, a theory that implies specific predictions about the conditions under which attitudes do change versus do not change in response to a persuasive message would be inconsistent with certain empirical outcomes, and thus falsifiable. In fact, counter to the common goal of constructing theories that capture a wide range of possible observations, the informational value of a scientific theory increases with the number of events that should not happen according to the theory. To illustrate this idea, imagine that the weather channel tells us that tomorrow will be either rainy or sunny. In this case, we clearly have not learned much. However, if the weather channel tells us that tomorrow will bring sunshine, rain is ruled out as a possible event, thereby increasing the informational value of the forecast. Another implication of Popper’s (1934) analysis is that the idea of science as establishing irrevocable truths is an illusion. There is no logical principle that could inductively verify the truth of a scientific theory. Even after painstaking accumulation of confirmatory evidence for a particular theory, it is always possible that a new study provides evidence that disconfirms the theory. Popper proposed the term fallibilism to describe the insight that any well-established theory could be disconfirmed by new evidence. To

6 BASICS

be sure, Popper’s rejection of inductivism as a basis for theory evaluation does not mean that scientists never engage in inductive thinking when they develop theories on the basis of previous observations. It simply means that individual observations can never guarantee the truth of a scientific theory. Whereas the former process refers to the context of theory construction as a creative activity (see Kruglanski & Higgins, 2004), the latter process refers to the context of theory evaluation as an epistemic activity that is constrained by the principles of classical logic. Popper’s falsifiability criterion further implies that scientific theories can be evaluated on the basis of their logical structure independent of any empirical evidence. Of course, logical analysis does not replace observation and experimentation as the empirical foundation of science. It simply means that scientific theories can be scrutinized for their logical structure without knowing whether they are consistent or inconsistent with the available evidence (Machado & Silva, 2007). For example, a minimum requirement for all scientific theories is that they should be logically coherent. If a theory is logically incoherent, it is essentially unfalsifiable because incoherent premises are consistent with any possible conclusion. For instance, if a given theory implies that self-relevance increases cognitive elaboration and, at the same time, that self-relevance decreases cognitive elaboration, it could be reconciled with any empirical outcome. Thus, to the extent that thorough conceptual analysis reveals that a given theory is logically incoherent, it does not matter that the theory is consistent with the available evidence because it would be consistent with any empirical observation. Although conceptual analyses for logical coherence are relatively rare in social psychology, there are a few examples in which well-established theories revealed logically incoherent assumptions, thereby challenging their status as scientific in terms of Popper’s criteria for theory evaluation (see Trafimow, Chapter 12, this volume). As a caveat, it is worth noting that logical incoherence can be rather difficult to detect, in particular for verbally formulated theories that rely only on the informal logic of natural syntax rather than mathematical formalizations (see Fiedler & Kutzner, Chapter 17, this volume; Klauer, Chapter 18, this volume; Smith & Beasley, Chapter 19, this volume). Johnson-Laird (2012) pointed out that a theory that includes n propositions can be inconsistent even if any n – 1 of them yields a consistent set. The implication for the identification of logical incoherence is illustrated by the fact that an exhaustive consistency assessment of a set of 100 propositions requires the consideration of 2100 possibilities. Even if each possibility could be examined in a millionth of a second, a comprehensive examination would take longer than the universe has existed. These issues suggest that theories are often more valuable if they include fewer rather than more assumptions. Theories involving a high number of propositions may give the superficial impression that they are precise and comprehensive. Yet, such theories would be useless according to Popper (1934) if they included unidentified inconsistencies.



Theory Evaluation 7

Another important criterion in the logical analysis of scientific hypotheses is whether they are empirical or tautological. For example, the general statement all bachelors are unmarried men is tautological because the term bachelor is semantically equivalent to unmarried man. Hence, it serves little purpose to empirically investigate whether all bachelors are unmarried men. The statement is true by virtue of the semantic meaning of the two concepts. Some prominent social-psychological theories have been criticized for being pseudo-empirical, in that quasi-tautological claims make them resistant to any kind of counterevidence (e.g., Greve, 2001; Smedslund, 2000; Wallach & Wallach, 1994). For example, the hypothesis difficult cognitive tasks require more cognitive resources than easy cognitive tasks may be criticized as pseudoempirical because the difficulty of a cognitive task is defined in terms of the cognitive resources it requires. If a cognitive task that was classified as difficult turned out to require fewer cognitive resources than a cognitive task that was classified as easy, we would not conclude that the proposed relation between task difficulty and cognitive resources is incorrect (i.e., cognitive tasks that are difficult actually require fewer cognitive resources than cognitive tasks that are easy). Instead, we would revise our beliefs about the relative difficulty of the two tasks, in that the task that was supposed to be difficult was in fact easy, and vice versa. Thus, in addition to identifying whether a theory is logically coherent, conceptual analysis is important to identify tautological claims that are true by definition instead of reflecting theoretical hypotheses that can be true or false.

The Pragmatics of Falsification and Holistic Theory Evaluation Popper’s (1934) deductive approach to theory evaluation is widely accepted in social psychology in that it serves as the conceptual basis for the practice of null-hypothesis testing. The basic idea is that tests for statistical significance do not verify a theoretically derived alternative hypothesis H1, but instead falsify the null hypothesis H0 (with an accepted alpha error probability of p < .05). Yet, the notion of falsification becomes much more complex when it is applied to the refutation of theories. Although the modus tollens provides a logical basis for deductive conclusions about the falsity of a general statement, the deduction and interpretation of observations are hardly ever based solely on a given theory. Rather, the deduction and interpretation of observations usually require the acceptance of numerous extra-theoretical background assumptions, including assumptions about the operationalization and measurement of the relevant theoretical constructs (McGrath, 1981; Proctor & Capaldi, 2001). Thus, it is not the theory alone that is subject to empirical test, but the theory in conjunction with all background assumptions that are required for the deduction and interpretation of a given observation (Duhem, 1908; Quine, 1953). Hence, if the prediction

8 BASICS

is disconfirmed, the empirical observation does not falsify the theory, but the conjunction of the theory and all background assumptions that are required to logically derive the predicted observation. In logical terms, this inference can be depicted in the following manner: (T ^ A1 ^ A2 ^ ... ^ Ai) → O ¬O Therefore, ¬(T ^ A1 ^ A2 ^ ... ^ Ai) Thus, what researchers learn from a disconfirmed prediction is that the conjunction as a whole is incorrect, but the disconfirmed prediction is insufficient to specify which particular component of the conjunction is incorrect. It might be the theory, but it could also be one of the background assumptions. In social psychology, researchers typically try to resolve this ambiguity by conducting additional tests of their background assumptions (e.g., manipulation checks to test the effectiveness of experimental manipulations; independent tests to establish the accuracy of measurement theories). On the basis of this practice, one might be tempted to conclude that a theory is falsified if all relevant background assumptions have been confirmed. Yet, what should be clear from Popper’s (1934) rejection of inductivism is that it is impossible to establish conclusively the truth of the background assumptions. In other words, although the modus tollens provides a logical basis for deductive inferences about the falsity of sets of theoretical assumptions, the impossibility of inductively verifying the truth of general statements makes it impossible to conclusively ascertain the falsity of a particular theory. Without verification, there is no conclusive falsification. In the philosophy of science, this insight is known as the Duhem–Quine thesis, with reference to its originators Pierre Duhem (1908) and Willard Van Orman Quine (1953). The impossibility of conclusively verifying or falsifying scientific theories led some philosophers to claim that science is an “anarchical” business that is guided by the principle of “anything goes” (e.g., Feyerabend, 1975). Yet, counter to such pessimistic views, others conceptualized the evaluation of scientific theories as consistency tests of broader sets of theoretical and empirical assumptions (e.g., Hempel, 1965; Quine & Ullian, 1978). Using the extended depiction of the modus tollens in its application to actual theory testing, one could argue that disconfirmed predictions signal the inconsistency of a broader set of assumptions that includes the theory, the relevant background assumptions, and the empirical observation. It is not possible to accept all of these propositions at the same time, because their inconsistency indicates that at least one of them must be wrong. To identify which proposition should be rejected, each of them can be scrutinized by testing the consistency of other sets that include one (or more) of the original propositions. To the extent that a given proposition is part of multiple sets that turn out



Theory Evaluation 9

to be consistent, it will likely be treated as “correct.” If, however, a proposition is part of multiple sets that turn out to be inconsistent, it will likely be rejected as “false.” Such consistency checks may involve earlier studies, for example when a theory previously led to predictions that were empirically confirmed or disconfirmed. Alternatively, consistency checks may involve the derivation of novel predictions, for example, when a post-hoc explanation for a disconfirmed prediction leads to the derivation of a new prediction (Lakatos, 1970). These consistency checks resemble Popper’s (1934) tests of logical coherence in that both are concerned with the logical consistency of a given set of propositions. Yet, they are broader in that holistic network tests involve not only the consistency of the theory itself, but the entire network of theoretical and empirical assumptions (Quine, 1953). According to Popper (1934), a logically incoherent theory should be rejected even if it is consistent with the available evidence. Moreover, the notion of holistic theory evaluation implies that a logically coherent theory may be rejected if it is inconsistent with the broader network of theoretical and empirical assumptions.

The Evaluation of Scientific Research Programs Holistic network checks for logical consistency provide an answer to the questions that arise when Popper’s (1934) deductive approach is applied to the pragmatics of theory testing in psychological science. However, a holistic reinterpretation of deductive theory testing also has important implications that are not evident from Popper’s original analysis. Because holistic network checks depend on the current network of accepted propositions, and because there is no possibility of conclusively verifying or falsifying any of these propositions, the outcome of any consistency check is contingent upon potential changes of the current network. For example, even if the current network suggests the rejection of a particular theory, new evidence may suggest a rejection of an involved measurement theory or an operationalization assumption. In other words, it is always possible that a theory that is accepted today will be rejected tomorrow and conversely that a theory that is rejected today will be resurrected tomorrow (see Aronson, 1992, and Kluger & Tikochinsky, 2001, for discussions of examples). Such developments of holistic networks over time are a central theme in Lakatos’s (1970) analysis of scientific research programs. Lakatos distinguishes between what he calls the hard core and the protective belt of a scientific research program. The hard core includes the central assumptions of a given theory; the protective belt includes a large set of background assumptions that are needed to derive testable predictions from the theory (e.g., measurement theories; operationalizations). According to Lakatos, researchers typically protect the theoretical core in the face of disconfirmed predictions by making adjustments in the protective belt. Thus, instead of interpreting the disconfirmed prediction as evidence against their theory, scientists tend to

10 BASICS

“neutralize” the negative evidence by searching for potential problems with the background assumptions of the protective belt. Whereas the protection of the theoretical core is described as a negative heuristic in response to disconfirming evidence, the revision of background assumptions in the protective belt is described as a positive heuristic. Although Lakatos’s analysis of how scientists deal with disconfirming evidence may sound like a textbook example of motivated reasoning (Kunda, 1990), it fully embraces the idea of rational theory evaluation. What matters according to Lakatos is whether revisions in the protective belt lead to novel predictions that survive empirical testing. To the extent that a disconfirmed prediction enforces revisions in the protective belt that lead to a novel prediction that can be empirically confirmed, the research program is characterized by what Lakatos calls a progressive problem-shift. If, however, the enforced revisions lead to novel predictions that are disconfirmed, the research program is characterized by a degenerative problem-shift. Similarly, degenerative problem-shifts are said to occur when revisions in the protective belt do not lead to any new predictions at all. Drawing on Popper’s (1934) analysis, such cases could even be described as nonscientific because they tend to involve ad-hoc claims that do not allow further empirical testing. Lakatos’s (1970) analysis has three important implications. First, his framework shifts the focus from individual theories (Popper, 1934) and current networks of theoretical and empirical assumptions (Quine, 1953) to changes in theorizing over the course of a research program. Second, the focus on changes in theorizing over the course of a research program implies that even well-established theories may eventually be rejected if they involve long-lasting degenerative problem-shifts that fail to inspire novel predictions that can be empirically confirmed. Third, the nature of problemshifts itself may change over time in that a degenerative research program may change into a progressive one, or vice versa. To illustrate such changes in the nature of problem-shifts, Lakatos (1970) described the example of a disconfirmed prediction about the course of a particular planet. To reconcile the conflicting observation with the theory, the researcher may postulate the existence of another planet that distorted the predicted course in line with the assumptions of the theory. If the researcher failed to detect any such planet, she might go on to postulate the existence of a third planet that concealed the view of the hypothesized second planet. Eventually, the researcher may be able to detect both hypothesized planets. Thus, what started as a degenerative research program eventually turned into a progressive one involving the discovery of two new planets that had been unknown before. Drawing on Lakatos’s (1970) conceptual framework, Ketelaar and Ellis (2000) reviewed some interesting examples of progressive and degenerative research programs in the field of evolutionary psychology (see also Ketelaar, Chapter 11, this volume). Addressing the common criticism that evolutionary theories tend to be unfalsifiable, Ketelaar and Ellis showed that some research



Theory Evaluation 11

programs in evolutionary psychology involved degenerative problem-shifts that ultimately led to their rejection, whereas others were characterized by progressive problem-shifts involving the generation of novel predictions that have led to interesting discoveries about the workings of the human mind.

Metatheoretical Criteria for Theory Evaluation Another important aspect of holistic conceptualizations of science is the empirical underdetermination of scientific theories (Quine, 1960). According to holistic conceptualizations, scientific theories are connected to empirical “facts” by the statements they imply about particular observations. Yet, in order to be non-tautological, these observation statements have to be implied by the theory without the theory being implied by the observation statements. If there were a bi-conditional relation between the theory and the observation statement (T → O and O → T), the two would be conceptually equivalent, making any claim about their relation tautological, and thus pseudo-empirical (cf. Popper, 1934). Importantly, the requirement that observation statements have to be implied by a theory without the theory being implied by the observation statements allows for the possibility that two (or more) theories imply the same set of observation statements, even when the theories themselves do not imply each other (Quine, 1981). In other words, two theories can be empirically equivalent (i.e., implying the same observation statements) without being just semantically different formulations of the same theoretical assumptions. In such cases, it is impossible to empirically decide which of two (or more) competing theories should be preferred, because there will never be any evidence that could distinguish between them (e.g., Greenwald, 1975). Although the underdetermination of scientific theories can make it difficult (and sometimes impossible) to empirically decide between competing theories, philosophers of science have proposed a number of metatheoretical criteria for the evaluation of scientific theories that are particularly useful when empirical data are unable to distinguish between competing theories (e.g., Harman, 1965; Quine & Ullian, 1978; Thagard, 1978). In addition, these criteria can provide a framework for the revision of theoretical networks in cases of disconfirming evidence. The underlying goal of these criteria is to maximize the ability to explain and predict events, which gives them the status of pragmatic heuristics (Quine & Ullian, 1978).

Conservatism When scientists are confronted with observations that conflict with the predictions of a given theory, the resulting inconsistency requires revisions in the network of currently accepted propositions. As outlined by Lakatos (1970), these revisions typically involve questioning one or more assumptions in the

12 BASICS

protective belt, rather than challenging the central theoretical assumptions of the hard core. Examples include the reinterpretation of previous findings and the introduction of new assumptions that are able to resolve the inconsistency. According to the criterion of conservatism, scientists should aim for theoretical revisions that minimize changes in the network of currently accepted propositions. The rationale underlying the quest for conservatism is to ensure the ability to explain and predict events, which would be undermined if researchers prematurely rejected well-established theories in the event of a disconfirmed prediction. As an example, imagine a case in which scientists respond to a disconfirmed prediction of an established theory by rejecting the theoretical assumptions underlying a commonly used measure. Such a rejection may resolve the inconsistency between the theoretically derived prediction and the empirically observed result, thereby protecting the theory from the disconfirming evidence. Yet, the rejection of the measurement theory would require reinterpretations of all empirical findings that were based on the relevant measure, which may produce even more inconsistencies that need to be resolved. Thus, to the extent that the employed experimental manipulation is relatively novel and less established than the measurement procedure, the principle of conservatism may lead scientists to attribute the disconfirmed prediction to inadequate assumptions about the experimental manipulation rather than inadequate assumptions about the employed measure. The principle of conservatism illustrates not only why well-established theories are often retained despite disconfirming evidence (cf. Lakatos, 1970); it also illustrates the importance of well-founded assumptions about methods (e.g., measurement, operationalization) that impose strong network constraints in the case of disconfirmed predictions. To the extent that methodrelated assumptions are weak, they will be an easy target in the resolution of inconsistency, thereby allowing for the retention of pretty much any theory (LeBel & Peters, 2011). According to Kuhn (1962), the set of method-related assumptions that is accepted by the scientific community constitutes a paradigm, which he considered a fundamental precondition for scientific progress. If there was no consensus about the methods that are suitable to study a particular phenomenon, researchers would have to justify every background assumption they rely on when deriving predictions about empirical observations, which undermines stringent tests of their theories. The principle of conservatism contributes to scientific progress by preventing premature rejections of paradigmatic assumptions about methods, thereby imposing stronger constraints on the hypotheses of the theoretical core.

Parsimony Although conservatism is important to ensure the ability to explain and predict events, it implies the risk that a theory is continuously protected from its disconfirmed predictions through a never-ending accumulation



Theory Evaluation 13

of ad-hoc assumptions. Such assumptions can undermine the ability to explain and predict events if they increase the complexity of the theoretical network to a point where it becomes unclear what empirical outcome one should expect under particular conditions. The criterion of parsimony counteracts such increases in complexity by favoring theories that require fewer assumptions to explain a particular empirical finding.2 If scientists would be willing to accept theories that make unnecessary assumptions to explain a given finding (i.e., when there are theories available that explain the same finding with less assumptions), the network of theoretical propositions might acquire a level of complexity that could ultimately undermine the possibility of relating its theories to empirical observations. At the macro level, potential conflicts between conservatism and parsimony are reflected in what Kuhn (1962) described as scientific revolutions, which are characterized by substantial revisions of theoretical core assumptions in favor of parsimonious theories. Because rejections of theoretical core assumptions often require revisions of related assumptions about methods, scientific revolutions are further characterized by paradigm shifts in that method-related assumptions that have been accepted in the past are replaced by a new set of paradigmatic assumptions. An often-overlooked aspect of parsimony is that it refers to the total number of theoretical propositions that are required to explain a given finding rather than the number of propositions of what might be considered the core of the relevant theory. As we outlined earlier in this chapter, statements about observations are never derived from a theory alone, but from the theory in conjunction with multiple background assumptions. The criterion of parsimony refers to the conjunction of a theory and the background assumptions that are needed to explain an empirical finding, not to the theory alone. An illustrative example is the ongoing debate between the proponents of single-process theories (e.g., Kruglanski, Erb, Pierro, Mannetti, & Chun, 2006) and dual-process theories (e.g., Deutsch & Strack, 2006) in social psychology (see also Deutsch, Chapter 7, this volume). Single-process theorists often appeal to the quest for parsimony, arguing that dual-process theories are less parsimonious than single-process theories because they postulate two qualitatively distinct processes rather than a single one. However, to explain a particular finding, single-process theories have to rely on a host of additional assumptions over and above the hypothesis that information processing is guided by a single process. For example, Kruglanski’s unimodel proposes that judgments are the outcome of a single process of rule-based inference that is modulated by five processing parameters (Kruglanski et al., 2006). Importantly, two of these processing parameters (i.e., accessibility, relevance) have a striking conceptual resemblance to what some dual-process theorists describe as associative and propositional processes (e.g., Gawronski & Bodenhausen, 2006, 2011; Strack & Deutsch, 2004). Thus, when evaluating theories on the basis of their parsimony, it does not suffice to count the number of propositions that may be regarded as the core of a given theory

14 BASICS

(e.g., single-process vs. dual-process), but the entire set of propositions that is required to capture a given finding.

Generality According to the criterion of generality, preference should be given to theories with higher rather than lower explanatory breadth. The larger the number of phenomena a theory is able to explain, the higher is its degree of generality. For example, a theory that offers explanations for the results of several operationalizations to study a particular phenomenon may be preferred to a theory that explains only the results of a particular operationalization. Similarly, scientists usually prefer theories that make predictions across a wide range of topics, rather than theories with a limited range of applicability. An illustrative example is the development of dual-process theorizing in social psychology over the last decades (for a review, see Gawronski & Creighton, 2013; see also Deutsch, Chapter 7, this volume). In the 1980s and 1990s, social psychologists proposed numerous dual-process theories for a wide range of phenomena, including persuasion (e.g., Chaiken, Liberman, & Eagly, 1989; Petty & Cacioppo, 1986), attitude–behavior relations (e.g., Fazio, 1990), prejudice and stereotyping (e.g., Devine, 1989), impression formation (e.g., Brewer, 1988; Fiske & Neuberg, 1990), and dispositional attribution (e.g., Gilbert, 1989; Trope, 1986). Although these phenomenon-specific theories were supported by a substantial body of evidence, the following decade was characterized by a remarkable shift toward domain-independent dualprocess theories (e.g., Smith & DeCoster, 2000; Strack & Deutsch, 2004). The latter kinds of theories are more general in the sense that they offer explanations for phenomena in a wide range of research areas. Importantly, domainindependent dual-process theories do not imply any predictions that would conflict with the predictions of the earlier phenomenon-specific theories. Yet, they do differ in terms of the hypothesized explanatory constructs, allowing them to provide explanations for a broader range of phenomena.

Refutability The quest for generality poses the risk of creating theories that explain everything, yet predict nothing. The criterion of refutability imposes constraints on such developments by emphasizing the predictive power of theories. Although refutability has a close resemblance to Popper’s (1934) notion of falsifiability, the refutability criterion is broader in that it considers theories in conjunction with the network of currently accepted propositions. According to Popper, a theory is unfalsifiable—and thus nonscientific—if it is consistent with any empirical observation. As we outlined earlier in this chapter, this would be the case for theories that are either logically incoherent or tautological. In addition to the rejection of unfalsifiable theories, Popper argued that scientists should prefer theories with a higher degree of falsifiability,



Theory Evaluation 15

which corresponds to the number of observations that are prohibited by a theory. As implied by the Duhem–Quine thesis, however, empirical observations are not implied by a theory alone, but by the conjunction of the theory and multiple background assumptions. This insight is captured in the criterion of refutability, which can be interpreted as a holistic reinterpretation of Popper’s (1934) quest to prefer theories with a higher degree of falsifiability. From a holistic point of view, refutability depends on the network of currently accepted propositions in that the number of observations that are prohibited by the theory is contingent on the background assumptions that are available to derive testable predictions (Quine & Ullian, 1978). Thus, different from the quest for logically coherent and non-tautological theories, the degree of refutability cannot be determined a priori on the basis of the logical structure of a given theory. Instead, refutability depends on the current state of scientific research. Hence, the refutability of a given theory could even change over time, in that the theory may seem irrefutable at an early stage of inquiry because of the absence of suitable background assumptions that would allow the derivation of testable predictions. Yet, the same theory may acquire a high degree of refutability when new research developments provide background assumptions that, in conjunction with the theory, prohibit specific observations. An illustrative example for such changes is the revival of psychoanalytic assumptions about unconscious motivational processes, which have been criticized as nonscientific on the basis of Popper’s falsifiability criterion (e.g., Grünbaum, 1986). From a holistic point of view, the low degree of refutability was rooted in the difficulty of either assessing or manipulating these processes. However, recent methodological developments offer some valuable tools that might capture at least some of the processes proposed by psychoanalytic theory. Thus, in conjunction with the network of currently accepted propositions, many assumptions about unconscious motivational processes may be refutable today, although they were irrefutable in the early days of psychoanalytic theory. For example, although not too long ago many psychologists would have rejected the idea of unconscious motivation as unfalsifiable, it has become a central concept in research on unconscious goal pursuit (Förster, Liberman, & Friedman, 2007; but see Newell & Shanks, 2014, for a critical discussion). Similar changes can be observed for many other constructs that are even closer to the original formulations of psychoanalytic theory (e.g., Erdelyi, 1985; Westen, 1998; Wilson & Dunn, 2004).

Precision An important means to achieve a high level of refutability is precision: The more precise the formulation of a given theory, the less ambiguous it will be which observations are prohibited by the theory in conjunction with available background assumptions. Thus, whereas the quests for parsimony and

16 BASICS

generality emphasize the explanatory power of scientific theories, refutability and precision are important criteria to ensure their predictive power. A useful example to illustrate the criterion of precision is provided by theoretical claims about the contribution of automatic versus controlled processes to social-psychological phenomena. In a strict sense, theoretical claims that a phenomenon is due to an automatic or a controlled process may be regarded as insufficient because such claims do not say what exactly the process is and in which particular sense the process is supposed to be automatic. For example, racial bias in weapon identification is often claimed to involve both automatic and controlled processes (for a review, see Payne, 2006), but such claims remain ambiguous about the exact nature of these processes (e.g., does the controlled process involve the identification of the target object or the inhibition of racially biased response tendencies?) and the particular sense in which these processes operate in an automatic versus controlled fashion (e.g., is the process claimed to require awareness, intention, or cognitive resources?). Precise formulations of these assumptions increase not only the informational value of the relevant theories; they also clarify which empirical observations are prohibited according these theories, thereby increasing their refutability.

Logical Fallacies So far, our chapter has focused on various criteria for the evaluation of scientific theories. Some of these criteria involve logical analyses of deductive relations, such as the quest for logically coherent and nontautological theories and the metatheoretical quests for parsimony and refutability. In the final section, we discuss a number of logical fallacies that can distort the outcome of such analyses. These fallacies are important not only for the evaluation of scientific theories, but also for the theoretical interpretation of empirical findings.

Affirming the Consequent One of the most common inferential fallacies in psychology is the fallacy of affirming the consequent, also known as reverse inference. In abstract terms, the fallacy can be described as the conclusion of X on the basis of the observation Y and the conditional “if X, then Y.”3 A useful example to illustrate this fallacy is the interpretation of the spreading-of-alternatives effect, which describes the phenomenon that choosing between two equally attractive alternatives leads to more favorable evaluations of chosen as compared to rejected alternatives (Brehm, 1956). The spreading-of-alternatives effect was discovered on the basis of a prediction by dissonance theory (Festinger, 1957), suggesting that people experience an aversive feeling of postdecisional dissonance when they recognize either (1) that the rejected alternative



Theory Evaluation 17

has positive features that the chosen alternative does not have, or (2) that the chosen alternative has negative features that are not present in the rejected alternative. To reduce this aversive feeling, people are assumed to emphasize or search for positive characteristics of the chosen alternative and negative characteristics of the rejected alternative, which in turn leads to more favorable evaluations of the chosen compared with the rejected alternative. Drawing on evidence for the contribution of postdecisional dissonance to the spreading-of-alternatives effect, some researchers have conversely interpreted the emergence of the spreading-of-alternatives effect (Y) as evidence for the presence of cognitive dissonance (X) on the basis of the theoretical conditional if dissonance, then spreading-of-alternatives (“if X, then Y”). For example, it has been argued that the mere emergence of a spreading-ofalternatives effect demonstrates postdecisional dissonance in children and monkeys (Egan, Santos, & Bloom, 2007) and in amnesic patients who cannot remember their choice (Lieberman, Ochsner, Gilbert, & Schacter, 2001). This inference is logically flawed because the conditional “if X, then Y” entails only that Y can be inferred from X, not that X can be inferred from Y. After all, Y may be implied by other premises that are not X (e.g., if Z, then Y). For example, spreading-of-alternatives effects have been shown to emerge in the absence of postdecisional dissonance as a result of mere ownership (Gawronski, Bodenhausen, & Becker, 2007) or simple methodological factors (Chen & Risen, 2010). In addition to the widespread equation of a behavioral outcome with a particular psychological mechanism (e.g., equation of the spreading-of-alternatives effect with the presence of postdecisional dissonance), reverse inferences are very common in the field of social neuroscience, where the activation of a particular brain area during the operation of a particular process is often used to draw the reverse inference that the process must be operating when there is evidence for activation in this brain area (see Beer, Chapter 9, this volume).

Appealing to Ignorance Another common inferential error in psychology is the fallacy of appealing to ignorance. Depending on whether the fallacy refers to positive or negative evidence, this fallacy can be reflected in two kinds of inferences: (1) there is insufficient evidence that X is true; therefore, X is false; and (2) there is insufficient evidence that X is false; therefore, X is true. An illustrative example is research and theorizing on unconscious processes, in which proponents of conflicting views often use weak evidence for the other theoretical view as support for their own view. For example, research on the contribution of unconscious processes to decision making has been criticized for offering weak evidence that the hypothesized processes indeed operate outside of conscious awareness (e.g., Gawronski & Bodenhausen, 2012; Newell & Shanks, 2014). Yet, the weakness of the available evidence for unconscious influences does not imply that the obtained influences operate under con-

18 BASICS

scious awareness. After all, absence of evidence for unconsciousness is not the same as evidence for consciousness.

Nominal Fallacy The nominal fallacy describes a variant of circular argumentation in which a given phenomenon is labeled as an instance of a particular category without providing an explanation of why the observed phenomenon occurred (also known under the phrase naming is not explaining). An illustrative example is Gigerenzer’s (1996) critique of Kahneman and Tversky’s research program on heuristics and biases (for an overview, see Kahneman, Slovic, & Tversky, 1982). To “explain” a variety of empirical phenomena that involve deviations from normative rules of inference, Kahneman and Tversky proposed three judgmental heuristics that were claimed to bias judgments in a systematic manner: (1) the anchoring-adjustment heuristic, which involves use of a numerical value as an anchor with insufficient adjustment (Tversky & Kahneman, 1974); (2) the availability heuristic, which involves probability or likelihood judgments on the basis of the ease by which relevant instances come to mind (Tversky & Kahneman, 1973); and (3) the representativeness heuristic, which involves the use of information on the basis of whether it seems representative rather than on the basis of its reliability or prior probability (Kahneman & Tversky, 1973). The three heuristics have been criticized for providing re-descriptions of the observed judgmental biases without explaining why these biases occur. According to Gigerenzer (1996), a theoretically convincing explanation would require a specification of the psychological computations that are responsible for the observed judgmental biases. Although such explanations were missing in the early stages of inquiry, several researchers worked toward overcoming the nominal fallacy by developing and testing theories about the mental operations underlying the observed judgmental biases. For example, Schwarz et al. (1991) argued that judgmental biases that have been attributed to the availability heuristic stem from the use of recollective experiences in retrieving relevant information from memory (instead of the content of the retrieved information) to solve the judgmental task (e.g., if it is difficult to recall, it cannot be typical). Such assumptions go beyond the criticized re-descriptions by providing clear specifications of the mental operations underlying the observed judgmental biases.

Denying the Antecedent A fourth inferential error is the fallacy of denying the antecedent, which involves rejection of the consequent of a conditional on the basis of rejecting the antecedent. In abstract terms, this fallacy can be described as the rejection of Y on the basis of the conditional “if X, then Y” and the rejection of X. However, the rejection of X in the conditional “if X, then Y” does not guarantee that there is no alternative Z that also implies Y. A typical exam-



Theory Evaluation 19

ple in psychology is the denial of a behavioral phenomenon that is implied by a theory that has been rejected as a viable explanation of human behavior. From a logical point of view, any such rejection presupposes that there is no alternative mechanism that could also produce the relevant phenomenon. For instance, the rejection of behaviorist reinforcement theories does not mean that the phenomena of stimulus-response learning predicted by these theories do not exist. After all, phenomena of stimulus–response learning could be mediated by cognitive learning mechanisms that are not part of the original behaviorist theories that led to the discovery of these phenomena.

Disjunctive Fallacy The disjunctive fallacy involves the inference that a given proposition must be incorrect, if there is evidence that supports an alternative proposition. In abstract terms, this fallacy can be described by the disjunction “either X or Y,” which may lead to the rejection of Y if there is evidence for X. An illustrative example is the theoretical debate about whether certain kinds of individual differences are the product of genetic or environmental influences (see Johnson & Penke, Chapter 10, this volume). For a long time, research in this area was framed as a question of either–or, and each piece of evidence supporting one theoretical view was interpreted as invalidating the opposing view. Yet, more recent accounts explicitly acknowledge the contribution of both genes and environment (as well as their interactions) as important sources of individual differences (Johnson, 2007). In fact, it seems as if contemporary social psychology has become much less prone to the disjunctive fallacy, given the increasingly widespread acknowledgment that many social-psychological phenomena are multiply determined.

Moralistic Fallacy The moralistic fallacy involves the assumption that the validity of a theoretical proposition is a function of its moral desirability. In abstract terms, the moralistic fallacy can be depicted by an inferential structure that resembles the modus tollens. Yet, its underlying inference is invalid in that it is based on a judgment of moral desirability rather than empirical observation. Specifically, the moralistic fallacy can be depicted by the inference: “if X, then Y”; Y is morally undesirable; therefore X is false.4 For example, claims relating to race and sex differences are often evaluated by reference to their moral palatability, even though such considerations obviously cannot count as scientific evidence (Pinker, 2003). Another interesting case to illustrate the moralistic fallacy is the response to a meta-analysis by Rind, Tromovitch, and Bauserman (1998) suggesting that the relation between child sexual abuse and psychopathology in adulthood is relatively weak and fully explained by the relation between family environment and psychopathology. The find-

20 BASICS

ing caused a major controversy in psychology, the public media, and government legislation, involving an unprecedented condemnation of Rind et al.’s meta-analysis by the U.S. Congress (Lilienfeld, 2002). Drawing on the moralistic fallacy, one could argue that this controversy was at least partially driven by the conviction that child sexual abuse is morally wrong. Therefore, any research suggesting that the negative psychological consequences of child sexual abuse are minor must be flawed. Although less explicit, similar arguments have been raised against evolutionary explanations of intergroup prejudice, which have been accused of justifying morally despicable behavior (see Ketelaar, Chapter 11, this volume).

Conclusion The main goal of the current chapter was to review criteria for the evaluation of scientific theories and to illustrate their relevance for theory evaluation in social psychology. We started by outlining the problems associated with logical incoherence and tautological assumptions for the falsifiability of social-psychological theories. Acknowledging pragmatic limits in the actual falsification of theories, we further explained the implications of the Duhem–Quine thesis for holistic consistency checks in theoretical networks and the evaluation of progressive and degenerative problem-shifts in scientific research programs. Finally, we illustrated the usefulness of various metatheoretical criteria for the evaluation of theories (i.e., conservatism, parsimony, generality, refutability, precision) and the inferential errors resulting from logical fallacies in the conceptual analysis of theories and the interpretation of empirical data. We hope that our review of these criteria provides a useful framework for both the evaluation of existing theories and the construction of new theories, thereby advancing research and theorizing in social psychology. Notes 1. Note that in a strict sense, general statements are falsified by statements about observations rather than observations in the sense of perceptual experiences (Popper, 1934). 2. The criterion of parsimony has also been described as simplicity (Thagard, 1978) or modesty (Quine & Ullian, 1978). In the philosophy of science, it is widely known as Occam’s razor. 3. Note that the inferential structure of the fallacy is equivalent to Popper’s (1934) rejection of inductive inferences as a basis for the evaluation of scientific theories. 4. A related inferential error is the naturalistic fallacy, which is essentially the reverse of the moralistic fallacy. Whereas the moralistic fallacy involves inferences about validity on the basis of moral desirability, the naturalistic fallacy involves inferences about moral desirability on the basis of what is the case.



Theory Evaluation 21

References Aronson, E. (1992). The return of the repressed: Dissonance theory makes a comeback. Psychological Inquiry, 3, 303–311. Brehm, J. W. (1956). Postdecision changes in the desirability of alternatives. Journal of Abnormal and Social Psychology, 52, 384–389. Brewer, M. B. (1988). A dual process model of impression formation. In T. K. Srull & R. S. Wyer (Eds.), Advances in social cognition (Vol. 1, pp. 1–36). Hillsdale, NJ: Erlbaum. Chaiken, S., Liberman, A., & Eagly, A. H. (1989). Heuristic and systematic processing within and beyond the persuasion context. In J. S. Uleman & J. A. Bargh (Eds.), Unintended thought (pp. 212–252). New York: Guilford Press. Chen, M. K., & Risen, J. L. (2010). How choice affects and reflects preferences: Revisiting the free-choice paradigm. Journal of Personality and Social Psychology, 99, 573–594. Deutsch, R., & Strack, F. (2006). Duality models in social psychology: From dual processes to interacting systems. Psychological Inquiry, 17, 166–172. Devine, P. G. (1989). Stereotypes and prejudice: Their automatic and controlled components. Journal of Personality and Social Psychology, 56, 5–18. Duhem, P. (1908). Ziel und Struktur physikalischer Theorien [The aim and structure of physical theory]. Leipzig: Barth. Egan, L. C., Santos, L. R., & Bloom, P. (2007). The origins of cognitive dissonance: Evidence from children and monkeys. Psychological Science, 18, 978–983. Erdelyi, M. H. (1985). Psychoanalysis: Freud’s cognitive psychology. New York: Freeman. Fazio, R. H. (1990). Multiple processes by which attitudes guide behavior: The MODE model as an integrative framework. Advances in Experimental Social Psychology, 23, 75–109. Festinger, L. (1957). A theory of cognitive dissonance. Evanston, IL: Row Peterson. Feyerabend, P. (1975). Against method: Outline of an anarchistic theory of knowledge. London: New Left Books. Fiske, S. T., & Neuberg, S. L. (1990). A continuum of impression formation, from categorybased to individuating processes: Influences of information and motivation on attention and interpretation. Advances in Experimental Social Psychology, 23, 1–74. Förster, J., Liberman, N., & Friedman, R. S. (2007). Seven principles of goal activation: A systematic approach to distinguishing goal priming from priming of non-goal constructs. Personality and Social Psychology Review, 11, 211–233. Gawronski, B., & Bodenhausen, G. V. (2006). Associative and propositional processes in evaluation: An integrative review of implicit and explicit attitude change. Psychological Bulletin, 132, 692–731. Gawronski, B., & Bodenhausen, G. V. (2011). The associative-propositional evaluation model: Theory, evidence, and open questions. Advances in Experimental Social Psychology, 44, 59–127. Gawronski, B., & Bodenhausen, G. V. (2012). Self-insight from a dual-process perspective. In S. Vazire & T. D. Wilson (Eds.), Handbook of self-knowledge (pp. 22–38). New York: Guilford Press. Gawronski, B., Bodenhausen, G. V., & Becker, A. P. (2007). I like it, because I like myself: Associative self-anchoring and post-decisional change of implicit evaluation. Journal of Experimental Social Psychology, 43, 221–232. Gawronski, B., & Creighton, L. A. (2013). Dual-process theories. In D. E. Carlston (Ed.), The Oxford handbook of social cognition (pp. 282–312). New York: Oxford University Press. Gigerenzer, G. (1996). On narrow norms and vague heuristics: A reply to Kahneman and Tversky (1996). Psychological Review, 103, 592–596. Gilbert, D. T. (1989). Thinking lightly about others: Automatic components of the social

22 BASICS inference process. In J. S. Uleman & J. A. Bargh (Eds.), Unintended thought (pp. 189– 211). New York: Guilford Press. Greenwald, A. G. (1975). On the inconclusiveness of “crucial” cognitive tests of dissonance versus self-perception theories. Journal of Experimental Social Psychology, 11, 490–499. Greenwald, A. G., Pratkanis, A. R., Leippe, M. R., & Baumgardner, M. H. (1986). Under what conditions does theory obstruct research progress? Psychological Review, 93, 216–229. Greve, W. (2001). Traps and gaps in action explanation: Theoretical problems of a psychology of human action. Psychological Review, 108, 435–451. Grünbaum, A. (1986). Is Freud’s theory well-founded? Behavioral and Brain Sciences, 9, 266–284. Harman, G. (1965). The inference to the best explanation. Philosophical Review, 74, 88–95. Hempel, C. G. (1965). Empiricist criteria of cognitive significance: Problems and changes. In C. G. Hempel (Ed.), Aspects of scientific explanation and other essays in the philosophy of science (pp. 101–122). New York: Free Press. Johnson, W. (2007). Genetic and environmental influences on behavior: Capturing all the interplay. Psychological Review, 114, 423–440. Johnson-Laird, P. N. (2012). Mental models and consistency. In B. Gawronski, & F. Strack (Eds.), Cognitive consistency: A fundamental principle in social cognition (pp. 225–244). New York: Guilford Press. Kahneman, D., Slovic, P., & Tversky, A. (1982). Judgment under uncertainty: Heuristics and biases. Cambridge, UK: Cambridge University Press. Kahneman, D., & Tversky, A. (1973). On the psychology of prediction. Psychological Review, 80, 237–251. Ketelaar, T., & Ellis, B. J. (2000). Are evolutionary explanations unfalsifiable? Evolutionary psychology and the Lakatosian philosophy of science. Psychological Inquiry, 11, 1–21. Kluger A. N., & Tikochinsky, J. (2001). The error of accepting the “theoretical” null hypothesis: The rise, fall, and resurrection of common sense hypotheses in psychology. Psychological Bulletin, 127, 408–423. Kruglanski, A. W., Erb, H.-P., Pierro, A., Mannetti, L., & Chun, W. Y. (2006). On parametric continuities in the world of binary either ors. Psychological Inquiry, 17, 153–165. Kruglanski, A. W., & Higgins, E. T. (2004). Theory construction in social-personality psychology: Personal experiences and lessons learned. Personality and Social Psychology Review, 8, 96–97. Kuhn, T. S. (1962). The structure of scientific revolutions. Chicago: Chicago University Press. Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108, 480–498. Lakatos, I. (1970). Falsification and the methodology of scientific research programmes. In I. Lakatos & A. Musgrave (Eds.), Criticism and the growth of knowledge (pp. 91–195). Cambridge, UK: Cambridge University Press. LeBel, E. P., & Peters, K. R. (2011). Fearing the future of empirical psychology: Bem’s (2011) evidence of psi as a case study of deficiencies in modal research practice. Review of General Psychology, 15, 371–379. Lewin, K. (1951). Field theory in social science: Selected theoretical papers. New York: Harper & Row. Lieberman, M. D., Ochsner, K. N., Gilbert, D. T., & Schacter, D. L. (2001). Do amnesics exhibit cognitive dissonance reduction? The role of explicit memory and attention in attitude change. Psychological Science, 12, 135–140. Lilienfeld, S. O. (2002). When worlds collide: Social science, politics, and the Rind et al. (1998) child abuse meta-analysis. American Psychologist, 57, 176–188. Machado, A., & Silva, F. J. (2007). Toward a richer view of the scientific method: The role of conceptual analysis. American Psychologist, 62, 671–681.



Theory Evaluation 23

McGrath, J. E. (1981). Dilemmatics: The study of research choices and dilemmas. American Behavioral Scientist, 25, 179–210. Newell, B. R., & Shanks, D. R. (2014). Unconscious influences on decision making: A critical review. Behavioral and Brain Sciences, 37, 1–19. Payne, B. K. (2006). Weapon bias: Split second decisions and unintended stereotyping. Current Directions in Psychological Science, 15, 287–291. Petty, R. E., & Cacioppo, J. T. (1986). The elaboration likelihood model persuasion. Advances in Experimental Social Psychology, 19, 123–205. Pinker, S. (2003). The blank slate: The modern denial of human nature. New York: Penguin. Popper, K. R. (1934). Logik der Forschung [Logic of scientific discovery]. Wien: Springer. Proctor, R. W., & Capaldi, E. J. (2001). Empirical evaluation and justification of methodologies in psychological science. Psychological Bulletin, 127, 759–772. Quine, W. V. O. (1953). Two dogmas of empiricism. In W. V. O. Quine (Ed.), From a logical point of view (pp. 20–46). Cambridge, MA: Harvard University Press. Quine, W. V. O. (1960). Word and object. Cambridge, MA: MIT Press. Quine, W. V. O. (1981). Empirical content. Theories and things (pp. 24–43). Cambridge, MA: Harvard University Press. Quine, W. V. O., & Ullian, J. S. (1978). The web of belief (2nd ed.). New York: McGraw-Hill. Rind, B., Tromovitch, P., & Bauserman, R. (1998). A meta-analytic examination of assumed properties of child sexual abuse using college samples. Psychological Bulletin, 124, 22–53. Schwarz, N., Bless, H., Strack, F., Klumpp, G., Rittenauer-Schatka, H., & Simons, A. (1991). Ease of retrieval as information: Another look at the availability heuristic. Journal of Personality and Social Psychology, 61, 195–202. Smedslund, G. (2000). A pragmatic basis for judging models and theories in health psychology: The axiomatic method. Journal of Health Psychology, 5, 133–149. Smith, E. R., & DeCoster, J. (2000). Dual-process models in social and cognitive psychology: Conceptual integration and links to underlying memory systems. Personality and Social Psychology Review, 4, 108–131. Strack, F., & Deutsch, R. (2004). Reflective and impulsive determinants of social behavior. Personality and Social Psychology Review, 8, 220–247. Thagard, P. (1978). The best explanation: Criteria for theory choice. Journal of Philosophy, 75, 76–92. Trope, Y. (1986). Identification and inferential processes in dispositional attribution. Psychological Review, 93, 239–257. Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5, 207–232. Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185, 1124–1131. Van Lange, P. A. M., Kruglanski, A. W., & Higgins, E. T. (Eds.). (2012). Handbook of theories of social psychology. Los Angeles: Sage. Wallach, L., & Wallach, M. A. (1994). Gergen versus the mainstream: Are hypotheses in social psychology subject to empirical test? Journal of Personality and Social Psychology, 67, 233–242. Westen, D. (1998). The scientific legacy of Sigmund Freud: Toward a psychodynamically informed psychological science. Psychological Bulletin, 124, 333–371. Wilson, T. D., & Dunn, E. W. (2004). Self-knowledge: Its limits, value and potential for improvement. Annual Review of Psychology, 55, 493–518.

2 Levels of Analysis in Social Psychology Jan De Houwer Agnes Moors

L

ike many aspects of human behavior, scientific research is a goaldirected activity. Hence, to understand scientific behavior, we need to understand the goals that drive scientists. Scientific goals first of all refer to an object that is under investigation. One could, for instance, argue that social psychology aims at understanding social behavior. Because many questions can be asked about any single object, a scientific goal can also specify the question that is asked about that object. For instance, social cognition researchers are specifically interested in the mental processes that underlie social behavior, whereas social neuroscientists might be interested primarily in the neural processes that underlie social behavior. In the present chapter, we focus on yet another way in which scientific goals differ: the level of analysis at which the scientific question is analyzed. Even a single question about a single object can be answered at different levels. We illustrate this point in the context of social cognition research. In doing so, we also clarify the way in which social cognition research is related to other approaches in social psychology, including functional and neuroscience approaches. This chapter is based on the influential ideas of David Marr (1982) about levels of analyses in cognitive psychology. In this chapter, we extend Marr’s 24

Levels of Analysis in Social Psychology 25

ideas by providing a more fine-grained analysis of the different levels and how these levels relate to each other. We specifically illustrate these ideas in the context of social psychology. We hope that reading this chapter will help researchers to make explicit the scientific goals that drive their research. Doing so not only sheds light on scientific practice but can also help facilitate research by exposing false scientific debates and revealing links between seemingly unrelated or contradictory lines of research.

Marr’s Three Levels of Analysis Marr’s (1982) ideas about levels of analysis were developed in the context of cognitive psychology. The goal of cognitive psychology is to understand the mental processes that underlie behavior—that is, the way in which humans and other animals acquire, process, and represent information and how this processing of information guides their behavior (Gardner, 1987). Social cognition research thus shares with cognitive psychology the question that is asked about behavior (i.e., what are the underlying mental processes) but differs from other cognitive research in that it focuses on social behavior rather than behavior in general. Hence, Marr’s ideas about the level of analysis of mental processes are particularly relevant for social cognition research, but as we will see later on in the chapter, they are also important for other traditions in social psychology. In his highly influential book, Marr (1982) argued that processes can be analyzed at three levels: the computational, the algorithmic, and the implementational. The computational level involves an analysis of what the process does. More specifically, it aims to specify what the output of a process will be given a certain input. In doing so, it provides insight into the function of the process. Take the example of a cash register (Marr, 1982). Its function is to add the price of different items. “Addition” is a function in that it specifies which input (e.g., 15 + 14) into the cash register leads to which output of the cash register (e.g., 29). Understanding that a cash register has the function to add up numbers provides an important insight into the nature of the cash register and thus constitutes an important level of analysis. Marr (1982) argued that, in a similar way, understanding the function of mental processes is vital to our understanding of those processes. It specifies the task that the processes carry out. The algorithmic level goes beyond the computational level by specifying the mechanism by which the input is transformed into the output. Again consider the example of a cash register. When entering the numbers 15 and 14 into the cash register, the output will be 29. But there are potentially different ways in which the input (15 and 14) can be transformed into the output (29). For instance, the numbers might be represented in Arabic form, first adding the units (i.e., 5 and 4), “carrying over” a number if this exceeds 10, and then adding the tens (i.e., 1 and 1). Another possible algorithm would

26 BASICS

be to represent the numbers in analog form (e.g., 15 dots and 14 dots) and then counting the total number of dots. An algorithm thus specifies a chain of intermediate steps that starts from the input and ends at the output. When applied to mental processes, algorithms can be conceived of as mental mechanisms (Bechtel, 2005, 2008). The mechanism is mental in that the organism is assumed to represent information at each step of the chain and to transform the information from one step to the next. The implementational level of analysis deals with the physical implementation of the algorithm. A cash register, for instance, can operate mechanically “as positions on a ten-notch metal wheel, or as binary coded decimal numbers implemented in the electrical states of digital logic circuitry” (McClamrock, 1991, p. 190). In a similar way, cognitive psychologists assume that a single information processing algorithm can be implemented in different physical systems (e.g., a human brain, an electronic computer; see Gardner, 1987). In sum, Marr (1982) proposed that understanding a mental process boils down to understanding what the process does (i.e., which input is transformed into which output), how the process achieves this function (i.e., the chain of intermediate information processing steps via which an input is transformed into an output), and how the process is physically implemented. The three levels can be separated because they are not entirely interdependent. That is, conclusions at one level do not necessarily imply conclusions at another level. This follows from the fact that one function (i.e., specified at the computational level) can be achieved by different algorithms (i.e., specified at the algorithmic level) that can each be implemented in different physical systems (i.e., specified at the implementational level). One of the biggest advantages of Marr’s (1982) distinction between levels of analysis is that it can expose false debates in psychology. As we pointed out before, analyses at different levels are guided by different goals. A computational analysis aims to uncover the relation between the input and output of a process, an algorithmic analysis is concerned with the mechanisms by which the input is transformed in the output, and an implementational analysis seeks to clarify how an algorithm is instantiated in a physical substrate such as the brain. Different theories may be compatible when they are directed at understanding different aspects of the same question. Hence, different answers to a scientific question (e.g., “what mental processes underlie social behavior?”) are not necessarily conflicting if they are formulated at different levels of analysis. To illustrate this point, let us consider research on causal attribution. Because we will return to this example several times during the course of this chapter, we will first provide some background about this type of research. Causal attribution research has a long history in social psychology. Perhaps most important in this respect is Kelley’s (1973) covariation theory of causal attribution. A central question in this line of research is: How do people infer that the behavior of another person (e.g., Mary refuses Jim’s

Levels of Analysis in Social Psychology 27

invitation to dinner on Friday) is caused by something about the person (e.g., Mary is impolite), something about the entity (e.g., Jim is boring), or something about the circumstances (e.g., bad timing). The central assumption in Kelley’s theory is that perceivers solve this inferential task by means of information about covariations between the presence or absence of the behavior, on the one hand, and the presence or absence of the person (i.e., consensus; e.g., do other people also refuse Jim’s invitations), the presence or absence of the entity (i.e., distinctiveness; e.g., does Mary also refuse invitations from others), or variations across time (i.e., consistency; e.g., does Mary refuse invitations for Jim at other points in time as well), on the other hand. In studies on causal attribution, participants typically see a series of trials on which cues (e.g., a particular person) and outcomes (e.g., a particular social behavior) can be present or absent. Most often, they are asked to judge the extent to which a particular cue causes a particular outcome (e.g., the extent to which a behavior is caused by a particular person). Some years ago, there was a debate between so-called statistical models, such as the delta P model, and association formation models, such as the Rescorla–Wagner model (see Cheng & Holyoak, 2011; De Houwer & Beckers, 2002; McClure, 1998, for reviews). The delta P model, for instance, postulates that causal judgments are a function of the difference between the probability of the outcome when a cue is present and the probability of the outcome when the cue is absent (i.e., P(O/C) – P(O/~C)). Association formation models such as the Rescorla–Wagner model, on the other hand, argue that causal judgments reflect the strength of associations that are formed between the representations of cues and the representations of outcomes in memory. The debate involved proponents of one theory raising doubts about the validity of the other theory by showing that people sometimes give causal judgments that seem to be in line only with the predictions of their own theory. In many cases, however, these attacks were countered by showing that some revised version of the criticized model could account for those results as well (see De Houwer & Beckers, 2002, for a review). Inspired by Marr’s (1982) ideas, researchers eventually realized that the two types of models are situated at different levels of analysis (Cheng, 1997; Shanks, 1995). Statistical models such as the delta P model are computational in nature because they specify which environmental input (i.e., the presence or absence of cues and outcomes) leads to which behavioral output (i.e., a causal judgment or causal attribution). They do not make any assumptions about the mental mechanisms by which the input leads to the output. Associative models do make assumptions about the mechanism by which events lead to causal judgments (i.e., the formation of associations) and are thus situated at the algorithmic level of analysis. The fact that both models do not contradict each other was even proven mathematically: Given certain assumptions about the parameters in the Rescorla–Wagner model, this model makes exactly the same predictions as the delta P model. In other words, both models are computationally equivalent in that they relate the same inputs to the same outputs.

28 BASICS

However, the Rescorla–Wagner model goes beyond the delta P model in that it not only specifies which input leads to which output but also incorporates assumptions about the mental mechanism by which an input results in an output (i.e., the formation of associations in memory). This example illustrates that it is important to know whether a model is directed at describing a process in terms of input–output relations or in terms of the mechanism by which input is transformed into output. Different models at different levels of analysis are therefore not necessarily incompatible. Of course, they can be incompatible (as would be the case when an algorithmic and computational model would make different predictions about which input leads to which output), but they can also complement each other.

Beyond Marr In the remainder of this chapter, we aim to extend Marr’s (1982) ideas both with regard to the number and nature of the levels of analysis that are distinguished and with regard to the relations between the different levels of analysis. First, we argue that the computational level actually encompasses two different levels: the behavioral functional level and the informational functional level. Second, we examine in more depth how the different levels of analysis are related. We explain these ideas in relation to social cognition research and discuss their implications for social psychology as a whole.

Two Functional Levels of Analysis In Marr’s (1982) cash register example, the input (i.e., entering numbers into the register) and output of the process (i.e., the number appearing on the display of the register) are both observable events in the physical world. This is also the case for mental processes that have environmental events as their input and behavioral events as their output. We refer to these mental processes as overarching mental processes in that they span the entire black box between an environmental input and a behavioral output. For instance, causal attribution could be regarded as an overarching mental process by which events in the world result in causal judgments. Other mental processes, however, have mental representations as their input and/or as their output. We refer to these processes as mental subprocesses because they constitute only one part in a chain of mental subprocesses by which environmental input influences behavior. For instance, the overarching mental process of causal attribution most likely entails the mental subprocesses of information storage and information retrieval. Information storage has environmental events as its input and mental (memory) representations as its output. Likewise, information retrieval has mental (memory) representations as its input and behavior or other mental representations as its output. We argued that two functional levels of analysis can be distinguished

Levels of Analysis in Social Psychology 29

based on the type of mental processes that are analyzed. The behavioral functional level reveals the function of overarching mental processes. The informational functional level, on the other hand, uncovers the function of mental subprocesses (i.e., mental processes that have mental representations as their input or output). Crucially, both levels are functional in that they describe which input leads to which output. They are silent with regard to the mechanism by which input is transformed into output. As in Marr’s (1982) framework, the analysis of the nature of the mechanism is still reserved for the algorithmic level. The distinction between the behavioral functional level and the informational functional level is meaningful because they do not overlap completely. Most importantly, one conclusion at the behavioral functional level is compatible with several competing conclusions at the informational functional level. Hence, understanding the behavioral functional level does not provide a full understanding of the informational functional level. This point can also be illustrated in the context of research on causal attribution, more specifically on the phenomenon known as discounting. Discounting relates to the ability to distinguish between genuine and spurious causes. It is related to many aspects of social behavior, including the fundamental attribution error (Ross, 1977), which describes the failure to discount the causal role of dispositional factors in the presence of situational factors that may cause the same behavior. A common effect of this failure is that perceivers draw correspondent inferences about people’s dispositions on the basis of observed behavior even when the behavior might have been caused by situational factors (i.e., the correspondence bias; see Gilbert & Malone, 1995). In studies on causal attribution, discounting refers to the observation that causal judgments are lower when relations are redundant than when they are not redundant (e.g., McClure, 1998). Assume that Person A and Person B most often win a difficult game when they play together. Some participants also see games in which Person A plays alone and is as likely to win as when Person A and Person B play together [i.e., P(win/A) = P(win/A.B)]. Discounting occurs when judgments about the ability of Person B to win the game when playing alone are lower when Person A is seen to win games than when Person A never plays alone. In the former case, the relation between B (cue) and winning (outcome) can be described as redundant in that the presence of B does not change the probability that the game will be won (Cheng & Novick, 1990). Hence, one can conclude at the behavioral functional level that causal attribution (as an overarching mental process) depends on the redundancy of relations in the environment. There are different causal judgments (output of the causal attribution process) for redundant relations than for nonredundant relations (input of the causal attribution process). For the present purposes, it is important to realize that this conclusion at the behavioral functional level has been attributed by some to functional properties of the information storage subprocess, whereas others have attributed it to the functional properties of the information retrieval subprocess.

30 BASICS

More specifically, some have argued that in case of redundancy, the occurrence of AB-win games (input of the information storage subprocess) does not result in the storage of information about the relation between B and winning (output of the information storage subprocess; e.g., Van Overwalle & Timmermans, 2005). Others, however, have argued that information about the relation between B and winning is stored in memory (and can thus serve as input of the information retrieval subprocess) but has little impact on causal judgments (output of the information retrieval subprocess) when the relation is redundant (see De Houwer, Beckers, & Vandorpe, 2005, for a review). This example illustrates that assumptions about relations between the input and output of mental subprocesses (i.e., analyses at the informational functional level) can vary, while assumptions about the input and output of overarching mental processes (i.e., analyses at the behavioral functional level) remain constant. Hence, both levels do not overlap. As noted above, both functional levels also do not overlap with the algorithmic level in that the latter level deals with how input is transformed into output, not merely with which input is transformed into which output.

How Are the Behavioral Functional Level, the Informational Functional Level, and the Algorithmic Level Related? We have already noted that the behavioral functional level, the informational functional level, and the algorithmic level do not overlap because a theory at one level can be compatible with multiple competing theories at other levels. Put differently, there is a one-to-many relation between certain levels. These three levels of analysis are interdependent in two important ways. First, in some cases, one theory at one level implies a single theory at another level. In other words, sometimes there is a one-to-one relation going from one level to another. Second, even when one theory at one level does not imply one theory at another level, often it does constrain the theories that are possible at another level. One-to-one relations can exist between different levels of analysis but only if certain conditions are fulfilled and only going from some levels to some other levels. Consider the algorithmic level. For any given input, an algorithm will produce an output. By applying the algorithm to a variety of inputs, one can therefore deduce a functional model of the process, that is, a description about what the input–output relations of the mental process are. Hence, provided that all elements of the algorithm are specified (e.g., all parameters are given a value) and certain background assumptions are made (e.g., that a person can see and hear), an algorithmic model implies a functional model. Actually, we already discussed one example of such a one-to-one relation, namely, the fact that the delta P model of causal attribution (which describes causal attribution at the behavioral functional level by specifying which environmental events lead to which judgments) can be derived from the Rescorla–Wagner model (i.e., an algorithmic model of

Levels of Analysis in Social Psychology 31

causal attribution). We now see that the expression “be derived from” actually means that, when the parameters of the Rescorla–Wagner model are fixed in a certain way, the Rescorla–Wagner model predicts for each possible set of learning events (input) the same causal judgments (output) as the delta P model. Likewise, assumptions at the informational functional level can sometimes imply conclusions at the behavioral functional level. For instance, based on the premises (made at the informational functional level) that (1) causal attribution involves information storage and (2) redundant events are not stored in memory, one can conclude (at the behavioral functional level) that redundant events should not influence causal judgments. Whether there can be one-to-one relations between levels of analysis also depends on the level at which the premises are stated and the level at which the conclusions are drawn. For instance, whereas making assumptions at the algorithmic level can allow one to deduce a single conclusion at the behavioral functional level, assumptions at the behavioral functional level (e.g., that redundant events do not influence causal judgments) do not imply a single conclusion at the algorithmic level (e.g., the absence of an impact of redundant events on judgments could be due to several algorithms). In other words, whereas there is a one-to-one relation starting from the algorithmic level and going to the behavioral functional level, there is a one-to-many relation starting from the behavioral functional level and going to the algorithmic level. Hence, the interdependence between different levels of analysis is asymmetrical. Figure 2.1 provides an overview of the types of relations between the different levels. Even when a theory at one level does not imply a single theory at another level, it still constrains the possible range of theories at the other level. In more technical terms, a one-to-many relation is not the same as a one-to-all relation. For instance, assumptions at the behavioral functional level (e.g., that redundancy influences causal judgments) constrain conclusions at the algorithmic level (e.g., algorithms need to explain the effect of redundancy), even though they do not allow one to infer a single set of conclusions at those other levels (e.g., there might be several possible mechanisms that produce redundancy effects). The previous paragraphs are related to two well-known rules of thumb from methodology classes: “theories can be used to derive predictions about data” (Principle 1) and “data can be used to reject theories but not to prove a theory” (Principle 2). In essence, our ideas go beyond these well-known principles by pointing out that (1) there are several levels of analysis at which theories can be formulated and (2) depending on the level at which it is formulated, a theory can imply a theory or merely constrain theories at other levels. Theories can address the function of an overarching mental process (i.e., a theory situated at the behavioral functional level), the function of a mental subprocess of this overarching mental process (i.e., a theory situated at the informational functional level), or the mechanism by which the function of a mental process (either overarching or subprocess) is real-

32 BASICS BF1

IF1

IF2

IF3

IF4

IF5 . . .

AL1

AL2

AL3

AL4

AL5 . . .

FIGURE 2.1.  Schematic overview of the relations between the behavioral function level, the informational functional level, and the algorithmic level of analysis. BF1 stands for one possible theory at the behavioral functional level. IF1 to IF5 stand for five different potential theories at the informational functional level. AL1 to AL5 stand for five different possible theories at the algorithmic level. Solid arrows symbolize one-to-one relations in that they point from a specific theory at a specific level of analysis to one theory at another level of analysis. Dotted arrows symbolize oneto-many relations in that they point from a specific theory at a specific level of analysis to several theories at another level of analysis.

ized (i.e., a theory at the algorithmic level). Hence, we can add to Principle 1 that some theories can be used not only to derive predictions about the data that will be obtained in a specific experiment but also to derive theories at other levels of explanation (e.g., the delta P model can be derived from the Rescorla–Wagner model). We can add to Principle 2 that the acceptance of certain theories at certain levels can constrain the theories that are possible at other levels without actually “proving” one specific theory at that other level (e.g., the claim that discounting is due to a lack of information storage, rather than information retrieval, excludes algorithms that store information regardless of redundancy). We would also like to add a third point that goes beyond traditional ideas about the relation between theory and data (also see De Houwer, 2011; De Houwer, Barnes-Holmes, & Moors, 2013). We agree with the axiom that data merely constrain rather than prove theory, but we also believe that data constrain theories at different levels of analysis to different degrees. Data constrain theories at the behavioral functional level to the largest extent because these theories specify only which aspects of the environment lead to which aspects of behavior. Whether something in the environment influences a behavior cannot be observed directly (which is why the behavioral functional level is a level of analysis rather than mere description), but it can be inferred from data generated from experiments. Hence, experimental research puts strong constraints on theories at the behavioral functional level and thus provides a good source for conclusions at that level (also see De Houwer, 2011).

Levels of Analysis in Social Psychology 33

Data put much less constraints on informational functional theories and even less on algorithmic theories. Whereas data are essentially behaviors that are observed under certain environmental conditions, informational functional theories contain assumptions about input–output relations that involve mental representations that intervene between the environment and behavior (De Houwer, 2011; Gardner, 1987). Hence, informational theories go beyond data in two ways: (1) they imply conclusions about which input leads to which output (as is the case for theories at the behavioral functional level); and (2) they make assumptions about the content of mental representations in the organism that performs the behavior (whereas data only specify environmental conditions and behaviors).1 Algorithmic models go even further beyond the data by adding assumptions about the mental mechanism by which an input is transformed into an output and the format of the representations on which these mechanisms operate. In addition, algorithmic models often include assumptions about how these mental representations are formed, transformed, and influence behavior. Hence, at the algorithmic level, even more needs to be inferred from the data. It seems reasonable to assume that the more that needs to be inferred from the data, the more difficult it is to arrive at firm conclusions. We introduce the term inductive penetrability to refer to the extent to which data constrain theories at the different levels of analysis: The behavioral functional level is more penetrable than the informational functional level, which is in turn more penetrable than the algorithmic level. Our analysis raises the question of how one can arrive at conclusions at the informational functional and algorithmic levels. One option is to look for so-called proxies of mental representations and mental mechanisms. Proxies are specific behavioral effects that can be treated as if they are a perfect reflection of some mental construct. For instance, ratings on an attitude scale can be taken as a proxy for mental attitudes. If the rating is positive, then the attitude is positive. If the rating is negative, then the attitude is negative. However, the use of proxies requires the unlikely assumption that the proxies are determined only by one mental construct (e.g., De Houwer, 2011). For instance, attitude ratings as proxies of attitudes make sense only if the ratings are determined solely by the attitudes in memory. However, it has been shown that attitude ratings depend on a wide variety of factors that are not related to attitudes, such as the order in which ratings have to be completed (e.g., Schwarz & Bohner, 2001). Many more proxies of attitudes have been proposed, but none provides a pure index of attitudes, probably because behavior is never determined only by one type of representation or process (see De Houwer, 2011; De Houwer, Gawronski, & Barnes-Holmes, 2013, for a detailed discussion of these issues in the context of attitude research). There is, however, another way to draw conclusions at the informational functional and algorithmic levels. We know that data constrain the behavioral functional level. We also know that the behavioral functional level constrains the informational functional level, which in turn constrains

34 BASICS

the algorithmic level. Hence, if we proceed carefully from the data to behavioral functional theories, conclusions at the behavioral functional level can be used to learn more about the informational functional level, which can then inform us about the algorithmic level. It will be difficult to arrive at definite conclusions at the informational functional and (even more so) algorithmic levels, but we can make progress at those levels by carefully considering how conclusions at these levels can be constrained by conclusions at other levels. As we shall see in the next section, these ideas have profound implications not only for social cognition research but for social psychology as a whole.

Implications for Social-Psychological Research Benefits of Focusing on the Functional Levels in Addition to the Algorithmic Level Many social cognition researchers try to formulate theories at the algorithmic level. This is very appealing if one’s goal is to understand the mental processes that underlie social behavior (as is the case for social cognition researchers). Not only does the algorithmic level offer an understanding of the mental mechanisms by which the input of a process is transformed into an output, but it also offers insight into the function of the mental (sub)process by allowing one to predict for any particular input into the algorithm what the output would be. As is often the case, however, the best things in life are the hardest to get. Our analysis suggests that algorithmic theories have a low inductive penetrability. In other words, it is difficult to arrive at firm conclusions at this level on the basis of empirical data. Illustrations of this problem in social psychology are manifold. Just think of the seemingly unresolvable debates between dual-process and single-process models of social information processing (e.g., Deutsch & Strack, 2006; Kruglanski et al., 2006), competing accounts of dissociations between implicit and explicit attitude measures (e.g., Greenwald & Nosek, 2008), and representational versus constructive theories of attitudes (e.g., Fazio, 2007; Schwarz, 2007). Some have even wondered whether any substantial progress has been made at the algorithmic level of understanding mental processes (e.g., Garcia-Marques & Ferreira, 2011; Meiser, 2011). Although we believe that social cognition researchers can continue to strive for progress at the algorithmic level, we also believe that it is important to simultaneously pay more attention to the behavioral and informational functional levels. First, these levels are inductively more penetrable than the algorithmic level of analysis, which is why it should be easier to reach conclusions at these levels and thus to make genuine scientific progress. Second, focusing on the two functional levels will also allow one to consolidate progress that is made at the algorithmic level. In those cases where empirical

Levels of Analysis in Social Psychology 35

data clearly contradict a specific algorithmic model, it is important to consider what these data imply at the informational and behavioral functional levels. Those data can be used not only to constrain the algorithmic level but also to make progress at the informational and/or behavioral functional levels. In this way, ideas at the algorithmic level can continue to fluctuate and evolve, while at the same time a stable base of knowledge is formed at the informational and behavioral functional level. This not only increases our understanding of the function of mental (sub)processes but also provides the best way to constrain future developments at the algorithmic level and thus make progress also at that level of analysis. Perhaps the informational functional level provides the most fertile ground for social cognition research. Just as there is merit in understanding environment–behavior relations without understanding the (mental) mechanisms by which elements in the environment lead to certain behaviors, there is merit in understanding which elements in the environment lead to which mental representations or which mental representations lead to which behaviors without understanding the mechanisms by which an input leads to an output. Importantly, this level combines some of the advantages of the other two levels. On the one hand, it has more deductive value than the behavioral functional level. That is, theories at the behavioral functional level in principle can be derived from theories at the informational functional level, whereas the reverse is not true. On the other hand, it is inductively more penetrable than the algorithmic level. Hence, the informational functional level can be a middle ground at which important and genuine progress can be made in understanding mental processes.

Different Levels of Analysis Are Mutually Reinforcing Provided That They Are Conceptually Separated Because different levels of analysis are interdependent, progress at one level can facilitate progress at another level. In other words, the different levels of analysis are mutually supportive. However, in order to be optimally supportive, the different levels need to be separated conceptually. It is particularly important that concepts at levels with low inductive penetrability (e.g., the algorithmic level) are not used in analyses at levels with high inductive penetrability (e.g., the functional levels). The lower the inductive penetrability, the more difficult it is to reach conclusions and thus the more likely it is that conclusions at that level will fluctuate while conclusions at other levels remain constant. If theories at a level with high penetrability are formulated in terms of theories at a lower level of penetrability, then changes in the latter theories might jeopardize the former theories. Consider the phenomenon of discounting in causal attribution. Discounting is sometimes also referred to as “blocking,” a term that stems from the idea that discounting it is due to a lack of association formation for redundant cue–outcome relations (e.g., Fanselow, 1998). The use of such

36 BASICS

a term might discourage researchers from considering algorithmic models that explain discounting not in terms of a lack of association formation but in terms of a lack of information retrieval. Moreover, if discounting is defined as blocking (i.e., a lack of association formation), then support for other explanations of blocking would imply that “real” discounting (i.e., discounting that is due to a lack of association formation) might not exist.

On the Relation between Different Approaches in Social Psychology As noted in the introduction, the unifying aim of social psychology is to understand social behavior. Until now in this chapter, we have focused on the social cognition approach in social psychology which tries to understand the mental processes underlying this social behavior. There are also, however, other approaches to studying social behavior. These other approaches differ from the social cognition approach with regard to the kind of question they ask about social behavior. In this section, we discuss two other approaches and examine how they are related to the social cognition approach. More specifically, we focus on the functional and the neuroscience approach in social psychology. The functional approach in social psychology aims to understand social behavior by identifying the environmental causes of that behavior. One example of this approach is provided by research on “the power of the situation” in the 1960s and 1970s, which primarily focused on how subtle features of social situations can have powerful effects on social behavior. The most prominent examples include Milgram’s (1963) studies on obedience, Zimbardo’s Stanford prison experiment (Haney, Banks, & Zimbardo, 1973), and Darley and Latané’s (1968) research on bystander intervention. Most of the initial research on these forms of social influence was not directed at understanding the mental processes by which the environment influences behavior (although in some cases subsequent research did attempt to do so). It is quite obvious that this functional approach in social psychology resonates well with the behavioral functional level of analysis in social cognition research. Both approaches try to uncover functional relations between environment and social behavior. In fact, the difference between the two merely concerns the goals to which the discovery of functional relations is a means. Whereas traditional functional researchers want to uncover functional relations between environment and social behavior in order to understand the environmental determinants of social behavior, social cognition researchers who engage in behavioral functional analyses want to uncover those same functional relations in order to better understand the overarching mental processes by which the environment influences social behavior. Now that we have made explicit the link between traditional functional research and the behavioral functional level of analysis within social

Levels of Analysis in Social Psychology 37

cognition research, it also becomes clear how the functional and the social cognition approach in social psychology are related. Because the functional approach and the behavioral functional analysis in the social cognition approach use the same means for different goals (i.e., they both look for functional relations between environment and social behavior, be it for different reasons), they can directly contribute to each other. The only requirement is that one reaches agreement about which overarching mental processes underlie which kinds of environment–behavior relations. For instance, one could agree that the overarching mental process that drives obedience in Milgram’s (1963) experiments is persuasion (in the sense of the overarching mental process by which persuasive messages influence behavior). In that way, by demonstrating how elements of the environment influence obedience behavior, traditional functional research also contributes to understanding of persuasion at the behavioral functional level. Vice versa, social cognitive research on persuasion as an overarching mental process can help increase our understanding of obedience behavior by uncovering new aspects of the environment that influence the output of the persuasion process. Because of this overlap between the traditional functional approach and the behavioral functional analysis within the social cognition approach, both approaches can also interact in other ways. As we have discussed above, behavioral functional analyses constrain analyses at the informational functional and the algorithmic level. The traditional functional approach can therefore also contribute to the informational functional and algorithmic levels of the social cognition approach by expanding knowledge at the behavioral functional level. In turn, the informational functional and the algorithmic levels can lead to new predictions and insights at the behavioral functional level. Testing these predictions can therefore also contribute to the traditional functional approach. In fact, by combining the traditional functional and the cognitive approach, a new functional-cognitive approach emerges that facilitates progress both at the functional and cognitive level (see De Houwer, 2011; De Houwer, Barnes-Holmes, & Moors, 2013; and De Houwer, Gawronski, & Barnes-Holmes, 2013, for more details). Another approach in social psychology that is becoming increasingly popular is the neuroscience approach. Social neuroscience aims to understand the neural processes that underlie social behavior. There are several ways in which it interacts with both the functional and cognitive approach in social psychology. From the perspective of a functional approach, the brain can be seen both as a part of the environment and as the locus at which neural behavior takes place. As such, within a functional approach, one can examine how properties of the brain (as part of the environment) influence behavior (e.g., by examining the effects of brain lesions on behavior) or how other parts of the environment influence the behavior of the brain (e.g., by examining the influence of stimuli on electrical activity or blood flow in the

38 BASICS

brain). This functional neuroscience approach actually exists under the name of behavioral (social) neuroscience (e.g., Carlson, 2010). From the perspective of social cognition research, neuroscience research can help us understand how mental processes are implemented in the brain. Hence, neuroscience can contribute to the implementational level of analysis as introduced by Marr (1982). As Marr already pointed out, the algorithmic level cannot be reduced to the implementational level because the same algorithm can in principle be implemented in different physical systems. In our terms, there is a one-to-many relation going from the algorithmic level to the implementational level. But there is also a one-to-many relation when going from theories at the implementational level to algorithmic theories. That is, a single set of conclusions about the physical system that underlies an algorithm (e.g., the involvement of certain brain regions) does not allow one to draw a single set of conclusions at the algorithmic level. Even if one would have perfect knowledge about the brain and how it guides behavior, we believe it would still be possible to come up with several different ideas about the information processing that the brain actually performs. What is most important for the present purpose is that these ideas imply that the (social) neuroscience approach will never fully replace the (social) cognition approach. Both approaches can certainly inform each other, but one cannot be reduced to the other (see Craver & Bechtel, 2007; Chiesa, 1992, 1994, for excellent discussions of how different questions about a single object relate to each other).

Conclusion Our scientific goals determine our research and the way in which we interact with our colleagues. We believe that making these goals explicit will promote scientific progress. In this chapter, we have argued not only that a scientific goal can specify an object and a specific question about that object, but also that a specific question about a specific object can be studied at different levels. Hence, to promote scientific progress, it is good to also be aware of the level of analysis at which a scientific question is examined and to know how the different levels relate. We hope that our chapter not only raises awareness of these issues but also helps researchers to engage in these analyses of their own research and that of their peers. Acknowledgments Preparation of this chapter was made possible by Methusalem Grant No. BOF09/01M00209 from Ghent University awarded to Jan De Houwer. We thank Bertram Gawronski and Galen Bodenhausen for their extensive comments on an earlier draft of this chapter, especially with regard to how we could illustrate the relevance of our ideas for social psychology.

Levels of Analysis in Social Psychology 39

Note 1. One could argue that analyses at the behavioral functional level also go beyond the observed environment and observed behavior. This is true because the observer always perceives the world in a way that is shaped by his or her concepts and theories. However, drawing conclusions about mental representations requires an extra level of subjectivity and uncertainty because mental representations, as nonphysical entities, can be inferred only by observing the physical environment and behavior.

References Bechtel, W. (2005). The challenge of characterizing operations in the mechanisms underlying behavior. Journal of the Experimental Analysis of Behavior, 84, 313–325. Bechtel, W. (2008). Mental mechanisms: Philosophical perspectives on cognitive neuroscience. London: Routledge. Carlson, N. R. (2010). Foundations of behavioral neuroscience. New York: Pearson. Cheng, P. W. (1997). From covariation to causation: A causal power theory. Psychological Review, 104, 367–405. Cheng, P. W., & Holyoak K. H. (2011). Causal attribution and inference as a rational process: The new synthesis. Annual Review of Psychology, 62, 135–163. Cheng, P. W., & Novick, L. R. (1990). A probabilistic contrast model of causal induction. Journal of Personality and Social Psychology, 58, 545–567. Chiesa, M. (1992). Radical behaviorism and scientific frameworks: From mechanistic to relational accounts. American Psychologist, 47, 1287–1299. Chiesa, M. (1994). Radical behaviorism: The philosophy and the science. Boston: Authors’ Cooperative. Craver, C. F., & Bechtel, W. (2007). Top-down causation without top-down causes. Biology and Philosophy, 22, 547–563. Darley, J. M., & Latané, B. (1968). Bystander intervention in emergencies: Diffusion of responsibility. Journal of Personality and Social Psychology, 8, 377–383. De Houwer, J. (2011). Why the cognitive approach in psychology would profit from a functional approach and vice versa. Perspectives on Psychological Science, 6, 202–209. De Houwer, J., Barnes-Holmes, D., & Moors, A. (2013). What is learning? On the nature and merits of a functional definition of learning. Psychonomic Bulletin and Review, 20, 631–642. De Houwer, J., & Beckers, T. (2002). A review of recent developments in research and theory on human contingency learning. Quarterly Journal of Experimental Psychology, 55B, 289–310. De Houwer, J., Beckers, T., & Vandorpe, S. (2005). Evidence for the role of higher-order reasoning processes in cue competition and other learning phenomena. Learning and Behavior, 33, 239–249. De Houwer, J., Gawronski, B., & Barnes-Holmes, D. (2013). A functional-cognitive framework for attitude research. European Review of Social Psychology, 24, 252–287. Deutsch, R., & Strack, F. (2006). Duality models in social psychology: From dual processes to interacting systems. Psychological Inquiry, 17, 166–172. Fanselow, M. S. (1998). Pavlovian conditioning, negative feedback, and blocking: Mechanisms that regulate association formation, Neuron, 20, 625–627. Fazio, R. H. (2007). Attitudes as object-evaluation associations of varying strength. Social Cognition, 25, 603–637.

40 BASICS Garcia-Marques, L., & Ferreira, M. B. (2011). Friends and foes of theory construction in psychological science: Vague dichotomies, unified theories of cognition, and the nex experimentalism. Perspectives on Psychological Science, 6, 192–201. Gardner, H. (1987). The mind’s new science: A history of the cognitive revolution. New York: Basic Books. Gilbert, D. T., & Malone, P. S. (1995). The correspondence bias. Psychological Bulletin, 117, 21–38. Greenwald, A. G., & Nosek, B. A. (2008). Attitudinal dissociation: What does it mean? In R. E. Petty, R. H. Fazio, & P. Brinol (Eds.), Attitudes: Insights from the new implicit measures (pp. 65–82). Hillsdale, NJ: Erlbaum. Haney, C., Banks, W. C., & Zimbardo, P. G. (1973). Interpersonal dynamics of a simulated prison. International Journal of Criminology and Penology, 1, 69–97. Kelley, H. H. (1973). The processes of causal attribution. American Psychologist, 28, 107–128. Kruglanski, A. W., Erb, H.-P., Pierro, A., Mannetti, L., & Chun, W. Y. (2006). On parametric continuities in the world of binary either ors. Psychological Inquiry, 17, 153–165. Marr, D. (1982) Vision. San Francisco: Freeman. McClamrock, R. (1991). Marr’s three levels: A re-evaluation. Minds and Machines, 1, 185– 196. McClure, J. (1998). Discounting of causes of behavior: Are two reasons better than one? Journal of Personality and Social Psychology, 74, 7–20. Meiser, T. (2011). Much pain, little gain? Paradigm-specific models and methods in experimental psychology. Perspectives on Psychological Science, 6, 183–191. Milgram, S. (1963). Behavioral study of obedience. Journal of Abnormal Psychology, 67, 371– 378. Ross, L. D. (1977). The intuitive psychologist and his shortcomings: Distortions in the attribution process. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 10, pp. 173–220). New York: Academic Press. Schwarz, N. (2007). Attitude construction: Evaluation in context. Social Cognition, 25, 638– 656. Schwarz, N., & Bohner, G. (2001). The construction of attitudes. In A. Tesser & N. Schwarz (Eds.), Intrapersonal Processes (pp. 436–457). Oxford, UK: Blackwell. Shanks, D. R. (1995). The psychology of associative learning. Cambridge, UK: Cambridge University Press. Van Overwalle, F., & Timmermans, B. (2005). Discounting and the role of the relation between causes. European Journal of Social Psychology, 35, 199–224.

3 Causal and Meaning-Based Explanation Yoshihisa Kashima

S

ocial psychologists often take for granted that social psychology is a discipline that constructs, evaluates, and applies causal explanations about social behavior. Our theories are meant to describe how events causally relate to each other and eventually produce a behavior of interest. In addition to causal explanation, however, a different type of explanation, which may be called meaning-based explanation, is increasingly offered in important subareas of social psychology. Roughly speaking, meaning-based explanation explains a social behavior in terms of its meaning, and the main point of meaning-relevant research is an insightful explication of meaningful relations among events and the behavior. Examples are often found in investigations of cross-cultural differences, folk or naïve theories of various knowledge domains, many studies of priming effects, and research on intention–action relations as I will point out later. Many of these studies whose main point is an explication of meanings may be called experimental semiotics (Kashima & Haslam, 2008), an experimental (and sometimes quasi-experimental or correlational) investigation of meaning. Confusing meaning-based explanations for causal explanations can lead us into a theoretical conundrum that we can do without. The purpose of this chapter is to characterize the meaning-based approach as clearly as possible, outline its historical background, warn 41

42 BASICS

against pitfalls of confusing the two, and suggest potential gains for keeping the distinction in mind in social-psychological research.

Meaning in Social Psychology Human sociality is saturated with meaning (Holtgraves & Kashima, 2008). Imagine you see a school boy rapidly contracting his eyelid (Ryle, 1971). What do you make of it? It could be that he has a facial neuralgia, an unfortunate state of affairs for which you might feel sorry on his behalf. Or he may be trying to wink, but only clumsily, so that it looks more like a facial neuralgia. You might then smile at the youthful attempt at trying to be an adult. Or he may be trying to mock another boy, who has a facial neuralgia. In this case, you might feel like giving this mischievous boy a piece of your mind, so that he wouldn’t make a mockery of the unfortunate predicament of another human being. Your reactions are likely to depend on which meaning you attribute to the same motor behavior of the rapid contraction of an eyelid (Geertz, 1973). It is the meaning of a motor behavior, rather than the motor behavior per se, that is critical for human social engagement. Nowhere is this more clearly observable than in language use. Imagine someone says to you: “先生の最近の論文を拝見しました。とても面白かったです。” He is making a couple of utterances in Japanese, and it would be obvious to you that he is moving his mouth, pushing in and out some air, opening or closing his vocal cords, etc., etc. These aspects of his behavior are what Austin (1962) called locutionary acts. However, if you do understand Japanese, you would respond to his illocutionary act, namely, what he meant to do in saying, “I read your recent article. It was very interesting.” (This is a translation of the above Japanese sentences in case you haven’t noticed.) So, in this case, you may see it as an act of praising, rather than producing a string of nonsense syllables. Or you might interpret it as flattering, or even ingratiating. You then may try to explain why he praised/flattered/ingratiated you— because he is an honest man, who rightly praises what is a brilliant paper, because he is friendly and trying to get acquainted with you, or because he is trying to get something from you? Just as you might try to determine the meaning of his behavior by referring to concepts such as his goals, personality, and other factors, the central explanandum for social psychology too is, more often than not, a meaningful social action. In fact, the notion of meaning has been an integral part of social psychology throughout its history. Since Asch’s (1946) classic study on impression formation, the social actor’s perception (i.e., from stimuli to psychological processes) has been regarded as the actor’s construction of meaning. More gen-

Causal and Meaning-Based Explanation 43

erally, Bartlett (1932), a psychologist often credited to have introduced the schema concept to psychology, regarded memory and cognition as effort after meaning. Meaning as a basis for human interpretation and understanding of a social event seems so much a part of social psychology that, according to Ross and Nisbett (1991), the principle of construal—an actor’s construal of an event, rather than the event per se, as a cause of social behavior—is one of the fundamental assumptions of our discipline. Meaning appears not only in social psychology of perception, but also in that of action (i.e., from psychological processes to behavior). At least from Lewin (1936) onward, social psychology of motivation has attempted to explain human action in terms of its purpose or goal. Feather (1982) noted that Lewin’s and many other theories of motivated action can be understood in terms of expectation and value; that is, the value of the goal and the expectation that an action can help reach the goal of motivating people to perform the action. In other words, the goal (i.e., its value) provides a reason or a purpose for performing the action. This expectancy × value formulation appears in numerous theories and models of social behavior, perhaps most prominently in the theories of reasoned action (Fishbein & Ajzen, 1975) and planned behavior (Ajzen, 1991). Bruner’s (1991) Acts of Meaning is a provocation; much like Bartlett’s (1932) effort after meaning, he urges us to consider human action as that of active meaning-making. More recently, taking the intellectual perspective of existentialism, Markman, Proulx, and Lindberg (2013) declared the arrival of a new science of meaning, while particularly emphasizing the purposive nature of human action. Finally, meaning has been central to social-psychological understanding of culture. In the very first chapter of the first volume of Advances in Experimental Social Psychology, Triandis (1964) provided a broad overview of culture and its psychological impact in terms akin to conceptual meaning shared by a group of people. To paraphrase him, culture is a group of people’s intersubjectively shared way of construing and acting on the world. Geertz (1973) in anthropology similarly regarded culture as a web of meaning only to be explicated through a thick description—a rich and textured description that makes the cultural meaning of events and actions intelligible to non-natives. Arguably, social psychology of meaningful perception and action describes how meaning is used, constructed, and perhaps shared in a generation; when it is passed on to the next generation, and to the next, and so on, it becomes the culture of the successive generations of people (see Kashima & Gelfand, 2012). In many ways, culture passed down from the previous generation and spread through interpersonal processes provides a basis of the meaning implicated in the social psychology of perception and action. In this sense, culture is a significant aspect of intersubjective meaning. Thus, social psychology has been, and will increasingly be, concerned with meaning. I suggest that in so doing, social psychology often needs to make use of meaning-based explanations.

44 BASICS

What Is a Meaning-Based Explanation? What is a meaning-based explanation in social psychology? To specify, let me make use of conceptual devices, which I call the standard and semiotic models of psychological explanations (Kashima, 2009). Figure 3.1 schematically presents these models. Suppose that there are event p, concept p, concept q, and behaviors q.1 to q.n, and that the occurrence of behavior q.1 is the explanandum while the occurrence of event p is offered as an explanans. To bring out the structure of the reasoning, we can write this as a simple if–then statement. If p, then q.1. p. Therefore, q.1. Here, p should read as a proposition like “event p occurs,” and q.1, “behavior q.1 occurs.” Social-psychological theories provide justification for the assertion, “If p, then q.1,” which is assumed to hold probabilistically. Because social psychology’s subject matter is typically meaningful social action, let us further suppose that behaviors q.1 to q.n can be described or interpreted by concept q, and let us assume that event p can also be described or interpreted by concept p. In this sense, the relationship between concept p and event p, as well as the relationship between concept q and behaviors q.1–q.n are extensional—the concept refers to an event or a behavior, and the latter acts as an example of the former. In addition, the relationship between concept p and concept q may be called intensional—one concept has some specifiable conceptual relationship with another concept. Roughly speaking, a concept’s extension is its referents in the world, whereas its intension is its associates, correlates, definitions, explications, and so on, by other concepts. Extensions connect concepts to the world outside the realm of concepts, whereas intensions connect concepts with other concepts—they reside inside the realm of concepts. The standard and semiotic models of psychological explanations differ in terms of the nature of extensional and intensional relations for the psychological concepts, p and q. The standard model suggests that event p, concept p, concept q, and behavior q.1–q.n are all causally related, as shown in the upper panel of Figure 3.1. It would run a causal story roughly as follows. Event p activates the concept p in the perceiver. The activation of this concept then spreads to the concept q, which is associated with the concept p in the perceiver’s memory. The concept q is then also activated, which then causally produces in the perceiver any one of the behaviors from q.1 to q.n. In this sense, the relations among event p, concept p, concept q, and behavior q.1–q.n are all causal. In contrast, the semiotic model suggests that the concepts p and q are used by theorists to interpret or describe the event p and behavior q.1–

Causal and Meaning-Based Explanation 45 Concept p

Event p

Concept q

……….. Behavior q.1

Behavior q.2

………..

Behavior q.n

Concept p

Event p

Various mediating events

Concept q

……….. Behavior q.1

Behavior q.2

………..

Behavior q.n

FIGURE 3.1.  Schematic representation of the standard (upper panel) and a semiotic (lower panel) representation of psychological theories. Solid lines indicate causal relations. Broken lines indicate meaning relations. In the lower panel, the parallel wavy dotted lines indicate that a variety of unknown mediating events may be occurring there, which provide the actual causal chains of events that connect event p and behaviors q.1 to q.n.

q.n, but they may or may not be actually used or “activated” by the perceiver. Instead, the concepts p and q act as a “gloss” or a theoretical summary to make the relation between event p and behavior q.1 intelligible; in this sense, the relations among event p, concept p, concept q, and behavior q.1–q.n. are all based on meaning. This is the sense in which an explanation based on the semiotic model may be meaning-based. In the discourse of social scientific metatheory, this type of “explanation” is often called interpretation, rather than meaning-based explanation. This is because the word explanation has historically been reserved for causal explanation, as we will see later. Now, a concise definition of meaning-based explanation may be given. It is an explanation of an action in terms of concepts that can be used to interpret or describe the action without necessarily causally producing it. In other words, some meaning-based explanations may be causal explanations, but

46 BASICS

some meaning-based explanations may not be. In this sense, meaning-based and causal explanations are overlapping sets. The intersection of the two can exist. However, first, because some meaning-based explanations may not be causal explanations, if a meaning-based explanation is given, and its semiotic reading is correct, it should not be mistaken as a causal explanation. Second, even if a certain meaning-based explanation is not a causal explanation, it can be a useful and legitimate part of social psychology because it can bring other social and cultural benefits.

Meaning-Relevant Research in Social Psychology: Exemplars To flesh out the above definition further, let me describe some examples of meaning-relevant research in social psychology.

Cultural Comparisons Kashima and Haslam (2008, p. 236) constructed an example based on Menon, Morris, Chiu, and Hong (1999) to illustrate the difference between causal and meaning-based explanations in cultural psychology. In a company, a group of coworkers was responsible for completing a very important project. The project itself involved few complications, but one problem constantly plagued the group. One coworker, who we will call “Z,” consistently showed up late for meetings and, worse, missed deadlines. Z had reasonable excuses for every incident. For example, in one case Z was tied up with an emergency personal situation, and in another Z came down with a bad flu. In the final analysis, Z’s work did not get done to the group’s satisfaction, and the group was often charged with the responsibilities that should have been Z’s. Group relations suffered, and the members of the group often lost their patience with Z and became sidetracked from the project. As a result of these issues, the final product did not meet expectations of quality.

Suppose two commentators, Mr. West and Mr. East, made the following utterances to their respective conversational partners: Mr. West: Z was in charge of his own actions and behaviors. Mr. East: The group was unsupportive, unable to handle internal problems.

According to Menon et al. (1999), North American students endorsed Mr. West’s statement more and Mr. East’s statement less than Chinese students. Here, our explanandum is Mr. West’s or Mr. East’s illocutionary act: They are performing an act of blaming or accusing Z or the group in saying what they said.

Causal and Meaning-Based Explanation 47

Supposing that Mr. West was born and raised in the United States, whereas Mr. East, in the People’s Republic of China, a typical explanation is to say that West (East) acted the way he did because he is from an individualist (collectivist) culture. Given that Mr. West (East) was born and raised in the United States (the People’s Republic of China), which is known to be an individualist (collectivist) culture, West (East) is likely to act individualistically (collectivistically). West’s (East’s) statement is individualistic (collectivistic) because it pertains to the individual person, Z (the group), and because it attributes blame to the individual (the collective). Therefore, we can explain West’s and East’s individual and group blaming illocutionary acts in terms of individualism and collectivism (e.g., Triandis, 1989, 1995) or independent and interdependent self-construal (Markus & Kitayama, 1990). This explanation, however, can be understood in standard or semiotic ways. Under the standard model, Mr. West’s individualistic value or independent self-construal (or Mr. East’s collectivistic values, interdependent self-construal, etc.) is concept q, his individual blaming (or group blaming) is behavior q.1, and concept q causally effects behavior q.1. Mr. West and Mr. East behave differently because they hold concept q with different strengths (alternatively, this can be read as accessibility, importance, etc.). In contrast, under the semiotic model, concept q does not causally produce behavior q.1, but is used by social psychologists to describe or interpret the behavior in question (Kashima, 2009). So, in this reading, individualism or independent self-construal (or collectivism or interdependent self-construal) is not a causal explanatory concept, but an interpretive concept that helps social psychologists describe cultural patterns. Arguably, Markus and Kitayama’s (1990) well-known theory of independent and interdependent self-construals can be regarded as not so much a theory of self-concepts as causal mechanisms, but as detailed explications of the extensional meanings of individualism (or independence) and collectivism (or interdependence). Two comments are worth making. First, whether one takes a standard or a semiotic reading of the explanation, both causal and meaning-based explanations can be used to predict Mr. West’s (East’s) behavior. If Mr. West (East) is born and raised in an individualist (collectivist) culture, he is likely to engage in an individualistic (collectivistic) behavior. In this way, both causal and meaning-based explanations can serve three of the four important functions of psychological theories—description, interpretation, and prediction. Concepts used in both explanations can describe data, interpret them, and even predict them. However—and this is the second comment—causal explanations speak to causality, whereas meaning-based explanations may not, and do not necessarily do so. The causal reading of individualism and collectivism (or other related concepts) suggests that the fact that West (East) was born and raised in an individualist (collectivist) culture is a distal causal factor that caused him to have concept q in his mind; the semiotic reading suggests that the fact that West (East) was born and raised in individualist (collectivist) culture is a theoretical gloss, which is a collection of a host of

48 BASICS

events that may have occurred in the past and that involve interpersonally and intrapersonally distributed cultural representations (Kashima, 2009). What difference does it make? Quite a lot. Kashima (2009) reviewed the cultural psychological literature to highlight puzzling findings. First, there is a lack of evidence for cultural coherence; namely, many of the measures purported to tap the same cultural constructs such as individualism and collectivism, independence and interdependence are not correlated with each other. For instance, Kitayama, Park, Sevincer, Karasawa, and Uskul (2009) reported a lack of correlations among implicit measures of the cultural constructs of independent and interdependent self-construal. Furthermore, there is also a lack of evidence for cultural causation; that is, measures of those constructs used to theoretically explain cultural differences (e.g., selfconstruals) often fail to statistically account for the cultural differences found on other psychological measures as Matsumoto (1999) pointed out sometime ago. Kashima (2009) pointed out similar issues for research on holistic and analytic thought systems proposed by Nisbett, Peng, Choi, and Norenzayan (2001). These findings are puzzling when the standard model is assumed. This is because most of the psychological tasks that tap cultural differences are measures of behaviors in specific contexts or domains, namely, behaviors q.1–q.n in Figure 3.1. They are supposed to be causally produced by the activation of concept q such as self-construal. If concept q causally produces q.1–q.n, these measures should correlate with each other because they have the same underlying cause. Furthermore, a measure of concept q should statistically mediate the effect of culture on any given behavior that the activation of concept q should affect. In contrast, under the semiotic model, the lack of cultural coherence and cultural causation is not surprising. Because there is no underlying latent concept q, which causally produces behaviors q.1–q.n, there is no reason to expect correlations among measures of these behaviors, nor is there any reason to expect a mediation of cultural differences by a measure of concept q. It is also important to remember that most mediating variables do not tap directly the underlying concept anyway. An implication of this line of thinking is that it may be a futile exercise to look for a measure of the explanatory concept q which can statistically account for cultural differences (Kashima, 2009). Note that this is not to say that there are no circumstances in which mediators or causal mechanisms can be found. As this involves some detailed discussions, I refrain from going further here. Interested readers are invited to see Kashima (2009) for these exceptions.

Priming as Meaning-Relevant Research Some of the studies of priming may be seen to be more about meaning than causality. Let us construct a hypothetical scenario to make the point. Suppose that Ms. Cash works for a company, and one day she is told to work on some difficult problems. She has on hand some excellent staff that

Causal and Meaning-Based Explanation 49 she can draw on for help, but she sets out to work on the problems on her own as she is aware her staff has other things to do as well. There is a computer on Ms. Cash’s desk, and it so happens that her staff has installed a screen saver that shows U.S. dollar bills flying about. She’s not even paying attention to the screen saver, but as dollar bills fly around, Ms. Cash persists without seeking help.

According to Vohs, Mead, and Goode (2006), Ms. Cash’s viewing bills can prime the concept of “money,” which can make her act in a more self-sufficient manner. So, she might have persisted without calling on help because she was acting self-sufficiently. In truth, Vohs et al. primed the concept of “money” not by screen savers, but by showing people monopoly money in their peripheral vision or getting them to unscramble scrambled sentences that included words related in meaning to money. They used a variety of behavioral measures of self-sufficiency and also helpfulness, and generally showed that money primes increase the likelihood of self-sufficient behaviors, but reduce that of helping behaviors. Note that an explanation can be given for Ms. Cash’s behavior under the standard and semiotic models. A typical explanation would use the standard model. That is, the event of seeing the screen saver (event p) primed the concept of “money” (concept p), whose activation has also activated the concept of “self-sufficiency” (concept q), which has in turn produced the behavior of individual persistence without seeking help (behavior q.1). The activation of the money concept activates the self-sufficiency concept because these concepts are associated in memory and there is a spreading activation from one to the other. In contrast, it is possible to construct a meaning-based explanation for Ms. Cash’s behavior. Under the semiotic model, although event p (exposure to the “money” screen saver) causes behavior q.1 (e.g., individual or group blaming), the concepts p and q (“money” and “self-sufficiency”) may not be involved in the causal process. Instead, the concepts of “money” and “selfsufficiency” act as summary concepts that help interpret the events and behaviors. Ms. Cash working on her own for hours is an instance of “selfsufficiency,” and so are other behaviors such as not helping an experimenter code data (Vohs et al., 2006, Experiment 3), spending less time helping one’s colleague (Experiment 4), and the like. Readers may wonder how this latter meaning-based explanation can even be tenable. So, let us try to construct alternative causal scenarios to the one sketched under the standard model. One possibility is the following. The “money” screen saver may have primed Ms. Cash’s experience of being rewarded by her parents for working alone on difficult problems when she was a child. This may have made her feel motivated to work harder at the task on hand. Note that there is no involvement of the concepts of “money” or “self-sufficiency” in this scenario, but only a chain of episodic memories. This scenario assumes the existence of a fairly high-level of analysis—a level at which symbolic descriptions are possible (see Kashima, Woolcock,

50 BASICS

& Kashima, 2000, for a discussion). If one adopts a view that assumes subsymbolic representations (e.g., distributed connectionist models), a different causal scenario needs to be generated. Furthermore, if one attempts to provide a causal explanation at a neural (or implementational; Marr, 1982; see De Houwer & Moors, Chapter 2, this volume) level, it might be difficult to identify the activation of any specific concept such as “money” or “selfsufficiency.” Note that I do not mean to commit to any of the above alternative causal scenarios. Which type of description—concept activation, episodic memory activation, subsymbolic distributed representations, or neural processes—is appropriate partly depends on the purpose of the investigation. Nor do I mean to claim that the implementational level of causal descriptions is in some sense better or worse than what Marr (1982) called algorithmic- or computational-level descriptions (see Kashima & Kerekes, 1994). The critical point here is that, in explaining the behavior in terms of priming, we may not know exactly what the intervening causal mechanisms are, and the concepts (e.g., concepts p and q) that are postulated to have mediated the cause–effect relationship may not be causally involved (although it is also possible that they are). This line of reasoning raises two salutary points. First, search for a mediator in some priming experiments may be futile. If indeed the semiotic model holds for a given situation, the effect of the prime (event p) on a given behavior (behavior q.1) may not be mediated by the activations of concepts p and q, and so even if one tries to measure these activations by some cognitive tasks, these measures may not show a clear statistical mediation. It is also worth reminding ourselves again that a measure cannot directly tap the activation of the purported mediating concept, but a behavioral manifestation or its outcome anyway. Second, these mediating concepts may be more like glosses that help social psychologists interpret the phenomenon on hand. So, what is intriguing about Vohs et al.’s studies is not so much about the fact that the priming caused a variety of behaviors (we have known that priming can cause behavior since the 1970s), but rather the observation that the priming of a host of experiences and concepts surrounding “money” can make people “behave more self-sufficiently.” It is the rather unexpected intensional (meaning) relation between “money” and “self-sufficient behavior” that delights us and encourages us to reflect on the meaning of money and its implications in contemporary social life. There is another advantage in taking meaning-based explanations and the semiotic model seriously. It provides a perspective on the so-called “many effects of one prime problem” (Bargh, 2006). This is a common issue in the priming literature. Depending on the circumstances, one prime can have different types of effects (e.g., Balota & Paul, 1996; Barsalou, 1982). By exposing people to words related in meaning to the concept of hostility (e.g., hit, punch), we may be able to lead them to identify a gun faster, form an impression of another as hostile, or behave in a more hostile manner (see Loersch & Payne, 2011). Nonetheless, from the perspective of the semiotic model, it

Causal and Meaning-Based Explanation 51

is not very surprising that any given prime (e.g., unscramble a scrambled sentence that contains a word like “hit”) can be interpreted differently in different circumstances (e.g., when given a chance to rate a fictitious person, when given a chance to give electric shocks), and the meaning of a prime differs as a function of the context. Just as any polysemous words (words with multiple meanings; e.g., ball, club, watch) have to be disambiguated by their context, the meaning of any concept needs to be disambiguated as to its implications for action. To put it differently, this perspective suggests that the effect of priming should depend on how people interpret the meaning of the prime for their action in situ. It is interesting to note that many recent theories about priming effects approach them in line with this perspective (e.g., Cesario, Plaks, & Higgins, 2006; Kay, Wheeler, & Smeesters, 2008; Loersch & Payne, 2011). These theories attempt to provide a causal explanation for the causal relationship between the prime and the observed behavior. In so doing, they stipulate that the priming event causally produces some form of psychological representations, which in turn are processed together with other psychological representations (e.g., prior knowledge) to produce the observed behavior. These processes are the purported events that causally connect the prime to the behavior. To put it differently, these theories tell causal stories of priming. At this point, meaning-based explanation and causal explanation begin to converge. That is, the meaning that the social perceiver construes (i.e., the perceiver’s representations) may become part of a causal explanation. To the extent that the perceiver’s meaning largely coincides with the theorist’s meaning (i.e., the perceiver’s representations are in fact what the theory says are being used in the process), the meaning-based explanation seems to be very close to a causal explanation. However, it is important to note that this implies that the perceiver and the theorist share the same (or sufficiently similar) meaning, or the same culture. Even then, however, the distinction between the standard and semiotic models (and standard and semiotic readings of an explanation) is useful in clarifying the aspects of the theory that are truly part of the causal story of cognition. For example, in Loersch and Payne’s (2011) theory of priming, the priming event gives rise to a psychological response, which is misattributed as coming from oneself, and this response is used to answer a question afforded by the situation. This causal explanation may be interpreted as stipulating that some form of self-concept (because the response has to be attributed to the self) and question asking (a meaning-seeking activity) are involved in the causal process. Here, an empirical question may be asked as to whether some self-concept is activated or whether a question is in fact asked in mind. Or are these stipulated cognitive events the theorists’ “glosses” and a more semiotic reading is preferred? Which aspects are glosses, and which aspects are not? Depending on the answers, the theory needs to be tested rather differently. More generally, because theorists need to communicate about their theory with other colleagues within the research

52 BASICS

community, the distinction between the standard and semiotic models can be an issue even if the theory is concerned with the perceiver’s meaning. In summary, in examining the effects of priming such concepts as money (Vohs et al., 2006; Zhou, Vohs, & Baumeister, 2008), independence and interdependence (Brewer & Gardner, 1996; Trafimow, Triandis, & Goto, 1991), free will (Vohs & Schooler, 2008), warmth (Williams & Bargh, 2008), and mortality (Greenberg, Pyszczynski, Solomon, Simon, & Breus, 1994), the studies may not be so much concerned about showing a causal relation between the prime and the behavior. Rather, the point of these studies is more about showing and explicating the meaningful relations between the prime and behavior. To give another recent prominent example, the point of Williams and Bargh’s (2008) studies is that something as trivial as holding a warm (vs. cold) cup can affect one’s impression of a person who is described by a series of adjectives like intelligent, skillful, industrious, determined, practical, and cautious as someone who is warm (vs. cold). It is the apparent existence of an intensional relation between holding a warm cup and forming a warm impression that surprises and delights us. Whether this relation is causal does not matter so much, but what matters is that the warmth of a cup of coffee can metaphorically mean interpersonal warmth for us.

Action Theories as Meaning-Relevant Research One of the most productive and well-known lines of research in social psychology is action theories. Here, I am following Greve’s (2001) terminology; action theories include the theories of reasoned action (Fishbein & Ajzen, 1975) and planned behavior (Ajzen, 1991) as well as the theory of implementation intention (Gollwitzer, 1999) and the ideomotor theory (Prinz, 1997). Despite their diversity, these theoretical orientations can be seen to have a common goal, namely, to investigate the psychological process underlying the production of action. As is well known, these theoretical orientations regard intention as a central concept in their conceptualization of action. Note that the spelling is intention with a t, not intension with an s as in intensional meaning. Suppose that Mrs. Tosca is an opera buff and has heard about a new production of a Puccini in town. She checks the schedule, finds that one is on offer on this Tuesday night, and forms an intention to go to the opera on Tuesday. The Tuesday comes, and Mrs. Tosca performs a whole host of diverse motor behaviors including walking out of the door, driving her car, parking near the opera house, having some drinks while chatting with her friends and acquaintances, etc. etc. as well as watching an opera and listening to music.

Mrs. Tosca’s intention to go to the opera on Tuesday is said to predict the action of actually going to the opera on the Tuesday night, unless unpredict-

Causal and Meaning-Based Explanation 53

able and unexpected events (e.g., her car breaks down, a major fire breaks out, or an earthquake shuts down the city) prevent her from doing so. The diverse behaviors are not directly related to watching an opera, and yet they are obviously meaningfully related to the intention to go to the opera. In this context, what do we make of the relation between intention and action? In terms of Figure 3.1, intention and action are concept q and behaviors q.1–q.n. The standard model would say that intention causes action; that is, concept q causes behaviors q.1–q.n. However, there is a well-known philosophical issue surrounding this conceptualization. The criticism goes along the semiotic model reading of the relation between concept q and behaviors q.1–q.n. That is, a particular action is typically defined in terms of the presence of the corresponding intention to carry out that action. So, “going to the concert this Tuesday” as an action is defined by the very fact that the actor intends to carry out the action of “going to the concert this Tuesday.” Attempts to define a meaningful action purely by behavioral terms have run into conceptual difficulties (see Greve, 2001, for a review of this literature), and the current consensus is that an action is distinguished from behaviors because there is a logical connection between the action and the corresponding intention to carry out that action (Greve, 2001). However, this logical connection between intention and action presents a problem in speaking about causality. This is because one of the dogmas of empiricism (Quine, 1953) is that causality should not be logical necessity, and that there should not be a logical connection between a cause and an effect. That is, because an action by definition logically necessitates a corresponding intention, the intention–action relation is one of logic, not of causality. Therefore, it is conceptually incoherent to say intention causes action (for a review of the literature, see Greve, 2001). This line of criticism highlights the distinction between causal explanation and meaning-based explanation. If one follows what Quine called a dogma of empiricism, intention does not provide a causal explanation of the production of action; however, intention provides a meaning-based explanation. Note that it is possible to abandon that dogma of empiricism and to adopt a different conception of causality; however, I will not pursue this line of argument here. To say intention provides a meaning-based, but not causal, explanation does not imply that empirical research on action theories is not legitimate or useless. Greve’s (2001) response is instructive. First, he reframes typical research that uses the theory of planned behavior. To paraphrase his argument, it is not so much about intention causing action as about a measure of prior intention being able to predict intention in action. The critical and worthwhile empirical question here is about stability of intention and validity of the measurement procedure. Greve further argues that the critical empirical questions are whether reasons offered for intention/action indeed cause the intention/action—the part that the expectancy–value model underlying the theories of reasoned action and planned behavior are dealing with (e.g., Feather, 1982; cf. Shah & Higgins, 1997, as a critical counterpoint)—and how

54 BASICS

the prior intention is implemented and translated into action—the critical questions the research of implementation intention and ideomotor theory is addressing.

Other Examples of Meaning-Relevant Research There are other domains of social psychological research that are often more concerned about meaning than causality. Research on naïve theories is a case in point. There has been systematic research into people’s untutored understandings about a variety of domains of knowledge, such as physics (e.g., Krist, Fieberg, & Wilkening, 1993; McCloskey & Kohl, 1983), biology (e.g., Atran, 1998; Medin & Atran, 2004), psychology (e.g., Malle, 1999), humanness (e.g., Haslam, Loughnan, Kashima, & Bain, 2008), race (e.g., Hirschfeld, 2001), stereotypes (e.g., Fiske, Cuddy, Glick, & Xu, 2002), and society and its change (e.g., Kashima et al., 2009). Not only is there research on specific domains of knowledge, but also there is a sizable literature on naïve ontology. Examples include research on entity versus incremental theories (e.g., Dweck, Chiu, & Hong, 1999; see Dweck, 2012, for a brief overview) and essentialism (e.g., Bastian & Haslam, 2006; Dar-Nimrod & Heine, 2011; Medin & Ortony, 1989). Depending on whether people regard something as changeable or unchangeable, they approach it differently, be it intelligence, personality, or morality. All these studies attempt to find out the content of people’s cognitive representations about particular domains of knowledge, and their behavioral implications. To the extent that these naïve theories are assumed to be shared within a group and passed on from generation to generation, they are part of the group’s culture. Arguably, the burgeoning work of experimental philosophy falls into this category of research, namely, explications of the intensional meaning of concepts—typically psychological concepts—than about causal explanations of events. Take, for example, a well-known controversy about a sideeffect effect (e.g., Guglielmo & Malle, 2010; Knobe, 2003). This work investigates people’s concept of intention. A typical example runs as follows: When a company director, Mr. Bad, implements a policy knowing its potential side effect of environmental damages, has he intentionally damaged the environment? When a company director, Mr. Good, implements a policy knowing its potential side effect of helping the environment, has he intentionally helped the environment?

Typically, people are more likely to attribute intention to Mr. Bad than to Mr. Good. In other words, people are more likely to say someone performed a behavior on purpose when its known side effect is morally repugnant than morally praiseworthy. This is an important moral and legal issue. Accord-

Causal and Meaning-Based Explanation 55

ing to the standard theory of morality, whether someone intentionally performed a behavior should determine the morality or otherwise of the behavior. However, the side-effect effect suggests that people’s judgment of intentionality itself may be biased by the moral implication of its side effect. Whether a side-effect effect exists as described by Knobe (2003) is still disputed (e.g., Guglielmo & Malle, 2010), and its various implications continue to be debated. Nonetheless, it should be clear that this research is again about the meaning of intentionality in contemporary discourse. Finally, there are other studies whose results would be difficult to interpret without having insights into the meaning of the concepts used in the research, although they are not explicitly about the investigation of concepts. A good example comes from Hardistry, Johnson, and Weber (2010). They examined Democrats’, Republicans’, and Independents’ support for products and policies that are more costly due to either the “carbon tax” or “carbon offset” in the U.S. context. In three studies, they found the level of support for more costly options due to “carbon offset” was fairly similar across respondents regardless of their political affiliations; however, Republicans’ and Independents’ support for the otherwise identical items plummeted when these items were framed in terms of “carbon tax.” Those who are familiar with the meaning of “tax,” “Republicans,” and “Independents” in the U.S. context can interpret these results quite readily, but a lengthy explication of the meaning of these concepts would be necessary in order to make sense of these findings for those who are unfamiliar with the current U.S. political culture. In many studies of social-psychological topics, it is often very difficult not to make use of meaning-based explanations at least in some aspects. These examples are, in my view, attempts at explicating the intensional meaning (naïve theories) or using it to interpret their results. In this type of research, meaning-based explanations are more likely to occur than in other types of work. It is important to retain a clear distinction between causal and meaning-based explanations.

Summary The standard model regards relations among events and concepts as causal, whereas the semiotic model regards the concepts as only interpretive. If meaning-based explanations in the semiotic sense are involved in psychological theories, these theories can serve the functions of a theory—to describe, predict, and interpret data. However, one of the most cherished functions—to explain causally—may not be served by these theories. To confuse meaning-based explanations for causal explanations may be a mistake. It can lead us down the garden path of futile search for nonexistent mediators and at the same time may give us a false sense of a job well done—falsely assuming that a causal explanation of an explanandum has been given when, in fact, it still eludes us without us so realizing.

56 BASICS

History, Meaning, and Experimental Semiotics If meaning-based explanations and meaning-relevant research are so prevalent in social psychology, why have they not been widely acknowledged? To sketch out an answer to this question, we need to go back in the intellectual history of psychology (for a brief exposition, see Kashima, 2000; Kashima & Gelfand, 2012). There are two, often competing, traditions of thought in the European intellectual history. One is usually called the Enlightenment tradition, which, since its origin, was strongly aligned with the British empiricism of Locke, Hume, and the like. It regards humans as part of the Newtonian universe, whose workings are governed by the natural law that stipulates cause–effect relations among events. The capacity to reason was regarded as a natural endowment for humanity, and its workings part of the natural law. Therefore, it takes natural science—especially physics—as a model of knowledge acquisition, and experimentation as a preferred method of investigation. On the other hand, there exists an intellectual tradition, well represented by Vico, Herder, and the like, which provides a counterpoint to this Enlightenment world view. In this Counter-Enlightenment (or sometimes called Romantic) tradition, humans are regarded as intentional beings, endowed with the ability to think, but more importantly to feel, and to construct their culture and society. Separating human sociocultural processes from the Newtonian universe, this school of thought has asserted that because humans create their culture and society, culture and society can be understood in terms of the creators’ design, namely, human mentality. In this case, experimentation is not a preferred method of investigation. Instead, hermeneutics— methods of interpretation developed in the context of textual analysis as in theology and philosophy—were to be used to gain a deeper understanding about the meaning of cultural and societal products. The notion of meaning-based explanation resonates with the CounterEnlightenment tradition, and the acceptance of meaning as a legitimate topic of investigation for psychology depends largely on the acknowledgment and acceptance of the Counter-Enlightenment thought in psychology. Then, how did the Enlightenment and the Counter-Enlightenment strands of thoughts play out in the history of psychology? These schools of thought appear perhaps most clearly in Wilhelm Dilthey (as an example of his writing, see a translation published in 1989). According to Dilthey, Geisteswissenschaften (cultural science; or often translated as human science or human studies) and Naturwissenschaften (natural science) differ in that Geisteswissenschaften seeks Verstehen, an empathetic and intuitive understanding of the subject matter, whereas natural sciences seek a causal explanation as a more detached, “external” sense making. For Dilthey, Verstehen is possible only for human action and its products, whereas it cannot be achieved for natural phenomena and only causal explanations can be offered. Verstehen involved an empathetic and intuitive understanding of what an agent means to do

Causal and Meaning-Based Explanation 57

with his or her beliefs, desires, happiness and fears, intentions and resolutions, or in short, what we may call the intentionality of a human agent. The founder of the first experimental laboratory in Leipzig, Wilhelm Wundt conducted a research program that also echoes this contrast. His early introspectionist psychology emphasized experimentation in the lab, whereas his later Völkerpsychologie made no use of experiments, but instead offered more or less hermeneutic analyses of linguistic and symbolic materials. The contrasting approaches to psychology can be schematically represented for the present purpose. On the one hand, the natural science model of psychology regards causal explanation as the central aim of psychological investigation, with its key concept of causality and the attendant method of investigation of experimentation. In contrast, the cultural scientific model of psychology treats understanding as the central aim of psychological investigation, with its key concept of intentionality and hermeneutics as a preferred method of investigation. Throughout history, mainstream psychology has favored the natural science model. Despite the founder of the first psychological laboratory, Wilhelm Wundt’s embrace of Völkerpsychologie as a legitimate topic of inquiry, Titchener dropped Völkerpsychologie from his version of introspectionist structuralism. Behaviorism with its dogmatic avoidance of mind concepts made the intellectual environment of the mainstream psychology less than congenial for the cultural science model due to its embrace of intentionality as a critical element of human activities. Even with the Cognitive Revolution of the 1960s, psychology failed to bring meaning back into its mainstream, according to one of its main architects, Jerome Bruner (1991). In social psychology, this very issue was debated in the 1970s, on the pages of the Journal of Personality and Social Psychology between Kenneth Gergen and Barry Schlenker with their respective titles, Social Psychology as History (Gergen, 1973) and Social Psychology and Science (Schlenker, 1974). Gergen generally took a cultural scientific model, whereas Schlenker generally adopted a natural science model of psychology (see Kashima, 2005, for a more general exposition of similar debates in the social sciences). It is fair to say that the mainstream social psychology has taken the natural science model as its foundation in the past. Despite social psychology’s concern about meaningful human action, neither meaning nor meaning-based explanation has been explicitly acknowledged as a central issue of social psychology. Part of the resistance may have come from the vexatious issue surrounding research methodology. As I noted earlier, the natural science model tends to advocate an experimental method, whereas the cultural science model tends toward qualitative methods. Following the Gergen–Schlenker debate, possibilities of qualitative research methodology were talked about in social psychology. Yet, this type of conversation often became an intractable debate about epistemology, taking side with either one methodology or the other. The then popular “postmodern” discourse, too, added fuel to this fire. With the mainstream social psychology’s adher-

58 BASICS

ence to experimentation as a primary method of inquiry (and research training), the cultural science model was largely marginalized in this era. The intellectual environment began to change only in the 1980s and 1990s (see Kashima & Gelfand, 2012). With the advent of cross-cultural research interests, social psychology began to take explicit notice of the significance of meaning in social action. This is not to say that social-psychological theories have been silent on the subject of meaning. On the contrary, many research traditions have tacitly acknowledged meaning, as I noted at the outset of this chapter. Cognitive consistency theories of the 1950s and 1960s were really addressing the question of conceptual coherence or consistency in meaning among cognitive elements, albeit in some narrower sense. Subsequent research in social cognition has often used the notion of schema (concept, construct, or whatever) consistency and inconsistency: It has mostly to do with consistency in meaning of information in relation to a given schema. For instance, in theorizing about the consistency of given information with stereotypes, the meaning of stereotypes is an inescapable part of theorizing. More generally, when social psychologists talk about process and content, what they mean by content is meaning. It is perhaps symptomatic of today’s social psychology that one of the more popular theories of stereotypes is the stereotype content model (Fiske et al., 2002), whose main point of theorizing is an explication of the meaning of stereotypes.

Concluding Remarks Meaning is central to social-psychological investigation. Explananda of social psychology are meaningful social action; explanations of such are often meaning-based as well. By explicitly acknowledging meaning and meaning-based explanation as a legitimate goal of social psychology, it is possible to free social psychology from largely futile debates about research methodology, futile search for causal “mediators,” and other conundrums in meaning-relevant research mentioned earlier in this chapter. Instead, I believe social psychology can endorse experimental semiotics (Kashima & Haslam, 2008)—an integration of methodological experimentalism and metatheoretical embrace of meaning—as a legitimate research program. It has a potential for bringing out some surprising insights into the meaning-rich world in which humans live. Cross-cultural investigations are one approach for doing so—relativizing one’s own culture in light of others. In addition, social psychology can engage in systematic investigations of naïve theories; their psychological effects—using priming and other experimental methods—are another powerful tool, while drawing on theoretical insights from qualitative investigations of meaning. Investigating the internal intricacies of meaning with experimental methods (in conjunction with other meth-

Causal and Meaning-Based Explanation 59

ods) seems to me to be a central advantage of explicitly acknowledging the class of explanations based on meaning. To avoid any misunderstanding, let me make two categorical statements here. First of all, I am NOT saying that causal explanations should not be sought or given. Social psychology can, and should, seek to clarify the causal structure of the social world. There is no dispute here. What I AM advocating is that social psychology often does and seeks to give meaning-based explanations, and taking meaning-based explanations for causal explanations may be a mistake. Sometimes, meaning-based explanations are more appropriately construed in semiotic terms (i.e., in line with the semiotic model). Even then, meaning-based explanations can serve significant functions by contributing to the public discourse. By pointing to the intensional relation between “money” and “self-sufficiency” (or even alienation in the form of lack of helpfulness), for instance, social-psychological findings can encourage the public to reflect on the meaning of money. This may be one way to expand the reach of social psychology as Kruglanski (2001) implored us to do. Second, I am NOT suggesting that a meaning-based explanation can never be a causal explanation. Rather, I AM saying that a meaning-based explanation may not be a causal explanation. Indeed, I suggest that meaning, appropriately construed, is a very important part of a causal story of the human social world (see also Kashima, 2009). If we can identify a particular set of meaning as part of the causal structure of the human social world, it is possible to intervene into this social process by intervening into meaning. It is my hope that, by explicitly acknowledging and embracing meaning-based approaches as a legitimate research program and by achieving a seamless integration of both qualitative and quantitative methodologies, social psychology may be able to have a paradigm shift away from the polarized view that pits natural and cultural science models of psychology to a new horizon of conceiving our science. Acknowledgments The preparation of this chapter was supported by a grant from the Australian Research Council (No. DP1095323) to Y. Kashima. I thank Simon Laham for valuable comments on earlier versions of the chapter.

References Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50, 179–211. Asch, S. E. (1946). Forming impressions of personality. Journal of Abnormal and Social Psychology, 41, 258–290.

60 BASICS Atran, S. (1998). Folkbiology and the anthropology of science: Cognitive universals and cultural particulars. Behavioral and Brain Science, 21, 547–609. Austin, J. L. (1962). How to do things with words. Cambridge, MA: Harvard University Press. Balota, D. A., & Paul, S. T. (1996). Summation of activation: Evidence from multiple primes that converge and diverge within semantic memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 827–845. Bargh, J. A. (2006). Agenda 2006: What have we been priming all these years? On the development, mechanisms, and ecology of nonconscious social behavior. European Journal of Social Psychology, 36, 147–168. Barsalou, L. W. (1982). Context-independent and context-dependent information in concepts. Memory and Cognition, 10, 82–93. Bartlett, F. (1932). Remembering. Cambridge, UK: Cambridge University Press. Bastian, B., & Haslam, N. O. (2006). Psychological essentialism and stereotype endorsement. Journal of Experimental Social Psychology, 42, 228–235. Brewer, M. B., & Gardner, W. (1996). Who is this “we”? Levels of collective identity and self-representations. Journal of Personality and Social Psychology, 71, 83–93. Bruner, J. (1991). Acts of meaning. Cambridge, MA: Harvard University Press. Cesario, J., Plaks, J. E., & Higgins, E. T. (2006). Automatic social behaviour as motivated preparation to interact. Journal of Personality and Social Psychology, 90, 893–910. Dar-Nimrod, I., & Heine, S. J. (2011). Genetic essentialism: On the deceptive determinism of DNA. Psychological Bulletin, 137, 800–818. Dilthey, W. (1989). Introduction to the human sciences. (R. A. Makkreel & F. Rodi, Eds. & Trans.). Princeton, NJ: Princeton University Press. Dweck, C. S. (2012). Mindsets and human nature: Promoting change in the Middle East, the schoolyard, the racial divide, and willpower. American Psychologist, 67, 614–622. Dweck, C. S., Chiu, C.-Y., & Hong, Y.-Y. (1999). Implicit theories and their role in judgments and reactions: A world from two perspectives. Psychological Inquiry, 6, 267–285. Feather, N. T. (Ed.). (1982). Expectations and actions: Expectancy-value models in psychology. Hillsdale, NJ: Erlbaum. Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior. Reading, MA: AddisonWesley. Fiske, S. T., Cuddy, A. J. C., Glick, P., & Xu, J. (2002). A model of (often mixed) stereotype content: Competence and warmth respectively follow from perceived status and competition. Journal of Personality and Social Psychology, 82, 878–902. Geertz, C. (1973). The interpretation of cultures. New York: Basic Books. Gergen, K. J. (1973). Social psychology as history. Journal of Personality and Social Psychology, 26, 309–320. Gollwitzer, P. (1999). Implementation intentions: Strong effects of simple plans. American Psychologist, 54, 493–503. Greenberg, J., Pyszczynski, T., Solomon, S., Simon, L., & Breus, M. (1994). Role of consciousness and accessibility of death-related thoughts in mortality salience effects. Journal of Personality and Social Psychology, 67, 627–637. Greve, W. (2001). Traps and gaps in action explanation: Theoretical problems of a psychology of human action. Psychological Review, 108, 435–451. Guglielmo, S., & Malle, B. F. (2010). Can unintended side effects be intentional? Resolving a controversy over intentionality and morality. Personality and Social Psychology Bulletin, 36, 1635–1647. Hardistry, D. J., Johnson, E. J., & Weber, E. U. (2010). A dirty word or a dirty world? Attribute framing, political affiliation, and query theory. Psychological Science, 21, 86–92. Haslam, N. O., Loughnan, S., Kashima, Y., & Bain, P. (2008). Attributing and denying humanness to others. European Review of Social Psychology, 19, 55–85.

Causal and Meaning-Based Explanation 61 Hirschfeld, L. A. (2001). On a folk theory of society: Children, evolution, and mental representations of social groups. Personality and Social Psychology Review, 5, 107–117. Holtgraves, T. M., & Kashima, Y. (2008). Language, meaning, and social cognition. Personality and Social Psychology Review, 12, 73–94. Kashima, Y. (2000). Conceptions of culture and person for psychology. Journal of CrossCultural Psychology, 31, 14–32. Kashima, Y. (2005). Is culture a problem for social psychology? Asian Journal of Social Psychology, 8, 19–38. (Reprinted in International Review of Qualitative Research, 1, 103–115.) Kashima, Y. (2009). Culture comparison and culture priming: A critical analysis. In R. S. Wyer, C.-Y. Chiu, & Y.-Y. Hong. (Eds.), Understanding culture: Theory, research, and application (pp. 53–77). New York: Psychology Press. Kashima, Y., Bain, P., Haslam, N., Peters, K., Laham, S., Whelan, J., et al. (2009). Folk theory of social change. Asian Journal of Social Psychology, 12, 227–246. Kashima, Y., & Gelfand, M. (2012). History of culture and psychology. In A. Kruglanski & W. Stroebe (Eds.), Handbook of the history of social psychology (pp. 499–520). New York: Psychology Press. Kashima, Y., & Haslam, N. O. (2008). Explanation and interpretation: An invitation to experimental semiotics. Journal of Philosophical and Theoretical Psychology, 28, 234– 256. Kashima, Y., & Kerekes, A. R. Z. (1994). A distributed memory model of averaging phenomena in person impression formation. Journal of Experimental Social Psychology, 30, 407–455. Kashima, Y., Woolcock, J., & Kashima, E. (2000). Group impressions as dynamic configurations: The tensor product model of group impression formation and change. Psychological Review, 107, 914–942. Kay, A. C., Wheeler, S. C., & Smeesters, D. (2008). The situated person: Effects of construct accessibility on situation construals and interpersonal perception. Journal of Experimental Social Psychology, 44, 275–291. Kitayama, S., Park, H., Sevincer, A. T., Karasawa, M., & Uskul, A. K. (2009). A cultural task analysis of implicit independence: Comparing North America, Western Europe, and East Asia. Journal of Personality and Social Psychology, 97, 236–255. Knobe, J. (2003). Intentional action and side effects in ordinary language. Analysis, 63, 190–194. Krist, H., Fieberg, E., & Wilkening, F. (1993). Intuitive physics in action and judgment: The development of knowledge about projectile motion. Journal of Experimental Psychology: Learning, Memory and Cognition, 19, 952–966. Kruglanski, A. W. (2001). That “Vision Thing”: The state of theory in social and personality psychology at the edge of the new millennium. Journal of Personality and Social Psychology, 80, 871–875. Lewin, K. (1936). Principles of topological psychology. New York: McGraw-Hill. Loersch, C., & Payne, B. K. (2011). The situated inference model: An integrative account of the effects of primes on perception, behavior, and motivation. Perspectives on Psychological Science, 6, 234–252. Malle, B. F. (1999). How people explain behavior: A new theoretical framework. Personality and Social Psychology Review, 3, 23–48. Markman, K. D., Proulx, T., & Lindberg, M. J. (Eds.). (2013). The psychology of meaning. Washington, DC: American Psychological Association. Markus, H. R., & Kitayama, S. (1990). Culture and the self: Implications for cognition, emotion, and motivation. Psychological Review, 98, 224–253. Marr, D. (1982). Vision. San Francisco, CA: Freeman. Matsumoto, D. (1999). Culture and self: An empirical assessment of Markus and Kita-

62 BASICS yama’s theory of independent and interdependent self-construals. Asian Journal of Social Psychology, 2, 289–310. McCloskey, M., & Kohl, D. (1983). Naïve physics: The curvilinear impetus principle and its role in interactions with moving objects. Journal of Experimental Psychology: Learning, Memory and Cognition, 9, 146–156. Medin, D. L., & Atran, S. (2004). The native mind: Biological categorization and reasoning in development and across cultures. Psychological Review, 111, 960–983. Medin, D. L., & Ortony, A. (1989). Psychological essentialism. In S. Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 179–195). New York: Cambridge University Press. Menon, T., Morris, M. W., Chiu, C.-Y., & Hong, Y.-Y. (1999). Culture and the construal of agency: Attribution to individual versus group dispositions. Journal of Personality and Social Psychology, 76, 701–717. Nisbett, R. E., Peng, K., Choi, I., & Norenzayan, A. (2001). Culture and systems of thought: Holistic versus analytic cognition. Psychological Review, 108, 291–310. Prinz, W. (1997). Perception and action planning. European Journal of Cognitive Psychology, 9, 129–154. Quine, W. V. O. (1953). From a logical point of view. Cambridge, MA: Harvard University Press. Ross, L., & Nisbett, R. E. (1991). The person and the situation. New York: McGraw-Hill. Ryle, G. (1971). The thinking of thoughts: What is “le Penseur” doing? In G. Ryle (Ed.), Collected papers (pp. 480–496). London: Hutchinson. Schlenker, B. (1974). Social psychology and science. Journal of Personality and Social Psychology, 29, 1–15. Shah, J., & Higgins, E. T. (1997). Expectancy × value effects: Regulatory focus as determinant of magnitude and direction. Journal of Personality and Social Psychology, 73, 447–458. Trafimow, D., Triandis, H. C., & Goto, S. G. (1991). Some tests of the distinction between the private self and the collective self. Journal of Personality and Social Psychology, 60, 649–655. Triandis, H. C. (1964). Cultural influences upon cognitive processes. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 1, pp. 1–48). New York: Academic Press. Triandis, H. C. (1989). The self and social behavior in differing cultural contexts. Psychological Review, 96, 506–520. Triandis, H. C. (1995). Individualism and collectivism. Boulder, CO: Westview Press. Vohs, K. D., Mead, N. L., & Goode, M. R. (2006). The psychological consequences of money. Science, 314, 1154–1156. Vohs, K. D., & Schooler, J. W. (2008). The value of believing in free will. Psychological Science, 19, 49–54. Williams, L. E., & Bargh, J. A. (2008). Experiencing physical warmth promotes interpersonal warmth. Science, 322, 606–607. Zhou, X., Vohs, K. D., & Baumeister, R. F. (2008). The symbolic power of money: Reminders of money alter social distress and physical pain. Psychological Science, 20, 700–706.

Part II Mental State Theories





4 Social-Cognitive Theories Bertram Gawronski Galen V. Bodenhausen

I

n naming our species in his biological taxonomy, Linnaeus (1758) chose Homo Sapiens, designating us as “the wise/knowing man.” Explicit in this choice is the belief that the construction of meaningful knowledge is the preeminent characteristic separating our species from its biological cousins. In Descent of Man, Darwin (1871) further underscored the unique role of subjective meaning in shaping human emotion and behavior. He described, for example, the profound feelings of revulsion and palpable physical symptoms that a devout Hindu man might instantaneously feel upon discovering that he has inadvertently eaten food that is considered “unclean” for religious reasons; however, a person with a different religious belief system might eat the very same food with great relish. Examples of this sort make it clear that human behavior is profoundly influenced by our subjective understandings of the world. The notion that social reality is mentally construed and humans act and react to it on the basis of this constructed understanding forms the core of social cognition research. In general terms, social cognition research seeks to understand the mental processes through which social meaning arises and exerts its influence on behavior. The scientific challenges inherent in studying these processes are formidable. While it may seem trivially obvious that religious beliefs, to use Darwin’s example, can indeed exert a powerful effect on behavior, carefully 65

66

MENTAL STATE THEORIES

unpacking the mental underpinnings of such effects is far from trivial. To study such matters, psychological measures must be devised. Psychometricians have studiously devoted themselves to the creation of such tools, but psychological scientists still lack universally shared, core metrics of the sort available in the physical sciences. Perhaps the closest psychologists have come to developing a universal currency for the study of meaning is Osgood, Suci, and Tannenbaum’s (1957) semantic differential approach, which empirically identified three core dimensions underlying the meaning of all concepts (i.e., evaluation, potency, and activity) and provided a standard method for measuring these dimensions. However, despite Osgood and colleagues’ ambition to provide a common conceptual and operational foundation for the psychological study of meaning, psychological researchers have made relatively little use of it (for a noteworthy exception, see the work on affect control theory; e.g., Heise, 2007). One likely reason for this neglect is the tendency for social cognition researchers to prefer to theorize in terms of general mechanistic processes through which knowledge representations influence behavior, rather than focusing on the semantic content of the representations per se (see Kashima, Chapter 3, this volume). Using Darwin’s example, social cognition researchers tend to be less interested in the specifics of what the Hindu man believes, compared to how his religious beliefs influence his actions and reactions. To understand such phenomena, social psychologists have proposed a general, yet functionally distinct, class of representations (i.e., sacred values) that exert systematic effects on information processing in a manner that generalizes across a variety of different specific beliefs (e.g., Tetlock, 2003). Typically, social-cognitive theories can be said to aim for content-independence in that they try to identify general principles that apply to all sorts of belief systems. For example, although attitude research can be described as being concerned with the psychological role of one of Osgood et al.’s (1957) core dimensions of meaning, evaluation, social-cognitive research on attitudes focuses primarily on questions that are content-independent, such as: (1) How are attitudes formed? (2) How are attitudes activated? (3) How do attitudes guide behavior? (4) How are attitudes represented? (De Houwer, Gawronski, & Barnes-Holmes, 2013).1 Yet, as we will see, there are many challenges in constructing scientifically useful theories of this kind. Fundamentally, psychological measures cannot directly assess mental processes and representations. At best, psychometrics can only tap the overt behavioral correlates of such inner mental states (De Houwer, 2011). The thorniness of this problem led the behaviorists to reject the scientific value of mental explanations in toto, but few social psychologists are similarly inclined to exclude mental explanations. In this chapter, we endeavor to spell out some of the biggest theoretical challenges facing researchers who propose mental accounts for observed patterns of behavior, and we offer some potential strategies for successfully tackling them.



Social-Cognitive Theories 67

Social Cognition as a Level of Analysis Since its inception in the 1970s, there have been recurring debates about the most appropriate way to conceptualize social cognition. Whereas some conceive of social cognition as a methodological approach to understanding social phenomena (e.g., Hamilton & Carlston, 2013), others argue that social cognition should be understood as a particular content area (e.g., Macrae & Miles, 2012). Adopting a metatheoretical view, we think that social cognition may be best characterized as a level of analysis in the study of social phenomena in that social cognition research aims at understanding social phenomena on the basis of their underlying mental processes. The metatheoretical implications of this conceptualization can be illustrated by means of Marr’s (1982) distinction of three levels of analysis: (1) the computational level, (2) the algorithmic level, and (3) the implementational level (see De Houwer & Moors, Chapter 2, this volume). According to Marr, the main goal of research at the computational level is to identify which types of inputs produce which kinds of outputs. In functional terms, the relevant inputs may include any type of environmental stimulus and the contextual conditions under which it is encountered, whereas outputs refer to overt behavioral responses that are elicited by a given stimulus. For example, a large body of research on behavioral priming can be described as computational in that it focuses on the particular behaviors that are elicited by exposure to various kinds of prime stimuli (for a review, see Bargh, 2006). Research of this kind differs from research at the algorithmic level, which is concerned with the mental mechanisms that translate inputs into outputs. This level of analysis resonates with the agenda of social-cognitive research, which aims at identifying the mental processes and representations underlying social phenomena. For example, expanding on the identification of input–output relations in studies on behavioral priming, a considerable body of research aimed at identifying the mental processes and representations that mediate the effects of prime exposure on overt behavior (e.g., Cesario, Plaks, & Higgins, 2006; Loersch & Payne, 2011; Wheeler, DeMarree, & Petty, 2007). Finally, research at the implementational level is concerned with the physical systems that implement the mechanisms identified at the algorithmic level. In social psychology, this approach is prominently reflected in the emerging field of social neuroscience, which aims at identifying the neural underpinnings of social phenomena (see Beer, Chapter 9, this volume). For example, expanding on mental process theories of prime-to-behavior effects (e.g., Cesario et al., 2006; Loersch & Payne, 2011; Wheeler et al., 2007), research at the implementational level may investigate the neural underpinnings of the mechanisms that mediate observed relations between prime stimuli and overt behavior. In terms of Marr’s (1982) framework, social-cognitive theories are located at the algorithmic level in that they are concerned with the mental processes and representations that mediate relations between socially rel-

68

MENTAL STATE THEORIES

evant inputs and outputs. Although this conceptualization may seem rather trivial, it helps to clarify the empirical phenomena social-cognitive theories aim to explain (explanandum) and the theoretical assumptions they propose to explain these phenomena (explanans). From an epistemological point of view, research at the computational level aims at explaining observed outputs by relating them to inputs that cause these outputs. Using the above example of behavioral priming, exposure to a particular stimulus may serve as an explanation for an observed behavioral response to the extent that the stimulus can be said to cause the behavioral response. In other words, the observed behavior represents the phenomenon that needs to be explained, and exposure to the prime stimulus serves as the event that is supposed to explain the behavior (causal explanation). However, stating that exposure to the prime explains the behavioral response does not say anything about how the prime caused the observed behavior. This question is central in research at the algorithmic level, in which the causal relation between prime exposure and behavior represents a phenomenon that is in need of further explanation (De Houwer, 2011). Research at the algorithmic level provides an answer to this question by identifying the mental mechanisms that mediate the link between prime exposure and overt behavior (mechanistic explanation). In this sense, social-cognitive theories offer explanations of identified input-output relations by specifying the mental mechanisms that translate inputs into outputs.2

Some Principles of Social-Cognitive Explanation A central requirement for any scientific explanation is that the explanans should be conceptually independent of the explanandum (Hempel, 1970). To illustrate this requirement, imagine that Sally is wondering why Bob is not married. Telling Sally that Bob is a bachelor does not provide a useful answer to her question, because the proposed explanans (i.e., Bob is a bachelor) has conceptual overlap with the explanandum (i.e., Bob being unmarried). To qualify as a useful explanation, any answer to Sally’s question should refer to something that is conceptually independent of the fact that Bob is not married (e.g., Bob’s personality). At the computational level, there is little confusion about this requirement because the concepts that are used to categorize inputs (i.e., stimuli) are rarely conflated with the concepts that are used to describe outputs (i.e., behavior). At the algorithmic level, however, the independence requirement is often violated when input–output relations are equated with the mental constructs that are proposed explain them. Using mental constructs to describe behavioral effects is problematic because it can lead to circular explanations and conceptual contradictions (De Houwer et al., 2013). As an example, consider research showing that a short period of distraction can lead to better decisions compared to an equally long period of delib-



Social-Cognitive Theories 69

eration (for a meta-analysis, see Strick et al., 2011). Drawing on Dijksterhuis’s (2004) original explanation, this phenomenon is often referred as the unconscious thought effect. From a meta-theoretical perspective, such descriptions are problematic because they equate the observed effect with the mental process that is supposed to explain the effect. That is, unconscious thought (the explanans) is empirically defined as the beneficial effect of distraction on decision quality (the explanandum). Such equations make explanations in terms of unconscious thought circular because the explanans involves the same concepts as the explanandum, thereby violating the requirement of conceptual independence. Descriptions of observed effects in terms of mental mechanisms can also lead to conceptual contradictions, for example, when differences between distraction and deliberation conditions are found to be driven by conscious overthinking in the deliberation condition (e.g., Payne, Samper, Bettman, & Luce, 2008). In this case, one would have to draw the paradoxical conclusion that the effects of unconscious thought are the product of conscious thought. Such theoretical pitfalls can be avoided by clearly distinguishing between the causal relations between inputs and outputs that need to be explained (e.g., effects of distraction on decision quality) and the mental constructs that are proposed to explain the identified input– output relations (e.g., unconscious thinking). Although a clear conceptual distinction between mental constructs and behavioral effects is a necessary precondition for scientifically sound explanations, it is not sufficient to prevent explanatory circularity. Another potential pitfall is the lack of a clear specification of the mental mechanisms that translate inputs into outputs. This limitation can lead to circular explanations even when there is a clear conceptual distinction between the behavioral effects that need to be explained and the mental constructs that are proposed to explain them. As an example, consider the distinction between System 1 processing and System 2 processing in prominent dualsystems theories of judgment and decision making (see Deutsch, Chapter 7, this volume). Although the distinction between the two kinds of processing subsumes several conceptually distinct dualities (e.g., Kahneman, 2003), it is sometimes boiled down to the distinction between resource-dependence and resource-independence. If a given effect is attenuated by time pressure or distracter tasks, it is explained in terms of System 2 processing. Conversely, if a given effect is unaffected (or enhanced) by time pressure or distracter tasks, it is explained in terms of System 1 processing (e.g., Dhar & Gorlin, 2013). Such explanations meet the above criterion of conceptual independence between explanans and explanandum to the extent that the to-beexplained behavioral effect is described without reference to the distinction between System 1 and System 2 processing. However, they may still be circular if there is no specification of System 1 and System 2 processing over and above the assumption about their differential resource-dependence. In the absence of such specifications, claims that a given effect is due to System 1 or System 2 processing do not provide anything beyond simple classifica-

70

MENTAL STATE THEORIES

tions of observed effects (Gawronski, Sherman, & Trope, 2014). To the extent that a given effect is resource-independent it will be categorized as being due to System 1 processing, but it will be attributed to System 2 processing if it is resource-dependent. Moreover, if an effect that was initially attributed to System 1 processing turns out to depend on cognitive resources, this effect would be recategorized as the product of System 2 processing, and vice versa. Without a clear specification of the mental operations that characterize System 1 and System 2 processing, explaining resource-independent effects in terms of System 1 processing and resource-dependent effects in terms of System 2 processing involves a circular explanatory structure. Such explanations may also be criticized as irrefutable because they do not imply any prediction that could be inconsistent with a given result (see Gawronski & Bodenhausen, Chapter 1, this volume). Thus, over and above the requirement that behavioral effects should not be described in terms of the mental constructs that are proposed to explain them (i.e., conceptual independence of explanans and explanandum), social-cognitive theories should provide clear specifications of the mental mechanisms that translate inputs into outputs to avoid the criticism that they provide circular and irrefutable explanations.

How Can We Test Social-Cognitive Theories? From a naïve point of view, one could argue that social-cognitive theories can be tested by measuring the hypothesized mental constructs and then testing whether these constructs account for the relations between inputs and outputs they are supposed to mediate. For example, drawing on the available statistical tools for testing mediation (e.g., Baron & Kenny, 1986; Preacher & Hayes, 2004), researchers may experimentally manipulate certain factors at the input level and then test whether their influence on a given output variable is statistically mediated by the measure of the hypothesized mental construct. Although this approach is rather common, it involves a number of metatheoretical problems, the most important being the fact that, as we have already noted, it is not possible to directly measure mental constructs. In a strict sense, psychological measures capture the behavioral outputs of mental processes and representations, but these outputs are conceptually distinct from their mental antecedents (De Houwer, 2011). We already discussed the difference between mental constructs and behavioral effects when we explained the requirement that explanans and explanandum have to be conceptually independent. Yet, the impossibility of measuring mental constructs goes beyond the problem of explanatory circularity in that it also involves the measurement of mental constructs that are independent of the input–output relations they are supposed to explain. The only requirement of the independence criterion is that the input–output relations that need to be explained are conceptually distinct from the mental constructs that are proposed to explain them. To the extent that the hypoth-



Social-Cognitive Theories 71

esized mental construct is measured independently of the relevant input– output relations, there would be no problem with the required independence of explanans and explanandum. Yet, the possibility of measuring mental constructs is undermined by a more fundamental problem: the absence of a bi-conditional relation between mental constructs and behavioral responses (De Houwer et al., 2013). In a strict sense, direct measurement of a mental construct by means of behavioral outputs presupposes that there is a one-toone relation between the mental construct and a particular behavior, such that variations in one unambiguously reflect variations in the other (i.e., if p, then q and, at the same time, if q, then p). Claims of such bi-conditional relations seem untenable because (1) there can always be conditions under which the mental construct does not produce the relevant behavior and (2) the same behavior may be produced by another mental construct. For example, self-reported evaluations do not unambiguously reflect mental attitudes because the impact of mental attitudes on evaluative judgments can sometimes be disrupted (e.g., when people are motivated to conceal their attitudes; see Fazio, 2007) and evaluative judgments can be influenced by various factors other than mental attitudes (e.g., incidental mood states; see Schwarz, 1990). Similar concerns apply to the use of less reactive measures. For example, evaluative priming tasks will provide unambiguous indices of mental attitudes only if variations in mental attitudes are both necessary and sufficient to produce variations in priming effects. Yet, evaluative priming effects are influenced by various factors other than mental attitudes (e.g., processes involved in response interference; see Gawronski, Deutsch, LeBel, & Peters, 2008) and the impact of mental attitudes on evaluative priming can be reduced under certain conditions (e.g., through strategic counteraction; see Teige-Mocigemba & Klauer, 2013). To be sure, both measures can be interpreted as capturing evaluative responses in the sense of behavioral outputs. However, in the absence of a bi-conditional relation between behavioral responses and a particular mental construct, it is not possible to treat these responses as a measure of this construct. Moreover, even if there were a uni-conditional relation such that variations in a mental construct always produce variations in a given behavior, inferring the mental construct on the basis of the behavior would involve the fallacy of affirming the consequent in that the conditional “if p, then q” is used to draw the logically invalid inference “if q, then p.” Such inferences are problematic, because they presuppose that the relevant behavior can never be produced by an alternative mechanism (see Gawronski & Bodenhausen, Chapter 1, this volume). If we can measure only the outcome of mental constructs, but not mental constructs per se, how is it possible to test social-cognitive theories? Does this mean that social-cognitive theories are unfalsifiable? To answer this question, it is useful to consider our earlier discussion of Marr’s (1982) levels of analysis and the mutual relations between the computational and the algorithmic levels. Specifically, we argued that social-cognitive theories at

72

MENTAL STATE THEORIES

the algorithmic level provide mechanistic explanations of causal relations between inputs and output at the computational level. From this perspective, social-cognitive theories can be tested by deriving predictions about input–output relations from their assumptions about mental processes and representations. To the extent that these predictions are empirically confirmed, researchers can treat this evidence as (preliminary) support for their theories. In this case, it makes sense to believe that the processing and representation of social information are characterized by the principles stated by the theory. However, if the predictions about input–output relations are disconfirmed, the conflict between prediction and data suggests that at least one of the assumptions that has been used to derive the prediction must be false (see Gawronski & Bodenhausen, Chapter 1, this volume). In this case, researchers are faced with the challenging task of identifying which component of this broader set of assumptions led to the conflict between prediction and data. Although regression-based approaches to testing mediation have been criticized for a variety of reasons (e.g., Jacoby & Sassenberg, 2011; Spencer, Zanna, & Fong, 2005), the current conceptualization also clarifies why experimental approaches are better suited to test hypotheses about mental processes and representations. Because psychological measures capture only the behavioral output of mental constructs, it is not feasible to measure mental mediators for regression-based mediation analyses (e.g., Baron & Kenny, 1986; Preacher & Hayes, 2004). Instead, theoretical assumptions about mental processes and representations have to be tested by deriving predictions about causal relations between inputs and outputs. In many cases, these predictions also include hypotheses about the boundary conditions of a given behavioral effect.3 Thus, theoretical hypotheses about mental mediation at the algorithmic level can be tested by deriving predictions about contextual moderation at the computational level (Jacoby & Sassenberg, 2011). Although this conceptualization may seem to blur the distinction between mediation and moderation, the two concepts retain their original meaning in that they refer to different levels of analysis. Whereas the term mediation refers to the mental mechanisms that mediate input–output relations (algorithmic level), the term moderation refers to contextual factors that moderate input–output relations (computational level). Another valuable insight that can be gained from this conceptualization is the mutually supportive relation between the computational and the algorithmic level of analysis (see De Houwer, 2011). On the one hand, computational research supports algorithmic research in that causal relations between inputs and outputs at the computational level provide the empirical phenomena that algorithmic research aims to explain. On the other hand, algorithmic research supports computational research to the extent that algorithmic theorizing about mental processes and representations can lead to new discoveries of previously undetected input–output relations and their boundary conditions at the computational level. Thus, although the



Social-Cognitive Theories 73

two levels of analysis are distinct in the sense that they differ in terms of their explanandum (i.e., behavioral outputs vs. input-output relations) and in terms of their explanans (i.e., environmental inputs that cause outputs vs. mental processes and representations that mediate input–output relations), their relation is mutually supportive in that progress at one level advances research at the other level (and vice versa).

Explanans and Explanandum of Social-Cognitive Theories In our discussion of Marr’s (1982) levels of analysis, we argued that socialcognitive theories are located at the algorithmic level in that they aim to identify the mental mechanisms underlying social phenomena. From this perspective, the explanada of social-cognitive theories are input–output relations that can be described as “social” in some sense. For example, input– output relations may be described as socially relevant if they involve either social stimuli as inputs (e.g., effects of target characteristics on judgments in research on person perception) or social behavior as outputs (e.g., effects of nonsocial stimuli on social behavior in research on behavioral priming). Although both lines of research have made valuable contributions to our understanding of social phenomena, some critics have raised concerns that social cognition researchers tend to focus primarily on the social nature of inputs, while ignoring the relevance of their nonsocial outputs (e.g., ratings, response times) for understanding social behavior in real-world settings (e.g., Baumeister, Vohs, & Funder, 2007; Macrae & Miles, 2012). Whether or not this criticism is justified is a matter of debate. Nevertheless, it is worth noting that research on the effects of social inputs on nonsocial outputs typically has a stronger impact on other fields when corresponding effects are demonstrated for social outputs. Although social-cognitive theories differ in terms of whether they focus on social inputs, social outputs, or both, a shared characteristic is their concern with the mental processes and representations underlying the phenomena of interest. In general, social-cognitive theories aim to provide answers to at least one of four questions: (1) How are mental representations formed? (2) How are mental representations activated? (3) How do activated mental representations guide behavior? (4) How is social information represented? Whereas the first three questions are concerned with the characteristics of mental processes, the fourth question is concerned with the nature of mental structures. Unfortunately, testing competing explanations in terms of process versus structure can be extremely difficult (Wyer, 2007). In many cases, a finding predicted by a process account may be reinterpreted in terms of competing representational accounts, and vice versa. An illustrative example is the debate about whether dissociations between implicit and explicit measures

74

MENTAL STATE THEORIES

reflect the operation of distinct mental processes (e.g., Fazio, 2007; Gawronski & Bodenhausen, 2006) or distinct mental representations (e.g., Rydell & McConnell, 2006; Wilson, Lindsey, & Schooler, 2000). Because it is not possible to directly measure either of the two, the informational value of socialcognitive theories depends on precise assumptions about how the proposed mental constructs are related to environmental inputs and behavioral outputs. For example, to provide a valuable account of social phenomena, theories focusing on the nature of mental structures should also specify how the proposed representations are activated and how activated representations guide behavior. In the absence of such assumptions, representation theories become susceptible to criticism of irrefutability because they may be able to explain any possible finding in a post-hoc fashion. Although such theories are characterized by a high level of generality in the sense that they can explain a wide range of empirical results, their predictive value is typically quite low in that it is difficult to identify which input–output relations can be expected on the basis of theory in an a priori fashion (see Gawronski & Bodenhausen, Chapter 1, this volume). Thus, to provide a valuable account of social phenomena, social-cognitive theories should address more than just one of the four questions, and ideally provide answers to all of them.

Scope and Refutability of Theoretical Claims In addition to addressing different subsets of the four central questions, social-cognitive theories vary greatly in their scope. As we have noted in our introductory chapter (Gawronski & Bodenhausen, Chapter 1, this volume), broader theories that have widespread application are generally considered more valuable than narrow ones that account only for a relatively limited range of phenomena. At the same time, broad theories can often be so general and encompassing that they essentially become irrefutable (Quine & Ullian, 1978). The dynamics of this trade-off are quite evident in social cognition research. Some explanatory accounts focus on a particular phenomenon, invoking a delimited subset of mental constructs that have direct relevance within the given domain. For example, Jones and Davis (1965) developed a theory of correspondent inference that was concerned with a very specific question: Under what conditions do social perceivers draw dispositional inferences about actors on the basis of the actors’ behavior? It is an important question, with significant applications, yet it is also a relatively narrow question. It provided a basis for clear, specific, falsifiable predictions, and indeed, a central assumption of the theory was empirically disconfirmed when it was discovered that the presence of strong situational constraints failed to adequately attenuate dispositional inferences, as the theory asserted it should (Jones & Harris, 1967). As a result, new theories of dispositional inference emerged to supersede correspondent inference theory (e.g., Gilbert, 1989; Trope, 1986). Although disconfirmed in a noteworthy respect, the



Social-Cognitive Theories 75

theory was ultimately quite valuable in generating explanatory progress, stimulated in large part by the interest value of its failed prediction and the previously unknown phenomenon it revealed: the correspondence bias (see Jones, 1990). This example illustrates that, counter to the common preference for theories that are consistent with larger sets of potential outcomes, theories can advance science even if their predictions have been empirically disconfirmed by stimulating novel research to understand and explain the unexpected discovery (Lakatos, 1970). Despite the scientific value of narrow theories that focus on particular phenomena, such theories involve a considerable risk of conceptual and theoretical fragmentation. This concern is prominently reflected in the desire for ambitious theories that attempt to provide very general accounts that are applicable in all domains of social cognition. One example is the theory of reasoned action (Fishbein & Ajzen, 2010), which adapts the notions of subjective expected utility and normative beliefs about social pressure into a relatively small set of assumptions about the factors guiding intention and action (see Trafimow, Chapter 12, this volume). The theory can be, and indeed has been, applied to a very wide variety of substantive topics. Another example is Anderson’s theory of information integration, described most recently in his book Unified Social Cognition (2008), which consists of a very small set of laws said to govern the use of information in forming judgments, choices, and intentions. In the book, Anderson argues forcefully for the necessity of developing broadly integrative theory, and he laments the state of fragmentation that results from focusing on narrow, domain-specific theories whose interest value, in his view, inevitably rises and falls faddishly. Both the theory of reasoned action and information integration theory are concerned with the process whereby individuals use multiple informational inputs to determine a response of some kind. They are similar in both proposing a valuation process whereby discrete pieces of relevant input are translated into a common subjective evaluative metric, but they differ in their assumptions about the process governing the combination of these evaluations into an overall response. Whereas Fishbein and Ajzen’s theory implies that informational inputs are translated into evaluative outputs in an additive fashion, Anderson’s theory implies a weighted averaging relation. This divergence immediately suggests the possibility of a competitive test, and many such tests have indeed been conducted. For example, Anderson (1965) showed that in forming social impressions, adding two pieces of moderately positive information to two pieces of very positive information about an actor resulted in a less positive impression, compared to when just the two very positive pieces of information were provided. Such an outcome is consistent with an averaging, but not with an adding, integration function. However, other findings have contradicted the averaging mechanism (e.g., Yamagishi & Hill, 1981), and it has been shown in fact that additive and averaging functions cannot easily be distinguished on the basis of the kinds of empirical tests that have been applied toward that end (Hodges,

76

MENTAL STATE THEORIES

1973). The problem, in a nutshell, is that somewhat more complex versions of each type of integration function can easily be generated to account for any given pattern of input–output relation. In other words, the competing assumptions of the two theories are essentially irrefutable because the hypothesized information integration functions are empirically ambiguous. Without the constraints of additional predictions about boundary conditions of functional relations, it would always be possible to propose different variants of additive or averaging functions that are consistent with a given set of unpredicted findings. However, such predictions are beyond the scope of the two theories because the hypothesized integration functions are assumed to underlie all input–output relations under any condition. The prospects for refuting unconditional claims about universal mental principles is cast further in doubt when considering the typical responses inspired by the discovery of disconfirming evidence. In the case of rational actor-type theories, one response to disconfirming evidence, common for applications of expected-utility theory within economics, is to define theoretically aberrant behavior as “irrational” and thus beyond the scope of a theory of rational choice (e.g., Thaler, 1990). An alternative to dismissing inconsistent evidence as irrelevant is the possibility of revising one’s assumptions about the contents of the hypothesized mediators. For example, if people behave in a manner that seems counterintuitive from the perspective of expected-utility theory, theoretical claims about the representations of value and probability are often revised to make the observed outcome consistent with basic tenets of the theory. However, such strategies make theoretical explanations in terms of expected utility circular, in that expected utility (the explanans) is merely inferred from the behavior that needs to be explained (the explanandum). In principle, a theorist could consistently resist the rejection of a favored theory by continuously revising or adding auxiliary assumptions in order to account for each new, unanticipated finding (Lakatos, 1970). Such theoretical fine-tuning can be justified only when new empirical implications can be derived from the modifications and subjected to potential falsification; otherwise, theorists are simply indulging in the patchwork quilt fallacy (Giere, 2005). The important point is that ad-hoc modifications cannot be justified merely on the basis of the original evidence that compelled them, but only on the basis of the new predictions they offer for potential falsification in novel tests. The strategy of multiple, successive ad-hoc modifications can provide an indefinite stay of execution for a cherished theory, pointing again to the problem of irrefutability. Only as long as such modifications generate novel insights, supported by new empirical tests, can the research program be considered progressive in the Lakatosian sense (see Gawronski & Bodenhausen, Chapter 1, this volume). It is very easy to sympathize with Anderson’s (2008) assertions about the desirability of broadly integrative theory in light of the undeniably fragmented state of knowledge in social psychology. Yet the refutability dilem-



Social-Cognitive Theories 77

mas associated with broad, general-purpose theories appear to be substantial. An alternative strategy for striving toward greater theoretical integration lies in an approach in which theories are initially developed more modestly, within a particular domain, and then their applicability in other domains is explored. The heuristic-systematic model (Chaiken, 1980), for example, was originally proposed as a theoretical account for understanding variations in the impact of persuasive messages, but in subsequent work, its implications in a diverse range of other domains were explored and a broader model was proposed (see Chen & Chaiken, 1999).

Relation to Other Theoretical Approaches Resonating with the quest for explanatory breadth, social cognition is often regarded as a general approach that is applicable to any content domain within social psychology (e.g., Hamilton & Carlston, 2013). Because several other approaches share this feature, an important question is how socialcognitive theories are related to other types of domain-independent theories. Does social cognition compete with other overarching approaches, or do they complement each other by accounting for different aspects of social phenomena? In our discussion of Marr’s (1982) levels of analysis, we already nodded to social neuroscience, which aims at identifying the neural underpinnings of social phenomena (see Beer, Chapter 9, this volume). In terms of Marr’s framework, social neuroscience can be located at the implementational level in that it is concerned with the physical systems that implement the mechanisms identified at the algorithmic level. From a metatheoretical perspective, the relation between social cognition and social neuroscience can be described in two ways. First, social neuroscience may be conceptualized as being concerned with the neural substrates of the mental processes and representations identified by social cognition. An illustrative example of this approach is research on brain mapping, which aims at identifying the brain regions that implement specific mental operations. Second, neural responses may be regarded as a particular kind of output next to overt behavior. This conceptualization resonates with the idea that well-understood neural responses may serve as alternative measures to test hypotheses about the mental mechanisms underlying social phenomena. An important difference between the two approaches is how data at one level constrain theoretical interpretations at the other level. Whereas in the first case behavioral data and their algorithmic interpretation constrain theoretical interpretations at the neural level, in the second case relations between inputs and neural outputs constrain algorithmic theories like any other behavioral outputs. However, a major issue in designing studies of the second kind is the problem of reverse inference, which can arise when a neural output is treated as a measure of a particular mental construct (Poldrack, 2006). This issue is structur-

78

MENTAL STATE THEORIES

ally equivalent to the concern about using behavioral responses as measures of mental constructs, such that it involves the logically invalid inference “if q, then p” from the conditional “if p, then q” (see Gawronski & Bodenhausen, Chapter 1, this volume). This problem does not imply that neural data cannot be used to test social-cognitive theories. Yet, studies of this kind require particular prudence in experimental design and theoretical interpretation to avoid the fallacy of affirming the consequent (see Beer, Chapter 9, this volume). Two other approaches that have a close relationship with social-cognitive theories are emotion theories (Manstead & Parkinson, Chapter 5, this volume) and motivation theories (Dunning, Chapter 6, this volume). Although social cognition has often been criticized for ignoring the roles of emotion and motivation, the relation between research on “hot” and “cold” processes has become much less contentious than it used to be a few decades ago (Schwarz, 2000). In fact, despite the emphasis on cognition in the term social cognition, many recent theories aim to integrate the unique and interactive roles of affect, cognition, and motivation (e.g., Gawronski & Cesario, 2013; Higgins, 1997; Strack & Deutsch, 2004). In recognition of this development, we generally avoided references to cognition in the current chapter, and instead talked about mental processes and representations, terms that were intended to subsume affective, cognitive, and motivational components. Although theories of emotion and motivation have to deal with some metatheoretical issues that are unique to these domains (see Dunning, Chapter 6, this volume; Manstead & Parkinson, Chapter 5, this volume), we would argue that the issues discussed in the current chapter are relevant regardless of whether the postulated mental constructs are affective, cognitive, or motivational. The same applies to theories that explain social behavior in terms of personality systems, to the extent that these theories characterize individual differences in terms of their affective, cognitive, and motivational underpinnings (Cervone, Caldwell, & Mayer, Chapter 8, this volume). Another approach that is often regarded as overarching in the sense that it aims at identifying “ultimate causes” of social behavior is evolutionary theory (see Ketelaar, Chapter 11, this volume). Expanding on Marr’s (1982) framework, one could argue that evolutionary theories constitute a fourth level of analysis that aims at explaining the historical processes that shaped the physical systems that implement the mental mechanisms identified at the algorithmic level (Conway & Schaller, 2002). However, when it comes to historical antecedents, evolutionary accounts often compete with cultural theories (Eom & Kim, Chapter 16, this volume), the most prominent example being theories that are concerned with historical changes in social structures (e.g., Wood & Eagly, 2012). Although the two approaches are quite different in terms of their explanans, they share the goal of identifying the historical antecedents of mental processes and representations. Yet, whereas evolutionary accounts attribute their historical retention to evolved psychological mechanisms and genes (see Johnson & Penke, Chapter 10, this volume),



Social-Cognitive Theories 79

cultural approaches tend to locate historical retention at the level of social groups and societies (Caporael, 2001). A final category of theories that deserves closer attention in a chapter on social-cognitive theories are formal theories, in particular computer simulations (Fiedler & Kutzner, Chapter 17, this volume) and mathematical models (Klauer, Chapter 18, this volume). Formal theories aim to simulate or quantify the role of multiple processes in the mediation of inputs and outputs. The significance of this agenda is reflected in the principle of equifinality, which refers to cases in which different combinations of processes can produce the same behavioral outcome. For example, in research on selfregulation two people may show the same behavioral response when (1) the initial impulse and inhibitory control are weak or (2) the initial impulse and inhibitory control are strong (Sherman et al., 2008). This question plays a central role in many dual-process theories, which seek to explain social phenomena in terms of the interplay of distinct automatic and controlled processes (Deutsch, Chapter 7, this volume). Formalized theories are able to capture such complex interplays by providing computer simulations and quantitative estimates of the hypothesized processes. As such, computer simulations and mathematical modeling procedures provide valuable tools for social-cognitive theorists in specifying and testing their theories about the mental mechanisms underlying input–output relations.

Summary The main goal of this chapter was to review the metatheoretical foundation and explanatory structure of social-cognitive theories. Drawing on Marr’s (1982) conceptual framework, we argued that social cognition can be described as a level of analysis, namely, the algorithmic concern with the mental processes and representations underlying social phenomena. In this sense, social-cognitive theories can be said to provide mechanistic explanations of causal relations between environmental inputs and behavioral outputs. On the basis of this conceptualization, we identified several metatheoretical criteria for evaluating social-cognitive theories. Specifically, we argued that the explanatory and predictive value of social-cognitive theories depends on: (1) a clear conceptual distinction between the input–output relations that need to be explained and the mental constructs that are proposed to explain them; (2) a clear specification of the hypothesized mental constructs; and (3) precise assumptions about how the proposed mental constructs are related to environmental inputs and behavioral outputs. To provide valuable accounts of social phenomena, we argued that social-cognitive theories should address four central questions: (1) How are mental representations formed? (2) How are mental representations activated? (3) How do activated mental representations guide behavior? (4) How is social information represented? To the extent that social-cognitive theories include precise and refutable answers

80

MENTAL STATE THEORIES

to these questions, they can provide invaluable insights through their ability to explain and predict causal relations between environmental inputs and behavioral outputs as well as their boundary conditions. Such insights are important not only for basic research on the mental underpinnings of social phenomena, but also for applications to real-world problems that aim at changing social behavior and improving social relationships. Notes 1. Some social-cognitive theories focus explicitly on the semantic content of social beliefs (e.g., Cuddy, Fiske, & Glick, 2008), but even in these cases, the emphasis is typically on general information processing mechanisms that act on broad categories of meaning, which can be instantiated in terms of any of a number of more specific beliefs. 2. To avoid potential confusion, it is worth noting that the distinction between causal and mechanistic explanation goes beyond Kashima’s (Chapter 3, this volume) conceptualization of causal and meaning-based explanation in that both causal and mechanistic explanation are subsumed under the term causal explanation in Kashima’s framework. 3. In some cases, predictions about boundary conditions derived from algorithmic theories also explain why a behavioral effect may be difficult to replicate at the computational level. An illustrative example is the controversy about Bargh, Chen, and Burrows’s (1996) finding that participants walked slower down the hall when they were primed with the stereotype of the elderly (see Doyen, Klein, Pichon, & Cleeremans, 2012). Drawing on an algorithmic theory of the mental mechanisms underlying behavioral priming effects, Cesario et al. (2006) argued that priming effects result from perceivers preparing themselves to interact with primed social group members. An empirically confirmed prediction of their account is that participants walk slower after “elderly” priming when they hold positive evaluations of the elderly, but they walk faster after “elderly” priming when they hold negative evaluations of the elderly (and vice versa for “youth” priming). To the extent that evaluations of the elderly are distributed evenly around a neutral value, the basic priming effect will seem impossible to replicate when its underlying mental mechanism is not taken into account.

References Anderson, N. H. (1965). Averaging versus adding as a stimulus-combination rule in impression formation. Journal of Experimental Psychology, 70, 394–400. Anderson, N. H. (2008). Unified social cognition. New York: Psychology Press. Bargh, J. A. (2006). What have we been priming all these years? On the development, mechanisms, and ecology of nonconscious social behavior. European Journal of Social Psychology, 36, 147–168. Bargh, J. A., Chen, M., & Burrows, L. (1996). Automaticity of social behavior: Direct effects of trait construct and stereotype activation on action. Journal of Personality and Social Psychology, 71, 230–244.



Social-Cognitive Theories 81

Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173–1182. Baumeister, R. F., Vohs, K. D., & Funder, D. C. (2007). Psychology as the science of selfreports and finger movements: Whatever happened to actual behavior? Perspectives on Psychological Science, 2, 396–403. Caporael, L. R. (2001). Evolutionary psychology: Toward a unifying theory and a hybrid science. Annual Review of Psychology, 52, 607–628. Cesario, J., Plaks, J. E., & Higgins, E. T. (2006). Automatic social behavior as motivated preparation to interact. Journal of Personality and Social Psychology, 90, 893–910. Chaiken, S. (1980). Heuristic versus systematic information processing and the use of source versus message cues in persuasion. Journal of Personality and Social Psychology, 39, 752–766. Chen, S., & Chaiken, S. (1999). The heuristic-systematic model in its broader context. In S. Chaiken & Y. Trope (Eds.), Dual-process theories in social psychology (pp. 73–96). New York: Guilford Press. Conway, L. G., & Schaller, M. (2002). On the verifiability of evolutionary psychological theories: An analysis of the psychology of scientific persuasion. Personality and Social Psychology Review, 6, 152–166. Cuddy, A. J. C., Fiske, S. T., & Glick, P. (2008). Warmth and competence as universal dimensions of social perception: The stereotype content model and the BIAS map. Advances in Experimental Social Psychology, 40, 61–149. Darwin, C. R. (1871). The descent of man, and selection in relation to sex. London: John Murray. De Houwer, J. (2011). Why the cognitive approach in psychology would profit from a functional approach and vice versa. Perspectives on Psychological Science, 6, 202–209. De Houwer, J., Gawronski, B., & Barnes-Holmes, D. (2013). A functional-cognitive framework for attitude research. European Review of Social Psychology, 24, 252–287. Dhar, R., & Gorlin, M. (2013). A dual-system framework to understand preference construction processes in choice. Journal of Consumer Psychology, 23, 528–542. Dijksterhuis, A. (2004). Think different: The merits of unconscious thought in preference development and decision making. Journal of Personality and Social Psychology, 87, 586–598. Doyen, S., Klein, O., Pichon, C., & Cleeremans, A. (2012). Behavioral priming: It is all in the brain, but whose brain? PLoS One, 7(1), e29081. Fazio, R. H. (2007). Attitudes as object-evaluation associations of varying strength. Social Cognition, 25, 603–637. Fishbein, M., & Ajzen, I. (2010). Predicting and changing behavior: The reasoned action approach. New York: Psychology Press. Gawronski, B. (2013). What should we expect from a dual-process theory of preference construction in choice? Journal of Consumer Psychology, 23, 556–560. Gawronski, B., & Bodenhausen, G. V. (2006). Associative and propositional processes in evaluation: An integrative review of implicit and explicit attitude change. Psychological Bulletin, 132, 692–731. Gawronski, B., & Cesario, J. (2013). Of mice and men: What animal research can tell us about context effects on automatic responses in humans. Personality and Social Psychology Review, 17, 187–215. Gawronski, B., Deutsch, R., LeBel, E. P., & Peters, K. R. (2008). Response interference as a mechanism underlying implicit measures: Some traps and gaps in the assessment of mental associations with experimental paradigms. European Journal of Psychological Assessment, 24, 218–225. Gawronski, B., Sherman, J. W., & Trope, Y. (2014). Two of what? A conceptual analysis of

82

MENTAL STATE THEORIES

dual-process theories. In J. W. Sherman, B. Gawronski, & Y. Trope (Eds.), Dual-process theories of the social mind (pp. 3–19). New York: Guilford Press. Giere, R. N. (2005). Understanding scientific reasoning. Stamford, CT: Cengage. Gilbert, D. T. (1989). Thinking lightly about others: Automatic components of the social inference process. In J. S. Uleman & J. A. Bargh (Eds.), Unintended thought (pp. 189– 211). New York: Guilford Press. Hamilton, D. L., & Carlston, D. E. (2013). The emergence of social cognition. In D. E. Carlston (Ed.), The Oxford handbook of social cognition (pp. 16–32). New York: Oxford University Press. Heise, D. R. (2007). Expressive order: Confirming sentiments in social actions. New York: Springer. Hempel, C. G. (1970). Aspects of scientific explanation and other essays in the philosophy of science. New York: Free Press. Higgins, E. T. (1997). Beyond pleasure and pain. American Psychologist, 52, 1280–1300. Hodges, B. H. (1973). Adding and averaging models for information integration. Psychological Review, 80, 80–84. Jacoby, J., & Sassenberg, K. (2011). Interactions do not only tell us when, but can also tell us how: Testing process hypotheses by interaction. European Journal of Social Psychology, 41, 180–190. Jones, E. E. (1990). Interpersonal perception. New York: Freeman. Jones, E. E., & Davis, K. E. (1965). From acts to dispositions. In L. Berkowitz (Ed.), Advances in experimental social psychology (pp. 219–266). New York: Academic Press. Jones, E. E., & Harris, V. A. (1967). The attribution of attitudes. Journal of Experimental Social Psychology, 3, 1–24. Kahneman, D. (2003). A perspective on judgment and choice: Mapping bounded rationality. American Psychologist, 58, 697–720. Lakatos, I. (1970). Falsification and the methodology of scientific research programmes. In I. Lakatos & A. Musgrave (Eds.), Criticism and the growth of knowledge (pp. 91–195). Cambridge, UK: Cambridge University Press. Linnaeus, C. (1758). Systema naturae per regna tria naturae, secundum classes, ordines, genera, species, cum characteribus, differentiis, synonymis, locis (10th ed.). Holmiae: Salvius. Loersch, C., & Payne, B. K. (2011). The situated inference model: An integrative account of the effects of primes on perception, behavior, and motivation. Perspectives on Psychological Science, 6, 234–252. Macrae, C. N., & Miles, L. K. (2012). Revisiting the sovereignty of social cognition: Finally some action. In S. T. Fiske & C. N. Macrae (Eds.), The SAGE handbook of social cognition (pp. 1–11). Los Angeles: Sage. Marr, D. (1982). Vision: A computational investigation into the human representation and processing of visual information. New York: Freeman. Osgood, C. E., Suci, G. J., & Tannenbaum, P. (1957). The measurement of meaning. Urbana: University of Illinois Press. Payne, J. W., Samper, A., Bettman, J. R., & Luce, M. F. (2008). Boundary conditions on unconscious thought in decision making. Psychological Science, 19, 1118–1123. Poldrack, R. A. (2006). Can cognitive processes be inferred from neuroimaging data? Trends in Cognitive Sciences, 10, 59–63. Preacher, K. J., & Hayes, A. F. (2004). SPSS and SAS procedures for estimating indirect effects in simple mediation models. Behavior Research Methods, Instruments, and Computers, 36, 717–731. Quine, W. V. O., & Ullian, J. S. (1978). The web of belief (2nd ed.). New York: McGraw-Hill. Rydell, R. J., & McConnell, A. R. (2006). Understanding implicit and explicit attitude



Social-Cognitive Theories 83

change: A systems of reasoning analysis. Journal of Personality and Social Psychology, 91, 995–1008. Schwarz, N. (1990). Feelings as information: Informational functions of affective states. In E. T. Higgins & R. M. Sorrentino (Eds.), Handbook of motivation and cognition: Foundations of social behavior (Vol. 2, pp. 527–561). New York: Guilford Press. Schwarz, N. (2000). Social judgment and attitudes: Warmer, more social, and less conscious. European Journal of Social Psychology, 30, 149–176. Sherman, J. W., Gawronski, B., Gonsalkorale, K., Hugenberg, K., Allen, T. J., & Groom, C. J. (2008). The self-regulation of automatic associations and behavioral impulses. Psychological Review, 115, 314–335. Spencer, S. J., Zanna, M. P., & Fong, G. T. (2005). Establishing a causal chain: Why experiments are often more effective than mediational analyses in examining psychological processes. Journal of Personality and Social Psychology, 89, 845–851. Strack, F., & Deutsch, R. (2004). Reflective and impulsive determinants of social behavior. Personality and Social Psychology Review, 8, 220–247. Strick, M., Dijksterhuis, A., Bos, M. W., Sjoerdsma, A., van Baaren, R. B., & Nordgren, L. F. (2011). A meta-analysis on unconscious thought effects. Social Cognition, 29, 738–762. Teige-Mocigemba, S., & Klauer, K. C. (2013). On the controllability of evaluative-priming effects: Some limits that are none. Cognition and Emotion, 27, 632–657. Tetlock, P. E. (2003). Thinking the unthinkable: Sacred values and taboo cognitions. Trends in Cognitive Sciences, 7, 320–324. Thaler, R. H. (1990). Anomalies: Saving, fungibility, and mental accounts. Journal of Economic Perspectives, 4, 193–205. Trope, Y. (1986). Identification and inferential processes in dispositional attribution. Psychological Review, 93, 239–257. Wheeler, S. C., DeMarree, K. G., & Petty, R. E. (2007). Understanding the role of the self in prime-to-behavior effects: The active-self account. Personality and Social Psychology Review, 11, 234–261. Wilson, T. D., Lindsey, S., & Schooler, T. Y. (2000). A model of dual attitudes. Psychological Review, 107, 101–126. Wood W., & Eagly, A. H. (2012). Biosocial construction of sex differences and similarities in human behavior. Advances in Experimental Social Psychology, 46, 55–123. Wyer, R. S., Jr. (2007). Principles of mental representation. In A. W. Kruglanski & E. T. Higgins (Eds.), Social psychology: Handbook of basic principles (2nd ed., pp. 285–307). New York: Guilford Press. Yamagishi, T., & Hill, C. T. (1981). Adding versus averaging models revisited: A test of a path-analytic integration model. Journal of Personality and Social Psychology, 41, 13–25.

5 Emotion Theories Antony S. R. Manstead Brian Parkinson

O

ur goal in this chapter is to analyze some issues confronting theorists seeking to explain the phenomenon of emotion. We start by examining a potential obstacle, namely, the lack of a theory-independent way of defining emotion. The result is that theorists who favor biologically based explanations tend to define emotion (either explicitly or implicitly) in ways that are compatible with that explanation; the same applies to those who favor more cognitive explanations or more social constructionist explanations. For this reason, theoretical “debates” are rarely debates about the same underlying phenomenon. We examine some of the key disagreements between emotion theorists about how the explanandum should be defined: whether it is adequately specified by prescientific common sense; whether it is fully constituted by subjective experience; whether it is reflected in coherence or coordination of response components; whether it is a state or a process; whether it is a passion or an action; and whether it should be represented in discrete categories or continuous dimensions. We go on to consider some of the different levels of analysis from which researchers have approached the investigation of emotion, including structural versus functional, computational versus implementational, individual versus social, and empirical versus conceptual. In the final two sections of 84

Emotion Theories 85

the chapter, we examine logical and inferential problems in emotion theory, including fallacies in interpretation of data and the nonfalsifiability of certain theoretical claims, before drawing some conclusions about what would be desirable in a good theory of emotion and what obstacles stand in the way of developing such a theory.

Characterizing Emotion In everyday English, the word emotion is used in a variety of senses, and its descriptive usages accommodate a wide range of phenomena. Looking for corresponding terms in other languages potentially increases this variety. How, then, can psychologists get a fix on the essential core meaning of emotion in order to coordinate their research activity on a common object? Should they even try? Even blind men clutching at different parts of an elephant can potentially combine their different impressions of smaller parts into a more general conception of the bigger object in the room. Psychological researchers face a less tractable problem because the “emotional” phenomena they address do not necessarily cohere into a unitary entity in the first place. Rather than addressing these potential complications and uncertainties, most theorists take it for granted that there are definable emotions awaiting investigation, even when there are no obvious unifying principles to their operational or implicit definitions. Theories often need to balance generality against precision (see Gawronski & Bodenhausen, Chapter 1, this volume). On one hand, the aim is to account for as wide a range of phenomena as possible. On the other hand, general-purpose theories tend to require additional principles before they can accommodate particularities of subsets of these phenomena. Emotion theories consequently vary in their level of abstraction and inclusiveness. Those that are more specific and less inclusive may overlap in the domains to which they apply, but rarely perfectly. Rather than acknowledging that their applicability is delimited to a subset of emotional phenomena, however, a common tendency is to adopt a formulation of “emotion” that corresponds to the theory’s range of application. For example, some theories attempting to explain emotion in terms of appraisal (a cognitive-evaluative process that detects the personal significance of events) have denied that apparently appraisal-independent emotions are emotions at all (Lazarus, 1982; also see the section on “Proximal Causes” below). Such strategies are difficult to undermine because of the wide disagreement about how the emotion domain should be demarcated (e.g., Leventhal & Scherer, 1987). However, they raise issues relating to tautology and circularity: Emotions are formulated as the events that a theory is equipped to explain, and the theory is defended on the basis of its explanation of those same events. Further, alternative theories may be rejected on the basis that they are explaining events

86

MENTAL STATE THEORIES

outside the relevant domain. This is obviously detrimental to the advancement of an integrated social psychological science of emotion. To break out of these circles, it is important either to develop clear, theoryindependent criteria for demarcating the emotion domain or to delimit the intended domain of application more explicitly. The following subsections therefore address this issue of what counts as an “emotion,” and consider whether such criteria can be consensually applied. If they cannot, then theories need retargeting.

Common Sense Like many other psychological concepts, emotion has a prior meaning in ordinary language. However, explicit specification of that meaning presents challenges, partly because of the flexible and context-dependent usages of the word (e.g., Edwards, 1999). Indeed, the diversity of phenomena apparently encompassed within the common-sense category of “emotion” can be bewildering. How psychologists cope with this diversity partly depends on their level of skepticism about the validity and applicability of ordinarylanguage concepts. Some theorists (e.g., Oatley, 1992) believe that everyday emotion categories provide a generally accurate characterization of psychological reality that forms a suitable basis for empirical investigation. In this case, careful conceptual analysis of everyday uses of emotion-related vocabulary provides one possible route to demarcation of the emotion domain. Others (e.g., Duffy, 1941; Russell, 2003) suggest that the usefulness of “emotion” as a psychological concept remains to be determined, or even that emotion talk is not an activity that we should attempt to calibrate with putative languageindependent emotional phenomena in the first place (Edwards, 1999). Our intermediate position is that “emotion” as conceptualized in ordinary language represents a place-holder (or “chapter heading,” Bentley, 1928; and see Mandler, 1975), requiring tighter specification for scientific purposes. This tighter specification might be achieved both by means of conceptual analysis of “emotional” language and by empirical investigation of the phenomena relating to emotion concepts (Scarantino & Griffiths, 2011). The hope is that over time these two approaches will interact to produce a convergent and coherent specification of “emotion.” However, this specification is unlikely to match all features of contemporary scientific or folk understandings, just as our present-day idea of “fish” no longer includes whales and dolphins (see Griffiths, 1997).

Subjective Experience At least as practiced in Western societies (see Lutz, 1988), common sense tends to see emotions as primarily subjective experiences, directly accessible only to those who are feeling them (Parkinson, 1995). However, people might

Emotion Theories 87

sincerely deny the experience of something that for others is self-evidently an emotion. As Averill (1980, p. 137) put it, “It is perfectly meaningful to say of a person that he is envious, even though that person might sincerely and vehemently deny the fact.” Similarly, I might scream out “I am not angry!” when arguing with someone who accuses me of losing my temper. According to Frijda (e.g., 1986, 2005), my angry experience may be constituted by my perception of the annoying qualities of your conduct rather than by any representation of my own mental state (see also Lambie & Marcel, 2002). More generally, theorists often characterize emotions as forms of action readiness rather than as simple internal experiences. Such a view suggests that measures relating to implicit attunements and response latencies might provide better information about emotion than self-reports. Use of these measures is consistent with the idea that emotions are not only about internally registered experience but also about a person’s orientation toward what is happening in the outside world.

Response Syndromes Although common sense tends to conceive of emotion as subjective, it is also thought (e.g., by James, 1898) to involve physical changes in the body that are potentially measurable. Different researchers have prioritized different aspects of bodily response when indexing participants’ emotions in research, but the most commonly used measures are of facial movement (e.g., Ekman & Friesen, 1978) and autonomic nervous system (ANS) activity (e.g., Kreibig, 2010). Both kinds of measure potentially reflect action readiness since the ANS conserves or releases metabolic energy required for behavior, and the face not only contains sensory organs that orient to objects requiring attention, but also muscles that tighten or relax in preparation for broader movements of the body. Perhaps the most popular psychological definition of emotion sees it as a syndrome of responses including these forms of bodily activity, as well as cognitive and attentional changes (implemented by the brain) and subjective experiences (indexed by self-report). In short, emotion is specified in terms of a number of coordinated components rather than by any single criterion. Empirical evidence suggests, however, that coordination of supposed response components is less than perfect (e.g., Mauss, Levenson, McCarter, Wilhelm, & Gross, 2005). For the purposes of measurement, then, it is necessary to collect more than one kind of information in order to get a fix on emotion. At a theoretical level, the lack of coordination suggests that emotions are at best flexible or open systems rather than fixed response patterns. The action readiness modes constituting emotions may involve mental and physical preparation for achieving an abstractly defined goal rather than preparation for any specific set of muscle movements. Indeed, different possible physical responses may serve the same emotional function on different occasions and in different contexts (see below). For some theorists, the

88

MENTAL STATE THEORIES

lack of coordination suggests a more radical conclusion: that “emotions” are not coherent response syndromes at any level in the first place (e.g., Russell, 2003).

States versus Processes Another problem with the response-syndrome idea is that it fails to specify the temporal patterning of different components and their responsivity to unfolding environmental events. Many theorists emphasize that emotions are dynamic processes rather than momentary states that can be captured fully by one-shot measures. Although it is certainly possible to conceive of “states of action readiness” (e.g., Arnold, 1960; Frijda, 1986), as manifested in postural positions, facial configuration, or mental attitude, clearer information about how someone is inclined to respond to events comes from observation of their changing actions and reactions, and how these relate over time to external changes (e.g., Krumhuber, Kappas, & Manstead, 2013). Indeed, the distinctive qualities of some emotions may depend intrinsically on these dynamic correspondences. In Wittgenstein’s (1953/2001, p. 174) example: ‘Grief’ describes a pattern which recurs, with different variations, in the weave of our life. If a man’s bodily expression of sorrow and joy alternated, say, with the ticking of a clock, here we should not have the characteristic formation of the pattern of sorrow or of the pattern of joy. ‘For a second he felt violent pain.’—Why does it sound queer to say: ‘For a second he felt deep grief’? Only because it so seldom happens?

In other words, some of the defining characteristics of certain emotion categories seem to depend not on immediately sensed qualities of experience (such as pleasure or pain) but rather on the way that these qualities change over time.

Passions versus Actions The idea of emotion as reactive rather than active is so deeply engrained that it seems to be encoded in the syntax and grammar of the English language. Emotions are most commonly represented as nouns or adjectives rather than as active verbs. Indeed, the verbs “anger,” “embarrass,” and “disgust” are about provoking emotions in others rather than the actor being emotional. In those cases where emotional experiences are directly conveyed using verbs (e.g., “X loves/hates Y”), causality is typically attributed to the object rather than the subject of the verb, that is, whatever the person is getting emotional about rather than the person getting emotional (Brown & Fish, 1983). However, if emotions are processes that serve functions (such as readying the mind and body for action), then it also seems possible that they are things

Emotion Theories 89

we do and not simply things that happen to us: actions rather than passions. Indeed, some theorists (Sarbin, 1986; Sartre, 1962; Solomon, 1976) undermine the so-called myth of the passions by arguing that the presentation of emotions as falling outside our responsibility serves certain social functions (e.g., Averill, 1980) and does not necessarily reflect reality. In fact, people may take on emotional roles or orientations toward events in more active or controlled ways. Most contemporary accounts acknowledge delimited agency in at least some emotion-related processes by reference to the concept of emotion regulation (e.g., Gross & Thompson, 2007). The usual implication is that some aspects of emotion are passive responses, but even these are potentially susceptible to some level of anticipatory or retrospective control. However, critics (e.g., Campos, Walle, Dahl, & Main, 2011; Kappas, 2011) have countered that emotions are regulatory processes in the first place, opening the possibility of interlocking higher- and lower-order control mechanisms. In either case, one clear implication of acknowledging the active nature of some emotional processes is that they cannot be explained as simple responses to stimuli, even if some of their components start out that way during their early development (see Leventhal & Scherer, 1987).

Dimensions versus Categories Debates throughout the history of psychology have considered whether emotion is better represented by categories or dimensions. Some emotional phenomena are clearly more similar than others, but is it also the case that all differences between them can be reduced to their values on a small number of underlying dimensions of affect, such as pleasure and arousal? The answer to this question depends partly on how emotions are characterized in terms of the distinctions discussed above. If emotions are momentary subjective states, then the features available for discriminating their categories are correspondingly limited. For example, if the only parameters of emotion came from immediate internal perception, excluding prior, current, or (anticipated) future context, the possibilities for differentiation would be correspondingly limited. However, if information about active relational orientation and temporal articulation supplements and contextualizes any subjective signals, values along only two or three affective dimensions are unlikely to specify the range of possible variations in the resulting phenomena. For similar reasons, recent dimensional accounts (e.g., Barrett, 2006; Russell, 2003) focus on “core affect” rather than more articulated “emotion” episodes, and incorporate additional principles to accommodate categorical perceptions of experience or self-attributions. Correspondingly, even those subscribing to the notion of discrete categorical emotions acknowledge that some aspects of emotional experience may be represented in dimensional terms, such as pleasure, arousal, and intensity. Thus, both sides of the debate

90

MENTAL STATE THEORIES

treat some aspects of emotion concepts as dimensional and others as categorical (Hamann, 2012). What remains in dispute, however, is exactly where the lines should be drawn.

Demarcating Emotion In addition to the issue of whether emotion instances can be categorically distinct from each other, there is dispute about whether the more inclusive category of “emotion” is genuinely separate from nonemotion, or whether any transition is simply a matter of degree (e.g., Clore & Ortony, 1991; Russell, 1991). If emotion is distinct and distinctive, the obvious inclination is to seek an ingredient or set of ingredients that makes it special. If emotion shades into nonemotion, then the characteristics of emotion often also characterize related phenomena. For categorical theorists who characterize emotion in terms of subjective mental state, the special ingredient that makes emotion emotional must come from within. For example, James (1898) concluded that the source of distinctively emotional experience was internal sensory information from the body. Subsequent theorists have extended this account to include other internal sources of affectively relevant information, such as pleasure and other kinds of information registered more directly by the brain (e.g., Damasio, 1994, and see below). For categorical theorists who acknowledge the relational basis of emotion, a better candidate for distinguishing emotion from nonemotion is some registration of the personal significance of what is happening, namely, appraisal (Arnold, 1960). This process is thought to coordinate a range of adaptive responses, including facial expressions and autonomic changes that constitute a state of action readiness oriented to the concern identified by appraisal. More generally, categorical approaches assume that emotion can be demarcated from nonemotion, and different emotions can be demarcated from each other, according to the presence or absence of certain components, potentially including appraisals, action tendencies, bodily responses, facial expressions, and subjective feelings. However, theorists disagree about whether all or any of these components are logically necessary for emotional experience (e.g., Averill & Nunley, 1992), whether they must reach an intensity threshold to qualify as emotion-relevant, and whether they need to be structured in particular temporal or synchronic patterns before attaining the status of emotion (e.g., Moors, 2009). For dimensional theorists who focus on “core affect,” the demarcation issue concerns how pleasure and arousal dimensions translate into apparently categorical experiences. The emotion/nonemotion distinction may be finessed by arguing that it is a fuzzy rather than sharp dividing line and/ or relates to how phenomena are conceptualized rather than how they are constituted (Russell, 2003).

Emotion Theories 91

Conclusions Evaluating emotion theories is difficult partly because it is difficult to specify what range of phenomena fall within their scope. Some theorists assume that a thorough conceptual analysis of the ordinary language category of “emotion” can lay the groundwork for demarcating the phenomenon, while others treat ordinary-language concepts as relevant only to the conceptualization of emotional phenomena, and not the phenomena themselves. The range of application of theories therefore varies not only in coverage of explananda at a single level of analysis but also in the different levels of analysis at which the theory is applied, which sometimes includes both emotional phenomena and their conceptualization (see section on “Concept versus Reality,” below). Some theories rely neither on ordinary language nor on reconstruction of the emotion domain based on conceptual or empirical analysis, but instead develop formulations of emotion based on theoretical principles. The danger with this strategy is that it may lead to circularity, where the theory’s inclusiveness is established by excluding theory-inconsistent observations from its domain of relevance. None of these problems is easily solved by dictating in advance what should be called an emotion and what should not. Ordinary language cannot do this job because of its inherent flexibility and variability (both contextual and historical). In addition, anthropological research shows that relevant terms do not translate perfectly across different languages. The alternative is to be more specific about which kinds of putatively emotional events are the focus of any theory and to establish empirically whether the theory may be applied beyond that immediate range of convenience. Ultimately, ordinarylanguage categories may adjust in response to theoretically driven scientific knowledge, leaving a more tractable common-sense category of emotion with which to work.

Levels of Analysis Structure versus Function Most theories of emotion work at both a structural and a functional level. On one hand, they address what the components of emotions are and how they fit together (over time and in terms of concurrent association). On the other hand, they explain these structural features by reference to the adaptive problems that the identified response syndromes are designed to solve. Thus, Scherer (e.g., 2001) argues that emotions involve response synchronization that is achieved by a sequential series of “stimulus evaluation checks” that have evolved to identify events requiring immediate but flexible response (the decoupled reflex). More generally, appraisal theories argue that the mental system needs a mechanism for detecting adaptive concerns and outputting specific modes of action readiness designed to deal with these concerns

92

MENTAL STATE THEORIES

(e.g., Cosmides & Tooby, 2001). What makes emotion distinctive, according to these accounts, is its plasticity in terms of elicitors and response modes: Rather than releasing a fixed action pattern in response to particular stimuli, appraisal identifies relational themes, then sets the mental and physical system in a mode that permits adaptively appropriate (and context-sensitive) deployment of resources to address these themes. Structural and functional accounts do not always mutually support one another, however. For example, thinking of emotions as tightly packaged response syndromes (e.g., innately programmed “basic emotions,” e.g., Ekman, 2003) works against the idea of their flexibility, leading some theorists to propose that they are “open” rather than “closed” systems (see Parkinson, 2009). Certainly, the evidence concerning the response coherence of emotion (referred to above) suggests that the level of structural integrity is generally low to moderate, with frequent dissociations between ANS activity, facial movement, appraisals, and subjective experience (e.g., Reisenzein, Studtmann, & Horstmann, 2013). From a functional perspective, some theorists might suggest that no single mode of action readiness would appropriately serve adaptive purposes and that the mental system merely needs to be attuned to the nature of the concern (without setting any default behavior, e.g., Frijda, 1986; see also Roseman, Wiest, & Swartz’s [1994] distinction between action readiness and “emotivational” goals, Cacioppo, Berntson, & Klein’s [1992] distinction between “tactic” and “strategy,” and the section on “Response Syndromes,” above). Thus Cannon (1927) argued that emotions are generally responses to emergencies requiring only that the body needs to be mobilized for nonspecific action using undifferentiated arousal. Here, the general conclusion would be that the functional requirements of emotion necessarily leave it relatively unstructured. Even focusing on structural aspects of emotion theories, the level at which emotion components are formulated can differ. Indeed, a problem for some theories lies in underspecification of the processes whereby components from different levels of analysis interact or combine. For example, Schachter’s (1964) theory argued that emotion reflects the combination of two factors, one physiological, the other cognitive, but did not make it clear whether the physiological factor served merely as a source of information for the cognitive part (e.g., Valins, 1966) or had a more basic motivational or subjective role (Reisenzein, 1983). Similar problems confront more recent structural theories that specify emotions as response syndromes, including appraisals, autonomic changes, facial activity, and subjective experience. These supposed response components cannot all be specified as measurable physical responses, because subjective experience seems to depend on introspection and appraisal apparently implies cognitive or perceptual operations. To complicate matters still further, some theorists (e.g., Lazarus, 1991; Moors, 2013) define the appraisal component in functional rather than structural terms, as whatever process converts input into a recognizable emo-

Emotion Theories 93

tional meaning (see next section). How then do functionally, introspectively, and physically defined events combine to produce a structurally coherent response syndrome?

Computation, Algorithm, and Implementation Over the years, several writers have distinguished levels of analysis depending on their inclusivity and specificity. Some explanations are more abstract than others. A currently popular characterization of some of these differences is based on Marr’s (1982) distinctions between computational, algorithmic, and implementational levels of analysis (see also De Houwer & Moors, Chapter 2, this volume). The computational level defines a particular kind of functional analysis where input–output relations are specified: It addresses what an information processing device does (and why it does it). The algorithmic level specifies operations (algorithms) and representations that mediate these relations. The implementational level concerns the physical (e.g., neural) processes that perform these algorithmic operations. According to Moors (2013), appraisal processes should be formulated at Marr’s computational level. The appraisal processes specify what needs to be achieved by the process (getting from informational inputs to particular emotionally relevant outputs; cf. Lazarus, 1991), but not what specific representations or operations are used to make these computations (algorithms), or what neural mechanisms underlie their operation (implementation). In Marr’s (1982, p. 24) terms, appraisal thus specifies only “a mapping of one kind of information to another.” However, some theorists have attempted to detail the range of algorithms that might mediate appraisal computations. For example, Leventhal and Scherer’s (1987) model distinguishes three kinds of algorithm (and associated representations) instantiating appraisal computations, ranging from sensorimotor through schematic to conceptual. With the increasing availability of technologies for measuring and manipulating neural activity in the central nervous system, it is becoming more common for emotion theorists to associate algorithmically defined processes with regions, circuits, or networks in the brain. Indeed, evidence about implementation is also used to rule in or rule out certain algorithmically or even functionally framed hypotheses. For example, Storbeck, Robinson, and McCourt (2006) use evidence about neural pathways in support of hypotheses about the functional dependence of emotion on cognition; this is in contrast to Zajonc’s (1980) earlier attempts to draw the opposite conclusion on the basis of the more limited neural evidence that was then available. How computations at the algorithmic level map onto neural processes at the implementational level remains controversial. For one thing, it is not clear that distinct phenomena or processes defined computationally or algorithmically necessarily correspond directly to consistent and distinguishable patterns of brain activity. For example, evidence from fMRI and PET studies designed to identify brain regions associated with discrete emotions

94

MENTAL STATE THEORIES

suggests that localization differs across participants and studies (Lindquist, Wager, Kober, Bliss-Moreau, & Barrett, 2012), but this does not necessarily undermine the idea of categorical emotions coordinated by other means. Even tracking neural pathways running from sensory receptors to brain regions does not directly support particular kinds of computation without prior understanding of exactly what psychological functions and algorithms these brain regions support. Further, because different parts of the brain interact with each other in complex ways involving resonance and feedback processes (e.g., Lewis, 2005), one-way accounts of information transmission are likely to be oversimplistic. Neuroscientific and psychological investigations of emotion can still be synergistic, however. The ideal is that specific psychological hypotheses are translated into parallel hypotheses about neural processes and the results of tests of each are mutually informative. Psychologically uninformed investigation of brain activity can never provide a complete understanding of emotion, but knowledge derived from psychologically informed neuroscientific research can help to refine psychological understanding, in turn leading to more focused neuroscientific research (see Beer, Chapter 9, this volume). Of course, Marr’s distinctions are not theoretically neutral. For him, “most of the phenomena that are central to us as human beings . . . are primarily phenomena of information processing” (1982, p. 4). This is not an uncontroversial assertion for those interested in phenomena outside the realm of individual perception. Nor does it necessarily apply regardless of the theorist’s explanatory goals. The three-way classification is partly designed to justify explanations that are tied neither to specifications of the problem nor to directly observable physical processes. It works from a metaphor based on computer systems where software can run equivalently on a range of different kinds of hardware. However, some theorists (e.g., Barsalou, 1999) take issue with the view that the medium of representation makes no difference to how processes unfold, and emphasize the role of embodied rather than abstract representations in emotion processes (Barrett, 2006; Niedenthal, 2007). Other theorists take inspiration from Gibson’s (1979) ecological approach and deny that representations or information processing need to mediate perception, let alone less obviously informational phenomena such as emotions (e.g., Baron & Boudreau, 1986; Parkinson, 1996). In short, Marr’s distinctions may only have direct relevance for theorists who see emotion as dependent on forms of meaning extraction such as appraisal or self-attribution (Parkinson, 2009).

Individually Localized versus Socially Distributed Both structural and functional accounts of emotion tend to be formulated at the individual and intrapsychic level. The idea of emotions as private events strongly suggests that each individual’s mental apparatus needs to work out which emotion to activate, rather than causal processes being dis-

Emotion Theories 95

tributed between individuals or between individuals and their practical and social environment. However, the idea that emotions and their associated appraisals may emerge from ongoing transactions and reciprocal feedback processes has gained some currency (e.g., Campos et al., 2011; Fogel, 1993; Lewis, 1996; Parkinson, 2001). De Rivera (1984) argued that emotions take place between rather than within people and are first manifested in individual development as movements toward or away from other people. More recently, several theorists have argued that emotions often serve communicative and interpersonal influence functions (e.g., Manstead & Fischer, 2001; Oatley, 1992; Parkinson, 1995). Furthermore, emotions are often oriented to group-based concerns and intergroup relations, rather than purely individual matters (e.g., Parkinson, Fischer, & Manstead, 2005; Tiedens & Leach, 2004). One implication of communicative approaches is that emotions are oriented to actual and anticipated responses from other people and are modulated by available interpersonal feedback, rather than by purely intrapsychic regulatory processes (Campos et al., 2011). A key issue is how these strategic aspects of emotion presentation relate to antecedent-directed causal accounts such as appraisal and basic emotion approaches. From an appraisal perspective, it may be that appraisals lead to emotions and emotions in turn lead to communicative and other-responsive action tendencies. From a basic emotion perspective, interpersonal factors may motivate the regulation of emotions that are already underway, as implied, for example, by Ekman’s (1972) notion of display rules. However, for theorists such as Fridlund (1994), facial displays are determined by social motives in the first place and are attuned to potential addressees from the outset. Such an account might potentially be extended to other aspects of the emotion syndrome, suggesting that a communicative orientation is primary rather than something subsequently superimposed on more “basic” emotional urges or action tendencies activated by appraisals (Parkinson, 1995).

Concept versus Reality Although most theories of emotion operate at the level of explaining the empirical characteristics of the phenomenon, there are also theories that address the concepts that people use to describe emotional experience. The relation between these two levels of analysis takes different forms depending on the theorist in question (see above). For many theorists, concepts of emotion are more or less transparent windows onto the underlying phenomenon, which may introduce minor distortions to perception but do not fundamentally alter the nature of the explanandum. Such an approach is compatible with reliance on self-reports of emotion as the central dependent variable in research. Other theorists attempt a logical reconstruction of the grammar of emotional terms intended to tidy up concepts and identify their underly-

96

MENTAL STATE THEORIES

ing meaning. For example, Ortony, Clore, and Collins (1988) formulated an organized set of classical definitions specifying the appraisal structure of different emotions. Where Ortony and colleagues saw anything counting as anger as necessarily caused by appraisals of blameworthiness, other appraisal theorists regarded it as an empirical discovery that anger depends on other–blame (e.g., Lazarus, 1991). To the extent that the usual criterion for demonstrating the presence of anger in appraisal research is based on selfreport, it is difficult to understand how the emotion–appraisal association goes beyond a conceptual one (Parkinson, 1997). For yet other theorists, categorizing experience in emotional terms helps to constitute the phenomenon. Schachter (1964) famously argued that labeling one’s physiological arousal in emotional terms as a function of selfattribution directly produced emotional experience. More recently, Barrett (2006) suggested that conceptualization of the experiences of pleasure and arousal gives different emotions their distinctive qualities. In both cases, it is not entirely clear whether the explanandum is an empirical emotion, an emotion concept, or merely a self-report. For other theorists, emotions can arise without being conceptualized in emotional terms (or even as embodied simulations of emotion concepts). Russell (2003) provides one of the most comprehensive accounts available of how affective phenomena and the representations applied to those phenomena might interrelate. In his view, prototypical emotion scripts structure our perception of affective phenomena, leading us to apprehend categorical structure that is not present in psychological reality, just as we perceive patterned constellations of stars scattered across the night-time sky. In reality, the subjective feelings associated with emotion scripts are reducible to combinations of different levels of pleasure and arousal (core affect). Categorical differences between supposedly distinct emotions (e.g., basic emotions) reflect extrinsic features represented in emotion scripts relating to events that precede, accompany, or follow these feelings. Thus, saying that I am “angry” simply means that I am experiencing negatively valenced arousal in a context resembling the prototypical “anger” episode in which someone has offended me, and I am inclined to retaliate (but I try to stop myself). However, if I had been socialized to apply a different set of emotion scripts, I might categorize my experience (and that of other people) differently. The idea that emotion concepts help to constitute emotional phenomena is most strongly associated with the social constructionist tradition. According to this approach, cultural categories shape emotional activity and not just emotion perception. For example, children might learn to enact conventionally defined emotion roles to serve specific cultural needs. Thus, Averill (1980) describes the Gururumba “illness” of “being a wild pig” (see Newman, 1964) as a socially sanctioned strategy for overcoming economic demands on males associated with recent marriage without undermining the legitimacy of these demands. Because wild-pig behavior is believed to

Emotion Theories 97

be caused by being bitten by the ghost of a recently deceased tribe member, sufferers are not responsible for their behavior. However, it still serves to encourage other tribes-people to provide resources to the “victims” of this “illness.” Similarly, Western “anger” is a transitory role that permits social actors to threaten retaliation for misdemeanors (thereby upholding equity norms) without taking full responsibility for deliberate antagonism (thereby avoiding breaking social rules about nonaggression). The fact that anger is interpreted as a passion rather than an action means that angry people are not sanctioned for their conduct as much as they would be if they were not angry. For discursive psychologists such as Edwards (1999), there is no access to a putative realm of unrepresented phenomenon, and researchers should focus instead on formulations that construct events as emotional. For him, the study of how people talk in emotional terms is not an adjunct to more direct exploration of emotional or affective reality but the primary explanandum. Correspondingly, the use of emotional language is never merely descriptive, but instead performs direct pragmatic functions in conversation. Saying that I am angry, for example, may be a way of diverting blame or allocating responsibility, and may be met by contestation or accession by another party to the dialogue. A further distinction between levels of analysis in emotion theory, then, concerns whether emotion representations are to be analyzed in terms of their semantic meaning or pragmatic function.

Logical and Inferential Issues Logical Fallacies in the Theoretical Interpretation of Empirical Data One logical threat to the interpretation of empirical data in emotion research stems from the dependence of researchers on ordinary-language concepts and self-report data. As noted earlier, the fact that participants report strong associations between (say) appraisals of an emotional stimulus and the strength and/or type of emotional experience may simply reflect the fact that both types of self-report measure draw on the same common-sense conceptualizations of emotion. For example, the everyday concept of “anger” is associated with a sense of being unfairly impeded by someone or something else in one’s progress toward a valued goal. It is therefore no surprise that experiences reported by participants as “anger” also tend to be characterized as involving appraisals of illegitimacy and other-responsibility (see Parkinson, 1997). Illegitimacy of events is also a central justification for getting angry in our society. For this reason, participants allowed to choose which “anger” episode they should report in a psychological study are more likely to select an instance that is justifiable in terms of an other-blame appraisal (Parkinson, 1999). Thus, associations between self-reports of anger and self-reports of appraisal provide no direct evidence about actual appraisals

98

MENTAL STATE THEORIES

associated with independently verifiable anger episodes, and less still about whether such appraisals cause such anger episodes. The upshot is that evidence about how emotions are conceptualized and justified is sometimes mistaken for evidence about empirical factors associated with or causing emotions. A related issue concerns research using judgment tasks to assess associations between emotions and facial expressions (e.g., Ekman, Sorenson, & Friesen, 1969). In these studies, participants allocate faces intended to indicate emotional states to linguistic emotion categories. Agreement between participants about the “correct” emotion category is often high and almost always better than chance. These results tell us that emotion-related information can be extracted from faces, perhaps partly because some of those faces are conventionally associated with emotion concepts. The findings do not directly support the claim that people actually adopt these facial configurations when experiencing the actual emotion in question. Research directly assessing facial behavior during emotional experiences suggests that associations between the two is far from perfect. For example, the same display can occur during contrasting emotional states (Aviezer, Trope, & Todorov, 2012) and even well-established methods of inducing emotional states such as surprise (Reisenzein, Bördgen, Holtbend, & Matz, 2006) and disgust (Fernández-Dols, Sánchez, Carrera, & Ruiz-Belda, 1997) do no always result in the facial displays that are taken to be indices of these emotions (see Reisenzein et al. for a review). If even relatively intense experiences of these emotions are not accompanied by the predicted facial displays, how useful are these displays as indices of the supposedly associated emotion? Researchers wishing to defend the theoretical claim that (some) facial configurations are specific expressions of particular emotions point to confirmatory evidence collected in particular contexts. There are certainly occasions when an experienced emotion is genuinely associated with the predicted facial expression (e.g., Matsumoto & Willingham, 2006). However, problems arise when general conclusions are drawn on the basis of particular confirmatory instances. Apparent emotion–expression associations may in fact be context-dependent and/or reflect processes that are only extrinsically related to emotions, such as action readiness (Frijda, 1986) or social motives (e.g., Fridlund, 1994). Reliance on confirmatory evidence is also a problem in some research using fMRI and other measures of brain activity. As Aue, Lavelle, and Cacioppo (2009, p. 12) have argued, “If a theory posits that a specific brain response (Φ) causes a psychological event (ψ), the manipulation of the psychological event and the observation of activation over the specified brain response may provide only limited support for the experimental hypothesis.” Again, the reason is that the effect of the emotion manipulation on its supposed index may be context-dependent or may reflect some aspect of the manipulation that is extrinsic to its emotional content. Aue et al. (2009, pp. 12–13) go on to argue that “if a given brain area (Φ) shows a hemody-

Emotion Theories 99

namic response as a function of an experimentally manipulated emotional state (ψ), it does not logically follow that the activation in that brain area in another condition or study marks the occurrence of that same emotional state.” It is therefore not wholly surprising to discover that there is disagreement in the fMRI literature about the extent to which there are consistent relationships between emotion and brain activation. Some meta-analysts (e.g., Vytal & Hamann, 2010) have concluded that neuroimaging studies do provide support for the view that certain emotions, at least, are characterized by consistent correlations with regional brain activations. Others (e.g., Lindquist et al., 2012) find little evidence that discrete emotion categories are consistently localized to distinct brain regions. At the very least, it seems premature to rely on brain-imaging measures to infer the presence or absence of an emotion.

Problems in Identifying Proximal Causes Although most theorists presuppose a back-story relating to distal cultural or evolutionary precursors, more proximal causes of emotion are of the most direct interest to emotion theorists working in the field of psychology. Antecedents of emotions are sought either within the time frame of ontogenesis, or, more commonly, within the immediate situation in which emotion occurs. In most cases, the explanandum is a direct reaction (conceived as a state or response syndrome) to a momentary event, and not an unfolding adjustment to changing circumstances (see section on “States versus Processes,” above). In terms of theory-testing, predictions concern associations between (personal and situational) proximal antecedents and emotion outcomes (components of the response syndrome). When experimental methods are used, the prediction is correspondingly that manipulations of these postulated antecedents should impact on dependent variables related to emotion components. However, many of the theorized causal processes are inherently difficult to track or modify directly. For instance, appraisals are typically measured using self-reports and modified using instructions or changes in information, but the underlying appraisal process is thought to operate at an unconscious as well as conscious level, making it potentially impenetrable to such methods. The indirectness of measures and manipulation makes the evidence ambiguous. In addition, the lack of temporal resolution is problematic given the presumably tight time scale of appraisal–emotion interactions (Scherer, 1993). Progress depends on developing and validating more direct and more precise measures and manipulations. For example, implicit measures of appraisal based on interpretation of ambiguous materials, stimulus sensitivity, or perceptual attunement (e.g., Joormann & Gotlib, 2006; Keltner, Ellsworth, & Edwards, 1993) show promise (although discriminant validity may be difficult to substantiate). Similarly, priming manipulations permit some modulation of appraisal-related processes (Moors, De Houwer, & Eelen, 2004; Neumann, 2000). Corresponding problems also apply when testing theories that explain

100

MENTAL STATE THEORIES

emotion as responses to perceived internal feedback from bodily responses (e.g., James, 1898; Laird & Bresler, 1992). Manipulating such feedback can draw participants’ attention to it so that it is used inferentially instead of implicitly (cf. Leventhal & Mace, 1970). Further, manipulated feedback that does not realistically track the emotional qualities of the unfolding situational context may be perceived as lacking credibility (Tomkins, 1981). If the only kind of peripheral feedback that can influence emotion is realistic feedback arising from spontaneous responses to emotional stimuli, then experimental methods cannot be used to test whether feedback genuinely has an independent impact during emotion elicitation (Ellsworth & Tourangeau, 1981). Similar issues arise when operationalizing dependent variables. According to the response syndrome view, no single criterion can establish the presence of emotion. Further, because people can be unaware of their emotions, self-reports cannot solve the problem either. To further complicate matters, establishing the causal precedence of postulated antecedents requires isolation of a definite moment prior to an emotion transition. Even if technology with sufficient temporal sensitivity were available, current conceptualizations do not permit specification of the criteria that it would need to detect in order to establish that such a transition had taken place. In this section, we have argued that some core assumptions of many theories of emotion cannot easily be tested. However, more specific hypotheses may be generated about relations between delimited antecedents and responses potentially associated with emotions. Cumulative evidence collected using this strategy may ultimately help to legislate between alternative general approaches.

Falsifiability An influential paper by Zajonc (1980) initiated what became known as the “affective primacy debate.” For Zajonc, an increasingly common theoretical assumption was that emotion was the output of a protracted chain of information processing. By contrast, earlier theorists had emphasized the immediate and unreflective nature of affective experience. Zajonc aimed to redress the balance and to reemphasize the distinctive qualities of cognition and emotion. The terms of the debate were subtly reframed by Lazarus (1982), who contended that an individual needs to apprehend the personal meaning of what is happening before reacting emotionally. In effect, the fact that emotion always involves taking a meaningful relational stance toward some object or event was taken to imply a prior meaning-generation process in the mental system of the person who was about to get emotional. If emotion is defined in terms of its associated relational meaning, and if any process that leads to this meaning qualifies as a cognitive process, this explanatory account becomes a circular one. Lazarus (1991) defined the causal process leading to emotion (appraisal) in terms of its consequences (“it is meaning that counts

Emotion Theories 101

in emotion, not how that meaning is achieved,” p. 160), making it seem as if his central claim is definitional rather than empirical. When Zajonc presented apparently falsifying instances of “emotion” without prior “cognition,” Lazarus argued either that they are not really emotions or that there was actually an undetected cognition preceding them. Without appraisal-independent criteria for establishing what counts as emotion, and emotion-independent criteria for inferring undetected appraisal, such claims seem empty. Correspondingly, without operational measures unbiased by theory, it is difficult to see how the theory makes definite predictions rather than simply explaining events after the fact. In practice, researchers have often assessed both appraisal and emotion by self-report, meaning that any correlations obtained reflect associations between representations of situations and representations of reactions to those situations rather than real appraisals and real emotions (e.g., Parkinson & Manstead, 1992). One way of circumventing these problems is to dismiss the abstract question of whether emotion always depends on cognition as a matter of “mere semantics,” and to turn research attention instead to the more specific processes that may lead to more specific emotion-related outcomes (Leventhal & Scherer, 1987). Instead of arguing about whether causes count as cognitions and effects count as emotions, researchers can delineate lower-level causal processes that may come together to yield emotionally meaningful patterns of response. These more specific causes and outcomes are more susceptible to precise operational definitions than diffuse and fuzzy categories such as “cognition” and “emotion.” For many appraisal theorists, a central feature of emotions is their relational meaning (specified by an intentional orientation toward some emotional object), and they therefore seek explanations for emotions by specifying how that meaning is constructed. For feedback theories, the central aspect of emotion is the experiential quality that defines the emotion as distinct, and they therefore seek to explain how people detect the distinctive qualities of their experience. In either case, explanation targets only one aspect of emotion, which is not necessarily consistently present across all cases (Parkinson, 2013). For example, following Sartre (1962), Frijda (1986) points out that emotional experience need not be defined by any sense that one is in a particular kind of mental state (reflexive emotional experience) but rather may be constituted by perceptions of the environment as emotionally colored and as calling forth certain kinds of action (irreflexive emotion). When angry, I am not always thinking, “I am angry,” but rather I am seeing events as unfair and demanding active contention (see also Frijda, 2005; Lambie & Marcel, 2002). Defining emotion in terms of its active attunement and orientation rather than relational or experiential meaning (see the section “Passions versus Actions” above) may solve a number of theoretical problems, breaking the interdependence between conceptual and empirical levels of explanation. Another prominent theory that is difficult to falsify is Fridlund’s

102

MENTAL STATE THEORIES

(1994) behavioral ecology view of facial displays. As we saw above, Fridlund argued that the facial movements typically assumed to be expressive of emotion are in fact displays of social motives, or social intentions. His experimental demonstration of the importance of sociality, rather than emotionality, in giving rise to facial displays (Fridlund, 1991) included an alone condition. Critics of Fridlund’s theory have drawn attention to the fact that participants in his “alone” condition smiled more while viewing amusing film clips than they did during baseline, without a film, despite the fact that there were no others present to whom social motives or intentions could be communicated. Fridlund responded to such objections by appealing to the notion of implicit sociality: Even when we are alone, others are present in our imagination. This explanation runs the risk of being impossible to falsify. Acknowledging this, Fridlund (1991, p. 238) noted that “it may be that this theory, like many others, is untestable in extremis, but as the present findings suggest, it can be exposed to falsification using experimentally manipulable ranges of sociality.” In other words, if variations in implicit sociality had not influenced smiling in his study, a central tenet of his theory would have been disconfirmed. However, the impracticality of achieving complete removal of even implicit sociality (so that a participant is completely alone in both reality and imagination) means that the ultimate test of Fridlund’s hypothesis that facial display requires an addressee cannot be implemented.

Conclusions A good theory of emotion should treat emotions as processes that unfold over time rather than as states that are either present or absent. It follows that such a theory should not seek to identify a single cause of emotion at a particular moment in time, and should acknowledge the dynamics of emotion: Appraisals change over time; emotional experiences wax and wane and bleed into each other; facial expressions vary in rise time, length of apex, and speed of offset. A good theory should also recognize that the sequence linking an emotional event to cognitions, physiological changes, subjective experience, and emotional behavior is likely to contain feedback loops that reflect the ways in which “effects” can also be “causes.” Finally, a good theory is one that acknowledges that emotions arise from interactions between an individual and his or her environment, and that the most significant emotional features of the environment are other people, whether as individuals or social groups. The main challenges to developing such a theory are the lack of agreement about how to define emotion, the difficulty of integrating explanations across multiple levels of analysis (central and peripheral nervous systems; thoughts and feelings; expressive and instrumental behavior; interpersonal, group, and intergroup behavior; culture), and the enduring tendency of Western theorists to favor linear explanations and linear (cause–effect) meth-

Emotion Theories 103

ods for testing those theories. It should therefore be obvious that developing a good theory of emotion is extraordinarily challenging. We do not (yet) have good theoretical constructs that map onto the processes involved in emotion partly because so many of these processes are difficult to represent in words. Even appraisal theory, with its emphasis on meaning, struggles to identify the categories or dimensions of appraisal. To conclude on a positive note, one advantage of not having advanced theoretical models of emotion processes that are rooted in language is that there is (perhaps) more openness to the embodied nature of these processes and to the extent to which they are influenced by social and cultural contexts. We have focused on theories that seek to understand the causes of emotion and have paid relatively little attention to theories concerned with the effects of emotion (e.g., Forgas, 1995; Schwarz & Clore, 1988). Many theories, especially those oriented toward emotion’s functions, address both causal and causative processes in an integrative manner. Indeed, as we have argued in this chapter, emotion is best treated as unfolding relational activity oriented toward dynamic events and not as a discrete stage in a simple input– output process. Emotion can be understood in relation to its practical and social context rather than as a determinate reaction to specific stimuli or a momentary stage in a unidirectional causal chain. More research on what getting emotional does should help to clarify what makes us emotional. Acknowledgments We thank Toon Kuppens, Elena Lemonaki, Marlon Nieuwenhuis, Sindhuja Sankaran, Joe Sweetman, Job van der Schalk, and Jason Vandeventer for comments on an earlier draft of this chapter.

References Arnold, M. B. (1960). Emotion and personality: Vol. 1. Psychological aspects. New York: Columbia University Press. Aue, T., Lavelle, L. A., & Cacioppo, J. T. (2009). Great expectations: What can fMRI research tell us about psychological phenomena? International Journal of Psychophysiology, 73, 10–16. Averill, J. R. (1980). A constructivist view of emotion. In R. Plutchik & H. Kellerman (Eds.), Theories of emotion (pp. 305–340). New York: Academic Press. Averill, J. R., & Nunley, E. P. (1992). Voyages of the heart: Living an emotionally creative life. New York: Macmillan. Aviezer, H., Trope, Y., & Todorov, A. (2012). Body cues, not facial expressions, discriminate between intense positive and negative emotions. Science, 338, 1225–1229. Baron, R. M., & Boudreau, L. A. (1987). An ecological perspective on integrating personality and social psychology. Journal of Personality and Social Psychology, 53, 1222–1228. Barrett, L. F. (2006). Solving the emotion paradox: Categorization and the experience of emotion. Personality and Social Psychology Review, 10, 20–46.

104

MENTAL STATE THEORIES

Barsalou, L. W. (1999). Perceptual symbol systems. Behavioral and Brain Sciences, 22, 577– 660. Bentley, M. (1928). Is “emotion” more than a chapter heading? In M. L. Reymert (Ed.), Feelings and emotions: The Wittenberg Symposium (pp. 17–23). Worcester, MA: Clark University Press. Brown, R., & Fish, D. (1983). The psychological causality implicit in language. Cognition, 14, 237–273. Cacioppo J. T., Berntson, G. G., & Klein, D. J. (1992). What is an emotion? The role of somatovisceral afference, with special emphasis on somatovisceral “illusions.” Review of Personality and Social Psychology, 14, 63–98. Campos, J. J., Walle, E. A., Dahl, A., & Main, A. (2011). Reconceptualizing emotion regulation. Emotion Review, 3, 26–35. Cannon, W. B. (1927). The James–Lange theory of emotions: A critical examination and an alternative theory. American Journal of Psychology, 39, 106–124. Clore, G. L., & Ortony, A. (1991). What more is there to emotion concepts than prototypes? Journal of Personality and Social Psychology, 60, 48–50. Cosmides, L., & Tooby, J. (2001). Evolutionary psychology and the emotions. In M. Lewis & J. M. Haviland-Jones (Eds.), Handbook of emotions (2nd ed., pp. 91–115). New York: Guilford Press. Damasio, A. (1994). Descartes’ error: Emotion, reason, and the human brain. New York: Putnam. De Rivera, J. (1984). The structure of emotional relationships. In P. Shaver (Ed.), Review of personality and social psychology 5: Emotions, relationships, and health (pp. 116–145). Beverly Hills, CA: Sage. Duffy, E. (1941). An explanation of “emotional” phenomena without the use of the concept “emotion.” Journal of General Psychology, 25, 283–293. Edwards, D. (1999). Emotion discourse. Cultural Psychology, 5, 271–291. Ekman, P. (1972). Universals and cultural differences in facial expressions of emotion. In J. Cole (Ed.), Nebraska Symposium on Motivation (pp. 207–283). Lincoln: University of Nebraska Press. Ekman, P. (2003). Emotions revealed. London: Weidenfeld & Nicolson. Ekman, P., & Friesen, W. V. (1978). The Facial Action Coding System. Palo Alto, CA: Consulting Psychologists Press. Ekman, P., Sorenson, E. R., & Friesen, W. V. (1969). Pan-cultural elements in facial displays of emotion. Science, 164, 86–88. Ellsworth, P. C., & Tourangeau, R. (1981). On our failure to disconfirm what nobody ever said. Journal of Personality and Social Psychology, 40, 363–369. Fernández-Dols, J. M., Sánchez, F., Carrera, P., & Ruiz-Belda, M. A. (1997). Are spontaneous expressions and emotions linked? An experimental test of coherence. Journal of Nonverbal Behavior, 23, 163–177. Fogel, A. (1993). Developing through relationships: Origins of communication, self, and culture. Chicago: University of Chicago Press. Forgas, J. P. (1995). Mood and judgment: The Affect Infusion Model (AIM). Psychological Bulletin, 117, 39–66. Fridlund, A. J. (1991). The sociality of solitary smiles: Effects of an implicit audience. Journal of Personality and Social Psychology, 60, 229–240. Fridlund, A. J. (1994). Human facial expression: An evolutionary view. San Diego, CA: Academic Press. Frijda, N. H. (1986). The emotions. Cambridge, UK: Cambridge University Press. Frijda, N. H. (1993). The place of appraisal in emotion. Cognition and Emotion, 7, 357–387. Frijda, N. H. (2005). Emotion experience. Cognition and Emotion, 19, 473–497.

Emotion Theories 105 Gibson, J. J. (1979). The ecological approach to visual perception. Boston: Houghton-Mifflin. Griffiths, P. E. (1997). What emotions really are. Chicago: University of Chicago Press. Gross, J., & Thompson, R. A. (2007). Emotion regulation: Conceptual foundations. In J. Gross (Ed.), Handbook of emotion regulation (pp. 3–26). New York: Guilford Press. Hamann, S. (2012). Mapping discrete and dimensional emotions onto the brain: Controversies and consensus. Trends in Cognitive Sciences, 16(9), 458–466. James, W. (1898). The principles of psychology (Vol. 2). London: Macmillan. Joormann, J., & Gotlib, I. H. (2006) Is this happiness I see? Biases in the identification of emotional facial expression in depression and social phobia. Journal of Abnormal Psychology, 115, 705–714. Kappas, A. (2011). Emotion and regulation are one! Emotion Review, 3, 17–25. Keltner, D., Ellsworth, P. C., & Edwards, K. (1993). Beyond simple pessimism: Effects of sadness and anger on social perception. Journal of Personality and Social Psychology, 64, 740–752. Kreibig, S. D. (2010). Autonomic nervous system activity in emotion: A review. Biological Psychology, 84(3), 394–421. Krumhuber, E., Kappas, A., & Manstead, A. S. R. (2013). Effects of dynamic aspects of facial expressions: A review. Emotion Review, 5, 41–46. Laird, J. D., & Bresler, C. (1992). The process of emotional experience. In M. S. Clark (Ed.), Emotion (Review of Personality and Social Psychology, Vol. 13, pp. 213–234). Thousand Oaks, CA: Sage. Lambie, J. A., & Marcel, A. J. (2002). Consciousness and the varieties of emotion experience: A theoretical framework. Psychological Review, 109, 219–259. Lazarus, R. S. (1982). Thoughts on the relations between emotion and cognition. American Psychologist, 37, 1019–1024. Lazarus, R. S. (1991). Emotion and adaptation. New York: Oxford University Press. Leventhal, H., & Mace, W. (1970). The effect of laughter on evaluation of a slapstick movie. Journal of Personality, 38, 16–30. Leventhal, H., & Scherer, K. R. (1987). The relationship of emotion and cognition: A functional approach to a semantic controversy. Cognition and Emotion, 1, 3–28. Lewis, M. D. (1996). Self-organizing cognitive appraisals. Cognition and Emotion, 10, 1–15. Lewis, M. D. (2005). Bridging emotion theory and neurobiology through dynamic systems modeling. Behavioral and Brain Sciences, 28(2), 169–194. Lewis, M. D., & Liu, Z. (2011). Three time-scales of neural self-organization underlying human basic and non-basic emotions. Emotion Review, 3, 416–423. Lindquist, K. A., Wager, T. D., Kober, H., Bliss-Moreau, E., & Barrett, L. F. (2012). The brain basis of emotion: A meta-analytic review. Behavioral and Brain Sciences, 35, 121–143. Lutz, C. A. (1988). Unnatural emotions: Everyday sentiments on a Micronesian atoll and their challenge to western theory. Chicago: University of Chicago Press. Mandler, G. (1975). Mind and emotion. New York: Wiley. Manstead, A. S. R., & Fischer, A. H. (2001). Social appraisal: The social world as object of and influence on appraisal processes. In K. R. Scherer, A. Schorr, & T. Johnstone (Eds.), Appraisal processes in emotion: Theory, methods, research (pp. 221–232). New York: Oxford University Press. Marr, D. (1982). Vision: A computational investigation into the human representation and processing of visual information. New York: Freeman. Matsumoto, D., & Willingham, B. (2006). The thrill of victory and the agony of defeat: Spontaneous expressions of medal winners of the 2004 Athens Olympic Games. Journal of Personality and Social Psychology, 91, 568–581. Mauss, I. B., Levenson, R. W., McCarter, L., Wilhelm, F. L., & Gross, J. J. (2005). The tie that

106

MENTAL STATE THEORIES

binds? Coherence among emotion experience, behavior, and physiology. Emotion, 5, 175–190. Moors, A. (2009). Theories of emotion causation: A review. Cognition and Emotion, 23, 625– 662. Moors, A. (2013). On the causal role of appraisal in emotion. Emotion Review, 5, 132–140. Moors, A., De Houwer, J., & Eelen, P. (2004). Automatic stimulus-goal comparisons: Support from motivational affective priming studies. Cognition and Emotion, 18, 29–54. Neumann, R. (2000). The causal influences of attributions on emotions: A procedural priming approach. Psychological Science, 11, 179–182. Newman, P. L. (1964). Religious belief and ritual in a New Guinea society. American Anthropologist, 66, 257–272. Niedenthal, P. M. (2007). Embodying emotion. Science, 316, 1002–1005. Oatley, K. (1992). Best laid schemes: The psychology of emotions. Cambridge, UK: Cambridge University Press. Ortony, A., Clore, G. L., & Collins, A. (1988). The cognitive structure of emotions. New York: Cambridge University Press. Ortony, A., & Turner, T. J. (1990). What’s basic about basic emotions? Psychological Review, 97, 315–331. Parkinson, B. (1995). Ideas and realities of emotion. London: Routledge. Parkinson, B. (1996). Emotions are social. British Journal of Psychology, 87, 663–683. Parkinson, B. (1997). Untangling the appraisal–emotion connection. Personality and Social Psychology Review, 1, 62–79. Parkinson, B. (1999). Relations and dissociations between appraisal and emotion ratings of reasonable and unreasonable anger and guilt. Cognition and Emotion, 13, 347–385. Parkinson, B. (2001). Putting appraisal in context. In K. R. Scherer, A. Schorr, & T. Johnstone (Eds.), Appraisal processes in emotion: Theory, methods, research (pp. 173–186). New York: Oxford University Press. Parkinson, B. (2005). Do facial movements express emotions or communicate motives? Personality and Social Psychology Review, 9, 278–311. Parkinson, B. (2009). What holds emotions together? Meaning and response coordination. Cognitive Systems Research, 10, 31–47. Parkinson, B. (2013). Comment: Journeys to the center of emotion. Emotion Review, 5, 180– 184. Parkinson, B., Fischer, A. H., & Manstead, A. S. R. (2005). Emotion in social relations: Cultural, group, and interpersonal processes. Philadelphia: Psychology Press. Parkinson, B., & Manstead, A. S. R. (1992). Appraisal as a cause of emotion. In M. S. Clark (Ed.), Emotion and social behavior (Review of Personality and Social Psychology, Vol. 13, pp. 122–149). Newbury Park, CA: Sage. Reisenzein, R. (1983). The Schachter theory of emotion: Two decades later. Psychological Bulletin, 94, 239–264. Reisenzein, R., Bördgen, S., Holtbend, T., & Matz, D. (2006). Evidence for strong dissociation between emotion and facial displays: The case of surprise. Journal of Personality and Social Psychology, 91, 295–315. Reisenzein, R., Studtmann, M., & Horstmann, G. (2013). Coherence between emotion and facial expression: Evidence from laboratory experiments. Emotion Review, 5, 16–23. Roseman, I. J., Wiest, C., & Swartz, T. S. (1994). Phenomenology, behaviors, and goals differentiate discrete emotions. Journal of Personality and Social Psychology, 67, 206–221. Russell, J. A. (1991). In defense of a prototype approach to emotion concepts. Journal of Personality and Social Psychology, 60, 37–47.

Emotion Theories 107 Russell, J. A. (2003). Core affect and the psychological construction of emotion. Psychological Review, 110, 145–172. Sarbin, T. R. (1986). Emotion and act: Roles and rhetoric. In R. Harré (Ed.), The social construction of emotions (pp. 83–97). Oxford, UK: Blackwell. Sartre, J.-P. (1962). Sketch for a theory of the emotions. London: Methuen. Scarantino, A., & Griffiths, P. (2011). Don’t give up on basic emotions. Emotion Review, 3, 444–454. Schachter, S. (1964). The interaction of cognitive and physiological determinants of emotional state. In L. Festinger (Ed.), Advances in experimental social psychology (Vol. 1, pp. 49–80). New York: Academic Press. Scherer, K. R. (1993). Neuroscience projections to current debates in emotion psychology. Cognition and Emotion, 7, 1–41. Scherer, K. R. (2001). Appraisal considered as a process of multilevel sequential checking. In K. R. Scherer, A. Schorr, & T. Johnstone (Eds.), Appraisal processes in emotion: Theory, methods, research (pp. 92–120). New York: Oxford University Press. Schwarz, N., & Clore, G. L. (1988). How do I feel about it? Informative functions of affective states. In K. Fiedler & J. P. Forgas (Eds.), Affect, cognition, and social behavior (pp. 44–62). Toronto: Hogrefe. Solomon, R. C. (1976). The passions: Emotions and the meaning of life. New York: Anchor Press/Doubleday. Storbeck, J., Robinson, M. D., & McCourt, M. E. (2006). Semantic processing precedes affect retrieval: The neurological case for cognitive primacy in visual processing. Review of General Psychology, 10, 41–55. Tiedens, L. Z., & Leach, C. W. (Eds.). (2004). The social life of emotions. New York: Cambridge University Press. Tomkins, S. S. (1981). The role of facial response in the experience of emotion: A reply to Tourangeau and Ellsworth. Journal of Personality and Social Psychology, 40, 355–357. Valins, S. (1966). Cognitive effects of false heart-rate feedback. Journal of Personality and Social Psychology, 4, 400–408. Vytal, K., & Hamann, D. (2010). Neuroimaging support for discrete neural correlates of basic emotions: A voxel-based meta-analysis. Journal of Cognitive Neuroscience, 22, 2684–2885. Wittgenstein, L. (1953/2001). Philosophical investigations. Oxford, UK: Blackwell. Zajonc, R. B. (1980). Feeling and thinking: Preferences need no inferences. American Psychologist, 35, 151–175.

6 Motivational Theories David Dunning

A

nyone who has played a long night of poker has experienced tilt. Tilt is that emotional state that engulfs a poker player after he or she has just unexpectedly lost a large bet due to bad luck. Players “on tilt” find themselves making large and risky bets to make up for their unanticipated loss as quickly as possible, often playing much too aggressively for their own good. Better poker players strive to control tilt. Even better players learn to take advantage of tilt in others. The best players learn to manipulate others into that state. In a sense, the many facets of tilt demonstrate how motivational states can influence the thoughts, emotions, and actions of people. By motivation, I refer to psychological states that energize and direct the human organism toward some goal, either in thought or deed, or both. The term is derived from the Latin verb movere, which translates into modern English as “to move.” The state of tilt often leads people to be motivated to try to win back their money as quickly as they can, often paradoxically “moving” them toward decisions that circumvent that goal. In more common parlance, motivation can refer either to the overall desire to reach some end state (e.g., She wants better grades) or the specific reason why the person moves to pursue that end state (e.g., She wants to impress her father with them). The term motive can also refer to specific reasons or incentives that energize a person to achieve some 108



Motivational Theories 109

end state. As such, I will use the terms motive and motivation interchangeably when referring to those specific bases for action. The issue of motivation pervades psychology in its social, personality, cognitive, educational, organizational, and health variants. Scholars in social psychology, for example, ask what motives inform, shape, and distort social beliefs (Kunda, 1990) down to even the realm of visual perception (Dunning & Balcetis, 2013). They also ask how people choose actions to help them achieve motivational aims they have identified in their everyday world (Fishbach & Ferguson, 2007; Locke & Latham, 1990). In organizational, educational, and health psychology, researchers and practitioners ask how to motivate employees to work harder and smarter (Pinder, 2010), students to apply more effort toward learning (Dweck, 1999; Elliot & McGregor, 2001), and patients to adhere more strictly to medical regimens (Rogers, 1975; Rosenstock, 1966) through the wise and judicious use of incentives, rewards, and motivational appeals. Motivations can be automatic in that some circumstances lead people to direct their thoughts and actions toward goals without much conscious intervention or effort (Ferguson, Hassin, & Bargh, 2008), such as when hunger and thirst make food or drink irresistible to an individual. The state of tilt is like this. People know that tilt leads to decisions they will regret later, but they still find themselves placing big bets on marginal hands. Motivation, however, can also be conscious, effortful, and strategic (Bandura, 1997). A soccer coach can discuss for long hours the goals and achievements he or she wants a team to reach, and then the coach and team can apply hours of practice to reach those goals.

What Motivation Explains Issues of motivation lie at the heart of the human organism. On the most basic level, one can merely point to the fact that no human would get out of bed in the morning unless there was something that organism wanted or needed to do. As such, to understand the human organism, one must understand the motivations the organism harbors. In all, there are two categories of research that scholars interested in motivation might pursue. The first is to try to determine and understand what motives govern human thought and action as people live their lives freely or ad lib. Making their own choices, embedded in their natural social environment, in what directions do people naturally and spontaneously take their thoughts and actions? What tactics or strategies do they use to try to succeed in the directions they set for themselves? The second type of research involves goals that the researcher or practitioner sets, with an eye toward motivating people in a direction he or she wants people to travel. In this avenue of research, the issue is not what people do naturally; rather, it is how much their motivation can be tapped to ben-

110

MENTAL STATE THEORIES

efit themselves or the groups around them. For example, how does a doctor motivate his or her patients to take their medicine or to work harder on their rehabilitation so that they can live self-sufficient lives sooner? How does a company make sure that it harnesses as much as it can from its employees’ talents and capabilities? With these considerations as background, it is clear that motivations are important to understand because they determine three basic aspects of thought and action: direction, persistence, and termination.

Direction First, for the researcher interested in people behaving in their ad lib lives, motivations direct which thoughts and actions a person will pursue, in that people will tend to think and act toward goals or states they which to attain. A person motivated toward a scholastic achievement will study rather than play sports. An athlete pursuing excellence may run up and down the stairs of football stands to enhance endurance, measuring her workout not in units of stairs or minutes, but in the number of “stadia” she has run rather than relax and socialize with others.

Persistence Second, for researchers interested in both the ad lib nature and potential engineering of motivation, motives determine just how long people will persist toward a goal, especially in the face of difficulty or challenge. The more strongly motivated person will work harder and longer even in the face of fatigue, boredom, or distraction. A motivated cook will try ingredient after ingredient in each iteration of his recipe until it is no longer improved rather than skimp with any cheap substitution. A writer dedicated to writing the great novel will scan the thesaurus just one last time to find the mot juste she needs to punctuate her story with feeling.

Termination Finally, motivations will determine when a person will terminate goaldirected behavior. People cease their goal-directed behavior once they believe they have satisfactorily achieved their goal. A student aiming for a “C” likely ends his studying before the student aiming for an “A.” A video game player trying mainly to get along with other players may practice just enough to be adequate at a game, but an individual going for the alltime highest score ever may continue practice well into every night, not stopping until he or she posts that highest score or decides that the goal is futile. Issues of persistence and termination are central to the understanding of motivation, but their study is often filed under different topic names. Some



Motivational Theories 111

researchers might file work on persistence and termination under the topic of self-control (Duckworth, 2011; Tangney, Baumeister, & Boone, 2004) or selfregulation (Baumeister & Vohs, 2004; Carver & Scheier, 2001). Such researchers may focus on the circumstances that allow people to persist rather than give up (e.g., Baumeister, Vohs, & Tice, 2007). Alternatively, they might focus on explicit tactics that make self-regulation more likely to succeed (e.g., Gollwitzer, 1999).

The Problem of Identification Whichever facet of motivation a scholar is most interested in, the one theoretical issue that causes the most thought and controversy is identification. Identification refers to recognizing which particular goal or motive, out of the many that are usually possible in any situation, are in play to move a person’s behavior. Often, whether or not a person persists or terminates depends importantly on what that person’s specific motivation is. As such, identifying the goal turns out to be the most important part of any motivational analysis. Correctly identifying the specific goal can be difficult. Consider the following: According to neoclassical economics, people are motivated to maximize their wealth (Becker, 1976; Pareto, 1909/1971). As such, on days in which they can gain money easily they should be more motivated to engage in economic activity. For example, cab drivers should be motivated to work longer hours on those days when there are plenty of tourists walking along the sidewalk looking for a cab. However, gaining money may not be the actual goal that governs a cabbie’s behavior. One specific study has shown that cab drivers in a major metropolitan area quit earlier on those days in which potential passengers are plentiful, relative to days in which fares are fewer and farther between. Apparently, the typical cabbie’s motive was not to enhance his or her financial wealth as much as possible, or to practice the virtue of hard work. Rather, these cabbies’ primary motivation seemed to have been to make just enough money so that they could live at some level of comfort, and then quit to enjoy that comfort immediately (Camerer, Babcock, Loewenstein, & Thaler, 1997). That is, the question of persistence and termination among cabbies depended importantly on identifying the exact goal those cabbies were pursuing However, therein lies the trick. Ask any 6-year-old why Johnny is chasing the neighbor’s dog and the child is likely to answer “Because he wants to.” Motivation is easily invoked as a cause in human behavior—perhaps too easily, for there is evidence that people are skilled at identifying motivations and intentions that are simply not there (Barrett & Johnson, 2003; Heider & Simmel, 1944). However, although easily asserted, verifying the operation of a particular motive in any scientifically conclusive way, even the most obvious one, can be vexingly difficult.

112

MENTAL STATE THEORIES

Errors in Affirming the Consequent One reason that motivations are easy to invoke but difficult to document is that motivational analyses are often the ones that entice theorists, both armchair and professional, into the fallacy of affirming the consequent (Lacey, 1996), as described in this volume by Gawronski and Bodenhausen (Chapter 1, this volume). People see behavior Y, and use that as evidence to infer some Y-motive that closely resembles it. One sees this very often, for example, in the area of self-serving judgment. Researchers see people make some self-serving claim, and argue those claims as direct evidence that people are motivated to bolster their sense of personal worth. People, for example, describe themselves as good leaders, parents, ethicists, and car drivers much more than can logically be true (Dunning, Heath, & Suls, 2004). This can lead theorists to the argument that people must be unreasonably motivated to see themselves in a rosy light—that the presence of unrealistically positive selfimpressions must be caused by a similarly unreasonable motivation to see the self as positive, that unrealistic X must be evidence of motive X. To be sure, such rosy self-impressions might be the product of a motive toward maintaining self-worth, but that is not the only positive view of themselves that people might harbor (Chambers & Windschitl, 2004). They might simply believe the flattery that other people heap on them to their face (Jones, 1964) and remain innocent of what those same people say behind their back. Or, they may simply not be skilled enough to identify flaws in their competence and character (Kruger & Dunning, 1999). Thus, although self-serving behavior X might be indicative of a motive X, one cannot decisively assume it. That is, in the usual causal terminology, the motive toward self-esteem might be a sufficient cause of self-serving behavior X at least some of the time in that the emergence of the motive produces the behavior associated with it. However, that motive may still fail to be a necessary cause of that behavior. By necessary cause, I mean that seeing the behavior must mean that the motive preceded it.

Methods to Implicate Motivation Indeed, one cannot even claim that motive X is a cause at all, given that behavior X can be produced via other means. Thus, to even establish that motive X is a cause, one must go further than merely observing behavior X. Common methods exist for going further and successfully establishing a role for motives in producing behavior. These strategies are essential to show that it is motivation, and not some nonmotivational process, that is the instigator.

Depriving versus Satisfying the Motive The first method is to make a motivation acute by depriving an individual of fulfilling some goal. For example, if one wants to establish a hunger motiva-



Motivational Theories 113

tion, one has to deprive a person of food (within ethical bounds, of course). If one believes that people are motivated to maintain their self-esteem, one must create a situation that deprives the person temporarily (within ethical bounds, of course) of the certainty of their self-esteem. A small failure experience on a task that matters to the participants, for example, often suffices (see Dunning, Leuenberger, & Sherman, 1995). If a person reacts by redoubling their motivational efforts—toward food or toward reaffirming their selfesteem—one has evidence of the motive in question. The second method, of course, is the opposite. Satisfy some motive and the person should become less energized to pursue more of it. For example, people will turn away a delectable dessert if they have already had too much to eat. In addition, asking people to complete a self-affirmation exercise in which they describe some aspect of themselves that makes them feel proud tends to quell any subsequent effort people may show to gain self-esteem (Sherman & Cohen, 2006). Indeed, they may even be more open to experiences that threaten their sense of self-worth (Cohen, Aronson, & Steele, 2000; Sherman, Nelson, & Steele, 2000).

Equifinality In a sense, this second method takes advantage of a common feature of motivation—that it is equifinal. Often, any number of actions may allow a person to fulfill a specific goal. Thus, if a person fails to follow an obvious course of action to fulfill a goal, one should not assume that the motivation is absent. Instead, the individual may fulfill that goal by some other behavioral means. For example, let’s suppose that the researcher wants to see if people are motivated to signal to others that they are altruistic individuals, and so gives a person a chance to generously donate $5 to some charity in the lab. The experimenter gives the participant a chance to share the money and finds that the participant demurs. Perhaps the participant fails to have the motive hypothesized, or instead it could be that the participant has already fulfilled the goal of affirming a positive self-identity by helping a little old lady across the street on the way to the laboratory, or by staying up until three in the morning helping another person with their conversational Spanish. Thus, motives may be satisfied by any number of means, and the researcher who limits his or her search to only a subset of those means may miss their operation.

Sufficiency versus Necessity One important note must be made about these two common methods of establishing motivational influence. Both are designed specifically to show that motivation is a sufficient cause of some behavior, but they do not show that it is a necessary cause. That is, making people unsure of their self-esteem might be adequate to make them boast about themselves, but it may not be the only reason that people boast. Other motives or psychological pro-

114

MENTAL STATE THEORIES

cesses can produce the same behavior. For example, sometimes people boast merely to impress others. Thus, establishing necessity requires a different research logic. A claim of necessity means that some behavior must have been preceded by a specific cause. Thus, to establish it, one would have to somehow show that some behavior is uniformly preceded by a deficient motivational state assumed to cause that behavior. For example, if a theorist wants to claim that persons boast about their attributes only out of a need to maintain self-esteem, that theorist would have to follow individuals as they carry out their lives to observe situations in which they went out of their way to boast about themselves. Are those episodes preceded by periods of uncertain or lowered self-esteem? If so, that would be evidence of necessity. That is, if boasting is always preceded by a period of low self-esteem, that is evidence that low self-esteem is necessary to prompt an episode of boasting.

The Temptation toward the Nominal Fallacy Upon further reflection, motivational inferences often tend to fail as explanations for phenomena. To say that a person does X because he or she wants to may seem, at first blush, like an explanation, but upon further thought it is merely a variant of the nominal fallacy (Gawronski & Bodenhausen, Chapter 1, this volume; Lacey, 1996). In the nominal fallacy, a person provides a pseudoexplanation for a phenomenon by simply redescribing it. Adding a motivation for a phenomenon seems to be just a slightly more advanced way of committing this fallacy. Although the behavior itself is not simply renamed, some underlying state of desiring to perform that behavior is named in its stead, without much of a gain in understanding. This type of theorizing is, however, seen in the behavioral sciences. For example, if an experimenter gives a participant $10 in the laboratory and gives that participant an opportunity to share any or all of it with another participant, the first participant often gives half, $5, to that second participant. Economists often assume that people share money with others, even when they are under no obligation to do so, because they are motivated, or more specifically have a taste for, bettering the circumstances of others. They are motivated to enhance the wealth of everyone, just not themselves (Rabin, 1993). X is explained by asserting a motivation for X. That may be the case, but the single act of giving the $5 can signify any number of other motivations. Perhaps people share because they think it is unfair that they receive all the money and they want to reestablish a fair situation closer to their ethical code (Fehr & Schmidt, 1999). Or perhaps they feel guilty about keeping all the money and want to dispel that guilt by giving some money away (Cialdini et al., 1987). Or perhaps they want to signal to the experimenter or the other person that they are giving persons (Milinski, Semmann, & Krambeck, 2002). Or maybe they want to signal to themselves that they are terrific individuals (Bodner & Prelec, 2003). Or maybe they just



Motivational Theories 115

think they are expected to give the $5. Note how all of these motivations are fulfilled by the same act of giving the $5.

Multifinality Thus, a central and often vexing problem facing the motivational theorist is identifying which single motivation or potential combination of motives actually instigates the action that they observe. The problem with behaviors is that they are multifinal. A single action can fulfill any number of motives and bring about any variety of outcomes that matter. Thus, it can be misleading to assume that the primary motive in play is the one associated with the most obvious outcome achieved. Research on volunteering in real-world settings most explicitly shows that any number of motivations underlie why people work for charities and service organizations. Some people want to make a difference. Others want to make friends. Still others want to burnish their resumes for future jobs. Others engage to understand the world better. Such diverse motivations—all producing the same behavior—matter in that they determine what specific choices people will make in volunteering, how long they will volunteer, and what appeals can be made to induce them to participate (Clary & Snyder, 1999).

Investigating Scope How is one to avoid the temptation toward the nominal fallacy? There are at least two different ways to do so. One is to consider the “scope” of the motive that one is considering. In the $10 example above, are people motivated specifically to create a fair situation, and thus are much more willing to give $5 and $2 to another person because the former donation creates a fair situation and the latter does little to alleviate the situation? If so, the scope of the motivation is narrow and concerns specifically producing a fair outcome. The scope of the motivation may be much broader, however. Instead, people may wish to signal that they are giving people. If so, having them give to a charity before they even face the $10 decision might prompt them to be less generous to the other person, indicating that they have a broader goal of signaling that they are a generous person. Or maybe the scope of the motive is broader still: People may just want to show that they are wonderful people by whatever means, so that succeeding on some important intellectual puzzle before reaching the $10 decision might make people less generous, thus showing that the primary motive is to burnish a more general self-image. The two empirical strategies described above, examining the effects of depriving versus satisfying a person’s motive, can often be exploited to assess the “scope” of the motivation in play—that is, to see how general the goal is that the human organism wishes to fulfill. For example, consider

116

MENTAL STATE THEORIES

esteem goals. If students fail an important math class how specific or general is the sense of self they wish to reaffirm? Do they become motivated to reestablish their academic self in general, so that a success in an English class would do? Or, do they perhaps want to repair an even more specific self, such as their math academic self, and only success in math will enhance esteem? Or, do they just want to reestablish an even broader positive self, and so scoring the winning goal for their intramural soccer team will do to reestablish a satisfactory sense of self. In the case of self-worth, the data suggest that people are motivated to maintain a general sense of self rather than anything more specific (Sherman & Cohen, 2006; Tesser, 2000). If a threat to self-esteem looms in one area of life (e.g., the classroom), a success elsewhere in any other area, no matter how irrelevant to the first (e.g., being elected to a leadership position in a social group), is sufficient to repair self-esteem. The scope of self-esteem motives appear to be about the self in general, not about specific components of the self. For example, making people fail on an intellectual task causes them to bolster their beliefs that they will succeed in marriage, suggesting that their motive is a general one (Dunning et al., 1995). In addition, having a person affirm his pride in his family life lowers his defenses when he hears about his risks for HIV infection (Sherman et al., 2000). When it comes to positive views of self, the “scope” of the motive seems to be a general one. People are motivated to maintain a general view that they are wonderful people, and so they can allow for specific instances that contradict that view just as long as some other self-view is affirmed.

Investigating Varieties of Motives The issue of avoiding the nominal fallacy, however, might be better served by reminding one’s self that many behaviors are multifinal and that the true goal of any investigation is to identify the stew of motives that might be satisfied by a behavior rather than trying to find the “one true” goal that produces it. Consider all the categories of motivations that may wrap around a single act that a person pursues. If one sits down and enumerates all the “flavors” of motives that might plausibly be relevant, one finds a theoretical pursuit that travels directly away from any nominal fallacy.

Instrumental versus Expressive Why do people vote, particularly in national campaigns? Their single vote has no hope of determining the outcome of an election, so why do they do it? More to the point, why are they more likely to vote in larger elections in which they will have no impact (i.e., national elections) than in elections (i.e., local) in which their vote has a much higher chance of being the one that tips the election? This pattern of voting is a paradox if one considers voting to be purely instrumental—that is, a behavior designed to bring about a particu-



Motivational Theories 117

lar outcome, such as one’s candidates winning, along with the politics and policies associated with those candidates. These outcomes are downstream in that they having nothing to do with the action of “voting” per se, but are outcomes they may obtain because one has voted, although those outcomes are not certain to follow from one’s vote. But voting is not only instrumental, in the sense of possibly bringing about some downstream outcome. It is also expressive in that the act of voting itself produces immediate outcomes just by the very nature of the act itself (Dunning & Fetchenhauer, 2013). By voting, people produce a host of immediate outcomes for themselves. They get to express their political values and reaffirm those values as an important part of themselves. They may get a rush of civic pride and a sense of pleasure in fulfilling an important social obligation. They get to wear “I voted” stickers that give them some prestige among their peers. They also have a chance to share an event with others, bringing themselves closer to other citizens (Brennan & Hamlin, 1998). Understanding whether an action is instrumental versus expressive can be important. Consider crime, which can be motivated by expressive or instrumental dynamics. Sex crimes, for example, are almost always expressive. Committing the crime itself brings an immediate pleasure, fulfillment, or reward that prompts the criminal to do it. Paying a person to do it, an instrumental reward, would likely not increase the likelihood that they would. Embezzlement, however, is almost never expressive. People do not embezzle in order to achieve some immediate intellectual or emotional release. Rather, they embezzle to gain the downstream, instrumental benefit of more money (Chambliss, 1966). If the money is not attached to the embezzlement, people will not do it. The expressive versus instrumental nature of crime matters in that attaching additional instrumental concerns to a crime—such as heavy fines or longer prison sentences—has a larger deterrent effect on instrumental crimes than on expressive ones. Thus, acts such as white-collar crime, forgery, career thievery, and parking law violations are deterred by harsher legal sanctions. Acts such as murder and those associated with drug addiction are more resistant to deterrence via tougher sanctions. For these crimes, the rewards are immediate, and downstream consequences fail to have any impact on their incidence (Chambliss, 1966).

Intrinsic versus Extrinsic A closely related distinction involves intrinsic versus extrinsic motivation. When one is intrinsically motivated to perform an activity, that means that one derives pleasure from the action itself. A person who is intrinsically motivated to play soccer, for example, simply enjoys playing the game. People are extrinsically motivated when there exists some external motivator for their action. A person might not necessarily enjoy playing soccer but may do so to please a parent or to get a college scholarship (Lepper, Greene, & Nisbett, 1973).

118

MENTAL STATE THEORIES

At first, the intrinsic/extrinsic distinction might sound much like the expressive/instrumental one, and there is some resemblance between the two distinctions. The two, however, do not completely overlap. A person can be expressively motivated, yet by extrinsic concerns. For example, a person who votes in order to wear that “I voted” sticker is doing so for an expressive reason in that the act of voting itself allows people to wear that reputationenhancing sticker. However, the person has not voted for the joy of doing so, which is the hallmark of intrinsic motivation. Rather, the person’s concern is in building up his or her social standing.

Proximal versus Distal Motivations can also explicitly exist along a dimension of time. Some dynamics are proximal in that they operate in the here and now, influencing a person’s motivation over a short time horizon. Other dynamics are more distal in that they influence behavioral patterns over longer stretches of time and across situations (Kanfer, 1992). An individual’s overall need for achievement might be one contributor to motivation at the distal end of the time spectrum (Kanfer, 1992; Kirk & Brown, 2003). At the proximal end, just articulating a goal explicitly can enhance effort and achievement in that it publicly creates a commitment that one wants to fulfill (Bandura & Schunk, 1981). Another name for this dimension rests on the distinction between immediate motives versus ultimate ones. Goals completed in the immediate now (e.g., a perfect haircut) can also be aimed at or relevant to more ultimate aims (e.g., making one physically attractive to other people in order to catch mates).

Approach versus Avoidance A chocolate cake and a snake are both motivating stimuli, but for different reasons and in different directions. A cake, like other pleasant objects, produces approach goals. People are energized because they have something desirable they can gain. A snake, on the other hand, is usually an undesirable object. Snakes produce avoidance goals in that people apply effort to avoid them. The approach/avoidance distinction in motivation has a long history and has been discussed under many guises (Elliot, 2006; Elliot & McGregor, 2001). In its most recent variant, Higgins and colleagues, in their regulatory focus theory, have described a promotion versus prevention distinction that sounds much like approach and avoidance (Higgins, 1997; Higgins et al., 2001). A promotion orientation is one toward possible gains and rewards. One is motivated by the size or availability of possible benefits to action. In contrast, avoidance goals are about evading loss. As such, one is more motivated by any possible penalties or risks to action. Studies show that people differ in their orientation toward promotion and prevention (Higgins et al., 2001). For example, Lee and Aaker (2004) discovered that people were more



Motivational Theories 119

motivated to drink a new grape juice product when the advertisement fit their usual orientation. Promotion-oriented individuals were more compelled by being told that the drink enhanced their energy, whereas prevention-oriented folk were more persuaded by ads that emphasized the drink’s disease-fighting characteristics. Researchers should recognize, at least in the realm of choice, that the approach/avoidance distinction means that there is an ambiguity whenever people face a choice between two courses of action. People make a choice, but is that because they are approaching one option or are merely avoiding the alternative? This ambiguity, it turns out, is sometimes an important one to resolve. Consider the decision to trust another person. In my lab, my colleagues and I have found that people trust complete strangers with money to an extent that cannot occur according to the tenets of neoclassical economics. In my experiment, people are asked if they want to give $5 to another person— a complete stranger—who can either decide to keep the money or give it back with a profit. Some people think they are slightly more likely to have their trust violated than honored, with the other person leaving with the participant’s money, yet a majority of participants in our experiment choose to trust that complete stranger (Fetchenhauer & Dunning, 2009). Why do they do it? How can a majority of people consider trusting another person to be attractive when their expectations about their peers are so cynical? Recent work suggests an answer. It is not that people harbor an approach motivation toward trusting another person. Rather, it appears that they have an avoidance motivation. They want to avoid the act of not trusting because forgoing trust makes them feel guilty, remorseful, tense, and anxious. In order to avoid these unpleasant feelings, they go ahead and hand their $5 over to a complete stranger, thinking more often than not that they will never see the money again. We have found that the degree of guilt and anxiety that people avoid by giving the money is by far the best predictor of who is trusting and who is not trusting, with such an avoidance goal trumping any economic variable that may be in play (Schlösser, Fetchenhauer, & Dunning, 2012). The approach/avoidance distinction complicates motivational analyses in that there are many different ways in which an action may be approach versus avoidant. Researchers can talk about a motivation in a hedonic sense. An outcome may be approach-oriented in that it evokes positive emotions such as pleasure or excitement, or it may be avoidant-oriented because it brings negative emotions to the fore. Or an outcome may be approach-oriented because it activates the organism to come closer to some goal regardless of emotion in the immediate sense, or avoidant because it inspires the organism to retreat. Or an outcome may be an approach-oriented one because the organism’s ultimate, and not immediate, goal is to bring the outcome closer to it. Or an outcome may be avoidant—in the ultimate sense—because the organism is energized to escape it.

120

MENTAL STATE THEORIES

These types of approach and avoidance tendencies need not be congruent with one another (Krieglmeyer, De Houwer, & Deutsch, 2011). A person might get angry at another person (avoidant in the hedonic or emotional sense) and approach that person out of aggression and hostility (approach in the immediate sense), with the ultimate goal of ridding herself of the other person forever (avoidant in the ultimate sense). Or, in dating, people may play “hard to get” (avoidant in the immediate sense) in order to attract the other person in the ultimate sense.

Homeostatic versus Maximizing Often motives also present a question about the nature of their fulfillment. Some motives are merely about returning the organism to some level or state. In that sense, they are homeostatic. Much like a thermometer is “motivated” to heat a house only when the temperature grows cold and ceases operation when some temperature level is achieved, homeostatic motivations are evoked and cease only when deviations from some benchmark are detected. In this way, psychological motivations operate metaphorically like biological ones. People eat until they feel full. They do not continue eating merely because food is available (Hull, 1943). Other motives can be construed as maximizing in that the organism continues to be motivated to gain more and more (or less and less, depending on circumstances). This is no set-point at which the drive to accumulate some object is satiated. Economics tends to work on such principles. Firms are not homeostatic in their desire to drive profits ever upward (Becker, 1976; Pareto, 1909/1971). Whatever level of profitability they reach, they are dutybound to do what they can to reach some higher level. In the identification of human motives, one can address the issue of whether the motive in play is homeostatic or maximizing. Take the acquisition of self-esteem. In American culture, at the very least, people are motivated to think well of themselves and to consider themselves people of some worth. But is this a homeostatic motivation or a maximizing one? People clearly show concerns about their self-image—but are they motivated mostly to maintain a certain level of self-esteem, or are they motivated to enhance it as much as they can? Curiously, the literature has not really focused on this question, although there is extant research suggesting an answer. That answer is that people are homeostatic in their motivations to manage their self-esteem. For example, under threat, people show more selfserving biases in the attribution of success versus failure. Without threat, this tendency evaporates (Campbell & Sedikides, 1999). Similarly, people denigrate others to recapture self-worth just after they have failed but not after they have succeeded. This suggests that people engage in motivated social thought to regain self-esteem called into question, but in the absence of that question they are much less motivated to acquire higher levels of self-worth (Beauregard & Dunning, 1998).



Motivational Theories 121

In contrast, individual differences in motivations to gain knowledge and expertise seem to follow the logic of maximization. People who display the most curiosity about a topic, and thus pay to learn more, tend to be those who already know more (Loewenstein, 1994). People more willing to buy a self-improvement book on emotional intelligence tend to be those who have just done well on a test on the ability. Those who do poorly show little motivation to buy the book (Sheldon, Ames, & Dunning, 2013).

Explicit versus Implicit Another distinction, increasingly important, is between explicit and implicit motives. Explicit motives are those that people are aware of, with people applying mental effort to strategize and work to fulfill them. Implicit motives guide effort and behavior without the person’s awareness of their operation (Schultheiss & Brunstein, 2010). How implicit motives operate is a question that is wide open for research and inquiry. To be sure, there already is a long tradition of measuring individual differences in implicit motives toward achievement, affiliation, and power by assessing the stories people generate to explain pictures that researchers show them. More recently, social psychologists have begun to prime people acutely to have goals they have no awareness they harbor (Bargh, Gollwitzer, Lee-Chai, Barndollar, & Troetschel, 2001; Fishbach & Ferguson, 2007), such as doing well on the crossword puzzle lying on the table in front of them. However, researchers are quite far from completely understanding how such subterranean goals, operating underneath awareness, manage to interact, supersede, or wax and wane in the face of explicit motives, individual needs, and situational circumstances. For anyone studying motives at the implicit level, two facts are important to keep in mind. First, there is no reason to suspect that implicit and explicit motives correlate within the individual. Research has long documented the idea that a person’s level of some explicit motive need not be tightly related to with the level of that motive at an implicit level (McClelland, Koestner, & Weinberger, 1989). Implicit and explicit motives also need not influence the same outcomes. Implicit motives seem tied to spontaneous and natural behaviors responding to incentives in the environment. Explicit motives are related more to conscious decisions people reach to pursue a goal when asked to, for example, by another person (McClelland et al., 1989). As a consequence, it is not surprising that implicit and explicit motives often have little to do with one another. Second, if one is going to study implicit motives, one has to make sure it is motives that one is studying. Researchers can assume they are priming people with the goal of achievement by exposing them to words such as smart and study, but are they priming the goal or merely the semantic content of the words (Förster, Liberman, & Friedman, 2007)? To ensure that they are priming motives, researchers must ensure that their priming produces

122

MENTAL STATE THEORIES

effects associated with striving toward goals. They must show that the priming causes people to work better, longer, or differently at a task, that they will resume the task more energetically if they are interrupted, and that the effect of the prime increases over time, rather than decreases, until a goal is satisfied (Bargh et al., 2001; Chartrand & Barth, 1996).

On the Pursuit of a Master Motive The problem of identification becomes writ large when considering a type of theoretical pursuit in which researchers at times engage. That pursuit is to describe one primary motive that subsumes all others. That pursuit concedes that people display a variety of motives at the surface, but all those more superficial motives are just a manifestation of a deeper, more fundament über or master motivation. Several candidates for the master motivation have been offered through the years. Just recently, Heine and colleagues asserted that people, first and foremost, are driven by a search for meaning. They wish to dismiss the nonsensical or the chaotic, preferring instead the regular and coherent (Heine, Proulx, & Vohs, 2006). Terror management theorists disagree, stating that people are driven first and foremost by a desire to deny the inevitability of death, that much if not all of human behavior—including the building of culture and civilization—is due to a disavowal of individual mortality (Greenberg et al., 1990). Instead, sociometric theorists consider the primary motive to be one of belonging. People wish to be respected and valued members of their social groups, and fear the “social death” of ostracism (Leary, Tambor, Terdal, & Downs, 1995). Such a pursuit to identify the one motive underlying all others is seductive, but it may also be logically undecidable. Let us imagine that we believed Motive A to be the one that underlay all others and that could manifest itself as Motive B or C. For example, the motive for meaning might be the underlying motivation that could manifest itself as fear of death (and its suggestion for meaninglessness) or as a need for belonging (which gives a person a coherent place in the world). However, if Motive A can easily manifest itself as B or C, why think it as causally prior to the others? Instead, it would remain the same motive underneath, having different faces depending on the circumstances. Thus, why claim A as the primary motive when B or C could just as easily be the chief motive? Why not, instead, simply claim it as an “all of the above” motive that can manifest itself in different ways? There is a historical precedence for this observation. For the pre-Socratic philosophers, one primary question was the identity of the “basic stuff” that made up the universe. The Milesian school claimed that the world was made up of water, which transformed itself to earth or air. Diogenes asserted that the primary stuff of matter was air, which transformed itself to earth or water. Heraclitus assigned the role of primary matter to fire, which could transform



Motivational Theories 123

into air, water, or the rest (Lloyd, 1970). What these philosophers missed was that if it was the same stuff that constituted water, fire, air, or earth, why claim this stuff was essentially one of the four when it could manifest itself equally well as any of the four. Why not just call it “stuff” and leave it at that?

Finding Relations among Motives However, a researcher sometimes comes to a question of understanding the relationships between or among goals, and so has to ask if Motive A comes “prior to” or underlies Motive B. Empirical strategies are available that could potentially answer that question—even up to and including claims for a master motive. That method would be to see if there is asymmetry in the relation of one motivated pursuit to another. Namely, if threatening Motive B incites a person to pursue Motive A, but threatening A does little to inspire the motivated pursuit of B, one can make the claim that Motive A is really the motive of concern for the individual. Further, if alleviating Motive A also quells a need to satisfy Motive B, but alleviating B does little to alter the pursuit of A, one can similarly claim A to be the more general motive in that satisfying A also satisfies B. To illustrate this logic, consider two motives that have received a good deal of attention over the past few years: One is the need for coherent meaning (Heine et al., 2006), and the other is the need to dispel the threat of death (i.e., terror management) (Greenberg et al., 1990). Suppose a theorist thought that the need for meaning was more general or causally prior to the need to dispel the threat of death. If that is the case, bringing up the threat of death should evoke its causally prior motive about meaning, and people should strive harder to find meaning and coherence in whatever they next consider. The opposite should not emerge (i.e., making people feel incoherent should make them strive more to deny death). However, alleviating a need for meaning, which is the more general motive, should prevent people from defending themselves so much against the threat of death. The more general and causally prior motive has been dealt with, leading to less of a need to deal with any “submotive” associated with it. Thus, in this way, researchers can better ascertain the relation between two motives. To the extent that alleviating Motive A and B alleviates each other symmetrically, it is likely that they share a common underlying motive, but there is no reason to claim it is Motive A or B—just something in common. However, if asymmetries arise between the operation of Motive A and B, then one can make the claim that one motive is more general and causally prior to the other.

Finding Distinctions among Motives The methods described above may also be used to determine whether motives are distinct. To the extent that they are distinct, then depriving one

124

MENTAL STATE THEORIES

motive should do little to affect the other. More importantly, alleviating one motive should do little to alleviate the other. For example, Knowles and colleagues (2010) examined the relationship between the need for self-esteem and the need to belong. They found that threatening a person’s intelligence inspired many esteem-repairing behaviors, many of which had nothing to do with intelligence—reaffirming the idea that people harbor a general goal to maintain a positive sense of self (Sherman & Cohen, 2006; Tesser, 2000). However, threatening the sense of interpersonal belonging by asking participants to recall a time someone else had rejected them prompted them specifically to want to repair a sense that they were loved and accepted by the people around them. It did not inspire them to try to repair self-esteem instead by, for example, writing about how intelligent they were. Further, allowing them to extol their intellectual abilities did little to quell any threat they felt about being rejecting by others. In short, the need for belonging appeared to be distinct from a general need for self-esteem (Knowles, Lucas, Molden, Gardner, & Dean, 2010). One should be mindful, however, that motives may be distinct, yet still be related. Although separate and independent, they may still reside in a hierarchy in which one motive must be satisfied before moving on to pursue the next motive in the hierarchy. Maslow’s (1943) hierarchical scheme of motives fits this description in which people must first satisfy their physiological needs of food and sleep before moving to chase after the security of shelter, before next pursuing affiliation needs of friendship and intimacy, and so on. The method for documenting such a hierarchy would be straightforward. One would merely have to show that people are disinterested in one goal (e.g., affiliation) until another goal beneath it in the hierarchy (e.g., security) was satisfied. In addition, one would have to show that the opposite pattern does not emerge, that fulfilling some higher-level goal would stop people from attending to motives lying lower in the hierarchy.

Related Concepts The notion of motivation is so central to the study of psychology that it is crowded in by any number of related terms that form the core of their own theoretical treatments. The relation of motivation to these several other terms, some forming the topics of other chapters in this book, must be made explicit.

Function First, theories of motivation must be distinguished from theories of function (Tetlock & Fincher, Chapter 13, this volume). Like motivational theories, functional theories suggest that the thoughts and actions of people serve to push them toward desirable end states. For example, hunger motivates peo-



Motivational Theories 125

ple to eat, and eating serves the function of physical survival, allowing both individual and species to survive. Although both motivational and functional theories focus on the end states the human organism wishes to achieve, they do differ along several dimensions. First, motivational theories range from the very specific to the quite general. A person on tilt, for example, may be motivated specifically to win back that pot of money in the poker game, or more generally may be motivated to maintain a positive balance in his checking account, or even more generally to preserve his social reputation. When functions are invoked, they tend to be more general in definition. For example, Katz (1960) listed four general functions that people might serve by forming political and social attitudes. They might strive to make sense of their world, to maintain a positive self-image, or to express their heartfelt values, and to ingratiate themselves with like-minded people around them. Second, functional theories tend, either explicitly or implicitly, to focus on functions that help the human organism to adapt to its environment. Thus, not any function will do, but rather only those functions that actually enhance the well-being and survival of the individual. As such, functions do not necessarily exist as a goal for the individual. Rather, they are benefits that certain behaviors produce even if people have no intention of producing those benefits via their choices. For example, a person might start a neighborhood poker game with the goal of alleviating boredom, but that behavior may prove adaptive or functional in enhancing that person’s standing in the eyes of the neighborhood.

Habit One problem associated with the identification of motives is that they need not necessarily be present to induce behavior. Motivation prompts the human organism to initiate thought and behavior, but it is not the only psychological mechanism that can do so. Habits, too, can initiate. Habits are associations that humans build up between environmental stimuli and possible responses (Wood, Quinn, & Kashy, 2002; Wood & Neal, 2007). Once those stimuli appear, they automatically cue a response from the individual. For example, the clock striking 10 o’clock at night might prompt a person to absentmindedly start making the evening martini. Habits are learned associations and do not require a goal to instigate action. Instead, driving to an intersection near one’s home might prompt a person to turn left toward the office (a morning habit), when the driver wanted to get to the supermarket instead (Ouellette & Wood, 1998). Although habits do not require motivations, motivated behaviors often do become habitual. To further one’s goals, a person might create an association between situational cue and behavior to make sure that behavior steers toward the goal. The gym bag may lie near the door of the office to cue the somewhat out-of-shape office manager to stop at the health club on the way

126

MENTAL STATE THEORIES

home. After a while, the manager picks up the bag without a second, or even a first, thought at the end of the workday and continues her daily trek toward a healthy goal (Wood & Neal, 2007).

Teleology Motivational theories, by definition, assume that a person has a purposedriven life. People act to achieve some result or outcome. In the grand scheme of Aristotelian causes, motivations by this definition are final causes in that they are the goal or purpose that drives and shapes thought and behavior. Another term for this analytical approach is teleology, or the belief that things and events arise due to the purposes one can imbue in them. Of course, since the time of Plato there has been a philosophical debate about whether teleological thinking and final causes actually exist in nature, or whether they should be admitted to scientific thought. That is, it is only natural to think about human behavior in its motivational—or rather teleological—terms. The problem is that people often overgeneralize this approach to topics where it obviously does not apply. As a consequence, explanations based on final causes, though perfectly appropriate in treatments of human motivation, are problematic in other areas of scientific study, and it is often considered “bad form.” According to the argument, final causes are beyond verification in that they cannot be judged true or false with precision, whereas more mechanistic explanation based on preexisting events and forces are easier to verify (Bacon, 1620/1878). A classic example involves the neck of giraffes. One can say that evolution worked to lengthen the neck of giraffes with a purpose in mind: to make it easier for giraffes to reach the tree leaves they eat to survive. However, imbuing evolution with a purpose-driven motive, much like those of the human organism, is an imprecise strategy in that there is no scientific way to verify whether or not nature works with a purpose. It is beyond human verification. As such, many scientific thinkers, such as evolutionary theorists, dispel any talk of purpose and focus instead on preexisting mechanisms and conditions that produce the results they see in the laboratory or in the natural world (see Ketelaar, Chapter 11, this volume). Why did giraffes through the generations grow such long necks? Because shorter giraffes, relative to their taller counterparts, failed to achieve an adequate diet and thus disproportionately died out, leaving predominantly taller giraffes to convey their genes to the next generation. Mechanisms like this, natural selection, can be directly observed and assessed precisely. As a result, they have been verified by scientific observation for over 100 years and thus are considered a better form of scientific argument. Hence, although it is perhaps more precise and appropriate to apply the logic of teleology to human motivation, one should be sensitive to the prejudice that other scientists may have against that logic. As a consequence, at times, it may be advantageous to couch motivational theories in a different logic focusing on preconditions. For example, instead of a motive, a theo-



Motivational Theories 127

rist might point instead to preconditions that bring about compensatory or repair behavior: such as deficit in blood sugar, or self-esteem, or well-being relative to their normal level. Or one can adopt the theoretical halfway house notion of teleonomy, couching motivational theories in what an organism expects to happen if they take a certain course of action (e.g., my course grade will be better if I read the textbook). This has been the strategy followed by disciplines such as economics, in which rational actors simulate the future to determine what outcomes are likely and rewarding, and then plan their behavior accordingly (see Trafimow, Chapter 12, this volume). One, of course, has to be mindful that, although human behavior may follow a teleonomic scheme, the rest of nature, such as the neck of a giraffe, must not be expected to likewise follow.

Concluding Remarks To say that human behavior is motivated is easy, but to be able to disentangle the Gordian Knot of motives that may come into play for any specific human behavior can be intrinsically difficult. Any human behavior can fulfill any number of motives, from the obvious to the counterintuitive. As such, the construction of any motivational theory that adequately explains human behavior is a noble task that runs the risk of becoming quixotic. In this chapter, I have attempted to lay out some issues that any motivational theorist must deal with, as well as some common avenues that one must consider when trying to identify the motives underlying human thought and action. It is hoped that in laying out these difficulties, I haven’t dampened the motivation of any future theorist. Instead, I hope the discussion serves to enliven anyone desiring to understand this most central facet of human life. The study of motivation is a theoretical feast, with a neverending menu of questions and puzzles to entice the appetite of anyone with sufficient curiosity and motivation. Acknowledgments The writing of this chapter was supported financially by National Science Foundation Grant No. 0745806. Any opinions, conclusions, or recommendations expressed in this chapter are those of the author and do not necessarily reflect the views of the National Science Foundation.

References Bacon, F. (1620/1878). Novum organum. New York: Macmillan. Bandura, A. (1997). Self-efficacy: The exercise of control. New York: Freeman. Bandura, A., & Schunk, D. H. (1981). Cultivating competence, self-efficacy, and intrinsic

128

MENTAL STATE THEORIES

interest through proximal self-motivation. Journal of Personality and Social Psychology, 41, 586–598. Bargh, J. A., Gollwitzer, P. M., Lee-Chai, A. Y., Barndollar, K., & Troetschel, R. (2001). The automated will: Nonconscious activation and pursuit of behavioral goals. Journal of Personality and Social Psychology, 81, 1014–1027. Barrett, J. L., & Johnson, A. H. (2003). The role of control in attributing intentional agency to inanimate objects. Journal of Cognition and Culture, 3, 208–217. Baumeister, R. F., & Vohs, K. D. (Eds.). (2004). Handbook of self-regulation: Research, theory, and applications. New York: Guilford Press. Baumeister, R. F., Vohs, K. D., & Tice, D. M. (2007). The strength model of self-control. Current Directions in Psychological Science, 16, 351–355 Beauregard, K. S., & Dunning, D. (1998). Turning up the contrast: Self-enhancement motives prompt egocentric contrast effects in social judgments. Journal of Personality and Social Psychology, 74, 606–621. Becker, G. (1976). The economic approach to human behavior. Chicago: Chicago University Press. Bodner, R., & Prelec, D. (2003). Self-signaling in a neo-Calvinist model of everyday decision making. In I. Brocas & J. Carillo (Eds.), Psychology of economic decisions (Vol. I, pp. 105–126). London: Oxford University Press. Brennan, G., & Hamlin, A. (1998). Expressive voting and electoral equilibrium. Public Choice, 95, 149–175. Camerer, C., Babcock, L., Loewenstein, G., & Thaler, R. (1997). Labor supply of New York City cab drivers: One day at a time. Quarterly Journal of Economics, 112, 407–411. Campbell, W. K., & Sedikides, C. (1999). Self-threat magnifies the self-serving bias: A metaanalytic integration. Review of General Psychology, 3, 23–43. Carver, C. S., & Scheier, M. F. (2001). On the self-regulation of behavior. New York: Cambridge University Press. Chambers, J. R., & Windschitl, P. D. (2004). Biases in social comparative judgments: The role of nonmotivated factors in above-average and comparative-optimism effects. Psychological Bulletin, 130, 813–838. Chambliss, W. (1966). The deterrent influence of punishment. Crime and Delinquency, 12, 70–75. Chartrand, T. L., & Bargh, J. A. (1996). Automaticity of impression formation and memorization goals. Journal of Personality and Social Psychology, 71, 464–478. Cialdini, R. B., Schaller, M., Houlihan, D., Arias, K., Fultz, J., & Beaman, A. L. (1987). Empathy-based helping: Is it selflessly or selfishly motivated? Journal of Personality and Social Psychology, 52, 749–758. Clary, E. G., & Snyder, M. (1999). The motivations to volunteer: Theoretical and practical considerations. Current Directions in Psychological Science, 8, 156–159. Cohen, G. L., Aronson, J., & Steele, C. M. (2000). When beliefs yield to evidence: Reducing biased evaluation by affirming the self. Personality and Social Psychology Bulletin, 26, 1151–1164. Duckworth, A. L. (2011). The significance of self-control. PNAS, 108, 2639–2640. Dunning, D., & Balcetis, E. (2013). Wishful seeing: How preferences shape visual perception. Current Directions in Psychological Science, 22, 33–37. Dunning, D., & Fetchenhauer, D. (2013). Behavioral influences in the present tense: On expressive versus instrumental action. Perspectives on Psychological Science, 8, 142– 145. Dunning, D., Heath, C., & Suls, J. (2004). Flawed self-assessment: Implications for health, education, and the workplace. Psychological Science in the Public Interest, 5, 71–106. Dunning, D., Leuenberger, A., & Sherman, D. A. (1995). A new look at motivated infer-



Motivational Theories 129

ence: Are self-serving theories of success a product of motivational forces? Journal of Personality and Social Psychology, 59, 58–68. Dweck, C. S. (1999). Self-theories: Their role in motivation, personality, and development. Philadelphia: Psychology Press. Elliot, A. J. (2006). The hierarchical model of approach-avoidance motivation. Motivation and Emotion, 30, 111–116. Elliot, A. J., & McGregor, H. A. (2001). A 2 × 2 achievement goal framework. Journal of Personality and Social Psychology, 80, 501–519. Fehr, E., & Schmidt, K. M. (1999). A theory of fairness, competition, and cooperation. Quarterly Journal of Economics, 114, 817–868. Ferguson, M. J., Hassin, R., & Bargh, J. A. (2008). Implicit motivation: Past, present, and future. In J. Shah & W. Gardner (Eds.), Handbook of motivation science (pp. 150–166). New York: Guilford Press. Fetchenhauer, D., & Dunning, D. (2009). Do people trust too much or too little? Journal of Economic Psychology, 30, 263–276. Fishbach, A., & Ferguson, M. J. (2007). The goal construct in social psychology. In A. W. Kruglanski & E. T. Higgins (Eds.), Social psychology: Handbook of basic principles (2nd ed., pp. 490–515). New York: Guilford Press. Förster, J., Liberman, N., & Friedman, R. S. (2007). Seven principles of goal activation: A systematic approach to distinguishing goal priming from priming of non-goal constructs. Personality and Social Psychology Review, 11, 211–233. Gollwitzer, P. M. (1999). Implementation intentions: Strong effects of simple plans. American Psychologist, 54, 493–503. Greenberg, J., Pyszczynski, T., Solomon, S., Rosenblatt, A., Veeder, M., Kirkland, S., et al. (1990). Evidence for terror management: II. The effects of mortality salience on reactions to those who threaten or bolster the cultural worldview. Journal of Personality and Social Psychology, 58, 308–318. Heider, F., & Simmel, M. (1944). An experimental of apparent behavior. American Journal of Psychology, 57, 243–259. Heine, S. J., Proulx, T., & Vohs, K. D. (2006). The meaning maintenance model: On the coherence of social motivations. Personality and Social Psychology Review, 10, 88–110. Higgins, E. (1997). Beyond pleasure and pain. American Psychologist, 52, 1280–1300. Higgins, E. T., Friedman, R. S., Harlow, R. E., Idson, L. C., Ayduk, O. N., & Taylor, A. (2001). Achievement orientations from subjective histories of success: Promotion pride versus prevention pride. European Journal of Social Psychology, 31, 3–23. Hull, C. L. (1943). Principles of behavior. New York: Appleton-Century. Jones, E. E. (1964). Ingratiation: A social psychological analysis. New York: Appleton-CenturyCroft. Kanfer, R. (1992). Work motivation: New directions in theory and research. In C. L. Cooper & I. T. Robertson (Eds.), International review of industrial and organizational psychology (Vol. 7, pp. 1–53). Chichester, UK: Wiley. Katz, D. (1960). The functional approach to the study of attitudes. Public Opinion Quarterly, 24, 163–204. Kirk, A. K., & Brown, D. F. (2003). Latent constructs of proximal and distal motivation predicting performance under maximum test conditions. Journal of Applied Psychology, 88, 40–49. Knowles, M. L., Lucas, G. M., Molden, D. C., Gardner, W. L., & Dean, K. K. (2010). There’s no substitute for belonging: Self-affirmation following social and non-social threats. Personality and Social Psychology Bulletin, 36, 173–186. Krieglmeyer, R., De Houwer, J., & Deutsch, R. (2011). How farsighted are behavioral ten-

130

MENTAL STATE THEORIES

dencies of approach and avoidance? The effect of stimulus valence on immediate vs. ultimate distance change. Journal of Experimental Social Psychology, 47, 622–627. Kruger, J. M., & Dunning, D. (1999). Unskilled and unaware of it: How difficulties in recognizing one’s own incompetence lead to inflated self-assessments. Journal of Personality and Social Psychology, 77, 1121–1134. Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108, 480–498. Lacey, A. R. (1996). Dictionary of philosophy (3rd ed.). New York: Routledge. Leary, M. R., Tambor, E. S., Terdal, S. K., & Downs, D. L. (1995). Self-esteem as an interpersonal monitor: The sociometer hypothesis. Journal of Personality and Social Psychology, 68, 518–530. Lee, A. Y., & Aaker, J. L. (2004). Bringing the frame into focus: The influence of regulatory fit on processing fluency and persuasion. Journal of Personality and Social Psychology, 86, 205–218. Lepper, M. R., Greene, D., & Nisbett, R. (1973). Undermining children’s intrinsic interest with extrinsic reward: A test of the “overjustification” hypothesis. Journal of Personality and Social Psychology, 28, 129–137. Lloyd, G. E. R. (1970). Early Greek science: Thales to Aristotle. New York: Norton. Locke, E. A., & Latham, G. P. (1990). A theory of goal setting and task performance. Englewood Cliffs, NJ: Prentice-Hall. Loewenstein, G. (1994). The psychology of curiosity: A review and reinterpretation. Psychological Bulletin, 116, 75–98. Maslow, A. H. (1943). A theory of human motivation. Psychological Review, 50, 370–396. McClelland, D. C., Koestner, R., & Weinberger, J. (1989). How do self-attributed and implicit motives differ? Psychological Review, 96, 690–702. Milinski, M., Semmann, D., & Krambeck, H. J. (2002). Reputation helps solve the “tragedy of the commons.” Nature, 415, 424–426. Ouellette, J., & Wood, W. (1998). Habit and intention in everyday life: The multiple processes by which past behavior predicts future behavior. Psychological Bulletin, 124, 54–74. Pareto, V. (1971). Manual of political economy (A. Schweir, Trans.). New York: A. M. Kelley. (Original work published 1909) Pinder, C. C. (2008). Work motivation in organizational behavior (2nd ed.). New York: Psychology Press. Rabin, M. (1993). Incorporating fairness into game theory and economics. American Economic Review, 83, 1281–1302. Rogers, R. W. (1975). A protection motivation theory of fear appeals and attitude change. Journal of Psychology, 91, 93–114. Rosenstock, I. M. (1966). Why people use health services. Milbank Memorial Fund Quarterly, 44, 94–127. Schlösser, T., Fetchenhauer, D., & Dunning, D. (2012). Against all odds: The role of immediate and anticipated emotions in trust. Unpublished manuscript, University of Cologne, Germany. Schultheiss, O. C., & Brunstein, J. C. (Eds.). (2010). Implicit motives. New York: Oxford University Press. Sheldon, S., Ames, D. R., & Dunning, D. (2013). Emotionally unskilled, unaware, and uninterested in learning more: Biased self-assessments of emotional intelligence. Unpublished manuscript, Rutgers University, New Brunswick, NJ. Sherman, D. K., & Cohen, G. L. (2006). The psychology of self-defense: Self-affirmation theory. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 38, pp. 183–242). San Diego, CA: Academic Press. Sherman, D. K., Nelson, L. D., & Steele, C. M. (2000). Do messages about health risks



Motivational Theories 131

threaten the self? Increasing the acceptance of threatening health messages via selfaffirmation. Personality and Social Psychology Bulletin, 26, 1046–1058. Tangney, J. P., Baumeister, R. F., & Boone, A. L. (2004). High self-control predicts good adjustment, less pathology, better grades, and interpersonal success. Journal of Personality, 72, 271–324. Tesser, A. (2000). On the plasticity of self-defense. Current Directions in Psychological Science, 10, 66–69. Tice, D. M. (1991). Esteem protection or enhancement? Self-handicapping motives and attributions differ by trait self-esteem. Journal of Personality and Social Psychology, 60, 711–725. Wood, W., & Neal, D. T. (2007). A new look at habits and the interface between habits and goals. Psychological Review, 114, 843–863. Wood, W., Quinn, J. M., & Kashy, D. (2002). Habits in everyday life: Thought, emotion, and action. Journal of Personality and Social Psychology, 83, 1281–1297.

7 Duality Models in Social Psychology Roland Deutsch

T

he present chapter discusses the structure and explanatory capacity of duality models in social psychology (for recent reviews and edited volumes, see Evans, 2008; Gawronski & Creighton, 2013; Sherman, Gawronski, & Trope, 2014; Strack & Deutsch, 2015). Duality models are a family of multiple individual theories sharing the idea that many psychological phenomena involve the operation of two mental faculties (i.e., two distinct processes, two distinct representations, or two distinct systems). Most duality models also make assumptions about how the two mental faculties relate to features of automaticity—such as intentionality, awareness, or efficiency (cf. Gawronski, Sherman, & Trope, 2014; Moors, 2014; Strack & Deutsch, 2015)—typically assuming one of the two faculties operating at a greater level of automaticity. Duality models have surged not only in basic research to explain psychological phenomena, but also in applied fields such as clinical psychology (e.g., Roefs et al., 2011; Stacy & Wiers, 2010), consumer behavior (e.g., Strack, Werth, & Deutsch, 2006), health psychology (e.g., Hill & Durante, 2011; Hofmann, Friese, & Wiers, 2011), and organizational behavior (e.g., Marquardt & Hoeger, 2009). As such, they seem to be of great scientific and applied interest. This chapter provides a deeper insight into the way duality models explain psychological phenomena (for a chapter with a similar focus, see 132

Duality Models in Social Psychology 133

Gawronski et al., 2014). First, the chapter will describe typical structural features, commonly suggested processes and moderators, as well as those phenomena that are most frequently explained by duality models. The chapter will then discuss what can be considered to be hot spots in theory evaluation and development.

Properties and Function of Duality Models Despite the fact that duality models differ on various dimensions, they share a set of features and ways in which they explain social phenomena. At the center, duality models contain the description of two mental faculties (i.e., processes, representations, systems) in terms of their operating principles and their operating conditions (Gawronski & Bodenhausen, 2009). On this basis, duality models allow explaining why organisms differentially respond to a situation depending on moderator variables.

Basic Terms and Assumptions Duality models have been criticized for often lacking clear definitions of basic terms (e.g., Keren & Schul, 2009). Indeed, there is considerable heterogeneity with regard to the precision of definitions and the concrete meaning of basic terms across models. Reviewing the heterogeneity is beyond the scope of this chapter. To provide a clear conceptual frame, however, I will outline the basic terms in a way that I consider compatible with the larger number of existing models.

Psychological Process The basic assumptions of duality models rest on the term psychological process, which refers to a hypothetical chain of mental events that “translate inputs into outputs” (Gawronski et al., 2014, p. 2). Based on other analyses (De Houwer & Moors, Chapter 2, this volume; see also Gawronski & Bodenhausen, 2009; Marr, 1982), this umbrella term can be dissected into several aspects, all of which contribute to a more elaborate description of a psychological process. The first two components are its possible inputs (i.e., the stimuli or mental representations that can influence the process), as well as the possible outputs (i.e., the overt responses or mental representations that result from the process). As De Houwer and Moors (Chapter 2, this volume) argued, considering mental representations as potential inputs and outputs of psychological processes is necessary if an overarching process is dissected into subprocesses, which “constitute only one part in a chain of mental subprocesses by which environmental input influences behavior” (p. 8). For example, the process of racial stereotyping presupposes the extraction of racial cues from the perceptual input. In other words, stereotyping

134

MENTAL STATE THEORIES

uses the mental representations generated by perceptual processes as input. Likewise, using a stereotype may sometimes result in overt responses. But on other occasions, a stereotype may be used to generate a judgment that is never verbalized or translated into other behaviors. Instead, it may merely result in a memory trace linking the stereotyped person with stereotypical traits (for further examples and an in-depth discussion, see De Houwer & Moors, Chapter 2, this volume). The third component is the functional relationship between inputs and outputs that characterizes the process. It describes what the process does (e.g., adding, subtracting), and this refers to the computational level of analysis (Marr, 1982). The fourth component is the concrete algorithm (Marr, 1982) that brings about the functional relationship (e.g., adding by counting fingers, adding by using Arabic numbers; De Houwer & Moors, Chapter 2, this volume). It describes how the process does what it does. The third and fourth component may also be labeled operating principles (Gawronski & Bodenhausen, 2009; Gawronski et al., 2014). The fifth component is the implementation of the process in biological systems, such as the human brain, or physical systems, such as a silicone based computer (Marr, 1982). The final aspect is the set of operating conditions of a process (Gawronski & Bodenhausen, 2009). Operating conditions “represent empirical claims about when that process is operating” (Gawronski & Bodenhausen, 2011, p. 73). More precisely, they specify conditions that facilitate versus inhibit the operation of a process. For example, in Gilbert’s (1989) model of attributional inference, the process of situational correction is hypothesized to heavily depend on cognitive resources, whereas the process of forming dispositional inferences is hypothesized to make minimal requirements in terms of cognitive resources. In general, most duality models refer to internal and external factors that are relevant for automaticity features (Bargh, 1994; Moors & De Houwer, 2006a) as operating conditions.1 The processes underlying the two mental faculties typically are theorized to differ in their speed, their efficiency (i.e., how easily they are disturbed by parallel activities), their dependence on intentions and motivation, whether they can be stopped or changed intentionally, and whether they operate outside of consciousness. As a consequence, the speed of responding, high versus low cognitive capacity, and the presence versus absence of general motivation and specific intentions to start or to stop a process are operating conditions in many duality models. Yet, individual models differ in the specific set of automaticity features and corresponding operating conditions that are specified. Although all duality models describe the operating principles of two mental faculties, they differ in the particular operating conditions, operation principles, and hence types of inputs and outputs that are ascribed to the two faculties (for reviews, see Evans, 2008; Gawronski & Creighton, 2013; Strack & Deutsch, 2015). An example besides Gilbert’s (1989) model of attributional inference is the heuristic–systematic model of persuasion (HSM; Chaiken, 1987; Chaiken, Liberman, & Eagly, 1989). It suggests that attitude change in response to persuasive communication is determined by

Duality Models in Social Psychology 135

thorough systematic or superficial heuristic thinking about messages and other aspects of the communication (e.g., the communicator). The operating principles of systematic processing involve analyzing the strengths of presented arguments. The operating principles of heuristic processing involve identifying the implications of heuristic cues present in the communicative context. As a last example, Haidt’s (2001) social intuitionist model of moral judgment contrasts moral intuitions and moral reasoning. The operating principles of moral reasoning imply that the “reasoner searches for relevant evidence, weighs evidence, coordinates evidence with theories, and reaches a decision” (Haidt, 2001, p. 818), whereas the operating principles of moral intuitions involve pattern matching in memory based on partially innate response tendencies.

Psychological System The concept of a system has a long tradition in psychology and comes with varying meanings (for a review, see Keren & Schul, 2009). For the present purpose, I deem it useful to draw on a generic dictionary definition, according to which a system can be thought of as “a regularly interacting or interdependent group of items forming a unified whole as a group of interacting bodies under the influence of related forces . . . a group of body organs that together perform one or more vital functions . . . a group of devices or artificial objects or an organization forming a network especially for distributing something or serving a common purpose” (Merriam Webster).2 In abstraction, this entails a set of objects or processes that interact because of common ends (e.g., respiratory system, heating system) and/or common mechanisms (e.g., the electric system of a car, the central nervous system). Applied to psychology, this suggests defining a psychological system as a regularly interacting or interdependent group of psychological processes and/or representations, forming a unified whole as a group that has common mechanisms and/or serves a common function. As such, the definition covers both structural and functional aspects (Keren & Schul, 2009). To postulate a system suggests that at least two processes empirically correlate under specified conditions and thus form a cluster that is functionally distinct from other processes or clusters of processes. The notion of a psychological system therefore is an empirical statement, and such notions are widespread in psychology. Take, for example, the now classic theory of working memory in cognitive psychology (Baddeley & Hitch, 1974). Here, the interaction of three components (central executive, visuospatial sketchpad, phonological loop) forms a system that serves the function to quickly store and manipulate information. Likewise, “there is unequivocal evidence that our senses constitute different systems” (Keren & Schul, 2009, p. 535). Not only do our senses correspond to distinct anatomical structures, but each of them also forms a bundle of subprocesses (e.g., processes at the level of receptors or at the level of higher sensory integration) serving the function of representing relevant physical and chemical states of the world. The

136

MENTAL STATE THEORIES

notion of systems is not unique to psychology or duality models. Systemnotions are part of many branches of the natural and social sciences, such as biological systems (e.g., circulatory system, lymphatic system), physical systems (e.g., the solar system, an atom), and social systems (e.g., democracy, monarchy). As such, dual-system models in psychology are based on an abstract concept frequently used in many branches of science. In the realm of duality models, early dual-process models have been complemented by dual-system theories (for a discussion, see Deutsch & Strack, 2006), which are based on the contrast between two processing systems, each bundling more than one process. On one hand, the system characterizations are made on the computational level by describing what the processes of each system are capable of doing. For example, Lieberman’s (2003, 2007) dual-system model of judgment processes assumes that the X-system incorporates processes such as similarity-detection, tagging affect to stimulus representations, and adjusting interconnected representations in a way that reduces cognitive dissonance (Lieberman, Ochsner, Gilbert, & Schacter, 2001). The C-system, on the other hand, is endowed with processes of formal logic reasoning, error detection, and response inhibition. As another example, the reflective-impulsive model (Strack & Deutsch, 2004) suggests that the impulsive system hosts processes such as knowledge activation through external stimuli, knowledge activation through the deprivation of basic needs, or the elicitation of approach/avoidance behavior. The reflective system, on the other hand, hosts processes such as categorization, decision making, and the maintenance of selected intentions (i.e., intending). On the other hand, system characterizations are often also made on the algorithmic level. A dominant distinction in dual-system models draws a line between algorithms based on associative processing versus rule-based/propositional processing (for a review, see Strack & Deutsch, 2015).

Supporting Assumptions In the empirical sciences, theories do not only rely on their core assumptions, but also on a rich network of additional, supporting theoretical assumptions (Lakatos, 1970; see also Gawronski & Bodenhausen, Chapter 1, this volume). One important area of supporting assumptions in psychological research is the operationalization of theoretical constructs. With few exceptions,3 psychological theories are formulated on the level of hypothetical, internal constructs (or variables) that describe psychological processes, systems, or representations (see middle column of Figure 7.1). For example, Devine’s (1989) dissociation model contains statements about stereotypes and prejudice, how they relate to operating conditions, and how they influence judgments. All these constructs are internal and cannot be observed or manipulated directly. As a consequence, theoretical assumptions are needed regarding how these constructs can be manipulated or measured by means of external, directly changeable, or observable variables (see outer columns of Figure 7.1). In psychology, these supporting assumptions can be quite elaborate and

duality Models in Social psychology

137

fIgURe 7.1. Predictive structure of a duality model. The core theory consists of operating principles (comprising algorithms potentially implying the computations) and operating conditions. Supporting assumptions link external variables to the internal constructs (i.e., the operationalizations). Lines with arrowed endings represent facilitation, and lines with rounded endings represent inhibition. Under Condition1.1, Output1.1 is predicted because Process 1 is facilitated and process-appropriate input is present in the situation (Pathway A). Under Condition1.2, Output2.1 is predicted because Process 2 is facilitated and process-appropriate input is present in the situation (Pathway B). Note that processes may have subprocesses that use the output of other processes as input and deliver their output to other processes (not represented in the figure).

must be validated in comprehensive research programs, as it has happened for indirect measures of attitudes, stereotypes, and self-esteem (Gawronski & Payne, 2010; Petty, Fazio, & Brinol, 2008). In the realm of duality models, the supporting assumptions are as diverse as the content-foci of the individual theories. What is, however, shared by most theories are assumptions about how the automaticity-related operating conditions (e.g., intentions, cognitive capacity) can be measured and manipulated. I will return to this discussion in a later section. These supporting theories and observations provide the rationale and evidence for the internal and external validity of experiments and other empirical studies conducted to test core theories. A problematic consequence of this fact is that it limits the conclusiveness of attempts to falsify core theories (cf. Gawronski

138

MENTAL STATE THEORIES

& Bodenhausen, Chapter 1, this volume). Particularly, if a prediction derived from a core theory is not matched by observations, this can be due to the fact that (1) the theory is false or (2) the supporting assumptions are false.

Basic Prediction and Explanation What kinds of phenomena can be specifically explained by duality models? The answer to this question is twofold. On one hand, the phenomena to be explained depend on the specific operating conditions and operating principles suggested by the theories. For example, early versions of the HSM (Chaiken, 1987) are limited to phenomena in the context of persuasion and attitude change; the social intuitionist model (Haidt, 2001) can explain phenomena in the realm of moral judgments; and Gilbert’s (1989) theory can explain phenomena in the realm of attribution and person perception. Other phenomena addressed by these theories are attitude-to-behavior processes (Fazio & Towels-Schwen, 1999; Strack & Deutsch, 2004), person perception (Brewer, 1988; Gilbert, 1989), intuition (e.g., Deutsch & Strack, 2008; Epstein, 1990), or self-control (Hofmann, Friese, & Strack, 2009; Metcalfe & Mischel, 1999). So there is great variety in the explained phenomena, which makes it impossible to provide a comprehensive review in this chapter (for reviews, see Evans, 2008; Gawronski & Creighton, 2013; Strack & Deutsch, 2015). On the other hand, the common and abstract theoretical structure of duality models results in an abstract predictive and explanatory structure, which is shared by a great number of theories and will be explained in the next sections. Deviations and complications relating to this structure will be addressed in the later section, Theoretical Hot Spots.

Predicting Moderation, Mediation, and Clustering Basic variants of duality models allow three kinds of predictions. First, the operating conditions and operating principles represent theoretical laws that, in an abstract, general manner, specify conditions under which Faculty 1 or Faculty 2 will determine responses to situations. When applied to concrete situational input and concrete operating conditions, these abstract theoretical laws allow predicting concrete outputs (see Figure 7.1). More specifically, when the concrete operating conditions vary, the abstract theoretical laws predict that the concrete operating conditions will moderate the concrete responses to the concrete inputs. For example, the HSM (Chaiken, Liberman, & Eagly, 1989) predicts that perceivers’ response to weak arguments presented by an attractive communicator will be rather different when processed while simultaneously counting backwards compared to a control condition without counting. Without counting, systematic processing will set in, analyze the argument strength, and, after recognizing the weak nature of the argument, maintain the original attitude and not give in. With parallel counting, heuristic processing will analyze the quality of heuristic cues.

Duality Models in Social Psychology 139

After having recognized the attractiveness of the communicator, attitudes will be adjusted into the direction of the message. The second prediction relates to the mediators of an observed functional relation. From a theoretical vantage point, a given input–output relation can be generated by an infinite number of mediating processes (cf. De Houwer & Moors, Chapter 2, this volume).4 Many duality models, however, contain theoretical assumptions suggesting that one or some few of these potential mediators are operating under specific conditions. Take, for example, the theoretical notion of the elaboration–likelihood model (Petty & Cacioppo, 1986; Petty & Wegener, 1999) that attitude change is mediated via the process of message elaboration under conditions of high motivation and capacity, whereas message elaboration is postulated to play a minor role when motivation or capacity is low. Under such low-elaboration conditions, peripheral processes such as evaluative conditioning or self-perception processes will dominate attitude change. When this general theoretical statement is applied to concrete situations, conditional mediation can be predicted. For example, the elaboration–likelihood model predicts that reading a message with six effectively positive and strong arguments in favor of lowering tuition will produce more positive attitudes toward lowering tuition. Under high motivation and capacity, this will be mediated through elaboration of the strength of the six arguments in relation to the conclusion, as well as processes of evaluative conditioning, which associate the positivity of the arguments with the message topic. When distracted by doing mental arithmetic in parallel, the elaboration–likelihood model predicts that attitudes toward lowering tuition will become more positive because of evaluative conditioning alone. Although the concept of mediation and statistical mediation–analyses are widespread and well established in psychology (e.g., MacKinnon, Fairchild, & Fritz, 2007), there are several problems associated with it (e.g., Spencer, Zanna, & Fong, 2005). One issue is particularly relevant for the present discussion. Specifically, empirical tests of the hypothesis that a given input–output relation is mediated by a certain process presupposes an independent measure of the process. Obviously, psychological processes cannot be observed in a direct manner. Instead, researchers typically rely on observable responses to physical stimuli to make inferences on psychological processes. For example, persuasion models predict that under high elaboration conditions, attitude change is determined by argument quality and mediated by message elaboration, especially by the favorable versus unfavorable thoughts regarding the arguments (Petty & Cacioppo, 1986). Therefore, responses in an attitude questionnaire should empirically vary as a function of argument quality (input–output relation). Also, if message elaboration is measured independently, for example, via thought listings (Greenwald, 1968), indices of favorable message elaboration should correlate with argument quality and attitude change. As such, in a very abstract manner, testing mediation–hypotheses empirically implies that one input–

140

MENTAL STATE THEORIES

output relation (e.g., argument strength → attitude change) is driven by another input–output relation (e.g., thought-listing instruction → valence of listed thoughts). Third, most duality models allow making predictions regarding the cooccurrence of processes or features. At their theoretical core, dual-system models typically assume processes that are part of a system to share the same operating conditions (e.g., high vs. low motivation and capacity; Strack & Deutsch, 2004), to share an abstract function (e.g., performing vs. troubleshooting; Lieberman, 2007), to relate to one another in specified ways (e.g., behavioral-decisions activate intending; Strack & Deutsch, 2004), and to share other important features, such as being evolutionary old versus young and operating in a parallel or serial manner (for an extensive review and summary of central dualities, see Evans, 2008). These abstract theoretical notions allow empirical predictions such that concrete outputs of the multiple processes constituting a mental faculty are predicted to correlate or overlap. For example, Lieberman’s theory of reflexive and reflective processes (e.g., Lieberman, 2003, 2007) contains the theoretical assumption that the function of the C-system is to represent exceptions from well-learned rules and to inhibit prepotent responses if adequate. Applied to a concrete instance of being confronted with an exception from a rule, indicators of processing such exceptions (e.g., activation of parts of the anterior cingulate cortex; see Botvinick, Braver, Barch, Carter, & Cohen, 2001) can be predicted to co-occur with indicators of response inhibition (e.g., activation parts of the frontal cortex; see Aron, Robbins, & Poldrack, 2004). The fact that duality models typically postulate the overlap of multiple dualities has been criticized for being hard to prove empirically and inconsistent with existing evidence (e.g., Keren & Schul, 2009; Kruglanski & Gigerenzer, 2011). At the same time, such postulates may also be considered to be a particular strength of duality models (Gawronski et al., 2014). As long as the overlap of multiple dualities comes in the form of a hypothesis instead of a definition (see the section Definition versus Hypothesis), increasing the number of hypothesized overlapping dualities increases the number of observations that are excluded by the theory, and hence the refutability of the theory in general (for a more detailed discussion, see Gawronski et al., 2014).

Explaining Empirical Phenomena According to the deductive-nomological schema of explanation (Hempel & Oppenheim, 1948),5 predictive and explanatory structures go hand in hand, with the main difference being what is given and what is inferred during the epistemic process. In prediction, the explanans (what is used to explain a phenomenon) and the conditional part of the phenomenon (what is hypothesized to be the cause of an effect) are given, and a particular observation is predicted to be made under the given conditions. In abstract terms, the core theoretical assumptions (middle column in Figure 7.1) represent the explan-

Duality Models in Social Psychology 141

ans, and a particular response to a particular combination of inputs and conditions (left and right column in Figure 7.1) is predicted. For example, by drawing on its core-theoretical assumptions, the HSM (Chaiken et al., 1989) predicts attitude change (the observation) under the condition of strong arguments and high motivational and cognitive resources by drawing on the core assumptions about the role of motivation and capacity for heuristic versus systematic processing (the explanans). In explanation, the explanandum (what is to be explained, typically an empirical phenomenon) is observed and theoretical statements are sought that would predict the observation under the given conditions. If such statements are found, they can be considered to represent an explanation (i.e., the explanans) of the phenomenon (i.e., the explanandum). In abstract terms, the typical explanandum in the realm of duality models is the interactive effect of variables related to operating conditions (e.g., motivation, intention, capacity) on the one hand and aspects of the processed stimuli on the other hand on observed responses. For example, one might observe that communicator expertise (stimulus aspect) influences attitude judgments (observed responses) under conditions theorized to induce low motivation, but much less so under conditions theorized to induce high motivation (ordinal interaction effect) (e.g., Petty, Cacioppo, & Goldman, 1981).6 To explain such an observation, one can draw on the core theoretical notions of the HSM (Chaiken et al., 1989), which predict that heuristic cues primarily have an effect under low processing motivation and capacity. To the degree that duality models predict moderation, mediation, and clustering, these models are capable of explaining observed, concrete moderator and mediator effects, as well as the observed clustering of variables indicating the operation of system-immanent processes. Two patterns of observations can be particularly well explained by duality models (cf. Strack & Deutsch, 2015). The first pattern of observations is that of a simultaneous contradictory belief (labeled Criterion S by Sloman, 1996), where a person reports having two conflicting responses at the same time. One prominent example is the conflict between knowledge and perception in visual illusions (Sloman, 1996; Strack, 1992) where the perceiver knows that the impressions generated by the visual system are wrong. Such a pattern can be explained by theories postulating the simultaneous operation of two processes with diverging operating principles. An exemplary social phenomenon representing simultaneous contradictory beliefs is a pattern of co-occurring prejudiced feelings and egalitarian judgments in cross-ethnic encounters (Wilson, Lindsey, & Schooler, 2000). The second pattern is that of an experimental dissociation (Dunn & Kirsner, 2003). There are multiple specific dissociation patterns, which have in common that the effect of inputs on outputs qualitatively differs depending on conditions. One rather prominent example is the uncrossed double-dissociation, which corresponds to “the observation that one variable affects the first task and not the second, and another variable affects the second task and not the first” (Dunn & Kirsner,

142

MENTAL STATE THEORIES

1988, p. 93). A social-psychological example would be the observation that implicit attitudes affect spontaneous behavior but not reflected behavior, while explicit attitudes affect reflected behavior but not spontaneous behavior (e.g., Hofmann & Friese, 2008). It is important to highlight the epistemological status of Criterion S and experimental dissociations. First, one might hope that such patterns provide an inductive proof of the existence of two processes, modules or systems. In other words, they might be interpreted as proof for the truth of the core theoretical assumptions of duality models. Second, one might consider Criterion S and experimental dissociations as particularly specific predictions that can be deduced from duality models. If these predictions are not falsified, they may be considered as support for the viability of the theory from which they were derived. I will briefly discuss both perspectives on dissociations. Generally, the inductive-proof perspective has been fiercely criticized (e.g., Dunn & Kirsner, 1988, 2003; Gigerenzer & Regier, 1996; Keren & Schul, 2009). Dunn and Kirsner provide detailed formal demonstrations that even experimental crossed double dissociations can also be generated from a single-process mechanism. Keren and Schul add further examples for how Criterion S and experimental dissociations may be explained by single-process alternatives. So, both forms of dissociations do not seem to be magic bullets that allow a strong inference on the existence of dualities. For at least two reasons, this is hardly surprising. First, in line with the reasoning underlying the hypothetico-deductive method (Popper, 1934), inductively proving the truth of a synthetic, general law is impossible (see Gawronski & Bodenhausen, Chapter 1, this volume). Second, as has been argued before (e.g., Gigerenzer & Regier, 1996; Sloman, 1996), the concept of rules has such high representational power that every observed functional relation can probably be formulated as a more or less complex rule (see also the section Computational Power) and therefore as a single-entity alternative. Likewise, an observed functional relation can always be explained by more than two processing entities (see note 5). Thus, for theoretical reasons, the hope for inductive proof of the exclusive truth duality–notions is ill fated to begin with, just as it is for any other nonanalytic general theoretical statement. The second perspective on the role of Criterion S and experimental dissociations is very well aligned with the hypothetico-deductive method. Most duality models can predict simultaneous contradictory belief as well as experimental double dissociations. The latter immediately follow from the core predictive structure depicted in Figure 7.1. The former presuppose additional assumptions, especially the notion of multiple-response channels, but such assumptions are part of many existing models. If experiments do not falsify these predictions, this can be interpreted as support for the core theory. As argued in the preceding paragraph, it is quite conceivable that a given pattern of observations may also be predicted with other theories, some of them perhaps being single-faculty models. This is no general threat to duality notions, as theories may not only compete in terms of their abil-

Duality Models in Social Psychology 143

ity to successfully predict empirical phenomena, but also on features such as parsimony, generality, conservatism, and generality (see Gawronski & Bodenhausen, Chapter 1, this volume). As such, there is no way to decide between singularity versus duality notions in general and purely based on whether one might construe a single-process explanation for a given dissociation. Most likely, this will always be possible. Instead, one must evaluate competing singularity and duality models on all relevant metatheoretical parameters. As such, duality models have no special place in the pantheon of psychological theories. Neither are there magic bullets to prove their truth, nor are they automatically doomed if a competing, singularity theory makes overlapping predictions. Just as is true of all other theories, their individual quality needs to be evaluated on multiple dimensions and in comparison to their contenders.

Interim Conclusion The basic explanatory and predictive structure of duality models is that of moderated mediations (Muller, Judd, & Yzerbyt, 2005).7 Duality models predict when and how two mental faculties influence responses. Hence, duality models use a well-established explanatory pattern that is based on simple, rarely disputed assumptions (i.e., process distinctiveness, conditional operation of processes). Just as other theories, however, they suffer from the necessity of adding supporting assumptions, which may weaken the inferences that can be drawn from empirical observations (Lakatos, 1970). What has been presented so far represents a common core that is part of most duality models. This family of theories, however, comes with a great variety in many structural and process features of the theories. This variability corresponds to distinct theoretical and methodological strengths and challenges, which also determine the explanatory capacity (see Gawronski & Bodenhausen, Chapter 1, this volume).

Theoretical Hot Spots The structure of duality models creates a number of challenges that, when not appropriately met, can compromise the explanatory capacity of duality models. In what follows, I describe the most important challenges as well as potential solutions.

Defining the Nature of the Duality Duality notions are at the heart of the present family of theories. At the same time, defining the nature of the duality presents a number of pitfalls. For at least two reasons, a duality theory is useful to the degree that the two supposed mental faculties allow making differential predictions. Without dif-

144

MENTAL STATE THEORIES

ferential predictions, assuming an additional faculty cannot be considered a parsimonious measure. And without differential predictions, the idea of two mental faculties can hardly be put to an empirical test. But there are challenges above and beyond this basic need for differential prediction, which will be discussed in the following sections.

Partial Dualities A problem may arise when the two mental faculties are specified only in terms of very few instead of most process components (i.e., input, output, input–output relation, algorithm, implementation, conditions; see above). Specifically, it is easily possible to create constellations in which there is a qualitative difference in input–output relations, but not in the underlying algorithms (computational distinctiveness, algorithmical identity). An example for this can be derived from the perspective taken by Kruglanski and colleagues (Unimodel; Erb et al., 2003; Kruglanski & Thompson, 1999) while analyzing dual-process interpretations of persuasion (Chaiken et al., 1989; Petty & Cacioppo, 1986). The Unimodel acknowledges that motivation and capacity moderate which inputs (heuristic cues, message content) determine outputs (attitude judgments). Hence, it concurs that there are different input–output relations depending on operating conditions. The Unimodel, however, further suggests that these input–output relations are driven by the same abstract algorithms. Specifically, the Unimodel suggests that attitude judgments in persuasive settings are driven by a universal mechanism of evidence processing and belief adjustment. Depending on motivation and capacity, this universal mechanism may select facets (e.g., heuristic cues vs. arguments) from the situational input (the persuasive setting), resulting in differing input–output relations. It is also possible to construe constellations in which there is a qualitative difference in underlying algorithms, but not in the input–output relations (computational identity, algorithmical distinctiveness). Take, for example, the function to reverse the meaning of negated statements (e.g., he is not friendly). Processing the negation would suggest the output is rude or negative. According to the reflective–impulsive model (Strack & Deutsch, 2004), such an input–output relation could be generated by two different mechanisms. In one processing system (which is theorized to operate only under high cognitive and motivational resources), negations can be processed by a mechanism that flexibly applies the abstract relation of a negation to a semantic concept and thereby generates the negated meaning. In the other processing system (which is theorized to operate also with low motivational or cognitive resources), negations can be processed if their meaning has been stored in long-term memory as a result of extended learning (such as in no way or no idea). Here, the mediating process is one of long-term memory retrieval that does not draw on the flexible application of the negation relation. Thus, although superficially both processes may result in the same input–output

Duality Models in Social Psychology 145

relation (e.g., inferring impossible from no way), the mediating mechanisms may be rather different. Partial dualities do not pose a fundamental threat to the explanatory value of duality models. But they can create considerable confusion and pointless debates when they are not made explicit. A second negative aspect of partial dualities is that they may result in weaker theories compared to cases where input–output relations and algorithms are theorized to differ. The reason for this is that a double duality makes more differential predictions—that is, predictable differences in input–output relations as well as in algorithms (as far as they can be measured independently). A way to minimize the negative impact of partial dualities, especially in the case of computational identity, is to analyze the algorithms for whether they might create different input–output relations under conditions neglected so far. Take the example of negation. The memory-retrieval mechanism can be predicted to depend more heavily on training and prior experience than the propositional reasoning mechanism. As a consequence, manipulating the degree of practice with negated expressions can put this interpretation to an empirical test (Deutsch, Gawronski, & Strack, 2006).

Computational Power A high computational power of particular algorithms poses a serious threat to differential predictions. Take, for example, the very general conceptualizations of rules and associations, which, in a broad sense, play an important role in many duality models (for a review, see Strack & Deutsch, 2015). Understood broadly, both rules and associations can be used to model a very large (if not infinite) set of input–output relations. Abstract rules formulated in terms of symbols and logic can also represent those input–output relations predicted by associative principles broadly (Gigerenzer & Regier, 1996; Sloman, 1996). A current manifestation of this notion is the propositional model of associative learning (Mitchell, De Houwer, & Lovibond, 2009), which reconstructs learning phenomena in the realm of conditioning paradigms as the result of nonautomatic reasoning instead of automatic association formation. This model also promotes the idea that every associative relation between stimuli (e.g., CS and US) can be translated into a proposition or rule (e.g., “associated (CS, US)” or “if CS, then US”). As a consequence, observing that an organism acts in harmony with what is predicted by the operation of a CS–US association is always open for the alternative interpretation that the organism applied if–then rules. To render rules versus associations empirically distinguishable therefore requires adding specific constraints to their operating principles and/or their operating conditions. For example, some theorists have suggested defining rules/propositions as being selectively able to represent the truth versus falseness of a state of affairs (e.g., Deutsch et al., 2006; Gawronski & Bodenhausen, 2006; Gilbert, Krull, & Malone, 1990; Strack & Deutsch, 2004), whereas associations simply link two or more men-

146

MENTAL STATE THEORIES

tal contents without containing information about whether this is an adequate characterization of the world (Mitchell et al., 2009). Constrained this way, the two concepts can be distinguished empirically by testing whether truth information affects responding. An unconstrained concept of associations may also go along with a very high representational power. More specifically, the computational power of associative (connectionist) networks allows simulating input–output relations generated by rules (Van Overwalle & Siebler, 2005), including abstract concepts (Plaut, 1995). It is therefore crucial to further constrain general, powerful concepts such as rules, associations, propositions, and symbols to ensure discriminability.

Definition versus Hypothesis The issue of definition versus hypothesis has recently received a lot of attention (Gawronski et al., 2014; Moors, 2014). It is not as specific to duality models as the other issues discussed so far, but it may still play an important role in theory generation and evaluation. In very abstract terms, the problem follows from the fact that definitional relations cannot be explanations for empirical relations (cf. Greve, 2001; Hartmann, 1993; Wallach & Wallach, 1998). For example, if the term Robin is defined as a bird with a red patch under the head, the categorization of a bird as Robin cannot count as an explanation for the red patch. Applied to duality models, if the features that make up Faculty 1 or Faculty 2 are linked via definitions, their explanatory capacity will be compromised. For example, if a theory defines heuristic processes as operating under low-capacity conditions, the categorization of a process as heuristic cannot count as an explanation for the observation that a process operates under low capacity. Similarly, observing that a process that was categorized as heuristic needs capacity does not falsify the definitions of heuristic processes as operating without capacity. Instead it suggests that there was something wrong either with the original categorization or with the later measurement. On the other hand, if the theory hypothesizes a certain relation to be true, the relation can later serve as an explanation for empirical observations. For example, if a theory hypothesizes that analyzing message arguments requires more cognitive capacity than analyzing heuristic cues, this relation may be proven either right or wrong by empirical research. In turn, the hypothesized law can serve as an explanation for observations that were predicted by the law. Besides compromising the explanatory capacity, definitional relations may prevent researchers from studying certain phenomena. For example, if rule-based processes are defined as nonautomatic, researchers may simply not engage in experimental tests for automatic rule application (Moors, 2014). Unfortunately, duality models are not always explicit as to whether relations between process components are definitional or hypothetical. At the same time, this is crucial to ensure their explanatory capacity and refutability. As discussed earlier, a major asset of duality models is their capacity to explain the moderating effect of external variables on the operation of cer-

Duality Models in Social Psychology 147

tain algorithms or the instantiation of certain input–output relations. More specifically, the theory links operating conditions and operating principles, such that external variables hypothesized to influence operating conditions can be predicted to trigger specific operating principles (see Figure 7.1). This explanatory pattern, however, presupposes the relation between operating principles and operating conditions to be hypothetical instead of definitional. Effective measures are available to minimize this problem (Moors, 2014). When constructing duality models, researchers should take great care to clarify which relations are supposed to serve an explanatory role and which are not. Hypothetical and definitional relations should be made explicit, which presupposes rigorous analyses of the theoretical concepts involved (Moors, 2014). As a consequence, empirical research should be conducted to test the viability of the hypothesized relations, whereas empirical research testing the definitional relations should be avoided as it will result in little advancement of knowledge. (For an elaboration of this conclusion, see Greve, 2001; Wallach & Wallach, 1998.) Moreover, to fruitfully test hypothesized relations between properties of Faculty 1 versus Faculty 2, methods to clearly identify these properties are necessary (Moors, 2014). Take, for example, a duality model that hypothesizes associative processes to be automatic. To test this hypothesis, clear empirical criteria for automaticity and associative processing are necessary. To facilitate empirical progress, theorists should work to provide such empirical criteria.

Specification of Operating Conditions Operating conditions play a crucial role in most duality models (Smith & DeCoster, 2000; Strack & Deutsch, 2015). They determine the conditions under which one or the other mental faculty is expected to determine responses. As such, operating conditions carry a great explanatory burden. Duality models regularly specify operating conditions that have to do with automaticity features, such as intentionality, awareness, or efficiency (Bargh, 1994; Moors & De Houwer, 2006a). The dimensional nature of many of these features (cf. Uleman, 1999) creates a problem that needs to be addressed in order to secure the refutability and predictive power of duality models (cf. Keren & Schul, 2009). Specifically, for speed, efficiency, and motivation, no a priori criteria can be determined that would result in binary categorizations into fast versus low, efficient versus inefficient, motivated versus unmotivated. Experimentally studying the moderating effect of continuous variables therefore requires the deliberate selection of two or more levels on the continuum in order to create experimental conditions.8 The need to choose levels of continuous moderators for experimental purposes poses a serious threat to the refutability of the underlying theory. Imagine a theory that links Process A with high efficiency and Process B with low efficiency. An empirical test would entail manipulating the available resources (perhaps by introducing a secondary task) to create conditions under which highly efficient processes would operate (few resources) but

148

MENTAL STATE THEORIES

inefficient processes would not. If a moderator effect is observed (i.e., Process A is affected by the manipulation of cognitive resources but Process B is not), this can be considered as evidence in support of the theory that generated the prediction. If, on the other hand, the manipulation of cognitive resources has an effect on neither Process A nor Process B, this cannot be interpreted as a refutation of the theory. Specifically, this pattern is compatible with at least two interpretations. First, the manipulation of the moderator might not have been intense enough to affect the processes. Second, the assumptions about the effect of the moderator on the process are wrong. Reporting a (successful) manipulation check for the moderator also does not help to establish a clear refutation. The experimental levels of the moderator might have been different enough to affect the manipulation check but might not have been different enough to affect the dependent variables. What would be needed is an a priori prediction on which relative levels of the moderator (or better: which distance between factor levels at a given absolute level) are sufficient to affect one of the two processes. This problem does not occur for binary features. As another, more concrete example, imagine a duality theory suggesting that the correction of stereotypical judgments in person–perception requires high motivation, whereas the activation of stereotypical associations in memory does not. Which pattern of data would be able to falsify this claim? Certainly, not a failure of measured or manipulated motivation to moderate judgmental correction. Such a null effect would always be compatible with the notion that motivation might not have been high enough for correction to set in, even if the manipulation was shown to be strong on an independent manipulation check. There is no easy way out of this problem. In the laboratory, one might work hard to find optimal levels of the moderator, but as long as the moderator fails to have an effect, the observations remain ambiguous. Therefore, establishing empirical standards for levels of distraction, speed, motivation, and so on, that qualify as fast, strong, or high would be highly desirable. When applying duality models to predict outcomes in a field setting, the situation is even grimmer. For example, when asking whether people in a particular office setting will stereotype or correct their judgments, the answer depends on whether motivation will be sufficient. Without clear standards for what is high motivation, a prediction will be rather problematic. It must be stressed, though, that this problem is not unique to duality models but applies to all research on automaticity in general. What is especially troublesome for duality models, however, is the notion that the two key moderators—motivation and capacity—are each multifaceted (e.g., Johnson & Eagly, 1989). Hence, optimal levels must be determined in a multidimensional parameter space.

Interactions A final factor influencing the explanatory capacity of duality models concerns potential interactions between the two faculties that are the basis for

Duality Models in Social Psychology 149

the duality. As shall be explained in the following section, the faculties may interact in a rather large number of possible ways (cf. Evans, 2008; Gilbert, 1999), and at least three parameters specify the relation between the two faculties. The first aspect of interactions is connected to the operating conditions specified in the core of a theory. Specifically, each of the two processes or systems may be hypothesized to operate either as obligatory or optional. For example, the reflective-impulsive model (Strack & Deutsch, 2004) suggests that the impulsive system always operates, whereas the reflective system only operates under particular specified conditions (cf. Lieberman, 2007). Other theories (e.g., Epstein, 1990; Sloman, 1996) emphasize parallel operation of the two faculties. The second aspect of interactions is whether one of the two faculties has the power to dominate the other faculty. Some theories suggest that one process is master to the second process. For example, Gilbert’s model of attributional inference (Gilbert, 1989) suggests that, if operating, the process of situational correction has the ability to overwrite the output of the first process (i.e., dispositional inferences). Other models suggest that the two faculties may dominate each other at different points in time or in different domains of processing (e.g., Epstein & Pacini, 1999; Metcalfe & Mischel, 1999). Some other models suggest no dominance relation between the two faculties, either because they are assumed to act in an either–or fashion or because they are assumed to fully act in parallel. Finally, models differ in their assumption about whether there is an informational influence across the two faculties. Different from dominating the other faculty, there might be information flow between the two faculties in either one or both directions. For example, the reflective-impulsive model (Strack & Deutsch, 2004) includes the possibility that outputs generated by the impulsive system enter the reflective system and are used for making judgments and decisions. This model also includes the notion that operations in the reflective system can co-activate corresponding representations in the impulsive system and may even generate new associations in this system (Deutsch et al., 2006). Similar assumptions regarding bi-directional or uni-directional informational influences are part of many duality models (e.g., Chen & Chaiken, 1999; Epstein, 1990; Gawronski & Bodenhausen, 2006; Metcalfe & Mischel, 1999; Smith & DeCoster, 2000). These different aspects of how interactions are specified in various theories go hand in hand with differences in what is predicted and what can be explained. Obviously, the nature of the specified interaction determines the predictive pattern of the given model in concert with its assumptions about the operating conditions and operating principles of the two faculties in isolation. More precisely, while operating conditions and operating principles specify what one of the two faculties does when in isolation, the assumptions about interactions specify what the two faculties ultimately do in general. As such, taking great care in describing theoretical assumptions about interactions is an important task. Another important aspect of theory construction and evaluation is that different combinations of the interaction features go along with greater

150

MENTAL STATE THEORIES

refutability, easier generation of predictions, and parsimony. For example, consider a slightly interactive duality model that assumes Process A to be obligatory, Process B to be optional but dominating Process A and no informational influence between the two. In this situation, knowing the operating conditions of the two processes allows a clear, straightforward prediction. Under the operating conditions that favor Process A, the functional relationships that specify Process A will dominate the outputs. Under the operating conditions that favor Process B, the functional relationships that specify Process B will dominate the output. Now, take as a contrasting example a highly interactive duality model suggesting two processes that operate obligatory, dominate each other under different conditions and content domains, and that have bi-directional informational influence. A first observation is that the highly interactive theory contains a greater number of assumptions than the slightly interactive theory, as the former theory would have to specify how and when exactly the informational influence occurs, and when which process dominates the other in addition to the basic operating conditions and operating principles spelled out in the slightly interactive model. In addition, the notion of bi-directional information flow between the two processes makes it hard to test the theory by analyzing input–output relations alone. Dominance relations and a high degree of information flow reduce the isolability of the mental faculties (cf. Keren & Schul, 2009). Instead, evidence for mediation by specific algorithms is necessary. Providing evidence for moderated mediation, however, is experimentally much more demanding and requires further supporting assumptions regarding the operationalization of the mediators (cf. De Houwer, 2011). A hypothetical example might help us understand this issue. Let’s assume a theory suggesting that (1) Process A can make additions of twodigit numbers, (2) Process B can make multiplications of two-digit numbers, and (3) both processes can translate them into verbal judgments. Finally, the theory further postulates that (4) Process B can read out the outputs of Process A and vice versa. Because of the bi-directional informational influence, the input–output relations of the two processes have become identical. Process A can return correct additions, but it may also return the multiplications carried out by Process B. Likewise, Process B can return correct multiplications, but it may also return the additions carried out by Process B. Without the assumption of information influence, the processes would differ on a functional level. But when bi-directional information influence is incorporated into a duality model, the two processes become indistinguishable on a functional level. Therefore, just looking at the input–output relations does not result in differential predictions in this case. Instead, it would be necessary to draw on methods that help assess the adding and multiplying itself instead of the mere input–output relations. This, however, might be quite challenging as it presupposes methods that are capable of distinguishing adding and multiplying independent of input–output relations (cf. De Houwer, 2011).

Duality Models in Social Psychology 151

These issues do not imply that slightly interactive models are per se better than highly interactive models. In fact, it might very well be that the loss in parsimony and the challenges that go hand in hand with the need to prove moderated mediation are in balance with the potential gains in generality or predictive power. To maintain refutability, however, it is necessary to take great care in specifying the supporting assumptions regarding the measurement of potential mediators.

Conclusion By definition, duality models incorporate two, sometimes opposing principles of processing. This dual nature is mirrored in the scientific reception of this family of models. On one hand, duality models have been received with great favor. This type of theorizing has become very popular in basic psychology as well as in applied fields. At the same time, there have been fierce critics of duality notions for a very long time, sometimes with relatively general doubts regarding the fruitfulness of the whole program (e.g., Gigerenzer & Regier, 1996; Keren & Schul, 2009; Kruglanski & Gigerenzer, 2011; Kruglanski & Thompson, 1999; Moors & De Houwer, 2006b). The present analysis is relevant to this debate in multiple ways. First, duality models share a common core, but they differ profoundly in various features. Debates on the advantages and disadvantages of duality models therefore should specify whether they refer to the core structure of duality models or whether they address more specific aspects of particular models. Discussing duality models as a whole bears the serious risk of oversimplification. Second, the core of duality models represents a very common conceptual and explanatory structure in psychology. As I have argued above, the basic assumptions of dual-process models are as follows: (1) establishing evidence for the mediation of input–output relations is possible; (2) establishing evidence for the moderation input–output relations and their mediation is possible; and (3) either type of evidence is possible for two processes in parallel. Dual-system models make the additional assumption (4) that it is possible to establish evidence for the systematic covariation of multiple processes (i.e., systems). The assumptions (1) to (4) are part of most cognitive theories in psychology and also represent a common explanatory structure in other branches of science. As such, there is a strong indication that the core structure is a fruitful, viable tool for theorizing. Third, the chapter discussed various theoretical hot spots that make duality models different from each other. Depending on how individual models score on these parameters, their predictive and explanatory capacity may be higher or lower. Variability in these parameters therefore is a plausible source of variability in the structural quality of duality models that may determine whether or not these models represent good empirical theories.

152

MENTAL STATE THEORIES

Theoretical analyses of duality models as well as theorists constructing new or improving existing models therefore should focus on these parameters. Notes 1. After the demise of automaticity as a unitary construct, the currently dominant view is that automaticity refers to a set of loosely correlated features. Although this may not even be appropriately fine-grained (Moors & De Houwer, 2006a), distinguishing four to five automaticity features can be considered a useful approximation (e.g., Moors, 2014). 2. Retrieved June 2, 2013, from www.merriam-webster.com/dictionary/system. 3. For example, foregoing mental constructs is part of many varieties of behaviorism (see e.g., Skinner, 1938; Staddon, 2001; Watson, 1913). 4. This is an insight from the automata theory (e.g., Hartmann, 1993). It can be formally derived, but an easy thought experiment helps understanding at least one reason for why this is true. Imagine a particular input–output function that can be explained by assuming a particular mechanism. One may now add a set of additional mechanisms that are without effect but increase the complexity of the calculations. For example, imagine observing an organism that responds to two digits x and y (the input) in a way that follows the function output = x + y. The easiest way to explain this would be to assume that the organism indeed engages in a mental arithmetic representing x + y. But one can easily come up with other calculations that still fulfill output = x + y. For example, the calculation ((x + y)*n))/n results in x + y, and so does the calculation ((x/n) + (y/n))*n. Occam’s razor would let us choose the simplest mechanism that can explain all observed data. For a detailed discussion of this problem, see also De Houwer and Moors (Chapter 2, this volume). 5. This, of course, is not the only schema of explanation, and it has been subject to considerable criticism (e.g., Cummins, 2000). Integrating other explanatory schemas would result in a more complex evaluation that would be beyond the space limits of this chapter. 6. Which observable situations induce high versus low motivation is determined by supporting assumptions inherent to the theory or the field of research. In the present example, the message was presented as being relevant for the participants (the contents were supposed to affect them personally) or irrelevant for the participants (the contents were supposed to affect only future generations). Personal relevance was theorized to increase processing motivation. 7. Such explanatory and predictive structures are often tested by measuring mediators, by measuring or manipulating moderators, and by then applying regression techniques to prove the basic effects to be mediated by the suspected mediator depending on conditions. Alternatively, such theoretical structures can also be tested by fully relying on experimental methods, possibly resulting in superior validity (e.g., Spencer, Zanna, & Fong, 2005). 8. If the moderator is studied as a continuous predictor in a regression design, this typically goes along with a restriction of the range of the moderator. This occurs either by deliberate choice (e.g., “let’s only study people with normal levels of intelligence, i.e., 85 > IQ < 115”) or by natural restriction in range in certain samples (e.g., no IQ < 100 in a sample of graduate students). In principle, the conse-

Duality Models in Social Psychology 153 quences of using dimensional moderators for the refutability of duality models are the same in experimental or regression designs, with the latter treating the moderator as continuous predictor.

References Aron, A. R., Robbins, T. W., & Poldrack, R. A. (2004). Inhibition and the right inferior frontal cortex. Trends in Cognitive Sciences, 8, 170–177. Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. A. Bower (Ed.), The psychology of learning and motivation (Vol. 8, pp. 47–90). New York: Academic Press. Bargh, J. A. (1994). The four horsemen of automaticity: Awareness, intention, efficiency, and control in social cognition. In R. S. Wyer & T. K. Srull (Eds.), Handbook of social cognition (Vol. 1, pp. 1–40). Hillsdale, NJ: Erlbaum. Brewer, M. B. (1988). A dual process model of impression formation. In T. K. Srull & R. S. Wyer (Eds.), Advances in social cognition (pp. 1–36). Hillsdale, NJ: Erlbaum. Botvinick, M. M., Braver, T. S., Barch, D. M., Carter, C. S., & Cohen, J. D. (2001). Conflict monitoring and cognitive control. Psychological Review, 108, 624–652. Chaiken, S. (1987). The heuristic model of persuasion. In M. P. Zanna, J. M. Olson, & C. P. Herman (Eds.), Social influence: The Ontario Symposium (Vol. 5, pp. 3–39). Hillsdale, NJ: Erlbaum. Chaiken, S., Liberman, A., & Eagly, A. H. (1989). Heuristic and systematic information processing within and beyond the persuasion context. In J. S. Uleman & J. A. Bargh (Eds.), Unintended thought (pp. 212–252). New York: Guilford Press. Chen, S., & Chaiken, S. (1999). The heuristic–systematic model in its broader context. In S. Chaiken & Y. Trope (Eds.), Dual-process theories in social psychology (pp. 73–96). New York: Guilford Press. Cummins, R. (2000). “How does it work?” vs. “What are the laws?” Two conceptions of psychological explanation. In F. Keil & R. Wilson (Eds.), Explanation and cognition (pp. 117–145). Cambridge, MA: MIT Press. De Houwer, J. (2011). Why the cognitive approach in psychology would profit from a functional approach and vice versa. Perspectives on Psychological Science, 6, 202–209. Deutsch, R., Gawronski, B., & Strack, F. (2006). At the boundaries of automaticity: Negation as reflective operation. Journal of Personality and Social Psychology, 91, 385–405. Deutsch, R., & Strack, F. (2006). Duality models in social psychology: From dual processes to interacting systems. Psychological Inquiry, 17, 166–172. Deutsch, R., & Strack, F. (2008). Variants of judgment and decision making: The perspective of the reflective–impulsive model. In H. Plessner, C. Betsch, & T. Betsch (Eds.), Intuition in judgment and decision making (pp. 39–53). Mahwah, NJ: Erlbaum. Devine, P. G. (1989). Stereotypes and prejudice: Their automatic and controlled components. Journal of Personality and Social Psychology, 56, 5–18. Dunn, J. C., & Kirsner, K. (1988). Discovering functionally independent mental processes: The principle of reversed association. Psychological Review, 95, 91–101. Dunn, J. C., & Kirsner, K. (2003). What can we infer from double dissociations? Cortex, 39, 1–7. Epstein, S. (1990). Cognitive-experiential self-theory. In L. A. Pervin (Ed.), Handbook of personality: Theory and research (pp. 165–192). New York: Guilford Press. Epstein, S., & Pacini, R. (1999). Some basic issues regarding dual-process theories from the perspective of cognitive‚ experiential self-theory. In S. Chaiken & Y. Trope (Eds.), Dual-process theories in social psychology (pp. 462–482). New York: Guilford Press. Erb, H.-P., Kruglanski, A. W., Chun, W. Y., Pierro, A., Mannetti, L., & Spiegel, S. (2003).

154

MENTAL STATE THEORIES

Searching for commonalities in human judgement: The parametric unimodel and its dual mode alternatives. European Review of Social Psychology, 14, 1–47. Evans, J. S. B. T. (2008). Dual-processing accounts of reasoning, judgment, and social cognition. Annual Review of Psychology, 59, 255–278. Fazio, R. H., & Towels-Schwen, T. (1999). The MODE model of attitude–behavior processes. In S. Chaiken & Y. Trope (Eds.), Dual-process theories in social psychology (pp. 97–116). New York: Guilford Press. Gawronski, B., & Bodenhausen, G. V. (2006). Associative and propositional processes in evaluation: An integrative review of implicit and explicit attitude change. Psychological Bulletin, 132, 692. Gawronski, B., & Bodenhausen, G. V. (2009). Operating principles versus operating conditions in the distinction between associative and propositional processes. Behavioral and Brain Sciences, 32, 207–208. Gawronski, B., & Bodenhausen, G. (2011). The associative–propositional evaluation model: Theory, evidence, and open questions. Advances in Experimental Social Psychology, 44, 59–127. Gawronski, B., & Creighton, L. A. (2013). Dual-process theories. In D. A. Carlston (Ed.), The Oxford handbook of social cognition (pp. 282–312). New York: Oxford University Press. Gawronski, B., & Payne, B. K. (2010). Handbook of implicit social cognition: Measurement, theory, and applications. New York: Guilford Press. Gawronski, B., Sherman, J. W., & Trope, Y. (2014). Two of what? A conceptual analysis of dual-process theories. In J. W. Sherman, B. Gawronski, & Y. Trope (Eds.), Dual-process theories of the social mind (pp. 3–19). New York: Guilford Press. Gigerenzer, G., & Regier, T. (1996). How do we tell an association from a rule? Comment on Sloman (1996). Psychological Bulletin, 119, 23–26. Gilbert, D. T. (1989). Thinking lightly about others: Automatic components of the social inference process. In J. S. Uleman & J. A. Bargh (Eds.), Unintended thought (pp. 189– 211). New York: Guilford Press. Gilbert, D. T. (1999). What the mind’s not. In S. Chaiken & Y. Trope (Eds.), Dual-process theories in social psychology (pp. 3–11). New York: Guilford Press. Gilbert, D. T., Krull, D. S., & Malone, P. S. (1990). Unbelieving the unbelievable: Some problems in the rejection of false information. Journal of Personality and Social Psychology, 59, 601–613. Greenwald, A. G. (1968). Cognitive learning, cognitive response to persuasion, and attitude change. In A. Greenwald, T. Brock, & T. Ostrom (Eds.). Psychological foundations of attitudes (pp. 148–170). New York: Academic Press. Greve, W. (2001). Traps and gaps in action explanation: Theoretical problems of a psychology of human action. Psychological Review, 108, 435–451. Haidt, J. (2001). The emotional dog and its rational tail: A social intuitionist approach to moral judgment. Psychological Review, 108, 814. Hartmann, D. (1993). Naturwissenschaftliche Theorien. Wissenschaftstheoretische Grundlagen am Beispiel der Psychologie. [Scientific theories. Theory of science’s foundations taking the example of psychology]. Mannheim, Germany: BI Wissenschaftsverlag. Hempel, C., & Oppenheim, P. (1948). Studies in the logic of explanation. Philosophy of Science, 15, 135–175. Hill, S. E., & Durante, K. M. (2011). Courtship, competition, and the pursuit of attractiveness: Mating goals facilitate health-related risk taking and strategic risk suppression in women. Personality and Social Psychology Bulletin, 37, 383–394. Hofmann, W., & Friese, M. (2008). Impulses got the better of me: Alcohol moderates the influence of implicit attitudes toward food cues on eating behavior. Journal of Abnormal Psychology, 117, 420–427.

Duality Models in Social Psychology 155 Hofmann, W., Friese, M., & Strack, F. (2009). Impulse and self-control from a dual-systems perspective. Perspectives on Psychological Science, 4, 162–176. Hofmann, W., Friese, M., & Wiers, R. W. (2011). Impulsive processes in the self-regulation of health behaviour: Theoretical and methodological considerations in response to commentaries. Health Psychology Review, 5, 162–171. Johnson, B., & Eagly, A. H. (1989). Effects of involvement on persuasion: A meta-analysis. Psychological Bulletin, 106, 290–314. Keren, G., & Schul, Y. (2009). Two is not always better than one: A critical evaluation of two-system theories. Perspectives on Psychological Science, 4, 533–550. Kruglanski, A. W., & Gigerenzer, G. (2011). Intuitive and deliberate judgments are based on common principles. Psychological Review, 118, 97–109. Kruglanski, A. W., & Thompson, E. P. (1999). Persuasion by a single route: A view from the unimodel. Psychological Inquiry, 10, 83–109. Lakatos, I. (1970). Falsification and the methodology of scientific research programmes. In I. Lakatos & A. Musgrave (Eds.), Criticism and the growth of knowledge (pp. 91–195). Cambridge, UK: Cambridge University Press. Lieberman, M. D. (2003). Reflexive and reflective judgment processes: A social cognitive neuroscience approach. In J. P. Forgas, K. D. Williams, & W. von Hippel (Eds.), Social judgments: Implicit and explicit processes (pp. 44–67). New York: Cambridge University Press. Lieberman, M. D. (2007). The X- and C-systems: The neural basis of reflexive and reflective social cognition. In E. Harmon-Jones & P. Winkelman (Eds.), Fundamentals of social neuroscience (pp. 290–315). New York: Guilford Press. Lieberman, M. D., Ochsner, K. N., Gilbert, D. T., & Schacter, D. L. (2001). Do amnesics exhibit cognitive dissonance reduction? The role of explicit memory and attention in attitude change. Psychological Science, 12, 135–140. MacKinnon, D. P., Fairchild, A. J., & Fritz, M. S. (2007). Mediation analysis. Annual Review of Psychology, 58, 593–614. Marquardt, N., & Hoeger, R. (2009). The effect of implicit moral attitudes on managerial decision-making: An implicit social cognition approach. Journal of Business Ethics, 85, 157–171. Marr, D. (1982). Vision: A computational investigation into the human representation and processing of visual information. New York: Freeman. Metcalfe, J., & Mischel, W. (1999). A hot/cool system analysis of delay of gratification: Dynamics of willpower. Psychological Review, 106, 3–19. Mitchell, C. J., De Houwer, J., & Lovibond, P. F. (2009). The propositional nature of human associative learning. Behavioral and Brain Sciences, 32, 183–198. Moors, A. (2014). Examining the mapping problem in dual process models. In J. W. Sherman, Y. Trope, & B. Gawronski (Eds.), Dual-process theories of the social mind (pp. 20–34). New York: Guilford Press. Moors, A., & De Houwer, J. (2006a). Automaticity: A theoretical and conceptual analysis. Psychological Bulletin, 132, 297–326. Moors, A., & De Houwer, J. (2006b). Problems with dividing the realm of processes. Psychological Inquiry, 17, 199–204. Muller, D., Judd, C. M., & Yzerbyt, V. Y. (2005). When moderation is mediated and mediation is moderated. Journal of Personality and Social Psychology, 89, 852–863. Petty, R. E., & Cacioppo, J. T. (1986). The elaboration–likelihood model of persuasion. Advances in Experimental Social Psychology, 19, 123–205. Petty, R. E., Cacioppo, J. T., & Goldman, R. (1981). Personal involvement as a determinant of argument-based persuasion. Journal of Personality and Social Psychology, 41, 847–855.

156

MENTAL STATE THEORIES

Petty, R. E., Fazio, R. H., & Brinol, P. (2008). Attitudes: Insights from the new implicit measures. New York: Psychology Press. Petty, R. E., & Wegener, D. T. (1999). The elaboration likelihood model: Current status and controversies. In S. Chaiken & Y. Trope (Eds.), Dual process theories in social psychology (pp. 41–72). New York: Guilford Press. Popper, K. R. (1934). Logik der Forschung [Logic of scientific discovery]. Vienna: Springer. Plaut, D. C. (1995). Double dissociation without modularity—Evidence from connectionist neuropsychology. Journal of Clinical and Experimental Neuropsychology, 17, 291–321. Roefs, A., Huijding, J., Smulders, F. T. Y., MacLeod, C. M., de Jong, P. J., Wiers, R. W., et al. (2011). Implicit measures of association in psychopathology research. Psychological Bulletin, 137, 149–193. Sherman, J. W., Gawronski, B., & Trope, Y. (Eds.). (2014). Dual-process theories of the social mind. New York: Guilford Press. Skinner, B. F. (1938). The behavior of organisms. New York: Appleton-Century-Crofts. Sloman, S. A. (1996). The empirical case for two systems of reasoning. Psychological Bulletin, 119, 3–22. Smith, E. R., & DeCoster, J. (2000). Dual-process models in social and cognitive psychology: Conceptual integration and links to underlying memory systems. Personality and Social Psychology Review, 4, 108–131. Spencer, S. J., Zanna, M. P., & Fong, G. T. (2005). Establishing a causal chain: Why experiments are often more effective than mediational analyses in examining psychological processes. Journal of Personality and Social Psychology, 89, 845–851. Stacy, A. W., & Wiers, R. W. (2010). Implicit cognition and addiction: A tool for explaining paradoxical behavior. Annual Review of Clinical Psychology, 6, 551–575. Staddon, J. E. R. (2001). The new behaviorism: Mind, mechanism, and society. Philadelphia: Psychology Press. Strack, F. (1992). The different routes to social judgments: Experiential versus informational strategies. In L. L. Martin & A. Tesser (Eds.), The construction of social judgment (pp. 249–275). Hillsdale, NJ: Erlbaum. Strack, F., & Deutsch, R. (2004). Reflective and impulsive determinants of social behavior. Personality and Social Psychology Review, 8, 220–247. Strack, F., & Deutsch, R. (2015). The duality of everyday life: Dual-process and dual-system models in social psychology. In M. Mikulincer, P. R. Shaver, E. Borgida, & J. A. Bargh (Eds.), APA handbook of personality and social psychology: Vol. 1: Attitudes and social cognition (pp. 891–927). Washington, DC: American Psychological Association. Strack, F., Werth, L., & Deutsch, R. (2006). Reflective and impulsive determinants of consumer behavior. Journal of Consumer Psychology, 16, 205–216. Uleman, J. S. (1999). Spontaneous versus intentional inferences in impression formation. In S. Chaiken & Y. Trope (Eds.), Dual-process theories in social psychology (pp. 141–160). New York: Guilford Press. Van Overwalle, F., & Siebler, F. (2005). A connectionist model of attitude formation and change. Personality and Social Psychology Review, 9, 231–274. Wallach, M. A., & Wallach, L. (1998). When experiments serve little purpose—Misguided research in mainstream psychology. Theory and Psychology, 8, 183–194. Watson, J. B. (1913). Psychology as the behaviorist views it. Psychological Review, 20, 158– 177. Wilson, T. D., Lindsey, S., & Schooler, T. Y. (2000). A model of dual attitudes. Psychological Review, 107, 101–126.

8 Personality Systems and the Coherence of Social Behavior Daniel Cervone Tracy L. Caldwell Nicole D. Mayer

I

n personality psychology’s recent history, some scientific challenges have been met successfully, whereas others present less of a success story. The successes are numerous: the identification of major dimensions of interindividual differences (e.g., Marsh et al., 2010) and distinctive patterns of intraindividual variability in social behavior (e.g., Mendoza-Denton & Ayduk, 2012); the documentation of both stability and change in personality across the course of development (e.g., Roberts, Walton, & Viechtbauer, 2006); and the devising of assessment methods and statistical tools that reveal patterning and organization of personality at the level of the individual case (e.g., Hamaker, 2012; Little & Gee, 2007), to name a few. But when it comes to the topic highlighted by the present volume, scientific explanation, the field sometimes has come up short. In this chapter, we point to these shortcomings and, more importantly, outline principles through which personality functioning and person–situation interaction can be explained in a manner that adheres to basic principles of scientific explanation. To achieve these goals, we will first outline these principles. Our focus on scientific explana 157

158

MENTAL STATE THEORIES

tion is central to this volume’s mission, evaluating theories, since the fundamental aim of theories in science is to provide explanation (Salmon, 1989).

Principles of Scientific Explanation and the Study of Personality The key place to turn for information about scientific explanation is philosophy’s literature on the topic (e.g., Woodward, 2003). Capitalizing on this literature does not require a PhD in philosophy. Basic principles of explanation can be readily identified and put to use by the personality psychologist. Philosophy’s explanatory principles primarily concern the explanans, that is, the intellectual framework that provides a scientific explanation of observed phenomena. Before turning to the explanans, we need to identify personality psychology’s main explanandum, that is, the central phenomenon the field seeks to explain. Fortunately, this is not hard; investigators commonly recognize that the central explanandum is individuals’ distinctive patterns of personality coherence.

The Explanandum: Distinctive Patterns of Personality Coherence If people’s thoughts and actions did not cohere—if there were no meaningful connections among psychological experiences occurring at different times and places—there would be no need for a psychology of personality. A psychology of social influence, or perhaps behavioristic principles of learning, would suffice. But, instead, coherence abounds. One person’s friendliness toward old friends may be related to her friendliness toward new acquaintances. Another’s anxieties at work may relate to his argumentativeness at home. Patterns of personality coherence are distinctive. When social thoughts and actions are studied across diverse physical and interpersonal settings, people are found to display distinctive patterns of variability in response (Caldwell, Cervone, & Rubin, 2008; Fournier, Moscowitz, & Zuroff, 2009; Shoda, Mischel, & Wright, 1994). Variable, yet coherent, patterns of response are “signatures” of personality that distinguish individuals from one another (Mischel & Shoda, 1995). Personality coherence refers, specifically, to three interrelated phenomena (Cervone & Shoda, 1999). One is coherence in overt psychological response. Across time and circumstance, people display patterns of behavior that are coherently interconnected. Different theorists have highlighted different types of cross-situational coherence: consistency in mean levels of behavior (Epstein, 1979), variance in responses that are related to a given trait category (Fleeson, 2001, 2007), and other statistical parameters that describe meaningful intraindividual variability in response (Moskowitz

Personality Systems and the Coherence of Social Behavior 159

& Zuroff, 2004). Despite different emphases, there is wide agreement that cross-situationally coherent patterns of response are a hallmark of personality, and thus a core phenomenon to be explained. A second aspect of personality coherence is the coherent interrelation among distinct personality structures and processes. Theorists have distinguished among cognitive, affective, and motivational subsystems of personality (Cervone & Pervin, 2013). In everyday psychological functioning, these “parts” of personality work together as coherent systems (e.g., Kuhl & Koole, 2004). Third, people’s lives have narrative coherence. Even if you’ve got two jobs, three hobbies, and four kids, if someone asks you to describe yourself, you likely will tell a story that weaves diverse aspects of life into a coherent whole. This life story reflects, and contributes to, an integrated sense of self (McAdams, 1993). Below, we focus on the first of these three explananda, cross-situational coherence in response. All three, however, are key to a complete psychology of personality.

Explanation: Basic Principles What principles of explanation should guide personality psychology’s efforts to explain distinctive patterns of personality coherence? Philosophers’ analyses of scientific explanation provide firm guidance. We outline some core principles here.

1. Causal Processes Scientific explanation commonly proceeds through the identification of causal processes. In “bottom-up” strategies of explanation (Salmon, 1989), one searches for a set of mechanisms, causal connections among them, and connections from these underlying mechanisms to the observed phenomena to be explained. In personality, this principle suggests that one should identify underlying causal mechanisms that give rise to observed patterns of personality coherence (Cervone, 1999). As Salmon (1989) explained, bottom-up strategies of explanation vary from “top-down” approaches. In the latter, observed phenomena (the explanandum) are explained by subsuming them under a purportedly lawful regularity (the explanans). The individual event is explained, then, merely by noting that it is statistically expectable according to the top-down principle. Investigators may formulate top-down explanations while being ignorant of causal mechanisms linking abstract, high-level principles to concrete, individual occurrences. The deductive-nomological model of the mid20th century—whose limitations have been detailed extensively (Salmon, 1989; Woodward, 2003)—was the paradigm case of top-down explanation. A similar distinction comes from Lewin. His classic distinction between

160

MENTAL STATE THEORIES

Aristotelian and Galileian strategies of explanation (Lewin, 1935) mirrors the subsequent top-down/bottom-up distinction (Salmon 1989)—which attests to the foresight of Lewin. As Lewin explained, Aristotelian strategies explain an entity’s action by reference to its essential nature, which persists in a fixed manner across environments. Any given entity is categorized as a member of a broad, abstract category. The qualities of the category are said to explain the actions of the individual case. Suppose you start a fire and notice that its flames go up (the explanandum). The explanation is that your fire is a member of a broad category of objects whose essential qualities include a tendency to go upward. The categorization of the individual flame makes its upward tendency expected and thus furnishes explanation. In this Aristotelian framework, individual occurrences that do not conform to what is expected lie “outside of the realm of the comprehensible” (Lewin, 1935, p. 14) and are disregarded. Galileo, Lewin explained, introduced a paradigm in which (1) explanatory concepts do not refer to objects’ typical dispositions (laws of gravitation make no claim about the frequency with which any given object falls) but to causal relations; (2) the causal relations refer to dynamic relations between an entity and the environment in which it acts, rather than to isolated essential qualities of the entity; and (3) scientists seek causal principles that explain not only regularities—average “tendencies”— but also rare, atypical occurrences. 1.A. The Proposed Causal Structures Cannot Feature the Phenomenon to Be Explained.  When identifying causal mechanisms to explain phenomenon X, theorists cannot include X itself in the explanatory mechanisms. If you explain why the sky is blue, you cannot include “blue molecules” or a “blueness dispositional tendency” in your causal story. “A fundamental explanation of the property” under consideration, Nozick (1981, p. 632) noted, “will not refer to things with that very same property.” Similarly, Hanson (1961, p. 94) earlier explained that “an event is explained when it is traced to other events which require less explanation; when it is shown to be part of an intelligible pattern of events.”

2. Manipulability Is Key to Explanation Woodward (2003) explains that manipulation is central to causal claims. If a causal process at Step A is claimed to contribute causally to an observed phenomenon, which occurs at Step B, then “a suitable manipulation at A should produce a change in B” (Woodward, 2003, p. 26). This manipulability criterion entails a temporal sequence; the manipulation occurs prior to the effect it produces. As Woodward notes, the philosopher’s criterion of manipulability dovetails with psychologists’ reliance on experimental manipulation as the standard for establishing causality. As Woodward explains, scientific theories sometimes identify causal mechanisms and processes responsible for a phenomenon (our Principle 1),

Personality Systems and the Coherence of Social Behavior 161

but they are only manipulable in principle, that is, through a hypothetical experiment. For example, we explain tides by reference to the causal impact of the moon’s gravitational force, and “in some hypothetical experiment . . . by varying [the moon’s] distance from the earth, the motion of the tides would change” (Woodward, 2003, p. 129). 2.A. Predictive Information Alone Is Insufficient to Establish Causation.  The manipulability criterion implies that the capacity to predict events is not sufficient grounds for claiming causation. This is true no matter how strong the prediction. As Toulmin (1961) related years ago, the Babylonians could predict astronomical events such as eclipses quite accurately, but made no headway in explaining why the events occurred (also see von Soden, 1985/1994). Ancient Ionians, by contrast, developed insightful ideas about how the universe was structured (they thought Earth resided in a tube surrounded by fire, with pinpricks in the tube being the light of stars). The Ionians were on the path to establishing explanations. As writers have explained (Salmon, 1989; Woodward, 2003), the limitations of predictive information alone highlight, in turn, limitations in the deductive-nomological approach to scientific explanation promoted in the first half of the 20th century (Hempel & Oppenheim, 1948; Popper, 1959). In that approach, occurrences were explained when they were shown to be expectable (Salmon, 1989), or predictable, according to an overarching law. The scientific laws were purportedly “true statements of universal form” (Giere, 1999, p. 86). The nomothetic laws thus provided explanation by virtue of their prediction of occurrences. A popular example which shows that lawful prediction cannot be equated with explanation involves two statements about a flagpole: (A) The length of a shadow cast by a flagpole is predictable from information about height of the flagpole and the angle of the sun. (B) The height of a flagpole is predictable from information about the length of the shadow it casts and the angle of the sun. In a deductive-nomological account, shadow length explains flagpole height to the same degree that flagpole height explains shadow length. But, in reality, shadow length is no explanation of flagpole height, which must be explained in terms of other factors (e.g., decisions made at a flagpole manufacturing company). A second example (cf. Salmon, 1971) is that the event “Joe, who took a birth control pill, did not get pregnant” can be explained by subsuming it under the general law, “birth control pills stop men from getting pregnant.” Despite fitting the general form of deductive-nomological explanation, that is not a good explanation. As Woodward (2003) concludes, the classic deductivenomological approach “does not state sufficient conditions” or “necessary conditions for successful explanation” and the strategy of explaining individual events by subsuming them under nomothetic principles according to which they are expected is “misguided” (p. 181). Such considerations move scientific explanation from a concern with prediction to a concern with causal processes.

162

MENTAL STATE THEORIES

3. Scientific Theories Provide Explanation via Conceptual Models Much work in philosophy highlights the role in scientific explanation of models (Morgan & Morrison, 1999). “Models,” Giere (1999, p. 5) suggests, “are the primary representational entities in science.” Harré (2002) provides a particularly valuable overview of conceptual models and their relevance to psychological science. Models are the scientists’ efforts to depict the world of nature. The manipulable causal elements in the model are meant to represent real-world mechanisms and processes. Inevitably, any model is a simplification; the scientist does not attempt to describe every aspect of the actual physical world—that, in general, is not possible. Instead, the model contains enough information to explain phenomena of interest. A classic case is Bohr’s model of the atom. Bohr surely did not expect that the atom would look precisely like his iconic planetary model. Nonetheless, the model usefully explained atomic and subatomic phenomena. Conceptual models generally portray processes and structures that are unseen. A system of unseen mechanisms, then, is the explanans that provides understanding of observed phenomena. In a connectionist model of cognition (Rumelhart & McClelland, 1986), for example, unseen connections among layers of processing units explain observed cognitive performance. 3.A. Conceptual Models Are Not “True” or “False.”  The philosopher Giere (1999) emphasizes that since models, by design, are simplified depictions of a complex world, they cannot be evaluated according to a true/false criterion. The world is always more complex than any theorist’s model. It thus is inevitable that the model—any model—is not literally true. This point becomes obvious when the model is visual (such as Bohr’s depiction of the atom). Just as a picture of a vacation scene is not true or false, the “images” in a visual scientific model “could not be literally true or false” (Giere, 1999, p. 121). Proctor and Capaldi make a similar point while highlighting pragmatic criteria for evaluating scientific theories. They argue against the idea (cf. Popper, 1959) that “if any prediction [of a theory] is falsified, no matter how many confirming instances there might be, one is duty bound to reject the theory” (Proctor & Capaldi, 2001, p. 766). They note that, in practice, no theoretical model will be perfectly accurate; all can be falsified in some manner. If one rejected an entire conceptual system based on one “falsifying” failure of prediction, one inevitably would reject all conceptual models, some of which might have great pragmatic value. Proctor and Capaldi (2001) explain that “it is not uncommon” for scientists “to accept theories in the absence of their entailing specific predictions if they seem to be useful explanatory devices” (p. 767).

Personality Systems and the Coherence of Social Behavior 163

3.B. Ontology of the Model.  The elements of a conceptual model must be ontologically plausible. There should, in other words, be some empirical grounds for claiming that each element of the model corresponds to something that actually exists. Two considerations are important here. First, proposed structures should be broadly consistent with accepted knowledge in the scientific field; a “gremlins model” is unacceptable because no empirical evidence supports the existence of gremlins. Second, the empirical evidence must be at the appropriate level of analysis. Research on interpersonal tension in group dynamics would not support the claim that all individuals possess a corresponding trait (“interpersonal tensionness”), since interpersonal tension was a phenomenon that only emerged at the level of the group.

Evaluating Personality Psychology According to the Principles The question, then, is whether personality psychology meets a minimal requirement: providing a scientific explanation of personality coherence that adheres to the above principles of explanation. Sometimes it does; we provide some examples below. Yet, at other times, the field falls short. We note two examples—one 40 years back in the field’s history and the other more recent. Forty years ago, personality psychology’s most renowned advancedlevel textbook was that of Wiggins’s (1973). The book achieves its goals with the greatest expertise. But, as Lamiell (1987; also see Lamiell, 2013) insightfully noted, the goals reveal the professional field’s limitations. The “basic paradigm” that Wiggins surveyed was “not deeply concerned with the problem of which of two correlated personality measurements ‘causes’ the other” (this and subsequent quotes in this paragraph from Wiggins, 1973, p. 12). The field apparently was so unaccustomed to concepts of causality that Wiggins placed the word “causes” in scare quotes! The field’s sole goal was prediction, which can proceed in “total ignorance” of causal relations among variables. Some investigators, Wiggins explains, chose to ignore “theoretical niceties” involving causality, whereas others were “openly intolerant” of them. As Wiggins outlines it, the field’s agenda was pre-Aristotelian; it was Babylonian in its interest in numerical prediction and simultaneous disregard of questions of causal explanation. More recently, the five-factor theory of McCrae and Costa (1996) attracted great attention among personality psychologists. It did so despite violating many of the explanatory principles outlined above. In five-factor theory (all quotes from McCrae & Costa, 1996), all individuals are said to possess a common set of five traits that constitute “the universal raw material of personality” (p. 66). The traits are “abstract dispositions” (p. 69), that is, “basic tendencies” (p. 68) to exhibit one versus another type of behavior. The dispositions have causal force; they “influence patterns of thoughts, feelings, and behaviors” (p. 72). Traits themselves, however, are impervious to

164

MENTAL STATE THEORIES

any causal influences other than inherited biology; at the level of the individual person, then, they are fixed qualities. The theory “is focused on the prediction of characteristic adaptations” (skills, social attitudes, personality beliefs), which “constitute the direct and living manifestation” (p. 76) of the traits. As McCrae and Costa recognize, the mechanisms through which traits influence adaptions is unspecified in their theory; it merely is claimed that “dynamic processes” (p. 75) of an undetermined nature mediate the impact of abstract tendencies on concrete thoughts and actions. Personality coherence, then, is explained by locating the individual within a universal system of abstract dispositional tendencies. If someone, for example, characteristically exhibits interpersonal behavior that is agreeable (e.g., avoids arguments; displays sympathy for the poor), his or her behavior is explained by reference to the influence of the high-level trait structure, “agreeableness.” The reader readily can see that five-factor theory’s explanatory strategy violates many of the above principles of explanation: 1.  Causal processes are unspecified. The actions of any given individual are explained by subsuming them under a nomothetic principle, the influence of a universal trait, but the theory does not identify processes linking the trait to the actions. Furthermore, the explanatory strategy is Aristotelian in that (1) individual action is explained by reference to essential properties of the individual that transcend time and place and (2) actions that violate expectations (e.g., occasions in which an extraverted person behaves in an introverted manner) lie outside of the explanatory system and are essentially disregarded. As Mischel and Shoda (1995) emphasized, within this trait framework, variability in action simply falls into the statistical error term. 1.A. The proposed causal structures feature the phenomenon to be explained. Agreeable actions are explained by the trait structure “agreeableness.” “Extraversion” explains the tendency to exhibit the actions that we call “extraverted.” This plainly violates the principle articulated by Hanson (1961) and Nozick (1981). An analogous difficulty arises at the level of mere prediction. In numerous research reports employing five-factor constructs, predictor and outcome variables overlap. The null hypothesis that the reports reject, that predictor and outcome are uncorrelated, is thus implausible. There are two versions of this problem. In one, identical (or semantically equivalent) test items appear in instruments measuring the predictor and the outcome. Consider a study in which mothers (and a second adult) rated their children on qualities including (1) “tends to be shy,” (2) “is very energetic,” and (3) “prefers quiet, inactive games” (items from Rowe & Plomin, 1977). Later in childhood, mothers (and, again, a second adult) rated them on qualities including (1) “shy” (2) “energetic,” and (3) “quiet.” Ratings at time 1 and 2 correlated positively. The results were interpreted as bearing on the ques-

Personality Systems and the Coherence of Social Behavior 165

tion of whether two qualitatively distinct types of constructs, temperament and personality variables, are related (Hagekull & Bohlin, 2003). Furthermore, the results were interpreted as indicating that two different temperament qualities bear on one personality trait; the first set of items were from measures of “sociability” (item 1) and “activity” (items 2 and 3), whereas the second set were from a measure of one personality trait, extraversion. The field’s most senior scholars have articulated the numerous problems entailed in such research practices and interpretations (Kagan, 1989; Mischel, 1968), to frustratingly little effect. In the second version of the problem, researchers document that selfrated traits predict behavior while failing to account for the possibility that the prediction results from the impact of unmeasured prior behavior on the trait ratings. People’s self-views reflect, at least in part, observations of their own behavior (Bem, 1972). One could expect, then, that personality researchers would routinely control for the impact of past behavior when relating trait ratings to concurrent or future behavior. Unfortunately, this is not routine in this subfield. Consider results claimed to “strongly implicate specific personality constructs as underlying determinants of specific complex behaviors” (Paunonen, 2003, p. 421). In the research, college students completed (1) Big Five personality trait measures and (2) a “Behavior Report Form” at one time point. There was thus no way to evaluate whether past behavior influenced current trait ratings and accounted for the correlation between traits and behavior. Despite having none of the information usually required for causality claims in science, in this branch of personality psychology the investigator is licensed to say that personality factors were “underlying determinants.” 2.  The manipulability criterion (Woodward, 2003) cannot be invoked, since the unique causal structure, traits, cannot be manipulated. This is not just a practical matter; it is not merely the case, for example, that it is unethical to manipulate five-factor trait levels. One cannot—even in principle, in a hypothetical experiment—manipulate theoretically specified causal processes because, as noted above, five-factor traits (McCrae & Costa, 1996) are abstract tendencies for which causal processes are unspecified. (This criticism, by contrast, would not apply to personality theories that specify biological mechanisms of temperament and their functioning in the personality system as a whole, for example, Cloninger, Švrakić, & Przybeck, 1993; Pickering & Corr, 2008). Borsboom, Mellenbergh, and van Heerden (2003) have noted a corollary problem: In five-factor theory, an individual’s actions are explained by reference to trait variables that are constants; a person’s level on a trait is viewed as a stable entity. This explanatory strategy thus violates another standard principle of causality, namely, that causes must covary with the phenomenon to be explained (Pearl, 2000). Since the trait is a constant at the level of the individual, it cannot covary with any individual-level occurrences.

166

MENTAL STATE THEORIES

2.B.  In the absence of manipulation, claims in support of the theory rest entirely on predictive information which, in general, is insufficient for causal claims. 3.B.  No empirical evidence supports the ontology of the model. Fivefactor theory claims that the traits are “raw material” (McCrae & Costa, 1996, p. 66) in the psyche of each individual. The five factors are not mere descriptors of between-person differences in the population at large; they are the structural elements of “a model of the person” (Costa & McCrae, 1998, p. 115), singular. However, empirically, the traits are latent variables that emerge from between-person analyses of interindividual differences in the population. There are no grounds for assuming that between-person latent variables correspond to psychological structures that exist, and have causal force, at the level of the individual case, as Borsboom et al. (2003) explain in detail (also see Cramer, Waldorp, van der Mass, & Borsboom, 2010; Molenaar & Campbell, 2009). Harré (1998) notes that five-factor theory also violates the distinction between phenomena and explanation. Dispositional constructs are descriptive; they summarize actions or occurrences that one observes when describing an entity. High-level dispositions, then, “are of the wrong logical type” (Harré, 1998, p. 80) to serve as scientific explanations. Observed phenomena must be explained in terms of underlying structures and processes with causal power, not in terms of abstract descriptions. To view dispositions as explanations “is to confuse dispositions (traits) with powers and liabilities” (Harré, 1998, p. 79). “Dispositions could not be unobservable, explanatory properties of anything. . . . the only explanatory concept that could be imported to explain personal dispositions would be personal powers” (Harré, 1998, p. 79). Explaining a person’s friendly actions by reference to the causal impact of agreeableness, then, is like explaining a car’s quick acceleration by reference to its sportiness. In everyday discourse, the terms (agreeableness, sportiness) are quite useful. But in scientific explanation, they are not. Five-factor theory is not unique in its violations of explanatory principles. In endeavoring to advance a “neo-Allportian” approach to personality (cf. Allport, 1937), Funder (1991) explicitly equated the tasks of description and explanation: “The recognition of a pattern of behavior is a bona fide explanation of each of the behaviors that comprise it . . . identification of a regularity in a person’s behavior is an explanation of the specific instances” (p. 35) of behavior. Funder did “not regard . . . as truly distinct” (p. 35) the tasks of description and explanation—a position at odds with each of the approaches to explanation reviewed earlier and, of course, the explanatory principle articulated by Hanson (1961) and Nozick (1981). The reader can now see why we opened this chapter by evaluating the field negatively. Personality psychology’s history is marked by cases in which central principles of scientific explanation were either ignored or vio-

Personality Systems and the Coherence of Social Behavior 167

lated (also see Caprara & Cervone, 2000), as other writers also have noted (Bandura, 1999; Kagan, 1989). This history underscores the importance of efforts such as the present volume. Fortunately, there is good news, too. Numerous conceptual models in personality psychology are fully consistent with standard principles of scientific explanation. There is no space here for a comprehensive review. However, we note some examples. Kuhl and colleagues’ personality–systems interaction theory (e.g., Kuhl & Koole, 2004) explains experience and action in terms of multiple cognitive systems, affective processes, and volitional modes through which individuals maintain and regulate the self. Epstein’s (1994, 1998) cognitive-experiential self theory posits dual processing modes that explain phenomena identified in both cognitive and psychodynamic conceptions of personality. Finally, in two approaches that are foundational to our own efforts, Bandura (1986, 1999) grounds his theory of personality in cognitive capabilities that function as personal determinants of action, and Mischel and Shoda’s (1995) cognitive-affective framework explains how personality coherence is an emergent property of interactions among elements of the underlying personality system.

Putting Explanatory Principles into Action: Three Personality-Psychological Examples In the remainder of this chapter, we provide three examples in which a phenomenon of interest to personality psychology can be conceptualized in a manner that is consistent with basic principles of scientific explanation.

Cross-Situational Consistency in Self-Efficacy Appraisal The first example involves perceived self-efficacy, that is, people’s perceptions of their capability to perform effectively in the face of challenges and threats (Bandura, 1977). In the study of perceived self-efficacy and neighboring performance-related expectancies and appraisals, two facts are self-evident. On the one hand, self-perceptions can vary markedly across situations and challenges; someone may have high perceived self-efficacy for math but low self-efficacy for ballroom dancing. On the other, any given individual may exhibit cross-situational consistency in self-efficacy appraisal that is meaningful, even if it may be idiosyncratically patterned. Many people may view academic and athletic performance as unrelated, yet in principle, an individual could have low self-efficacy perceptions in both domains if she sees herself as being unable to handle stress (which can arise in either setting). Findings indicate that self-efficacy interventions can generalize from one domain to another (Weitlauf, Cervone, Smith, & Wright, 2001). A phenomenon that requires explanation, then, is cross-situational generalization in perceived self-efficacy. (This phenomenon is a special case

168

MENTAL STATE THEORIES

of the more general questions of cross-situational consistency in experience and behavior.) One explanatory strategy it to posit a high-level variable: generalized self-efficacy (Luszczynska, Scholz, & Schwarzer, 2005). Generalizations across situations would then be viewed as manifestations of the influence of the high-level, generalized tendency to appraise one’s capabilities positively or negatively. This explanatory strategy plainly violates the explanatory principles of explanation outlined above. An alternative strategy grounds explanation in a model of intraindividual personality systems, the knowledge-and-appraisal personality architecture (KAPA) model (Cervone, 2004a). A key principle of the KAPA model is the distinction, deriving from Lazarus and colleagues (Lazarus, 1991; Smith & Lazarus, 1990), between two aspects of cognition: knowledge and appraisal. Knowledge is “our understanding of the way things are and work” (Lazarus, 1991, p. 144), including enduring mental representations of the attributes of entities. Appraisals, in contrast, are dynamic, “online” evaluations of the meaning of encounters for oneself. In the KAPA model, knowledge structures shape appraisals processes through principles of knowledge activation identified in the study of social cognition (Higgins, 1996; Markus & Wurf, 1987). In particular, any given knowledge structure, such as a schematic belief about the self (Markus, 1977), may come to mind in multiple situations, shape self-appraisals in a similar manner in those situations, and thus give rise to cross-situationally consistent patterns of response. The KAPA model’s assessment strategy differs from that outlined by Wiggins (1973), which disregarded causality and aimed for prediction through “literally, anything we can employ to increase the accuracy of forecasts” (p. 12). In the KAPA model, assessments aim to identify psychological structures and processes that causally contribute to observed patterns of personality coherence (Cervone, 2004b; Cervone, Shadel, & Jencius, 2001). In the primary research paradigm (Cervone, 2004a), we employ a battery of assessments designed to identify beliefs and the social circumstances that activate them. Assessments are idiographically tailored. Idiographic assessments of beliefs about the self and social situations are found to predict participants’ subsequent appraisals of self-efficacy for coping with challenges in designated situations. Specifically, our idiographic assessments predict betweenperson and within-person variation in self-efficacy appraisal, as well as the speed with which people appraise their efficacy for performance (Cervone, 1997, 2004a; Cervone, Orom, Artistico, Shadel, & Kassel, 2007; Orom & Cervone, 2009; also see Shadel, Cervone, Niaura, & Abrams, 2004; Wise, 2009). Critically, priming manipulations that vary the cognitive accessibility of selfreferent beliefs alter people’s subsequent appraisals of their efficacy for performance (Cervone et al., 2008). This combination of evidence—(1) predicting cross-situational consistency in the content self-appraisal, (2) predicting the speed of self-appraisal, and (3) documenting the impact of priming on

Personality Systems and the Coherence of Social Behavior 169

self-appraisals—supports the contention that self-schemas causally contribute to self-efficacy appraisal. Note that explanations that reference the concepts of “schema” or “selfschema” are consistent with Principle 1.A, that causal structures cannot contain the phenomenon to be explained. Cognitive scientists have modeled schematic processing in terms of networks of propositional information (Rumelhart & Ortony, 1978), connectionist systems of distributed knowledge (McClelland & Rumelhart, 1986), and embodied cognitive systems (Wilson, 2002). In each case, no individual element of the system corresponds directly to (i.e., “contains”) the phenomenon that the given investigator seeks to explain. Fraley’s (2007) connectionist model of attachment working models shows how personality-relevant schematic processing can be understood through a system of distributed knowledge. The “raw material” of that system does not contain a “disposition to attach”—the raw material is merely a set of 40 linked processing units—and thus does not violate Principle 1.A. The KAPA-based theory and research we reviewed focused on crosssituational consistency in response. The other two aspects of personality coherence, we note, also must be explained while adhering to standard principles of scientific explanation. On these fronts, much progress already has been made. For example, Cloninger and colleagues and Lewis have long shown how the self-organizing properties of complex personality systems inherently address the challenge of explaining the coherent interrelations among personality structures and processes (Cloninger, Švrakić, & Švrakić, 1997; Lewis, 1997). Thagard (2000) explains how narrative coherence emerges from processes of constraint satisfaction.

Personality-in-Context: Humor as a Strategy in Social Interaction Personality develops, and reveals itself, in social interaction (MendozaDenton & Ayduk, 2012). One challenge for personality psychology, then, is to explain aspects of personality coherence which reveal themselves in the strategies that people select to navigate the interpersonal world. A domain in which we have taken up this challenge is the interpersonal strategy of humor use. We will note prior work in this area; evaluate it through the lens of explanatory strategies; and review a recent reconceptualization. Although individuals use humor for a variety of reasons, its assessment has historically focused on uses that are potentially adaptive (e.g., humor used to cope with adversity; for reviews, see Martin, 2001, 2004, 2007). In response, Martin, Puhlik-Doris, Larsen, Gray, and Weir (2003) designed an assessment tool, the Humor Styles Questionnaire (HSQ) to increase the comprehensiveness of humor assessment. The HSQ assesses four styles: affiliative (used to bring people together), self-enhancing (used to cope with adversity), aggressive (used to put others down), and self-defeating (used to gain the approval of others by putting oneself down). Individuals use a

170

MENTAL STATE THEORIES

7-point Likert scale to rate their agreement with items such as “If I am feeling depressed, I can usually cheer myself up with humor” (from the selfenhancing subscale). The sum of each of their subscale scores indicates how high or low they are in that particular humor style. Between-person variations in each of these scores is then used to predict between-person variations in various outcomes, including health and well-being (for reviews, see Lefcourt & Thomas, 1998; Martin, 2001, 2004, 2007). Martin et al.’s (2003) HSQ admirably succeeds in increasing the comprehensiveness of humor assessment. Yet, viewed as an approach to explaining humor use by the individual, it fails to conform to the explanatory principles outlined above. First, theirs is a top-down approach that treats humor as a higher level principle that subsumes disparate behaviors (Principle 1, above). For example, the tendency to tease others and the tendency to be unconcerned with how others perceive one’s humor (both assessed by items on the aggressive humor subscale) are each subsumed under the category “aggressive humor,” without regard for whether aggressive intentions actually drove either behavior. Second, variations in the explanandum—humor— involve an explanans that itself includes humor (I.A.); that is, Martin et al. explain variations in the use of aggressive humor by assessing variations in aggressive humor, thereby conflating the two activities. Third, because there is no causal model of the structures and processes that give rise to humorous tendencies (3), and also no ontologically plausible “parts” (3.B.), it fails on the manipulability criterion (2); that is, it is not clear what one could manipulate to causally explain, for example, why any given person is prone to using humor aggressively. Ideally, one would assess humor by measuring not only its behavioral indicators (e.g., doing or saying something humorous) but also the relevant personality structures that underlie those behaviors—structures that may be activated by a unique array of situations for each individual. An alternative approach to the assessment of humor—one that does not violate the principles of scientific explanation—would begin with a causal model that could guide the choice of which psychological structures to assess. This was the approach taken by Caldwell, Cervone, and Rubin (2008), who relied on the KAPA model to explain coherent yet potentially idiosyncratic within-person variations in the use of humor. The architecture outlined in the KAPA model is one that fosters the exploration of what processes may contribute to consistent patterns of within-person variability in humorous behavior. Within the KAPA approach, most behaviors are assumed to be preceded by ongoing appraisals of encounters (e.g., the decision to engage in a challenging behavior is preceded by the appraisal that one is capable of performing that behavior). The strategy suggested by the KAPA, then, is to assess the dynamic appraisals that precede the use of humor, as well as the stable self-knowledge structures that foster coherence in those appraisals. The general strategy adopted by Caldwell et al. (2008) was to assess the self-knowledge that contributes to the use of humor, to assess the contexts in which that knowledge was likely to be activated, and to use that informa-

Personality Systems and the Coherence of Social Behavior 171

tion to predict the appraisals of the likelihood of using humor across those same situations. Self-knowledge was assessed idiographically by asking participants, via open-ended question, to describe their own reasons for using humor. This technique invited individuals to describe their own schematic beliefs without the constraints inherent in nomothetic assessment. However, because it was possible that the HSQ does in fact assess the full range of humor (thereby precluding the need for idiographic assessment), participants’ humor styles were also assessed nomothetically by having them fill out the HSQ. To learn of the situations that would likely activate participants’ reasons for using humor, the researchers gave participants questionnaires presenting them with a list of 53 social situations in which they might find themselves (e.g., You are shopping by yourself in a distant city and unexpectedly see an acquaintance from school or work) and the participants’ task was to rate the extent to which a given type of self-knowledge was relevant. For example, if a participant had noted that she used humor to “make people laugh and be happy,” one of her questionnaires would include the instruction, “Please rate, for each situation, how likely it is that you would want to make people laugh and be happy.” They did this for two idiographically identified reasons for using humor and for the four HSQ-identifed reasons. For the same set of situations, the researchers assessed the dependent variable by asking participants to rate the likelihood that they would use humor. Caldwell et al. (2008) predicted that there would be within-person variability in the self-rated likelihood of using humor across situations and that this variability would be predicted by variations in the relevance of self-knowledge to those situations. This hypothesis was supported; withinperson patterns of variability in appraisals of the likelihood of using humor were idiosyncratic, and these idiosyncrasies were explained by within-person variations in the self-knowledge activated by each situation. Importantly, this was true of participants’ idiographically identified reasons for using humor, but not their HSQ-identified reasons for humor predicted their use of humor, even if they were high scorers on any of the four humor styles. Learning that an individual scored higher, relative to others, on the four HSQ subscales did not translate into being able to explain his or her use of humor in any given situation. The Caldwell et al. (2008) assessment of humor serves as an example of assessment that meets the criteria for scientific explanation. It is informed by a causal model (3) whose processes are well specified (1) and whose structures (e.g., beliefs about the self) are simultaneously ontologically plausible (3B) and manipulable (2); one could theoretically run an experiment in which schematic self-knowledge about one’s reasons for humor would cause variations in appraisals of the likelihood of using humor, in contexts relevant to that self-knowledge. In so doing, one would move away from predictive models that, by virtue of their focus on between-person variations, do not enable one to make causal claims at the level of each individual (2.A). Finally, this assessment technique assesses its explanandum, variations in humor-

172

MENTAL STATE THEORIES

ous behavior, without resorting to including humorous behavior itself as the explanans (I.A).

Rethinking Reliability A third domain in which explanatory principles can be brought to bear is the conception of reliability of measurement. We usually are taught that reliability is “a property of a test” (Traub & Rowley, 1991, p. 44), perhaps with the qualification that it references “a property of a test” when that test is “administered to a particular population of examinees under certain conditions” (Miller, 1995, p. 256). Either way, the term traditionally describes a property of a measuring device. This conception of reliability has enormous practical utility and thus is of enduring value. Yet it looks unusual when viewed through the lens of causality-based scientific explanation. A test, in and of itself, is not an agent with causal powers. The test cannot produce the pattern of responses that is classified as “reliable” or not. The agent who is responsible for those responses is, of course, the individual (or population of individuals) taking the test. This obvious point has the following implication. Let us take, as our explanandum, the reliability of test responses. If people respond to a test in a reliable manner, we need to explain why; we need, in other words, to determine what led individuals to respond reliably to the test items. What is required, then, is an explanation of the behavior of the people who took the test. Reliability, in other words, can be reconceptualized as a property of persons, rather than a property of tests, per se; the statement “the test is reliable” is a statement about the behavior of people when they responded to test items. This reconceptualization shifts one’s analysis to causal properties that function at the level of the individual person, rather than merely to properties of the test. Since this conceptualization is novel, we will reflect on the more traditional conceptualization. The phenomenon of reliability consists, descriptively, of relations among test item responses. One may compute relations among responses at any given time or, if people take the test more than once, one may compute total test scores and examine the correlation of test scores across occasions. Consistency of total scores across time would indicate that a test possesses test–retest reliability, whereas strong interrelations among test items (commonly indexed by coefficient alpha; Miller, 1995) would indicate that the test is internally reliable, or internally “consistent” (Onwuegbuzie & Daniel, 2002). Why would people respond to different test items in a consistent manner, or respond consistently on a test across testing occasions? In the traditional psychometric treatment of reliability, such reliable responses are explained (if a scientific explanation is offered) by reference to the psychological construct that the test is measuring. People respond reliably to a test of authoritarianism, for example, because authoritarianism somehow guides

Personality Systems and the Coherence of Social Behavior 173

their responses to test items. Unreliability—variability in response—is treated as statistical error that reflects limitations in the test. This strategy of explanation is, again, a top-down approach in which causal processes linking the abstract psychological attribute to concrete actions by the individual are unspecified. Researchers have, of course, identified cognitive processes that occur during test responding (Blair & Burton, 1987; Schwarz, 1999). Our point is that the psychometric property of reliability has not, to our knowledge, been explained in terms of a causal model of such processes. Before positing an explanatory mechanism, we note a phenomenon that would further motivate the reconceptualization considered here. It is possible that different people will display different levels of reliability when taking the same test. One person or subgroup of people may respond reliably, whereas another may not. If so, the position that reliability is “a property of a test” is plainly unhelpful in explaining, predicting, or even describing the observed phenomenon. If reliability were a property of the test, per se, then different individuals and groups should show similar levels of reliability when completing the same test items. One could retain the traditional concept while claiming that reliability refers to consistency when “a particular population of examinees” takes a test. But that position is little more than an after-the-fact description of the phenomenon (differential reliability on the same test, among different persons). We suggest that one psychological mechanism that might causally contribute to reliable test responding is schematicity. People who possess a schema (an elaborate cognitive structure that guides information processing) in a domain that is being assessed may respond more reliably to test items tapping that domain. Our reasoning is based on the principle that people’s response to any target of judgment—including a test item—may occur through any of a variety of cognitive processes (Forgas, 1995). When judgments are formed through either heuristic or deliberate processing, any of a variety of context effects may influence a person’s response. Since these context effects could vary from one testing occasion to another, or one item of a test to another (if contexts include recently answered questions; see Schwarz, 1999), they would lower the psychometric reliability of test responses. However, cognitive processing that is schema-driven would be relatively impervious to such contextual effects. Responses thus could be more consistent, increasing reliability. Similar reasoning was advanced by Nasby (1989). He reasoned that people who are high in private self-consciousness, and thus are likely to possess a self-consciousness schema, will respond more consistently across testing sessions than people lacking a schema. Results supported this hypothesis and demonstrated the phenomenon noted above; namely, different subgroups displayed different levels of test–retest reliability when completing the same measuring instrument (Nasby, 1989). More evidence is required to verify the link between schematicity and reliability (Mayer, 2012). For now, our point merely is that the concept of

174

MENTAL STATE THEORIES

reliability can be recast. Borsboom, Mellenbergh, and van Heerden (2004), we note, provide a similar recasting of the concept of validity. Those authors observe that researchers traditionally have conceptualized validity in correlational terms; a measure validly indexes a construct if empirical results display expected correlational patterns within a theorized nomological network. In contrast, Borsboom et al. turn to causality: Measurement instruments are valid if the attribute being assessed exerts a causal impact on the measuring instrument. Borsboom et al.’s concept of validity thus moves the field from unexplained correlations to causal-process-based explanations— exactly the move we are calling for in this chapter.

Conclusion Throughout the past half century, personality psychologists have frequently reflected on their field’s merits and flaws. Yet the self-assessments commonly have been incomplete. Partly in response to Mischel’s (1968) critique, investigators turned their attention to technical issues involved in solving an empirical problem: how best to predict behavior. While doing so, they devoted lesser attention to deeper conceptual issues involving the ontology of personality psychology’s constructs. The result, as the great historian of psychology, Kurt Danziger, has explained, was a field in which “historical amnesia took over” (Danziger, 2013, p. 60). The use of consensually accepted “instruments . . . transformed conceptual questions into technical issues. As a result, some of the most basic assumptions of the field disappeared from view” (Danziger, 2013, p. 60). The topic of this chapter and this volume, scientific explanation, inherently brings some of those assumptions back into the field’s line of sight. In light of the field’s history, we close with the following point. Our critique of explanatory strategies rests on conceptual issues, not empirical findings. Enhanced behavioral prediction does not solve the conceptual problems. Is the correlation between rated “conscientiousness” of persons and the actual behavior of those persons .3 or .7? Is the correlation between rated “reliability” of cars and actual future performance of those cars .3 or .7? If you’re selecting employees for a job or cars from a lot, you’ll want to know the answer. But if you’re building an explanatory model of a person or a car, neither correlation licenses you to treat the constructs as raw material with causal force. If you do so, you’ve got “a lot of explaining to do.” References Allport, G. W. (1937). Personality: A psychological interpretation. New York: Holt. Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behavioral change. Psychological Review, 84, 191–215.

Personality Systems and the Coherence of Social Behavior 175 Bandura, A. (1986). Social foundations of thought and action. Englewood Cliffs, NJ: Prentice Hall. Bandura, A. (1999). Social cognitive theory of personality. In D. Cervone & Y. Shoda (Eds.), The coherence of personality: Social-cognitive bases of consistency, variability, and organization (pp. 185–241). New York: Guilford Press. Bem, D. J. (1972). Self-perception theory. Advances in Experimental Social Psychology, 6, 1–62. Blair, E., & Burton, S. S. (1987). Cognitive process used by survey respondents to answer behavioral frequency questions. Journal of Consumer Research, 14, 280–288. Borsboom, D., Mellenbergh, G. J., & van Heerden, J. (2003). The theoretical status of latent variables. Psychological Review, 110, 203–219. Borsboom, D., Mellenbergh, G. J., & van Heerden, J. (2004). The concept of validity. Psychological Review, 111, 1061–1071. Caldwell, T. L., Cervone, D., & Rubin, L. H. (2008). Explaining intra-individual variability in social behavior through idiographic assessment: The case of humor. Journal of Research in Personality, 42, 1229–1242. Caprara, G. V., & Cervone, D. (2000). Personality: Determinants, dynamics, and potentials. New York: Cambridge University Press. Cervone, D. (1997). Social-cognitive mechanisms and personality coherence: Self-knowledge, situational beliefs, and cross-situational coherence in perceived self-efficacy. Psychological Science, 8, 43–50. Cervone, D. (1999). Bottom-up explanation in personality psychology: The case of crosssituational coherence. In D. Cervone & Y. Shoda (Eds.), The coherence of personality: Social-cognitive bases of personality consistency, variability, and organization (pp. 303– 341). New York: Guilford Press. Cervone, D. (2004a). The architecture of personality. Psychological Review, 111, 183–204. Cervone, D. (2004b). Personality assessment: Tapping the social-cognitive architecture of personality. Behavior Therapy, 35, 113–130. Cervone, D., Caldwell, T. L., Fiori, M., Orom, H., Shadel, W. G., Kassel, J., et al. (2008). What underlies appraisals? Experimentally testing a knowledge-and-appraisal model of personality architecture among smokers contemplating high-risk situations. Journal of Personality, 76, 929–967. Cervone, D., Orom, H., Artistico, D., Shadel, W. G., & Kassel, J. (2007). Using a knowledgeand-appraisal model of personality architecture to understand consistency and variability in smokers’ self-efficacy appraisals in high-risk situations. Psychology of Addictive Behaviors, 21, 44–54. Cervone, D., & Pervin, L. A. (2013). Personality: Theory and research (12th ed.). Hoboken, NJ: Wiley. Cervone, D., Shadel, W. G., & Jencius, S. (2001). Social-cognitive theory of personality assessment. Personality and Social Psychology Review, 5, 33–51. Cervone, D., & Shoda, Y. (1999). Social-cognitive theories and the coherence of personality. In D. Cervone & Y. Shoda (Eds.), The coherence of personality: Social-cognitive bases of consistency, variability, and organization (pp. 3–33). New York: Guilford Press. Cloninger, C. R., Švrakić, D. M., & Przybeck, T. R. (1993). A psychobiological model of temperament and character. Archives of General Psychiatry, 50, 975–990. Cloninger, C. R., Švrakić, N. M., & Švrakić, D. M. (1997). Role of personality self-organization in development of mental order and disorder. Development and Psychopathology, 9, 881–906. Costa, P. T., Jr., & McCrae, R. R. (1998). Trait theories of personality. In D. F. Barone, M. Hersen, & V. B. Van Hasselt (Eds.), Advanced personality (pp. 103–121). New York: Plenum Press.

176

MENTAL STATE THEORIES

Cramer, A. O. J., Waldorp, L. J., van der Maas, H. L. J., & Borsboom, D. (2010). Comorbidity: A network perspective. Behavioral and Brain Sciences, 33, 137–193. Danziger, K. (2013). Historical psychology of persons: categories and practice. In J. Martin & M. H. Bickhard (Eds.), The psychology of personhood: Philosophical, historical, socialdevelopmental, and narrative perspectives (pp. 59–80). New York: Cambridge University Press. Epstein, S. (1979). The stability of behavior: On predicting most of the people much of the time. Journal of Personality and Social Psychology, 37, 1097–1126. Epstein, S. (1994). Integration of the cognitive and the psychodynamic unconscious. American Psychologist, 49, 709–724. Epstein, S. (1998). Cognitive-experiential self-theory. In D. Barone, M. Hersen, & V. B. Van Hasselt (Eds.), Advanced personality (pp. 211–238). New York: Plenum. Fleeson, W. (2001). Toward a structure- and process-integrated view of personality: Traits as density distributions of states. Journal of Personality and Social Psychology, 80, 1011–1027. Fleeson, W. (2007). Situation-based contingencies underlying trait-content manifestation in behavior. Journal of Personality, 75, 825–861. Forgas, J. P. (1995). Mood and judgment: The affect infusion model (AIM). Psychological Bulletin, 117, 39–66. Fournier, M. A., Moskowitz, D. S., & Zuroff, D. C. (2009). The interpersonal signature. Journal of Research in Personality, 43, 155–162. Fraley, R. C. (2007). A connectionist approach to the organization and continuity of working models of attachment. Journal of Personality, 75, 1157–1180. Funder, D. C. (1991). Global traits: A neo-Allportian approach to personality. Psychological Science, 2, 31–39. Giere, R. N. (1999). Science without laws. Chicago: University of Chicago Press. Hagekull, B., & Bohlin, G. (2003). Early temperament and attachment as predictors of the five factor model of personality. Attachment and Adult Development, 5, 2–18. Hamaker, E. L. (2012). Why researchers should think “within-person”: A paradigmatic rationale. In M. R. Mehl & T. S. Conner (Eds.), Handbook of research methods for studying daily life (pp. 43–61). New York: Guilford Press. Hanson, N. R. (1961). Patterns of discovery: An inquiry into the conceptual foundations of science. Cambridge, UK: Cambridge University Press. Harré, R. (1998). The singular self: An introduction to the psychology of personhood. London: Sage. Harré, R. (2002). Cognitive science: A philosophical introduction. London: Sage. Hempel, C. G., & Oppenheim, P. (1948). Studies in the logic of explanation. Philosophy of Science, 15, 135–175. Higgins, E. T. (1996). Knowledge activation: Accessibility, applicability, and salience. In E. T. Higgins & A. W. Kruglanski (Eds.), Social psychology: Handbook of basic principles (pp. 133–168). New York: Guilford Press. Kagan, J. (1989). Unstable ideas: Temperament, cognition, and self. Cambridge, MA: Harvard University Press. Kuhl, J., & Koole, S. L. (2004). Workings of the will: A functional approach. In J. Greenberg, S. L. Koole, & T. Pyszczynski (Eds.), Handbook of experimental existential psychology (pp. 411–430). New York: Guilford Press. Lamiell, J. T. (1987). The psychology of personality: An epistemological inquiry. New York: Columbia University Press. Lamiell, J. T. (2013). Critical personalism: On its tenets, its historical obscurity, and its future prospects. In J. Martin & M. H. Bickhard (Eds.), The psychology of personhood: Philosophical, historical, social-developmental, and narrative perspectives (pp. 101–126). New York: Cambridge University Press.

Personality Systems and the Coherence of Social Behavior 177 Lazarus, R. S. (1991). Emotion and adaptation. New York: Oxford University Press. Lefcourt, H. M., & Thomas, S. (1998). Humor and stress revisited. In W. Ruch (Ed.), The sense of humor: Explorations of a personality characteristic (pp. 179–202). Berlin: Walter de Gruyter. Lewin, K. (1935). The conflict between Aristotelian and Galileian modes of thought in contemporary psychology. In K. Lewin (Ed.), A dynamic theory of personality: Selected papers (pp. 1–42). New York: McGraw-Hill. Lewis, M. D. (1997). Personality self-organization: Cascading constraints on cognition– emotion interaction. In A. Fogel, M. C. D. P. Lyra, & J. Valsiner (Eds.), Dynamics and indeterminism in developmental and social processes (pp. 193–216). Mahwah, NJ: Erlbaum. Little, B. R., & Gee, T. L. (2007). The methodology of personal projects analysis: Four modules and a funnel. In B. R. Little, K. Salmela-Aro, & S. D. Phillips (Eds.), Personal project pursuit: Goals, action and human flourishing (pp. 51–93). Mahwah, NJ: Erlbaum. Luszczynska, A., Scholz, U., & Schwarzer, R. (2005). The general self-efficacy scale: Multicultural validation studies. Journal of Psychology, 139, 439–457. Markus, H. (1977). Self-schemata and processing information about the self. Journal of Personality and Social Psychology, 35, 63–78. Markus, H., & Wurf, E. (1987). The dynamic self-concept: A social psychological perspective. Annual Review of Psychology, 38, 299–337. Marsh, H. W., Lüdtke, O., Muthén, B., Asparouhov, T., Morin, A. J. S., Trautwein, U., et al. (2010). A new look at the big-five factor structure through exploratory structural equation modeling. Psychological Assessment, 22, 471–491. Martin, R. A. (2001). Humor, laughter, and physical health: Methodological issues and research findings. Psychological Bulletin, 127, 504–519. Martin, R. A. (2004). Sense of humor and physical health: Theoretical issues, recent findings, and future directions. Humor, 17, 1–19. Martin, R. A. (2007). The psychology of humor: An integrative approach. Burlington, MA: Elsevier. Martin, R. A., Puhlik-Doris, P., Larsen, G., Gray, J., & Weir, K. (2003). Individual differences in uses of humor and their relation to psychological well-being: Development of the humor styles questionnaire. Journal of Research and Personality, 37, 48–75. Mayer, N. (2012). The concept of reliability: Putting the “psyche” into psychometric. Master’s thesis, University of Illinois at Chicago. McAdams, D. P. (1993). The stories we live by: Personal myths and the making of the self. New York: Morrow. McClelland, J. L., & Rumelhart, D. E. (1986). A distributed model of human learning and memory. In D. E. Rumelhart, J. L. McClelland, & PDP Research Group (Eds.), Parallel distributed processing: Explorations in the microstructure of cognition: Vol. 2. Psychological and biological models (pp. 170–215). Cambridge, MA: MIT Press. McCrae, R. R., & Costa, P. T. (1996). Toward a new generation of personality theories: theoretical contexts for the five-factor model. In J. S. Wiggins (Ed.), The five-factor model of personality: Theoretical perspectives (pp. 51–87). New York: Guilford Press. Mendoza-Denton, R., & Ayduk, O. (2012). Personality and social interaction: Interpenetrating processes. In K. Deaux & M. Snyder (Eds.), The Oxford handbook of personality and social psychology (pp. 446–466). New York: Oxford University Press. Miller, M. B. (1995). Coefficient alpha: A basic introduction from the perspectives of classical test theory and structural equation modeling. Structural Equation Modeling, 2, 255–273. Mischel, W. (1968). Personality and assessment. New York: Wiley. Mischel, W., & Shoda, Y. (1995). A cognitive-affective system theory of personality: Recon-

178

MENTAL STATE THEORIES

ceptualizing situations, dispositions, dynamics, and invariance in personality structure. Psychological Review, 102, 246–286. Molenaar, P. C. M., & Campbell, C. G. (2009). The new person-specific paradigm in psychology. Current Directions in Psychology, 18, 112–117. Morgan, M. S., & Morrison, M. (1999). Models as mediating agents. In M. S. Morgan & M. Morrison (Eds.), Models as mediators: Perspectives on natural and social science (pp. 10–37). New York: Cambridge University Press. Moskowitz, D. S., & Zuroff, D. C. (2004). Flux, pulse, and spin: Dynamic additions to the personality lexicon. Journal of Personality and Social Psychology, 86, 880–893. Nasby, W. (1989). Private self-consciousness, self-awareness, and the reliability of selfreports. Journal of Personality and Social Psychology, 56, 950–957. Nozick, R. (1981). Philosophical explanations. Cambridge, MA: Belknap Press of Harvard University Press. Onwuegbuzie, A. J., & Daniel, L. G. (2002). A framework for reporting and interpreting internal consistency reliability estimates. Measurement and Evaluation in Counseling and Development, 35, 89–103. Orom, H., & Cervone, D. (2009). Personality dynamics, meaning, and idiosyncrasy: Identifying cross-situational coherence by assessing personality architecture. Journal of Research in Personality, 43, 228–240. Paunonen, S. V. (2003). Big five factors of personality and replicated predictions of behavior. Journal of Personality and Social Psychology, 84, 411–424. Pearl, J. (2000). Causality: Models, reasoning, and inference. Cambridge, UK: Cambridge University Press. Pickering, A. D., & Corr, P. J. (2008). J. A. Gray’s reinforcement sensitivity theory (RST) of personality. In G. J. Boyle, G. Matthews, & D. H. Saklofske (Eds.), The Sage handbook of personality theory and assessment: Vol. 1. Personality theories and models (pp. 33–55). Los Angeles: Sage. Popper, K. R. (1959). The logic of scientific discovery. New York: Basic Books. Proctor, R. W., & Capaldi, E. J. (2001). Empirical evaluation and justification of methodologies in psychological science. Psychological Bulletin, 127, 759–772. Roberts, B. W., Walton, K. E., & Viechtbauer, W. (2006). Patterns of mean-level change in personality traits across the life course: A meta-analysis of longitudinal studies. Psychological Bulletin, 132, 3–27. Rowe, D. C., & Plomin, R. (1977). Temperament in early childhood. Journal of Personality Assessment, 41, 150–156. Rumelhart, D. E., McClelland, J. L., & PDP Research Group. (1986). Parallel distributed processing: Explorations in the microstructure of cognition: Vol. 1. Foundations. Cambridge, MA: MIT Press. Rumelhart, D. E., & Ortony, A. (1978). The representation of knowledge in memory. In R. C. Anderson, R. J. Spiro, & W. E. Montague (Eds.), Schooling and the acquisition of knowledge (pp. 99–135). Hillsdale, NJ: Erlbaum. Salmon, W. C. (1971). Statistical explanation and statistical relevance. Pittsburgh, PA: University of Pittsburgh Press Salmon, W. C. (1989). Four decades of scientific explanation. In P. Kitcher & W. C. Salmon (Eds.), Minnesota studies in the philosophy of science: Vol. XIII. Scientific explanation (pp. 3–219). Minneapolis: University of Minnesota Press. Schwarz, N. (1999). Self-reports: How the questions shape the answers. American Psychologist, 54, 93–105. Shadel, W. G., Cervone, D., Niaura, R., & Abrams, D. B. (2004). Developing an integrative social-cognitive strategy for personality assessment at the level of the individual:

Personality Systems and the Coherence of Social Behavior 179 An illustration with regular cigarette smokers. Journal of Research in Personality, 38, 394–419. Shoda, Y., Mischel, W., & Wright, J. C. (1994). Intraindividual stability in the organization and patterning of behavior: Incorporating psychological situations into the idiographic analysis of personality. Journal of Personality and Social Psychology, 67, 674–687. Smith, C. A., & Lazarus, R. S. (1990). Emotion and adaptation. In L. A. Pervin (Ed.), Handbook of personality: Theory and research (pp. 609–637). New York: Guilford Press. Thagard, P. (2000). Coherence in thought and action. Cambridge, MA: Bradford Books/MIT Press. Toulmin, S. (1961). Foresight and understanding: An enquiry into the aims of science. Bloomington: Indiana University Press. Traub, R. E., & Rowley, G. L. (1991). Understanding reliability. Educational Measurement: Issues and Practice, 10, 37–45. von Soden, W. (1994). The ancient Orient: An introduction to the study of the ancient Near East (D. G. Schley, Trans.). Grand Rapids, MI: Eerdmans. (Original work published 1985) Weitlauf, J., Cervone, D., Smith, R. E., & Wright, P. M. (2001). Assessing generalization in perceived self-efficacy: Multidomain and global assessments of the effects of selfdefense training for women. Personality and Social Psychology Bulletin, 27, 1683–1691. Wiggins, J. S. (1973). Personality and prediction: Principles of personality assessment. Reading, MA: Addison-Wesley Wilson, M. (2002). Six views of embodied cognition. Psychonomic Bulletin and Review, 9, 625–636. Wise, J. B. (2009). Using the knowledge-and-appraisal personality architecture to predict physically active leisure self-efficacy in university students. Journal of Applied Social Psychology, 39, 1913–1927. Woodward, J. (2003). Making things happen: A theory of causal explanation. New York: Oxford University Press.



Part III Biological Theories





9 Cognitive Neuroscience of Social Behavior Jennifer S. Beer

S

ocial and personality psychologists seek to understand the psychological processes of the mind, so why might they care about social neuroscience, which focuses on neural function? In contrast to the dualist view that mind and body are separate, a fundamental principle of cognitive neuroscience is that the psychological processes of the mind are associated with operations measured at the level of neural activity (Gazzaniga, Ivry, & Mangun, 2008). Social neuroscientists, in particular, are interested in understanding the neural function that is associated with social, cognitive, affective, and personality processes. There is much debate and history regarding the precise way in which operations occurring at the neural level of analyses may relate to the psychological processes of the mind. A major point of contention is the degree to which brain regions are specialized for particular mental processes or whether mental processes are relatively distributed across the brain. Some of the more extreme original theories have been dismissed. For example, little focus is placed on testing the central premises of phrenology such as whether there is a one-to-one correlation between the function of a specific brain structure and a specific mental process or whether the size of brain structures can be determined from the surface of the skull (Gazzaniga et al., 2008; Kanwisher, 2010). Modern perspectives conceptualize psychological processes as arising from patterns of brain activation in which particular regions may play a role in particular kinds of computations, repre 183

184

BIOLOGICAL THEORIES

sentations, or the transfer of information when working within a particular network. Therefore, cognitive neuroscience theories suggest that the level of analysis of interest to social neuroscientists is related to the level of analyses of interest to social and personality psychologists. This chapter considers the ways in which social neuroscience research can be valuable for social and personality psychologists who are interested in psychological processes rather than the brain function per se. A common refrain is whether psychologists learn anything from studies showing an association between brain activation and a psychological process. Not every experiment from the field of social neuroscience is motivated by a research question about the psychological process. However, particular experimental designs can help advance knowledge about psychological processes rather than simply creating a map of the brain. Specifically, researchers interested in psychological processes will find that social neuroscience methodology may provide opportunities to (1) manipulate processes that are difficult to manipulate using more traditional social–personality processes and (2) examine the similarities and differences in the computations that underlie behavioral manifestations of social and personality processes. The chapter illustrates these benefits and provides the reader with information about the types of neuroscience methodologies that have the potential inform social and personality psychological processes, the relation of experimental design to the research question, and the interpretation issues associated with these methodologies and experimental designs.

Neuroscience Methodologies Commonly Used to Investigate Social and Personality Psychological Processes There are numerous neuroscience methodologies that allow researchers to understand the associations between the mind and the neural level of analyses. Currently, three methodologies are most often employed in the type of social neuroscience research that may have value for social and personality psychologists: patient population approaches, functional magnetic resonance imaging (fMRI), and event-related potentials (ERPs). This section describes each methodology and points out each methodology’s unique set of constraints for experimental design (see Table 9.1). The value of each approach for research questions focused on the level of psychological process will be discussed in the next section of the chapter (see For Researchers Interested in Developing Theories of Psychological Processes).

Patient Population Approaches One of the earliest methodologies used to understand how the brain relates to the mind is the study of patient populations (Beer, 2009; Kolb & Whishaw, 2003; Rorden & Karnath, 2004). A premise of patient methodology is that def-

Cognitive Neuroscience of Social Behavior 185

TABLE 9.1.  Interpretations Associated with Patient, fMRI, and ERP Approaches Method

Interpretation

Patients with Psychological changes psychiatric or associated with diseased or medical disorders poorly functioning brain

Focal lesion patients

Limitations Difficult to isolate particular psychological processes Generalization to healthy populations is problematic because brain may develop differently, lack of random assignment

Psychological changes Impairment may result from critically related to localized region’s function or lack of damage to brain communication to/from another structure Lacks random assignment

fMRI

Regions that show differential activation in association with psychological process

Cannot conclude whether region is necessary or sufficient for psychological process, only an association

EEG/ERP

Provides good temporal resolution of how brain activation relates to psychological processes

Source localization procedures need more development in order to identify brain region that generates signal of interest

icits can be used to understand the relation of brain function to psychological function. Experiments are typically motivated by the question of how a specific area of brain damage impairs (or does not impair) specific behaviors. Therefore, experiments typically compare patients with damage of interest, patients with damage of noninterest, and a healthy control group. This ensures that effects are associated with the area of interest rather than brain damage in general. This approach typically involves patients with disorders (e.g., depression, autism, epilepsy, dementia) or those who have sustained focal lesions (e.g., traumatic brain injury, tumor resection). For social and personality psychologists, this approach tends to be most helpful for investigating the psychological processes that are undermined by these disorders. For example, if we want to understand how negative mood affects attitudes, then patients with major depression may provide the opportunity to understand naturally occurring, chronic negative mood. In this way, the study of patient populations may provide convergent information for laboratory manipulations of these processes in healthy populations or may provide unique information if the process in question is difficult to manipulate in healthy populations.

186

BIOLOGICAL THEORIES

Investigations of Patients with Disorders Social neuroscientists often conduct research on populations characterized by psychiatric and medical disorders that involve impairment in social and personality processes. For example, social and personality processes change in the context of disorders such as depression, autism spectrum disorders, Williams syndrome, and frontotemporal dementia (e.g., Bryson, Rogers, & Fombonne, 2003; Doyle, Bellugi, Korenberg, & Graham, 2004; Greening, Osuch, Williamson, & Mitchell, 2013; Perry et al., 2001). Patients with disorders are typically more accessible to researchers because there is a greater prevalence of these disorders (than patients experiencing focal lesions). It should be noted, however, that psychiatric and medical disorders tend to be associated with relatively diffuse impairments at the psychological level of analyses (as well as the neural level of analyses). Therefore, it can be difficult to isolate a particular process within the overall psychological impairment. It is often challenging to find a suitable comparison patient population to control for the conglomeration of problems. For example, researchers may be interested in understanding negativity biases in cognition, a central feature of major depression. While the participants with major depression may have pronounced negativity biases in comparison to nondepressed participants, they are likely to also have a host of other issues (Disner, Beevers, Haigh, & Beck, 2011). Therefore, the study of major depression does not truly isolate the effect of negativity bias but instead examines the effects of negativity bias in the context of a number of other differences with the nondepression population. The study of patient populations may be most useful for understanding which psychological impairments tend to arise together rather than manipulating a specific psychological process of interest.

Studies of Patients with Focal Lesions Social and personality processes may also be impaired when people sustain an injury to the brain. In contrast to populations characterized by psychiatric or medical disorders, populations characterized by traumatic brain injury may have relatively more circumscribed brain damage and relatively more specific impairments in social or personality processes. For example, patients often sustain focal lesions as the result of trauma from accidents, although lesions can also arise from tumor resection and stroke (e.g., Beer, 2009; Gazzaniga et al., 2008; Kolb & Whishaw, 2003). For example, the orbitofrontal cortex may be selectively damaged because it resides adjacent to the jagged ridges that hold the eye orbits in place. High-speed collisions may result in the brain bouncing against the back of the skull and then shooting forward against the jagged ridges. While patients with orbitofrontal damage may have very few problems walking, talking, and remembering, they do tend to have social problems such as social disinhibition (e.g., Beer, Heerey, Keltner, Scabini, & Knight, 2003).

Cognitive Neuroscience of Social Behavior 187

General Design Considerations for Patient Studies Social neuroscience research will only be valuable to psychologists if it is well controlled. There are two major points to consider when reading published patient studies or planning a patient study (see Table 9.1 for further considerations). First, an optimal patient study compares patients with disorders or focal lesions of interest, patients with disorders or focal lesions in a nonarea of interest, and healthy comparison subjects. The advantage of investigating patients with lesions is that impairments tend to be relatively circumscribed; this advantage is bolstered by identifying patients who have lesions that have strong overlap in both their volume and location. Even though a typical design focuses on between-group comparisons, it is important that length and severity of disorder are considered as well. These variables may explain dependent variables beyond group. Second, it is also important to consider time and population-size constraints when studying patient populations. Many social and personality studies include large numbers of participants because expected effects tend to be small. Sample size and session length for studies with patients may be very limited. The incidence of the population may be low or the patients may tire easily. Finally, patients may be referred from a Veterans Administration hospital or circumscribed catchment area, which can result in populations that are homogeneous in terms of ethnic, educational, and economic backgrounds. It is important that control participants are matched on these factors to rule out the possibility that effects are explained by demographic variables.

Brain-Imaging Approaches In contrast to the focus on deficits in patient studies, brain-imaging approaches are activity-focused. Brain activity can be measured in relatively direct or indirect ways to understand how neural activation changes in relation to the performance of particular mental process. An ever growing number of brain-imaging techniques are available to researchers, such as fMRI, electroencephalographic/event-related potential recordings (EEG/ERPs), transcranial magnetic stimulation (TMS), magnetoencephalogram (MEG), and intracranial recording in surgical patients (e.g., Gazzaniga et al., 2008; Harmon-Jones & Beer, 2009; Huettal et al., 2004; Kolb & Whishaw, 2003; Poldrack, Mumford, & Nichols, 2011). Two of the most commonly used and widely available methods are fMRI and EEG/ERPs.

Functional Magnetic Resonance Imaging fMRI is one of the most commonly used imaging techniques and is widely accessible to researchers (for further description, see Aguirre & D’Esposito, 2000; Buxton, 2002; Desmond & Glover, 2002; Friston, Zarahn, Josephs, Hen-

188

BIOLOGICAL THEORIES

son, & Dale, 1999; Harmon-Jones & Beer, 2009; Huettal, Song, & McCarthy, 2004; Moonen & Bandettini, 2000; Poldrack et al., 2011). A first step in understanding fMRI is some familiarity with nuclear magnetic resonance imaging (technically NMRI but more commonly known as MRI). MRI is a technique that manipulates the orientation of hydrogen nuclei in brain tissue in order to create an image of the brain’s structure (Horowitz, 1995). Hydrogen nuclei align themselves to the strong magnetic field of the brain scanner. The strength of the magnetic field of the scanner is measured in Tesla (T), and scanners are often referred to by their strength (i.e., “3T” is a 3 Tesla magnet). During the experimental session, a radio frequency pulse (RF pulse) is applied which shifts the nuclei from their original position. Once the pulse stops, the nuclei shed the energy injected by the RF pulse and “relax” back into the orientation imposed by the external magnetic field. Differences in the relaxation rates create the MRI images. For example, the hydrogen nuclei in water molecules in blood relax at a different rate than water molecules in other tissues, making it possible to depict blood versus brain tissue. When MRI is used to create images of brain activity rather than structural images of the brain, it is referred to as functional magnetic resonance imaging. fMRI uses the blood oxygenation level dependent (BOLD) contrast effect to derive images of brain activity (Ogawa, Lee, Nayak, & Glynn, 1990, 1992). Activation in a brain region is associated with a hemodynamic response that reflects the delivery of oxygen and other nutrients via changes in blood flow. The proportion of deoxygenated blood to oxygenated blood creates a signal known as the BOLD contrast effect. In this way, fMRI does not directly measure brain activation but does measure the aftermath of changes in blood flow associated with neural activation. In fMRI data analyses, images of activity from experimental and control conditions are statistically compared to determine significant activity. The images of activity are registered, or overlaid, on a structural image of the brain in order to identify where significant activity is generated. The exact relation between neural activity and the BOLD signal is not currently known. It also is important to note that the temporal resolution of fMRI is not very precise. Once a change in brain activity occurs, it takes time (roughly 16 seconds) for the hemodynamic response to unfold and create the BOLD signal, which is the basis of fMRI measurement. However, researchers have developed methods of presenting experimental stimuli in particular timing patterns to try to account for the slowness of the BOLD signal (e.g., Aguirre & D’Esposito, 2000; Poldrack et al., 2011).

Electroencephalographic Recording/Event-Related Potential The imaging of brain activation is also possible with EEG and ERP recording. EEG/ERP recordings are relatively more direct measurements of electrical brain activity than fMRI (for further description, see Fabiani, Gratton, &

Cognitive Neuroscience of Social Behavior 189

Coles, 2000; Harmon-Jones & Beer, 2009). A first step in understanding ERPs is familiarity with EEG. EEG recordings typically involve pairs of electrodes that are connected to amplifiers and placed on the participants’ scalp. In order to contrast with the scalp, an additional reference is also chosen (e.g., the ear). The number of amplifiers increases the sensitivity of the recordings, and publications typically note how many amplifiers (sometimes called channels) are involved. The electrodes are strategically placed over various brain regions and used to detect the summed electrical potentials from large bodies of neurons in those brain regions. The output of this measurement is a wave for each electrode of continuous activity across time. Significant change in brain activation is inferred when a wave’s magnitude or rhythm statistically differs from the ear recording (where there are no electrical potentials). Four major types of continuous rhythmic sinusoidal EEG waves have been identified (alpha, beta, delta and theta). Beta waves are most likely to be encountered in research that would be of interest to social and personality psychologists; they are associated with a participant who is awake with his or her eyes open and attending to external stimuli. ERPs are changes in the EEG signal in response to a particular stimulus. It is difficult to detect momentary changes in the EEG signal because they occur amidst so many other electrical responses in the brain during the continuous recording. Therefore, dozens to thousands of trials are averaged to garner reasonable ERP estimates. It should also be noted that trial length must be very brief, even more so than in fMRI studies. In the ERP approach, timing information is often assessed by recording how long it takes to see a change in magnitude of a wave after a stimulus has been presented. In this way, ERP provides superior temporal resolution when compared to fMRI. For scientists interested in psychological function, a number of ERPs have been identified and are theorized to reflect different cognitive functions. The nomenclature of ERPs typically indicate whether they are negative or positive (i.e., shift upward or downward from baseline, respectively) and a number derived from the time they are produced in relation to stimulus presentation. Although the neural regions associated with particular ERPs remain controversial (see Fabiani, Gratton, & Coles, 2000; Harmon-Jones & Beer, 2009), a growing body of research suggests how some of the more commonly studied ERPs relate to activity in specific brain regions. For example, one of the most commonly studied ERPs is the P300 or P3. The P300 is a positive ERP (P) that is produced 300 ms after a stimulus is presented (300). The P300 is theorized to indicate orientation or decoding of the meaning of the stimulus (e.g., Polich, 2004; Soltani & Knight, 2000); it has been split further into the P3a and P3b because each has a distinct cognitive function and may arise from a different neural location (e.g., Polich, 2004). The P3a is involved in automatic detection of novel stimuli and is associated with the frontal lobes and hippocampus. The P3b is involved with orienting toward a target and is associated with temporal and parietal areas. Another commonly stud-

190

BIOLOGICAL THEORIES

ied ERP is the N400, which is thought to indicate error detection and may be associated with the sylvian fissure, orbitofrontal, dorsolateral prefrontal, superior temporal sulcus, and temporal pole (Halgren et al., 1994; McCarthy, Nobre, Bentin, & Spencer, 1995; Nobre et al., 1994). The P600 is a later emerging ERP that is involved in encoding as well as recognition. There is controversy over whether recognition reflects effort exerted for recognition or the success of recognition. Candidate brain regions for this wave include the hippocampus, parahippocampal gyrus, temporal pole, orbitofrontal cortex, fusiform, cingulate, medial temporal lobe, paralimbic areas, and inferior frontal gyrus (Halgren et al., 1994; Guillem, Rougier, & Claverie, 1999). These are just a few examples of the most commonly (and perhaps better understood) ERPs.

Summary Of the numerous neuroscience methodologies, three tend to be most often used in the research that has potential value for social and personality psychologists. As outlined above, each approach has a unique set of considerations (see Table 9.1). Patient population approaches, particularly studies of patients with lesions, may be helpful for examining naturally occurring impairment in psychological processes. Access to patients may be restricted, and it is important to consider whether effect sizes can be examined in smaller samples. fMRI and ERP studies of healthy participants provide the opportunity to understand the brain activation patterns associated with psychological processes of interest. However, each has design constraints (i.e., timing, number of trials needed), so the psychological processes that can be studied are somewhat limited. ERP provides more information about the timing of brain activation underlying a psychological process, whereas fMRI provides relatively more information about where the changes in brain activation are taking place.

For Researchers Interested in Developing Theories of Psychological Processes, What Is the Explanatory Power of Understanding Neural Function? Cognitive neuroscientists have posited that psychological processes are rooted in neural activation and have developed methodological tools based on this theory, but does all of this have value for psychologists? Whereas researchers interested in developing theories of psychological processes may not dispute that the mind is rooted in brain function, they often raise the question of whether understanding the relation between the neural level of analyses and the psychological level of analyses has any potential to inform theories of psychological processes (e.g., Beer, Mitchell, & Ochsner, 2006; Coltheart, 2013; Mather, Cacioppo, & Kanwisher, 2013; Moran & Zaki, 2013;

Cognitive Neuroscience of Social Behavior 191

Poldrack, 2011). In other words, can tests of the relation between brain activation and psychological processes contribute to our understanding of psychological processes, or do they simply create a map of the brain? Many social neuroscience experiments are motivated by a desire to understand the architecture of the brain because that is one epistemic goal of social neuroscience. However, there are also ways to set up an experiment in which social neuroscience methodologies will benefit our understanding of psychological processes. These methodologies have the potential to provide (1) information about the similarity and differences between two psychological processes and (2) the opportunity to manipulate psychological processes that may otherwise be difficult to manipulate in the laboratory.

Similarities and Differences between Psychological Processes Researchers who are interested in understanding the commonalities and differences between psychological processes may test whether these psychological processes draw on similar or different neural networks (i.e., patterns of neural activation). The rationale is that psychological processes that have similar underlying computations, representations, or transfer of information will be reflected in similar brain activation patterns, whereas different patterns indicate different underlying mechanisms. This rationale has been applied to a number of questions about social and personality processes. For example, consider our understanding of the distinction between appraisal, emotion, and emotion regulation (e.g., Ellsworth & Scherer, 2004; Gross, 1998). Although emotion and emotion regulation are typically described as distinct psychological processes, appraisals are theorized as the underlying mechanism of both (see Manstead & Parkinson, Chapter 5, this volume). Appraisal theories take the perspective that the environment is constantly appraised along various dimensions (e.g., positivity, novelty, ability to control events) and particular combinations of these appraisals are the central components that distinguish one emotion from another. People may regulate which emotion they feel by reappraising their environment, that is, appraising it in a different manner than when they are not trying to regulate their emotions (e.g., Gross, 1998). If appraisals are used to generate emotion and accomplish emotion regulation goals, then what does it mean to say someone is having an emotion versus regulating an emotion? Are emotion and emotion regulation really distinct psychological processes in terms of their underlying mechanics? Studies find that there are distinctions between the neural systems modulated by initial appraisals of emotional stimuli compared to reappraisals made with the goal of emotion regulation (e.g., Ochsner et al., 2009). Therefore, the findings are consistent with the hypothesis that emotion and emotion regulation are somewhat distinct. Appraisals used to generate an emotion experience are not wholly redundant with appraisals used to generate a new emotional experience.

192

BIOLOGICAL THEORIES

Similarly, many popular theories in psychology point to a dual-system perspective in which there is a relatively more emotional, low-level, automatic system of reasoning that can work in parallel or combination with a system that is relatively more rational, elaborate, and controlled. This perspective has generated useful delineations at the behavioral level (Chaiken, 1980; Gawronski & Bodenhausen, 2006; Tversky & Kahneman, 1974). However, research has shown that while this division provides a powerful explanation at the behavioral level, dual systems of reasoning are not neatly reflected at the level of neural activity. For example, neural regions such as the orbitofrontal cortex and insula have been implicated in both emotional and rational decision making, suggesting that there is not a clean division between these types of reasoning at the neural level (e.g., Beer, Knight, & D’Esposito, 2006; DeMartino, Kumaran, Seymour, & Dolan, 2006; Greene, Sommerville, Nystrom, Darley, & Cohen, 2001; Sanfey, Rilling, Aronson, Nystrom, & Cohen, 2003). These are just a few examples of how understanding brain function in relation to psychological processes can inform questions about the similarities and differences of psychological processes.

Manipulation of Psychological Variables Additionally, lesion studies can inform questions about psychological processes by providing a manipulation for individual differences that are difficult to manipulate in the lab. For example, one theory of self-conscious emotions is that they are helpful in motivating people to avoid social mistakes (e.g., Beer & Keltner, 2004; Lewis, 1993; Tangney & Fischer, 1995). It is difficult to manipulate social mistakes in a lab setting. Healthy adult participants are often highly concerned about their performance, so social mistakes in the lab can be rare. Some advances have been made by introducing embarrassing situations, but because they are clearly not the fault of the participant, this approach sacrifices a good deal of ecological validity (e.g., asking participants to suck on baby pacifiers; Brown, 1970). However, patients with orbitofrontal cortex damage have self-regulatory deficits in their interpersonal behavior (e.g., Beer et al., 2003; Beer, Shimamura, & Knight, 2004). By comparing orbitofrontal patients with other patients and healthy controls, it is possible to examine the relation between self-regulation failures in social settings (i.e., social mistakes) and self-conscious emotion. In fact, research has shown that the social mistakes of patients with orbitofrontal damage are associated with disrupted self-conscious emotion. Specifically, social mistakes in this population are associated with pride and a lack of embarrassment, even though patients retain the ability to experience embarrassment in other settings (Beer et al., 2003). Further research has shown that social mistakes in this population may arise because self-insight, a component theorized to be critical for the generation of self-conscious emotion (Lewis, 1993; Tangney & Fischer, 1995), is impaired in cases of orbitofrontal cortex damage (Beer, John, Scabini, & Knight, 2006).

cognitive neuroscience of Social Behavior

193

A Matter of Interpretation Experiments that test the relation of brain function and psychological processes can have the potential to advance our understanding of psychological processes. The benefits will be greatest when researchers are sensitive to a number of issues that arise when interpreting results from brain-imaging or lesion studies. It is important to understand what kinds of conclusions can be made about the relation of brain function and psychological processing. Many of the issues are similar to issues associated with behavioral experiments. For example, experimental manipulations free of confounds are also important in neuroimaging and lesion studies. However, there are a number of issues with interpreting results from these neuroscience methodologies that are distinct from the more traditional behavioral methods used to study psychological processes.

Smokers and Lamps “Light Up,” but the Brain Does Not One of the leading misunderstandings in interpreting fMRI study results arises when researchers conclude that regions showing activation (e.g., see Figure 9.1, Panel A, Region X) have “lit up” in relation to a task. The only conclusion that you can make upon seeing such a brain map is that the region is differentially activated across the tasks that have been contrasted. The brain activation map alone does not provide enough information to conclude that the activation was increased during the task of interest, that it was the only region where activation changed in relation to the task of interest, or that it was significantly activated above a comparative baseline (e.g., activation when a person is not performing a task but resting and looking at fixa-

fIgURe 9.1. Depiction of activation patterns that result in significant and nonsignificant activation clusters in an fMRI contrast map.

194

BIOLOGICAL THEORIES

tion point). Further interpretation is possible if the activation in the region of interest is plotted in the context of the two tasks that have been compared to one another. Different interpretations of the meaning of a brain activation map can be illustrated with a simple hypothetical example: A participant performs two tasks such as viewing outgroup faces (i.e., Task A) and viewing ingroup faces (i.e., Task B). In order to understand how brain activation changes across these tasks, a contrast map will be created to compare brain activation associated with viewing outgroup faces (Task A) to brain activation associated with viewing ingroup faces (Task B). Let’s consider the example where Region X shows significant activation on the contrast map but Region Y does not (Figure 9.1, Panel A). Activation in these regions is estimated in relation to a control condition. The control condition is included in the experiment to model what the brain is doing when it is at rest or performing a lowlevel task (e.g., looking a fixation point). After plotting activation in Region X and Region Y associated with viewing outgroup faces (Task A) and viewing ingroup faces (Task B), we can further understand the meaning of the findings for both Region X and Region Y (Figure 9.1, Panels B–F). It may be that Region X does exhibit increased activation when viewing outgroup faces (Task A) but not when viewing ingroup faces (Task B) and that activation in Region Y is not engaged by viewing faces from either group (Figure 9.1, Panel B). This would be the pattern that is implied when researchers state that a region has “lit up.” However, the same contrast map (Figure 9.1, Panel A) could also mean that Region X increases its activation when the participants view outgroup faces (Task A) compared to when they view ingroup faces (Task B) whereas Region Y increases its activation for viewing both kinds of faces (Figure 9.1, Panel C). If a region is activated (i.e., “lit up”) above baseline in both tasks to a similar extent, it will not appear on a contrast map. This pattern illustrates one reason why it is not possible to conclude that the regions that appear on a contrast map are the only ones where activation is increased in relation to the task. Regions that do not show differential activation in a contrast map may have no change in activation associated with each of the experimental tasks, or they may be activating or even deactivating to a similar extent for each experimental task. Furthermore, it is also possible that Region X may not exclusively show activation when viewing outgroup faces (Task A) or may not even show activation at all in that condition. Again, even though the contrast map depicts changes in activation using bright colors such as red or yellow, the map only speaks to differential activation (Figure 9.1, Panel A). Region X could be activating when viewing both kinds of faces but is significantly more activated when viewing outgroup faces (Figure 9.1, Panel D). It is also possible that Region X appears on the contrast map because it slightly activates when viewing outgroup faces but then slightly deactivates when viewing ingroup faces. However, in absolute terms, neither the activation nor the deactivation is statistically different from the activation seen in the control condition. In

Cognitive Neuroscience of Social Behavior 195

this case, Region X appears on the contrast map because a comparison of its activation values across the two tasks is statistically significantly different from one another (Figure 9.1, Panel E) and not because activation while viewing faces is any different than viewing a fixation point. It may also be that Region X deactivates to a lesser extent when viewing outgroup faces (Task A) than ingroup faces (Task B) (Figure 9.1, Panel F). These last two examples illustrate why it is not possible to conclude that regions appearing on a contrast map indicate “lighting up” because those regions may be accounted for by nonsignificant activation or even deactivation in relation to the task of interest.

Critical Involvement? Or an Association? Even when a brain region shows differential activation in relation to psychological tasks, it is not possible to conclude that the brain region is necessary or even sufficient for supporting the relevant psychological process. For example, take two common findings in the social neuroscience literature: amygdala activation in response to ingroup and outgroup member faces is modulated by individual differences in implicit racial attitudes (Beer et al., 2008; Phelps et al., 2000), and medial orbitofrontal cortex activation is related to correcting unrealistically positive evaluations of self and close others (Beer & Hughes, 2010; Beer, Lombardo, & Bhanji, 2010; Hughes & Beer, 2012a, 2012b). However, even individuals with amygdala damage express implicit prejudice (Phelps, Cannistraci, & Cunningham, 2003), and individuals with orbitofrontal damage can correct their erroneous self-views when presented with videotaped evidence of their actual behavior (Beer et al., 2006). As these examples illustrate, changes in amygdala activation and medial orbitofrontal cortex activation are associated with these psychological processes, but their function is not required or even sufficient in order for the processes to operate. Even in the absence of their function, the psychological processes in question persist so they are not necessary. It is also impossible to determine whether their function is sufficient as these methodologies do not make it possible to determine whether the changes in their activation alone are enough to support the operation of the psychological processes in question. Instead, a brain region is only considered to be necessary for performing a psychological process that is “critically involved,’’ when the region’s absence or failure to function undermines the operation of the psychological process. But even if researchers were to find that damage to a region or disruption to a region’s function (e.g., TMS; see Harmon-Jones & Beer, 2009) precluded a particular psychological process, it is impossible to know whether “critical involvement” means that area is important for sending, receiving, or relaying a necessary signal. The psychological process may be undermined by damage or disruption to a region because that region is necessary for sending or receiving signals that directly affect the execution of the psychological process. Or the psychological process may be undermined

196

BIOLOGICAL THEORIES

because fibers of passage that are necessary for relaying a message between two brain areas reside in the area of damage or disruption. Both of these possibilities should be acknowledged in the interpretation of results from studies involving patients with specific lesions and studies involving TMS. Patient studies further cloud the association between a particular brain region with a particular psychological process in a number of ways. Patients with disorders have the advantage of being more accessible to researchers (compared to the availability of patients with focal lesions), but there are potential drawbacks. Psychiatric disorders and progressive medical disorders raise the possibility that individuals who are impacted have different developmental neural trajectories, are taking medications that affect brain function, and usually have diffuse areas of damage. Differential development and medication make it difficult to generalize findings from these populations to healthy populations. If the goal is to understand how a particular brain region is implicated in a specific mental process, then this approach is not ideal. For example, Urbach–Wiethe is a rare disease that causes calcification of the amygdala in almost 50% of cases (e.g., Siebert, Markowitsch, & Bartel, 2003). In other words, patients with Urbach–Wiethe provide a naturally occurring model of amygdala damage. However, the lack of random assignment to this disease makes it unclear whether findings will generalize to normal populations with normal developmental trajectories. Additionally, in some cases, the calcification spreads beyond the amygdala. The diffuse damage associated with most disorders makes it difficult to isolate behavioral deficits in relation to a specific brain region. Although patients with focal lesions may be less subject to the issue of atypical development, there are still a number of things to consider when interpreting how the area of damage is associated with impaired psychological processing. People who sustain focal lesions are not randomly assigned. If focal lesions result from trauma, it is important to consider whether patients with lesions prospectively differed on personality dimensions (e.g., risk-taking) which may have made them more likely to incur brain damage. These same individual differences may account for differences in psychological processes rather than neural function.

The Problem of Reverse Inference Another common problem in interpreting neuroimaging or lesion study results is the tendency to invoke modern-day phrenology—that is, to use a brain region to infer the psychological process in operation. This approach, often called reverse inference (see Gawronski & Bodenhausen, Chapter 1, this volume), assumes that the function of a particular brain region has a one-to-one correspondence with a specific psychological function. This premise ignores the fact that it is more likely that complex psychological functions are based in complex patterns and networks of brain activation (Kanwisher, 2010; Sporns, 2013). Additionally, it ignores the fact that the field

Cognitive Neuroscience of Social Behavior 197

of social neuroscience is far from understanding how the brain supports the complex psychological processes that are of interest to social and personality psychologists (Beer, Mitchell, & Ochsner, 2006; Moran & Zaki, 2013). It is understandable that early investigations may turn to speculation when discussing results, and this speculation may involve reverse inference. However, it is important to recognize reverse inference and acknowledge its limitations. Furthermore, a number of steps can be taken to build knowledge and eliminate the temptation of speculative reverse inference. For example, one of the first well-replicated discoveries in social neuroscience was that a portion of the medial prefrontal cortex shows differential activation to judging one’s own attributes compared to the attributes of a familiar, but nonintimate person (i.e., a politician) (for reviews, see Murray, Schaer, & Debbané, 2012; Ochsner et al., 2005; Qin & Northoff, 2011; Roy, Shohamy, & Wager, 2012). One unfortunate side effect of this discovery was a subsequent tendency to interpret any changes in medial prefrontal cortex activation as reflections of self-processing regardless of the experimental paradigm. While it might be the case that the differential activation of medial prefrontal cortex does reflect self-processing in some contexts, there are a number of reasons why it is not possible to definitively draw this conclusion. First, any involvement the medial prefrontal cortex may have in selfprocessing does not extend to all types of self-processing, nor is medial prefrontal cortex only implicated in self-processing. Research has identified a number of other brain areas that are associated with a host of self-processes, such as autobiographical memory and motivated evaluation of self-relevant information (e.g., Flagan & Beer, 2013; Lou et al., 2004; Roy et al., 2012; Rui, Rotshein, & Humphreys, 2013). Furthermore, the medial prefrontal cortex is involved in a number of psychological processes that are not specifically about the self (e.g., reward, learning, task difficulty: Daw, O’Doherty, Dayan, Seymour, & Dolan, 2006; Rolls, Grabenhorst, & Deco, 2010). These examples highlight a problem with reverse inference: The functions of brain regions do not correspond to one specific psychological process, and, therefore, the operation of a psychological process cannot be inferred from changes in the activation of a particular region. However, stronger inferences can be made when experimental tasks isolate a psychological process of interest and test its relation to changes in brain activation (referred to as forward inference in fMRI studies: Henson, 2005). As in behavioral research, the interpretation of results from neuroimaging and lesion studies is only as good as the experimental manipulation. If the psychological process of interest is not experimentally manipulated without confounds, there is no way to gauge whether brain activation is related to that psychological process. This is easier said than done, and it is important to understand that much is left to be understood about the relation between the brain and complex psychological processes. Take the example of the medial prefrontal cortex. Subsequent research has shown

198

BIOLOGICAL THEORIES

that the same region that showed differential activation to self-evaluation is also differentially modulated by evaluations of close others (e.g., Murray et al., 2012; Ochsner et al., 2005; Qin & Northoff, 2011). It might be tempting to look at the pattern of results and conclude that medial prefrontal cortex activation reflects evaluation of something that is emotional or important to one’s self. However, when researchers explicitly asked people to evaluate themselves on traits and to also indicate whether the trait was important to their self-view, they found that only a subportion of the medial prefrontal cortex was related to importance (D’Argembeau et al., 2012). The remainder of the region was related to the participant’s certainty about the evaluation (D’Argembeau et al., 2012). In fact, it is possible that the medial prefrontal cortex subregion that is so often labeled as the “self” region reflects greater certainty in evaluation (Flagan & Beer, 2013). In this interpretation, medial prefrontal cortex activation changes in relation to self- and close-other evaluation (when compared to evaluation of a politican) because we are more certain about self-evaluations and evaluations of close others whom we know than about our evaluations of a politican whom we’ve only seen in the news but have never met. As this example illustrates, even when careful attention is paid to experimental manipulation, it can be difficult to understand how brain activation relates to very complex psychological processing because such different levels of analyses are being bridged. Therefore, more research that cleanly manipulates the psychological process of interest is needed to build our understanding of how changes in brain activation and function relate to psychological processes. As this information becomes more robust, the temptation of superficial reverse inference will be eliminated. On a related note, stronger inferences about the psychological significance of brain activation will also be made possible as more information is gathered about the networks and patterns of brain activations that are expected in relation to psychological processes (e.g., Hutzler, 2014). Currently, three different approaches are moving in this direction. First, statistical techniques are now being developed so that the results of fMRI studies can be characterized in terms of functional connectivity between brain regions. Functional connectivity indicates that activation between brain regions covaries with one another, but it does not test whether the regions are physically connected to one another. For example, a common technique for evaluating connectivity is psychophysiological (PPI) analyses. PPI analyses can conceptually test whether brain activation in a region of interest covaries with activation in other brain regions as a function of an experimental task (i.e., an interaction with task). The results of such an analysis can give a picture of which regions are covarying together (or more strongly together) for a certain task (i.e., functionally covarying). However, it does not necessarily mean that the regions are communicating through a direct physical pathway. Instead, covariation might occur because of a direct physical pathway, a distal connection through one or more brain regions, or because a third region is sending input to the two regions in question.

Cognitive Neuroscience of Social Behavior 199

Although functional connectivity does not speak to the specific reason why activation is covarying across brain regions, it does bring researchers a step closer to understanding how the brain supports psychological processes. For example, the role of the medial orbitofrontal cortex in self-processing initially appeared to be paradoxical. Sometimes this region’s activation was associated with correcting unrealistically positive social cognition (Beer et al., 2006; Beer & Hughes, 2010; Hughes & Beer, 2012a), and sometimes this region’s activation was associated with promoting unrealistically positive social cognition (Hughes & Beer, 2013). However, when the existing datasets were reanalyzed using PPI, it became clear that the medial orbitofrontal cortex was operating in distinct networks when it either corrected or promoted unrealistically positive social cognition (Flagan & Beer, 2013). This example illustrates how understanding functional connectivity rather than simply the main effects of tasks on brain activation within one region makes it possible to better characterize the relation between brain and behavior. Second, in addition to functional connectivity within a task, statistical techniques are being developed to understand the psychological significance of brain activation across tasks. Instead of simply contrasting the magnitude of activation within brain voxels across tasks, these techniques are aimed at understanding how patterns persist across contexts. For example, memory researchers have distinguished the encoding and retrieval of a memory and theorize about their relation to one another. Is there something to be learned by examining these processes at the neurobiological level? For example, is there some pattern of brain activation at encoding that is then re-created when the encoded memory is successfully “retrieved”? Simply looking at a brain map to see if the same regions are activated at encoding and retrieval is not enough to infer that the patterns of brain activation are the same. The regions might be working within different networks. Even if the brain regions associated with retrieval were found to show similar functional connectivity at encoding, it is still unclear whether the covariation happens to a similar degree. Therefore, it is necessary to test the similarity of patterns of brain activation at encoding and retrieval. A technique called multivariate representational similarity analyses (MRSA) allows researchers to test the distance between brain activation patterns across tasks. Using this technique and others like it, researchers have discovered that brain activation patterns at encoding are indeed re-created at the time of successful retrieval (Danker & Anderson, 2000); the similarity of patterns of activation can be distinguished down to the level of the particular item that will be remembered rather than just successful retrieval in general (Ritchey, Wing, LaBar, & Cabeza, 2013). Finally, several databases are now being developed that have the capability of conducting meta-analyses on existing studies about a particular topic. One of the most popular and deeply researched database is called Neurosynth (neurosynth.org: Yarkoni, Poldrack, Nichols, Van Essen, & Wager, 2011). It includes the results of studies investigating a wide variety of psycho-

200

BIOLOGICAL THEORIES

logical topics. Researchers can input a particular psychological process (i.e., “pain”) and the database will generate regions associated with that process. This database is a huge advance in integrating individual fMRI studies. However, it should be noted that it is still in development. As in meta-analyses of behavioral studies, the information that comes out is only as good as the information that goes in. If researchers underreport their findings (i.e., by focusing only on regions of interest) or conduct studies with poor experimental manipulations, the database has no way to account for these issues.

The Consistency Fallacy Finally, it is important to avoid what researchers have called the “consistency fallacy” when interpreting fMRI data (Coltheart, 2013). The consistency fallacy is interpreting experimental results as proof of a theory simply because they are consistent with a theory, yet the experiment had no means of testing a plausible, but inconsistent, outcome. The consistency fallacy applies to experiments that do not use neuroimaging techniques, but it is of particular concern because the timing constraints of fMRI typically restrict the kinds of conditions that are included. Specifically, conditions that provide a strong test of alternative outcomes may not be included. For example, behavioral studies are often characterized by a 2 × 2 design in which each variable has two levels. This design allows researchers to see whether a variable’s impact is specific to the conditions predicted by a theory or extends more broadly to other levels included in the design. However, neuroimaging studies tend to contrast two conditions rather than include a full 2 × 2 design because of timing and monetary constraints. Therefore, it is important to evaluate results on the basis of whether a condition was included that could test an outcome that would falsify the theory in addition to testing conditions that are consistent with a theory.

Summary The central premise of cognitive neuroscience theories is that the mind (i.e., psychological processes) is associated with neural function. This premise has not been widely controversial among researchers interested in understanding psychological processes; however, there have been questions about whether this premise has any value in developing theories of psychological processes. Widely available techniques such as fMRI and EEG/ERPs, as well as relatively less accessible techniques such as studying patients with disorders or focal lesions, do have the potential to advance psychological knowledge. Studies can probe the similarities or difference of psychological processes by examining the similarity of their underlying neural associations. Studies with patients may provide unique opportunities to manipulate processes that are otherwise challenging to do in a lab environment. The value of these approaches will be maximized to the extent that researchers

Cognitive Neuroscience of Social Behavior 201

avoid the potential pitfalls of interpreting the meaning of neuroimaging and lesion results outlined in this chapter. References Aguirre, G. K., & D’Esposito, M. (2000). Experimental design for brain fMRI. In C. T. W. Moonen & P. A. Bandettini (Eds.), Medical radiology: Functional MRI (pp. 369–380). New York: Springer. Beer, J. S. (2009). Patient methodologies for the study of personality and social processes. In E. Harmon-Jones & J. S. Beer (Eds.), Methods in social neuroscience (pp. 148–169). New York: Guilford Press. Beer, J. S., Heerey, E. H., Keltner, D., Scabini, D., & Knight, R. T. (2003). The regulatory function of self-conscious emotion: Insights from patients with orbitofrontal damage. Journal of Personality and Social Psychology, 85, 594–604. Beer, J. S., & Hughes, B. L. (2010). Neural systems of social comparison and the “aboveaverage” effect. NeuroImage, 49, 2671–2679. Beer, J. S., John, O. P., Scabini, D., & Knight, R. T. (2006). Orbitofrontal cortex and social behavior: Integrating self-monitoring and emotion–cognition interactions. Journal of Cognitive Neuroscience, 18, 871–879. Beer, J. S., & Keltner, D. (2004). What is unique about self-conscious emotions? Comment on Tracy & Robins’ “Putting the self into self-conscious emotions: A theoretical model.” Psychological Inquiry, 15, 126–129. Beer, J. S., Knight, R. T., & D’Esposito, M. (2006). Integrating emotion and cognition: The role of the frontal lobes in distinguishing between helpful and hurtful emotion. Psychological Science, 17, 448–453. Beer, J. S., Lombardo, M. V., & Bhanji, J. P. (2010). Roles of medial prefrontal cortex and orbitofrontal cortex in self-evaluation. Journal of Cognitive Neuroscience, 22, 2108–2119. Beer, J. S., Mitchell, J. P., & Ochsner, K. N. (2006). Multiple perspectives on the psychological and neural bases of social cognition [Special issue]. Brain Research, 1079, 1–3. Beer, J. S., Shimamura, A. P., & Knight, R. T. (2004). Frontal lobe contributions to executive control of cognitive and social behavior. In M. S. Gazzaniga (Ed.), The newest cognitive neurosciences (3rd ed., pp. 1091–1104). Cambridge, MA: MIT Press. Beer, J. S., Stallen, M., Lombardo, M. V., Gonsalkorale, K., Cunningham, W. A., & Sherman, J. W. (2008). The quadruple process model approach to examining the neural underpinnings of prejudice. NeuroImage, 43, 775–783. Brown, B. (1970). Face-saving following experimentally induced embarrassment. Journal of Experimental Social Psychology, 6, 255–271. Bryson, S. E., Rogers S. J., & Fombonne, E. (2003). Autism spectrum disorders: Early detection and intervention, education, and psychopharmacological management. Canadian Journal of Psychiatry, 48, 506–516. Buxton, R. B. (2002). Introduction to functional magnetic resonance imaging: Principles and techniques. New York: Cambridge University Press. Chaiken, S. (1980). Heuristic versus systematic information processing and the use of source versus message cues in persuasion. Journal of Personality and Social Psychology, 39, 752–766. Coltheart, M. (2013). How can functional neuroimaging inform cognitive theories? Perspectives on Psychological Science, 8, 98–103. Danker, J. F., & Anderson, J. R. (2010). The ghosts of brain states past: Remembering reactivates the brain regions engaged during encoding. Psychological Bulletin, 136, 87–102. Daw, N. D., O’Doherty, N. D., Dayan, P., & Dolan, B. J. (2006). Cortical substrates for exploratory decisions in humans. Nature, 441, 876–879.

202

BIOLOGICAL THEORIES

D’Argembeau, A., Jedidi, H., Balteau, E., Bahri, M., Phillips, C., & Salmon, E. (2012). Valuing one’s self: Medial prefrontal involvement in epistemic and emotive investments in self-views. Cerebral Cortex, 22, 659–667. DeMartino, B., Kumaran, D., Seymour, B., & Dolan, R. J. (2006). Frames, biases and rational-decision making in the human brain. Science, 313, 684–687. Desmond, J. E., & Glover, G. H. (2002). Estimating sample size in functional MR (fMRI) neuroimaging studies: Statistical power analyses. Journal of Neuroscience Methods, 118, 115–128. Disner, S. G., Beevers, C. G., Haigh, E. A., & Beck, A. T. (2011). Neural mechanisms of the cognitive model of depression. Nature Reviews Neuroscience, 12, 467–477. Doyle, T. F., Bellugi, U., Korenberg, J. R., & Graham, J. (2004). “Everybody in the world is my friend”: Hypersociability in young children with Williams syndrome. American Journal of Medical Genetics, 124A, 263–273. Ellsworth, P. C., & Scherer, K. R. (2003). Appraisal processes in emotion. In R. J. Davidson, K. R. Scherer, & H. H. Goldsmith (Eds.), Handbook of affective sciences (pp. 572–595). New York: Oxford University Press. Fabiani, M., Gratton, G., & Coles, M. G. H. (2000). Event-related brain potentials. In J. T. Cacioppo, L. G. Tassinary, & G. G Berntson (Eds.), Handbook of psychophysiology (2nd ed., pp. 53–84). Cambridge, UK: Cambridge University Press. Flagan, T., & Beer, J. S. (2013). Three ways the midline regions contribute to self-processing. Frontiers in Human Neuroscience, 7, 1–12. Friston, K. J., Zarahn, E., Josephs, O., Henson, R. N. A., & Dale, A. M. (1999). Stochastic designs in event-related fMRI. NeuroImage, 10, 607–619. Gawronski, B., & Bodenhausen, G. V. (2006). Associative and propositional processes in evaluation: An integrative review of implicit and explicit attitude change. Psychological Bulletin, 132, 692–731. Gazzaniga, M. S., Ivry, R. B., & Mangun, G. R. (2008). Cognitive neuroscience: The biology of the mind (3rd ed.). New York: Norton. Greene, J. D., Sommerville, R. B., Nystrom, L. E., Darley, J. M., & Cohen, J. D. (2001). An fMRI investigation of emotional engagement in moral judgment. Science, 293, 2105–2108. Greening, S. G., Osuch, E. A., Williamson, P. C., & Mitchell, D. V. (2013). Emotion-related brain activity to conflicting socio-emotional cues in unmedicated depression. Journal of Affective Disorders, 150, 1136–1141. Gross, J. J. (1998). The emerging field of emotion regulation: An integrative review. Review of General Psychology, 2, 271–299. Guillem, F., Rougier, A., & Claverie, B. (1999). Short- and long-delay intracranial ERP repetition effects dissociate memory systems in the human brain. Journal of Cognitive Neuroscience, 11, 437–458. Halgren, E., Baudena, P., Heit, G., Clarke, J. M., Marinkovic, K., Chauvel, P., et al. (1994). Spatio-temporal stages in face and word processing. 2. Depth-recorded potentials in the human frontal and Rolandic cortices. Journal of Physiology, 88, 51–80. Harmon-Jones, E., & Beer, J. S. (Eds.). (2009). Methods in social neuroscience. New York: Guilford Press. Henson, R. (2005). What can functional neuroimaging tell the experimental psychologist? Quarterly Journal of Experimental Psychology, 58, 193–233. Horowitz, A. L. (1995). MRI physics for radiologists: A visual approach. New York: SpringerVerlag. Huettal, S. A., Song, A. W., & McCarthy, G. (2004). Functional magnetic resonance imaging. Sunderland, MA: Sinauer Hughes, B. L., & Beer, J. S. (2012a). Orbitofrontal cortex and anterior cingulate cortex are modulated by motivated social cognition. Cerebral Cortex, 22, 1372–1381.

Cognitive Neuroscience of Social Behavior 203 Hughes, B. L., & Beer, J. S. (2012b). Medial orbitofrontal cortex is associated with shifting decision thresholds in self-serving cognition. NeuroImage, 61, 889–898. Hutzler, F. (2014). Reverse inference is not a fallacy per se: Cognitive processes can be inferred from functional imaging data. NeuroImage, 84, 1061–1069. Kanwisher, N. (2010). Functional specificity in the human brain: A window into the functional architecture of the mind. Proceedings of the National Academy of Sciences USA, 107, 11163–1117. Kolb, B., & Whishaw, I. Q. (2003). Fundamentals of human neuropsychology. New York: Worth. Lewis, M. (1993). Self-conscious emotions: Embarrassment, pride, shame, and guilt. In M. Lewis & J. M. Haviland (Eds.), Handbook of emotions (pp. 353–364). New York: Guilford Press. Lou, H. C., Luber, B., Crupain, M., Keenan, J. P., Nowak, M., Kjaer, T. W., et al. (2004). Parietal cortex and representation of the mental self. Proceedings of the National Academy of Sciences USA, 101, 6827–6832. Mather, M., Cacioppo, J. T., & Kanwisher, N. (2013). How fMRI can inform cognitive theories. Perspectives on Psychological Science, 8, 108–113. McCarthy, G., Nobre, A. C., Bentin, S., & Spencer, D. D. (1995). Language-related field potentials in the anterior-medial temporal lobe: I. Intracranial distribution and neural generators. Journal of Neuroscience, 15, 1080–1089. Moonen, C. T. W., & Bandettini, P. A. (2000). Medical radiology: Functional MRI. New York: Springer. Moran, J. M., & Zaki, J. (2013). Functional neuroimaging and psychology: What have you done for me lately? Journal of Cognitive Neuroscience, 25, 834–842. Murray, R. J., Schaer, M., & Debbané, M. (2012). Degrees of separation: A quantitative neuroimaging meta-analysis investigating self-specificity and shared neural activation between self-and other-reflection. Neuroscience and Biobehavioral Reviews, 36, 1043–1059. Nobre, A. C., Allison, T., & McCarthy, G. (1994). Word recognition in the human inferior temporal lobe. Nature, 17, 260–263. Ochsner, K. N., Beer, J. S., Robertson, E. R., Cooper, J. C., Gabrieli, J. D., Kihsltrom, J. F., et al. (2005). The neural correlates of direct and reflected self-knowledge. NeuroImage, 28, 797–814. Ochsner, K. N., Ray, R. R., Hughes, B. L., McRae, K., Cooper, J. C., Weber, J., et al. (2009). Bottom-up and top-down processes in emotion generation: Common and distinct neural mechanisms. Psychological Science, 20, 1322–1331. Ogawa, S., Lee, T., Nayak, A., & Glynn, P. (1990). Oxygenation-sensitive contrast in magnetic resonance image of rodent brain at high magnetic fields. Magnetic Resonance in Medicine, 14, 68–78. Ogawa, S., Tank, D. W., Menon, R., Ellerman, J. M., Kim, S. G., & Merkle, H. (1992). Intrinsic signal changes accompanying sensory stimulation: Functional brain mapping with magnetic resonance imaging. Proceedings of the National Academy of Sciences USA, 89, 5951–5955. Perry, R. J., Rosen, H. R., Kramer, J. H., Beer, J. S., Levenson, R. L., & Miller, B. L. (2001). Hemispheric dominance for emotions, empathy and social behaviour: Evidence from right and left handers with frontotemporal dementia. Neurocase, 7, 145–160. Phelps, E. A., Cannistraci, C. J., & Cunningham, W. A. (2003). Intact performance on an indirect measure of race bias following amygdala damage. Neuropsychologia, 41, 203–208. Phelps, E. A., O’Connor, K. J., Cunningham, W. A., Funayama, E. S., Gatenby, J. C., Gore, J. C., et al. (2000). Performance on indirect measures of race evaluation predicts amygdala activation. Journal of Cognitive Neuroscience, 12, 729–738.

204

BIOLOGICAL THEORIES

Poldrack, R. A. (2011). Inferring mental states from neuroimaging data: From reverse inference to large-scale decoding. Neuron, 72, 692–697. Poldrack, R. A., Mumford, J. A., & Nichols, T. E. (2011). Handbook of fMRI data analysis. New York: Cambridge University Press. Polich, J. (2004). Neuropsychology of P3a and P3b: A theoretical overview. In N. C. Moore & K. Arikan (Eds.), Brainwaves and mind: Recent developments (pp. 15–29). Wheaton, IL: Kjeilberg. Qin, P., & Northoff, G. (2011). How is our self related to midline regions and the default mode network? NeuroImage, 57, 1221–1233. Ritchey, M., Wing, E. A., LaBar, K. S., & Cabeza, R. (2013). Neural similarity between encoding and retrieval is related to memory via hippocampal interactions. Cerebral Cortex, 23, 2818–2828. Rolls, E. T., Grabenhorst, F., & Deco, G. (2010). Choice, difficulty, and confidence in the brain. NeuroImage, 53, 694–706. Rorden, C., & Karnath, H. O. (2004). Using human brain lesions to infer function: A relic from a past era in the fMRI age? Nature Reviews Neuroscience, 5, 813–819. Roy, M., Shohamy, D., & Wager, T. D. (2012). Ventromedial prefrontal–cortical systems and the generation of affective meaning. Trends in Cognitive Sciences, 16, 147–156. Rui, J., Rotshein, P., & Humphreys, G. W. (2013). Coupling social attention to the self forms a network for personal significance. Proceedings of the National Academy of Sciences USA, 110, 7607–7612. Sanfey, A. G., Rilling, J. K., Aronson, J. A., Nystrom, L. E., & Cohen, J. D. (2003). The neural basis of economic decision-making in the Ultimatum Game. Science, 300, 1755–1758. Siebert, M., Markowitsch, H. J., & Bartel, P. (2003). Amygdala, affect and cognition: Evidence from 10 patients with Urbach–Wiethe disease. Brain, 12, 2627–2637. Soltani, M., & Knight, R. T. (2000). Neural origins of the P300. Critical Reviews in Neurobiology, 14, 199–224. Sporns, O. (2013). Making sense of brain network data. Nature Neuroscience Methods, 10, 491–493. Tangney, J. P., & Fischer, K. W. (1995). Self-conscious emotions: The psychology of shame, guilt, embarrassment, and pride. New York: Guilford Press. Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185, 1124. Yarkoni, T., Poldrack, R. A., Nichols, T. E., Van Essen, D. C., & Wager, T. D. (2011). Largescale automated synthesis of human functional neuroimaging data. Nature Methods, 8, 665–670.

10 Genetics of Social Behavior Wendy Johnson Lars Penke

I think the usual discussion [of] “vitalism-mechanism” puts the question upside down by asking “if we start with physics and chemistry can we explain the whole of biology?” Whereas the real question should be “if we analyse biological systems shall we come across anything which physics and chemistry cannot eventually accommodate?” This second form of question itself implies of course that physics and chemistry are themselves growing and developing subjects. There is a lot of biology which is [sic] at present is as far from basic physics as the gas laws are from the dynamics of the individual gas molecules. . . . The field of natural selection and evolution is one example and . . . the morphogenesis of large scale structures such as bones will quite likely turn out in the same category. . . . New bodies of theory will have to be developed to deal with such phenomena, but this does not imply . . . that the new theories cannot be fully incorporated into an expanded body of physics. . . . Looking at a few pieces of wire and plastic from the point of view of ordinary physics, I would not easily come to the conclusion that they could beat one at chess. Suitably assembled and programmed they could do so, and their behavior then is not “non-physical” but is I should say “super (conventional) physics.”     —Conrad Hal Waddington, letter to      Francis Crick, December 27, 1967

S

ubstitute biology and genetics for physics and chemistry and psychology of social behavior for biology in the above. You have to change the examples too, of course. Maybe Mendel’s laws of inheritance for the gas laws, 205

206

BIOLOGICAL THEORIES

and gene expression for the dynamics of individual gas molecules? Maybe the Big Five model of personality for natural selection and evolution, and self-esteem for bones? We’ll leave the chess-playing computer example at the end alone. It’s perfect as it is. Whether or not those particular substitutions for Waddington’s examples from physics, chemistry, and biology float your boat, together the substitutions make Waddington’s passage strikingly relevant to the role of genetics in social behavior. That is, because human beings are biological organisms and all biological processes involve gene expression of some kind, there is a level at which nothing about social behavior makes sense except in light of biology, or more specifically, genetics. At present, however, we frankly don’t know much about that level. The problem is that many people, even many working in the life sciences, assume we do know a lot about it. They believe that outcomes in which genes are involved emerge through fixed sets of genetic processes beyond control of the individual and basically independent of the environment. They consider these genetic influences to be the fundamental cause of the outcome, so that presence of the relevant genetic variants automatically assures presence of the outcome, and absence of the relevant genetic variants precludes it. They assume that each gene is linked one-to-one to some trait(s) and that everyone who shares a genetically influenced trait to the same degree carries the same combination of involved genetic variants, expressed in the same ways (so-called isomorphism). All this tends to make the outcome seem natural, inevitable, and predetermined, so that moral and ethical considerations must accommodate its presence (which, even if genetic determinism were true, is Moore’s (1903) natural fallacy; “ought” does not follow from “is”). And it tends to make many consider all those who share a particular genetically influenced characteristic to be generally homogeneous in other ways as well. None of this, however, is the case. If none of this is the case, why is this interpretation of genetic influences so pervasive? Some, such as Dar-Nimrod and Heine (2011), have argued that this interpretation is a form of psychological essentialism, or the human tendency to group the world into categories defining the causes, immutable features, and innate potential of objects and organisms (Gelman, 2003). This kind of thinking emerges in early childhood and is thought to be highly useful to young children in learning language and the fundamental concepts of their cultures. Overused, however, essentialism is thought to be one of the primary sources of bias and stereotype (e.g., Bastian & Haslam, 2006). It is, of course, possible that essentialism contributes to the deterministic interpretation of genetic influences, but it is not necessary that it do so (and this proposal, ironically, could in itself be an example of overuse of essentialist thinking). An alternative explanation for the pervasiveness of the deterministic interpretation of genetic influences is that many people, again even those working in the life sciences, believe this interpretation to be scientifically accurate. After all, this interpretation has been standard teaching in biology classes and science textbooks and even journal articles for many years. And it is

Genetics of Social Behavior 207

not completely wrong even today, though its limitations have become glaringly clear. For example, humans carrying an extra chromosome 23, or even just particular parts of an extra chromosome 23, do express a series of quite specific facial features, health and growth problems, and intellectual disabilities known as Down syndrome, and we do not know of any way to alter this, though we have learned much about how to help these people achieve greater intellectual skills and remediate some of the health problems. Those who carry at least some minimum number of DNA base sequence repeats in a particular position on chromosome 4 will, at least at this stage of our medical knowledge, inevitably develop the serious condition Huntington’s disease in middle adulthood. Even timing of the onset of the disease is quite reliably associated with the number of repeating DNA sequences in excess of that minimum. But the vast majority of traits and conditions that show genetic influences, which includes virtually every behavioral tendency that can be reliably measured (Turkheimer, 2000), do not do so in this kind of oneto-one completely penetrant and deterministic fashion. To be more specific, we know of no social behavior that follows this pattern. Social behaviors are genetically influenced, but not “hard-wired,” genetically determined, or innate. One way to reduce the pervasiveness of the assumption that the empirical presence of genetic influences on social behaviors means that their manifestation is determined may simply be to update educational programs to reflect current understanding more accurately. This takes time to work into the social fabric, but progress in doing so could be accelerated through greater care than is often taken in discussing new findings about genetic influences in the media, and even in presentation of such findings by scientists authoring journal articles.

How to Think about Genetic Influences Instead If not determinism, what understanding should be conveyed about genetic influences on social behaviors? Among genetically influenced medical conditions, only about 2% can be considered monogenic, or caused by single genetic variants (Jablonka & Lamb, 2006). In the other 98%, many, even perhaps thousands, of genetic variants are involved. Genetic variants are any differences in the DNA of individuals within a species, be they substitutions of base pairs (“letters”) of the DNA or small structural differences. The frequencies of the different variants (alleles) in the population can be anywhere from common to extremely rare (even specific to single families or individuals). The same individual allele can be expressed differently in different environmental circumstances, with the environment of any one allele including the rest of the organism’s genetic variants. These conditions emerge through transactions among many genetic variants in some environmental circumstances but not others, and those as well as other genetic variants themselves can influence the kinds of environments

208

BIOLOGICAL THEORIES

individuals seek, end up experiencing, and which subsequently influence later genetic expression. This is the pattern of genetic influence shown by social behaviors. There is not one social behavior that appears to act like the 2% of genetically influenced medical conditions that are considered monogenic. This general situation was long suspected by many developmental geneticists (Johnson, 2012, 2013), but hard evidence began to creep in about 15 years ago as linkage and candidate gene studies failed to replicate, and studies observing gene–environment interactions (G × E) began to pile up. The advent about 10 years ago of genomewide association studies (GWAS) that survey quite closely spaced markers located throughout the genome has made the overall situation very clear (Johnson, Penke, & Spinath, 2011): Though genetic influences on social behaviors are pervasive in some general sense, it is almost impossible that they determine any social behavior in any cohesive way. A dramatic source of evidence for this comes from, of all things, corn oil. One of the longest-running experiments of all time has provided evidence that has upset traditional understanding of genetics (Johnson, 2010; Le Rouzic, Siegel, & Carlborg, 2007). Since 1896, geneticists at the University of Illinois have been studying corn’s response to artificial selection for oil content (Hill, 2005; Laurie et al., 2004). Like many social behaviors, corn oil production varies continuously across individual plants, is influenced by many different genetic variants, and individual plants thrive though producing many different levels of oil. In this experiment, geneticists cannot control variation in seasonal temperature and rainfall from year to year, but all of the corn lines have been planted in the same area, so they have been subject to very similar conditions, particularly in any single generation. There was clear and quite steady response to selection from both groups. Oil content increased from about 5% to about 13% in the group bred for high oil production over 50 years, and kept right on going to over 20% after 100 years. It fell from about 5% to about 1% over 50 years in the group bred for low oil production, and the corn was no longer viable as oil content reached effectively 0% after around 85 years. But in the portions of both groups in which selection was reversed after 50 years, response was reversed as well, and at effectively the same pace as it had gone before. Selection was again reversed after another 5 years in some of the corn that had been selected for high oil production for 50 years and then for low oil production. Once again, oil production increased, and at the same rate as it had originally. The diversity of genetic variants involved in corn’s production of oil had apparently not been diminished by the 50 and 55 years of directional selection. This could only have taken place if some genetic variants contribute to oil production against some genetic backgrounds and not others, or if there is a steady infusion of new gene mutations that contribute to oil production. But all the evidence to date about genetic mutation rates indicates that they are way too low and unsystematic for this to occur (Le Rouzic, Siegel, & Carlborg, 2007). Moreover, after 100 years of selection, the

Genetics of Social Behavior 209

level of oil production in the corn lines that first experience selection for high oil production and then for low oil production was almost identical to that for the lines that had experienced the reverse. Given the very different selection histories in these two lines, it is very likely that few if any of the same genetic variants contributed to these levels at 100 years of selection. Such genetic heterogeneity might be ubiquitous. There is no reason to suspect that the genetics of corn oil are in any way unique, and if anything, social behaviors should be more complex traits than corn oil production. This means that it is very likely that two people can display the same levels of genetically influenced social behaviors, but share none of the genetic variants that contribute to individual differences in these variables in humans. This example of genetic selection in corn for oil production is dramatic for its clarity and consistency, but many recent experiments of much shorter direction point in very similar directions. Their results suggest that the genome is very “deep” and its expression very flexible in many ways. By “deep” we mean that the genome contains considerable redundancy: One genetic variant may be actively involved in some trait when it is present, but if it is not, some other genetic variant may serve the same role, perhaps even just as well. This other genetic variant could be “silent” or inactive if the other variant is there, or its product could simply be “surplus” to this trait and used in some other trait. By “flexible” we mean that environmental conditions independent of the genome do matter too. Year-to-year jags in the lines can be substantial. For example, swings in environmental conditions within periods of 1 to 2 years in the corn lines appear capable of creating as much difference in oil production as about 30 years of selection, but overall there was strong stability in the rate of response. Many geneticists are currently interested in understanding genetic involvement in physical and mental illness as well as behavioral traits. In recent years they have conducted technically powerful GWAS of as many as 1 million common genetic markers with a wide variety of traits and conditions of interest. Although these studies have turned up a few important “hits,” they have not generated anywhere near the expected harvest of understanding about the particular genetic variants involved in anything. Every study has generated a pile of possibilities, but the pile has generally differed quite dramatically from study to study of the same condition. Moreover, even within a single study, the pile of possibilities cannot account for anything close to the existing heritability estimates. This has led to frequent discussion of reasons for the “missing heritability” (Maher, 2008). Many suggestions have been offered, but consensus remains distant. Our own take is that, if the situation in corn is at all common, which we believe to be likely, this kind of genomic flexibility and density, along with some inflation of heritability estimates due to the presence of gene–environment interaction and correlation as discussed below, could easily account for a substantial amount of the missing heritability. The rare variants that GWAS studies do not examine no doubt also contribute.

210

BIOLOGICAL THEORIES

What Does This Imply for Genetic Influences on Social Behaviors? Of course, we do not practice genetic selection on humans as has been done in this experiment in corn, so our evidence of genetic influences on human social behavior is less direct. We can derive similar evidence from species considerably closer than corn, however. For example, most of the major dog breeds have been established and maintained based at least as much on breeding for behavior as for appearance. Breeding for behavior was described specifically in books dating back to 1576 and 1656 (Plomin, DeFries, McClearn, & McGuffin, 2008), and such breeding continues today. For example, in England, most dogs have been bred for hunting, and there are 26 breeds of hunting dogs, each specialized to a particular kind of hunting. In addition, it took only about 40 generations of breeding to create a new breed of fox that is tame enough to have become a popular house pet in Russia (Trut, 1999). Many laboratory experiments with rodents have also demonstrated genetic influences on behavior. Most of these experiments involve selection for high and low levels of the behavior of interest, as well as unselected control lines. For example, large individual differences among mice become evident when they are placed in a large, brightly lit box that has come to be known as the “open field.” Some freeze, defecate, and urinate, while others run around actively exploring the box. Breeding across 30 generations for high and low levels of activity in the open field generated responses strikingly similar to those in corn oil (DeFries, Gervais, & Thomas, 1978). Most of the evidence for the presence of genetic influence on social behavior in humans comes from twin studies. These studies rely on the observation of greater behavioral similarity in mono- than in dizygotic twins. Monozygotic (MZ) twins are effectively genetically identical, as they develop when a single fertilized egg divides early in gestation to go on to produce two individuals rather than the usual one. Dizygotic (DZ) twins are as genetically similar as full singleton siblings who share on average half the genetic variants (i.e., the genetic variance among humans), because they develop when a woman emits more than one egg during a single ovulatory cycle and two of them are independently fertilized. Much has been written about the limitations of twin studies in establishing genetic influences on behaviors, but these limitations, though real, are much more relevant to the magnitudes of estimates of genetic influences than they are to the question of presence versus absence of genetic influences (Johnson, Penke, & Spinath, 2011; Johnson, Turkheimer, Gottesman, & Bouchard, 2009). The implication of this is that, surprising as it may seem, the most relevant difference between corn and humans with respect to understanding and measuring genetic influences is not that we routinely breed corn however we please and develop heebie-jeebies at even the prospect of doing any breeding at all in humans. Rather, the most important difference is that humans move through and actively and passively, consciously and unconsciously, select among

Genetics of Social Behavior 211

various environmental opportunities throughout their lives (Johnson, 2007, 2012, 2013; Penke, 2010), while corn is stuck with whatever environmental conditions into which its seed happens to fall. To understand the implications of this, it is helpful to run through the assumptions on which twin studies rely, and thus the limitations on interpretations of their results. This is especially the case because these assumptions and their attendant limitations are often glossed over in twin study reports. The first assumption is that twins are representative of the more general, mostly singleton, population. There are questions about this assumption for some traits, especially in early childhood, as twins are often born a little early and at lower birth weights than singletons, and as a result are more likely to encounter birthing complications (Crosignani & Rubin, 2000). Most of the effects of these complications however, are generally outgrown by age 6 or so (Christensen et al., 2006). There has been evidence that twins may on average have slightly lower IQs than singletons (Voracek & Haubner, 2008), but the most recent and careful studies contradict this contention (Webbink, Posthuma, Boomsma, de Geus, & Visscher, 2008), offering at least two reasons. With improvements in medical care, birth complications that used to leave lasting effects may no longer do so, and the more recent studies have had the advantage of comparing twin and singleton IQs within families, making for more accurate assessment (Webbink et al., 2008). Past early childhood, there is little evidence for any differences between twins and singletons in personality, which is probably the area closest to social behavior in which the subject has been addressed. One of the largest studies (Johnson, Krueger, Bouchard, & McGue, 2002) found no differences except for the trait of social closeness, for which mean levels were slightly higher in twins. There is evidence that the twin relationship may be particularly close (for perhaps obvious reasons; Segal, 2000), but little or no evidence that twins respond differently than singletons to social stimuli when they are on their own. This is what would be most relevant to the question of the representativeness of twins in studying social behavior in the larger general population. Another assumption is that environmental influences act to make MZ and DZ twins similar to the same degree. This assumption has generally been considered valid (e.g., Plomin et al., 2008), but it also has primarily been tested by measuring whether MZ and DZ twins are treated alike to the same degree by parents and others around them, and whether these treatments impact on relevant behavior patterns. The assumption is actually quite a bit broader than this, and the way in which this is true makes it more likely violated. That is, the assumption includes the implication that whatever is “done” to MZ and DZ twins by the environment will affect them to the same degree. As genetic influences are expressed through environmental contexts (including other genetic background) and DZ twins are less genetically alike than MZ twins, this implication often does not hold true. Twin studies also generally rely on the assumption that people do not mate assortatively; that is, they do not end up with mating partners who are

212

BIOLOGICAL THEORIES

systematically more similar or dissimilar to them than chance on the trait in question. This allows the consequent assumption that DZ twins on average share 50% of their genetic variants. This latter assumption appears overall to be very accurate (Visscher et al., 2006). The accuracy of the assumption of absence of assortative mating for social behaviors, however, is a wide-open question. Most studies have shown little assortative mating for personality (Plomin et al., 2008), but generally considerable assortative mating for intelligence (spousal correlations on the order of .3 to .4) (Plomin et al., 2008). One of the primary reasons it takes place has implications for the potential for assortative mating for social behaviors. Assortative mating for intelligence occurs at least in part because most societies are rather stratified by educational level and people have tended to meet their mating partners while engaged in their day-to-day activities; what makes up those day-to-day activities is always genetically as well as environmentally influenced. Educational programs such as universities tend to accept students on the basis of measured intelligence, either directly, through test scores, or indirectly, through performance in lower-level educational programs. Many of these programs also demand enough investment of time and energy that they inevitably extend to include social activities, so that they tend to become quite central to participants’ lives and thus primary sources of mating partners. The net result is that genetic influences on presence occur in particular kinds of situations, such, for example, as attendance at a university campus party or presence in a bar frequented by auto production-line workers. This kind of sorting process likely applies to presence in all situations, though we may not have identified or developed measures for the traits that influence people’s participation in them. For example, social attitudes also show substantial assortative mating (spousal correlations on the order of .5–.6; Coventry & Keller, 2005). The sorting processes are less formal and do not generally involve qualification processes or explicit enrollment in any kind of program, as is the case with intelligence, so they are more difficult to track. Some studies have begun to address them, however. It is likely that many other social behaviors show similar though perhaps weaker patterns, but this remains speculation at present. This relates directly to the final and most important assumption: that genetic and environmental influences act independently of each other. At this point, we know that this assumption is often violated. It has two major components, one of which is much more commonly investigated than the other. The more commonly investigated component is G × E. It is worth taking a moment to consider the definition of G × E in some depth. This definition can be considered from two contrasting perspectives: genetically controlled differential sensitivity to environmental circumstances, and environmental control of genetic response (Purcell, 2002). The point of using either perspective is that, when G × E takes place, people who differ genetically from one another in some relevant way experience some environmental circumstance differently. But which perspective is more relevant in some particular situ-

Genetics of Social Behavior 213

ation matters greatly. When the first perspective of genetic control is more relevant, it is likely that individuals have some (conscious or unconscious) hand in whether or not they experience the kind of environmental circumstance to which they would be genetically sensitive. If they would be positively genetically sensitive, they are more likely to experience it; if they would be negatively genetically sensitive, they are less likely to experience it. A key here is that the reasons for this difference in probability of experiencing the relevant environmental circumstance are at least partially genetic: The relevant genetic background leads an individual to seek or avoid, as appropriate, and to the degree possible, the relevant environmental circumstance. In contrast, when the second perspective of environmental control is more relevant, individuals likely have little control over whether or not they experience the relevant environmental circumstance. Thus, two people with the same genetic variants involved in a trait may wind up expressing very different levels of the trait if their relevant environmental experiences have been different. Most studies of G × E have considered it in isolation from gene– environment correlation (rGE), which can also be defined from two perspectives: genetic control of exposure to the environment, or environmental control of genetic expression (Purcell, 2002). The reason for the detailed focus on the two perspectives encompassed by the definition of G × E should now be clear: It is through rGE that G × E will be manifest whenever individuals have some (conscious or unconscious) control over whether or not they experience the relevant environmental circumstance (Johnson, 2007). Because individuals almost always do have some control, conscious or not, over the environments they experience, G × E and rGE will generally be linked to at least some degree. Therefore, as we come increasingly to recognize that G × E is common, we must also begin to acknowledge that rGE is also common.

Why Does rGE Matter in Studying Social Behavior? rGE is the elephant in the room, to use the idiomatic expression for a problem or risk or ugly possibility no one wants to bring up, let alone discuss. There may be a too-apt analogy between social scientists’ behavior and that of the detectives in Mark Twain’s The Stolen White Elephant (Twain, 1882), which is likely one major source of the expression: In that story a white elephant in transit from India to Britain as a gift to the Queen is lost in New Jersey. The local police force springs into action, with detectives ineptly searching far and wide in ridiculous places, when in fact the elephant had never left the spot at all. Given Twain’s usual snide use of humor, there is little question that the fact that the elephant in the story was white was no accident: “white elephant” is an idiomatic expression for a possession that is too valuable simply to junk but that entails a burden of upkeep way beyond its worth, so its owner cannot get someone else to take it on for a price that would recoup

214

BIOLOGICAL THEORIES

its value. This only makes the analogy with social scientists’ behavior that much more apt: If we were to find that rGE is behind any of the social ills we experience, it would pose a huge social burden that we could not easily unload, one that is considered politically poisonous. But why would rGE be so bad? The answer to this question can be made clear through further definitions associated with rGE. These gene–environment correlations are often classified as passive, evocative, or active (Scarr & McCartney, 1983). They are passive when individuals both inherit genetic variants influencing behavior and tend to find themselves, through no action of their own (the second perspective in the definition above, environmental control of genetic response), in environmental circumstances that reinforce that behavior. Examples used to illustrate this form of rGE usually focus on childhood, when most people live with the biological families into which they are born, and parenting activities: For example, children of antisocial parents are likely to inherit genetic variants associated with antisocial behavior from those parents and are also more likely to experience at their hands maltreatment that reinforces the children’s own propensities toward antisocial behavior. In contrast, rGE is active when people actively (though not necessarily consciously) seek out and experience environments that reinforce genetic inclinations (the first perspective in the definition above, genetically controlled differential sensitivity to the environment). Common examples are tendencies for bright people of any age to engage in intellectually stimulating activities that build and reinforce already genetically influenced intellectual skills (and the less felicitous opposite tendency for less bright people to do things like watching TV instead). Evocative rGE falls in between the two definitional perspectives: Something about the genetically influenced ways an individual behaves tends to evoke certain kinds of responses from the environment that reinforce the genetic influences on that behavior. So, for example, people genetically inclined to be friendly make more friendly overtures to others, which in turn elicits friendly responses from these others, so that these people end up with more friends than those who do not make so many overtures. Similarly, children genetically inclined to act out anger in disruptive ways get punished for this behavior by parents and teachers and other caretakers, often making them yet angrier. Or people expressing genetic variants involved in higher physical attractiveness might be treated better by others, possibly resulting in the development of a greater sense of entitlement and tendency to be demanding in social interactions (Sell, Tooby, & Cosmides, 2009). As these examples highlight, all these forms of rGE involve reinforcement by the environment of expression of preexisting genetic differences. This means that, over time in the presence of rGE, influences of genetic variation that originally make only very small differences will tend to make increasingly large differences through development and, yes, learning and behavior reinforcement schedules. With time, these differences may become quite large and every bit as firmly stable and hard to change as if they were geneti-

Genetics of Social Behavior 215

cally “hard-wired.” To the extent that this is the case, it will tend to mean that societies become stratified for genetic variants influencing the traits the societies reward and punish. For most industrialized societies, this means positive stratification for genetic variants influencing traits such as intelligence, physical attractiveness, and ability to market one’s skills and abilities to others; and negative stratification for traits such as antisocial behavior, vulnerability to mental illness, and difficulty controlling impulses. Such a situation would undermine all our aspirations to an open society that makes equal opportunity available to all thus, even acknowledging the possibility is about as politically incorrect as one can get. Yet nature does not “care” about our ideals of equality or our standards of political correctness. So if the elephant is in the room, it will “poo” on us (exert its effects) whether or not we are willing to acknowledge its presence. And it will be a white elephant—valuable but “high maintenance”—in that the effects of rGE will potentially be very powerful but difficult to control. Complicating the situation is the fact that it would currently be almost impossible to refute the all-too-likely possibility of rGE and its attendant population genetic stratification. Doing so would involve demonstrating that there are no individual genetic markers associated reliably with socially positively and negatively valued behavioral traits that differ in frequency in various segments of the population to the appropriate degrees, and this is well beyond current technology. Any prospect of demonstrating this is only made more difficult given our present understanding that behavioral traits are influenced by a mixture of many common genetic variants of individually small effects that accumulate to rather substantial effects and genetic variants with potentially substantial effects that are extremely rare in the population (Gibson, 2012). In addition, any of these genetic variants may be expressed in some circumstances and not in others, to the point where two individuals could have the same level of some trait, say intelligence, but share none of the genetic variants that contribute to that trait. Moreover, the statistical techniques and twin studies that have demonstrated the general presence of genetic influences on behavior cannot refute rGE anywhere near as clearly. One of the best of these studies is the discordant twin study (also called twin difference design) that explores whether twins, especially MZ twins, discordant for some environmental condition are also discordant for some outcome that is associated with that condition in the more general population. This provides much stronger evidence than possible in any general population sample that the environmental condition is actually causal for the outcome. The problems are that most twin samples are quite short on MZ twin pairs discordant for environmental conditions, so power is generally low (McGue, Osler, & Christensen, 2010). Moreover, most such studies to date have found that the association between the environmental condition and the outcome is much smaller in MZ twins than in the general population, providing evidence for the presence of rGE. It is almost as

216

BIOLOGICAL THEORIES

impossibly difficult at present to find evidence clearly supporting the presence of rGE as well. This is because we do not know of any individual genetic variants that are reliably associated with behavioral traits, and it would take even more statistical power to demonstrate that any such genetic variants differ in frequency across social strata than it would to identify them in the first place. All this only hinders us from acknowledging the elephant that is all too likely in the room.

Another Reason rGE Matters in Studying Social Behavior But there is another important reason that rGE matters in studying social behavior. It involves the kind of lab experiments dear to the hearts of social psychologists and generally considered the gold standard for causal inference. Such experiments typically assign participants to artificially constructed laboratory situations. Generally, this is done purportedly randomly, under the assumptions that it does not matter who is placed where because on average everyone would respond the same to any given situation. It is considered ecologically valid under the implicit assumption that situations develop out of aggregated actions of many circumstances over which no one individual has any control. Increasingly, these experiments are being designed with recognition that not everyone responds the same way, but then a single measured difference in response tends to be the focus of study, examined as a statistical interaction. If G × E is common, however, the assumption that everyone will respond the same way is often violated, and specific interaction effects incorporated in the study design will likely not capture all of it. Moreover, the links between G × E and rGE and the common occurrence of rGE in its own right mean that situations do not just develop out of aggregated actions over which no one individual has any control. Instead, individuals exert considerable control over whether and how a situation develops at all, as well as how it turns out. After all, we can think of life as no more or less than an ongoing series of situations. The control may not be conscious and may be the outgrowth of long-standing patterns of behavior that have little overt association with entry into any particular situation. An example may help to clarify this notion. Wendy and Lars, the two authors of this chapter, were walking back toward downtown Edinburgh and their respective homes one Saturday night around 3:00 a.m. from a friend’s party. The walk took about an hour, which they spent engaged in animated conversation about, what else, some aspect of psychology. Perhaps about halfway there, someone at least slightly drunk approached and tried to pick a confrontation with Lars, striking him on the arm in the process. Lars responded to put him off, ignored the fact that he had been struck, the potential confrontation fizzled, and the man, and Lars and Wendy, continued on their respective ways. It might be easy to

Genetics of Social Behavior 217

see how Lars exerted control over the outcome of this situation, but how did he exert control over finding himself in it? Well, Lars is a rather large man, both tall in height and large in build. He may not have had any kind of direct control over this development, but size shows large genetically influenced individual differences, and individuals learn to accommodate their behavior to their sizes and the responses in others which their sizes create (Sell et al., 2009). In Lars’s case, we can speculate that some of his other characteristics have been at least partially accommodations to his size. Whether or not this is true, Lars’s other, also genetically influenced, characteristics relevant to his entering this situation include enjoyment of parties with friends, willingness and even eagerness to stay at them till late at night and undertake long walks home through all parts of town, never-ending interest in animated discussion, a distinctive style of dress (always all in black with his long hair tied back in a ponytail), self-confidence and assertiveness that shows in how he carries his body, a habit of talking with assurance in quite a loud voice, and, though very fluent, an obvious German accent in his English speech. He would definitely never have gotten into this situation that night if he had gone home much earlier with his wife when she took their son home in their car, and he very well might not have encountered it if he did not have as many habits of speech, dress, and carriage that elicit the attention of others. There is little question as well that, once entered, this situation could have had much different outcomes if Lars was not also willing to let small affronts just pass. Lars did not think consciously about entering this situation in any way, and he would not even say that he consciously contemplated response alternatives once in the situation, but both his presence in the situation and his response to it reflected long-term behavior patterns that show large genetically influenced individual differences. Wendy’s presence (or rather, really, absence) in this situation was similarly characteristic of her. She was happy to have Lars’s company for the long walk back to town, but she would have done it on her own if he had not been there and indeed she did proceed alone to her own flat when their routes parted downtown. She was completely ignored by the man during his encounter with Lars, and the fact that this is typical is no doubt part of what has led her to feel, over many years, free to walk around alone at night as a woman: She is small but not short, far from glamorous in appearance or dress but also far from decrepit, carries herself as if she is just on her own business, seems not to attract much attention at all as she goes along, and almost reflexively deflects what little attention she does attract. (Of course, no one controls the situations they enter into or do not so completely that she should truly feel safe, but that is also where the genetically influenced characteristic of being willing to accept some risk comes in.) Everyone has analogous patterns of behavior and preferences that contribute directly, but often not consciously, to the situations in which they enter and do not enter.

218

BIOLOGICAL THEORIES

Control over entry into situations is never complete, and people have many different, often contradictory, goals and motivations. This means that they commonly find themselves, consciously or unconsciously, having entered into situations they do not prefer. Some examples, such as being rearended at a stop light while a passenger in the car, are clear, but presence in many situations involves more complex combinations of choice and lack of choice. An extreme but not uncommon example might be a not particularly naturally nurturant husband who becomes caretaker when his wife develops Alzheimer’s disease. As has been well documented, this role is a very stressful role for most people. Few would choose it, but many accept it at least for some period of time out of love, loyalty, and sense of duty. Entering into any nonpreferred situation means tolerating the stress generated by the need to behave in ways appropriate to it, which always has individually characteristic repercussions on genetic expression that may influence other aspects of a person’s life and even physical and mental health as well (Schmalhausen, 1949). Waddington (1953), quoted in the opening of this chapter, provided one of the first experimental demonstrations of this principle in fruit flies, and we can see its traces in humans (e.g., Hicks, South, DiRago, Iacono, & McGue, 2009; Johnson & Krueger, 2005a, 2005b; Johnson et al., 2010). At present, however, we have little understanding of any specific mechanisms, and associations involving individual specific genetic variants are likely to measure out as at best wobbly (e.g., Duncan & Keller, 2011; McGuffin, Alsabban, & Uher, 2011; Risch et al., 2009; Uher & McGuffin, 2007). Lab experiments make a mess of these “normal” conditions of social situations. They commonly make use of university students, a segment of the population selected for genetically influenced traits of intelligence, conscientiousness, absence of antisocial behavior, and other traits (Henrich, Heine, & Norenzayan, 2010; Sears, 1986). These samples are at best not representative of the general population for these observed traits, but they are also likely not representative of the general population for whatever genetic variants are involved in these traits. Moreover, they are assigned during the experiment to artificial situations they might never experience or choose, consciously or unconsciously, to enter into on their own. In the lab, they know the circumstances are artificial and have received the blessing of some ethical review panel as being unlikely to confer any lasting real discomfort on them. In addition, they often involve academic credit that may prime them to be motivated to please the experimenters by behaving in whatever way they think the experimenter expects (Klein et al., 2012). This means that the behaviors these individuals display may be different in ways that are all too relevant to whatever is under study from any behaviors they would usually display. At the same time, it means that whatever behaviors they are displaying may not be representative samples of those the full population would display. And the lab experiment does not follow them to gather any information about how they deal with whatever stress this odd combination of choice and lack of choice of situation has generated. Your guess of what this means

Genetics of Social Behavior 219

for interpreting results is as good as ours, but the likelihood that it often has relevance should not be ignored.

Implications for Theory and Explanation in Social Psychology All this has two kinds of implications for theory and explanation in social psychology. The first concerns how to understand and interpret presented evidence that the heritability of some social behavior is some amount like 35% (which is very typical of the levels of heritability that can be expected for all social behaviors; Johnson, Penke, & Spinath, 2011). Such estimates confirm the presence of genetic influences, and by now there is no doubt that such influences are ubiquitous (Turkheimer, 2000). They directly imply that explanations of differences between individuals in social behavior cannot ultimately be completely environmental. But they say nothing about exactly how genetic differences are involved. They refer only to the variance in the trait across the population, not to anything about trait levels in any individual or even within the sample used to generate them. The specific magnitude of any estimate is very likely wrong because it was calculated based on assumptions that probably did not hold. It is not possible to say whether an estimate is high or low, however, without knowing which assumptions were violated and how. Even simple psychometric aspects, like how reliably the behavior was measured, influence the sizes of heritability coefficients. This really does not matter though because the particular magnitude of any estimate does not tell us much of anything about the trait. This is because heritabilities are ratios of genetically influenced variance to total variance, with environmentally influenced variance and measurement error making up the rest of the total. Heritability can be high simply because relevant environmental variance happens to be low temporarily or in the sample in question, or because violation of the assumptions about the independence of genetic and environmental influences happens to be great, rather than because the trait is inherently genetically determined (Johnson, Penke, & Spinath, 2011). Thus, the primary value of heritability estimates is the evidence they provide that genetic influences transmissible from one generation to the next are important and substantial, but not much more. Currently, epigenetic phenomena are commonly offered as possible ways in which it could appear that genetic influences are transmitted from one generation to the next when in fact it is an environmental influence that is transmitted. Epigenetic phenomena involve differences in gene activation and expression that are not tied to genotypic variation. For example, mice (mus muscalis) commonly show coats of either yellow or agouti (ticked, with different colors on single hairs) fur. Yellow color is not due to variation in the gene that controls coat color, but rather is due to suppression of expression of that gene. The suppression of expression is controlled in turn by a genetic

220

BIOLOGICAL THEORIES

variant lying upstream along the DNA from the color gene. But it can also be induced by feeding the mice a particular kind of otherwise-nutritious food. Female mice that have yellow coats due to consuming this food are more likely to have offspring with yellow coats through at least the next two generations (Blewitt, Vickaryous, Paldi, Koseki, & Whitelaw, 2006). Such epigenetic phenomena are likely involved somehow in G × E and rGE, but at present we know very little about how or to what degree. Epigenetics is sometimes discussed as if it were an alternative to genetic influences, or as a mechanism by which environmental influences could override genetic differences and make environmental influences on behavior transmissible across generations. These claims are grossly overstated. We do have evidence for environmental influences that affect gene regulation and translate to behavioral differences, mostly from model organism studies (Hoffmann & Spengler, 2012). Little is known reliably about similar influences in humans, though it is highly unlikely such influences are absent. What we do know points toward interactions between genetic differences and epigenetic influences, which are often genetically influenced themselves, not to epigenetic influences as alternatives to genetic influences (Murgatroyd & Spengler, 2012). Furthermore, while inheritance of epigenetic differences is possible, it is likely not very stable across generations and not a good explanation for the observed genetic influences on psychological traits and behaviors (Slatkin, 2009). So what do we get from estimates of genetic influences? Well, that brings us to the second kind of implications. These implications concern how the ubiquitous presence of genetic influences, G × E, and rGE should be taken into consideration in developing theories in social psychology. It means recognizing that the ways in which individuals differ genetically will tend to be expressed most freely when their behavior is least constrained by social norms and conscious goals. This might mean they may tend to be expressed more clearly in priming and in implicit rather than explicit attitudes. It also means that genetic differences likely distort the representativeness of behaviors in lab experiments. It means their influence on traits such as intelligence which contribute to what counts as success or failure in any situation must be taken into consideration in theories about the development and maintenance of self-esteem. Similarly, it means that their influence on temperamental traits that contribute to emotional experience and response must be taken into consideration in understanding the development of early attachment and its influences on later social relationships. In short, their traces pervade all areas of social psychology, but we are only beginning to understand this, never mind just how. As Waddington said so well in the introductory quote, this is going to mean developing new theories, as well as new methods to test them. Of course, social-psychological phenomena are remote from the level of gene action, but social-psychological theories still need to be consistent with our understanding of genetic variation and the ways in which their actions transact with those of the environment. The end result could well

Genetics of Social Behavior 221

transform understanding in the field, as much as appropriate assembly and programming transform the capabilities of wire and plastic when we build them into a computer that can be programmed to beat us at chess. References Bastian, B., & Haslam, N. (2006). Psychological essentialism and stereotype endorsement. Journal of Experimental and Social Psychology, 42, 228–235. Blewitt, M. E., Vickaryous, N. K., Paldi, A., Koseki, H., & Whitelaw, E. (2006). Dynamic reprogramming of DNA methylation at an epigenetically sensitive allele in mice. PLoS Genetics, 2, 399–405. Christensen, K., Peterson, I., Skytthe, A., Herskind, A. M., McGue, M., & Bingley, P. (2006). Comparison of academic performance of twins and singletons in adolescence: Follow-up study. British Medical Journal, 333, 1095–1097. Coventry, W. L., & Keller, M. C. (2005). Estimating the extent of parameter bias in the classical twin design: A comparison of parameter estimates from extended twin-family and classical twin designs. Twin Research and Human Genetics, 8, 214–223. Crosignani, P. G., & Rubin, B. L. (2000). Multiple gestation pregnancy. Human Reproduction, 15, 1856–1864. Dar-Nimrod, I., & Heine, S. J. (2011). Genetic essentialism: On the deceptive determinism of DNA. Psychological Bulletin, 5, 800–818. DeFries, J. C., Gervais, M. C., & Thomas, E. A. (1978). Response to 30 generations of selection for open-field activity in laboratory mice. Behavior Genetics, 8, 3–13. Duncan, L. E., & Keller, M. C. (2011). A critical review of the first 10 years of candidate gene-by-environment interaction research in psychiatry. American Journal of Psychiatry, 168, 1041–1049. Gelman, S. (2003). The essential child: Origins of essentialism in everyday thought. New York: Oxford University Press. Gibson, G. (2012). Rare and common variants: Twenty arguments. Nature Reviews Genetics, 13, 135–145. Henrich, J., Heine, S. J., & Norenzayan, A. (2010). The weirdest people in the world. Behavioral and Brain Sciences, 33, 61–83. Hicks, B., South, S. C., DiRago, A. C., Iacono, W. G., & McGue, M. (2009). Environmental adversity and increasing genetic risk for externalizing disorders. Archives of General Psychiatry, 66, 640–648. Hill, W. G. (2005). A century of corn selection. Science, 30, 683–684. Hoffmann, A., & Spengler, D. (2014). DNA memories of early social life. Neuroscience, 264, 64–75. Jablonka, E., & Lamb, M. J. (2006). Evolution in four dimensions: Genetic, epigenetic, behavioral, and symbolic variation in the history of life. Cambridge, MA: MIT Press. Johnson, W. (2007). Genetic influences on behavior: Capturing all the interplay. Psychological Review, 114, 423–440. Johnson, W. (2010). Understanding the genetics of intelligence: Can height help? Can corn oil? Current Directions in Psychological Science, 19, 177–182. Johnson, W. (2012). Developmental genetics and psychopathology: Some new feathers for a fine old hat. Development and Psycholopathology, 24, 1165–1177. Johnson, W. (2013). What do genes have to do with cognition? In S. Kreitler (Ed.), Cognition and motivation (pp. 192–214). New York: Cambridge University Press. Johnson, W., & Krueger, R. F. (2005a). Genetic effects on physical health: Lower at higher income levels. Behavior Genetics, 35, 579–590.

222

BIOLOGICAL THEORIES

Johnson, W., & Krueger, R. F. (2005b). Higher perceived life control decreases genetic variance in physical health: Evidence from a national twin study. Journal of Personality and Social Psychology, 88, 165–173. Johnson, W., Krueger, R. F., Bouchard, T. J., & McGue, M. (2002). The personalities of twins: Just ordinary folks. Twin Research, 5, 125–131. Johnson, W., Kyvik, K. O., Mortensen, E. L., Skytthe, A., Batty, G. D., & Deary, I. J. (2010). Education reduces the effects of genetic susceptibilities to poor physical health. International Journal of Epidemiology, 39, 406–414. Johnson, W., Penke, L., & Spinath, F. M. (2011). Heritability in the era of molecular genetics: Some thoughts for understanding genetic influences on behavioural traits. European Journal of Personality, 25, 254–266. Johnson, W., Turkheimer, E., Gottesman, I. I., & Bouchard, T. J. (2009). Beyond heritability: Twin studies in behavioral research. Current Directions in Psychological Science, 18, 217–220. Klein, O., Doyen, S., Leys, S., Magelhaes de Saldanha da Gama, P. A., Miller, S., & Cleeremans, A. (2012). Low hopes, high expectations: Expectancy effects and the replicability of behavioral experiments. Perspectives on Psychological Science, 7, 572–584. Laurie, C. C., Chasalow, S. D., Le Deaux, J. R., McCarroll, R., Bush, D., Haug, B., et al. (2004). The genetic architecture of response to long-term artificial selection for oil concentration in the maize kernel. Genetics, 168, 2144–2155. Le Rouzic, A., Siegel, P. B., & Carlborg, O. (2007). Phenotypic evolution from genetic polymorphisms in a radial network architecture. BMC Biology, 5, 50. Maher, B. (2008). Personal genomes: The case of the missing heritability. Nature, 456, 18–21. McGue, M., Osler, M., & Christensen, K. (2010). Causal inference and observational research: The utility of twins. Perspectives on Psychological Science, 5, 546–556. McGuffin, P., Alsabban, S., & Uher, R. (2011). The truth about genetic variation in the serotonin transporter gene and response to stress and medication. British Journal of Psychiatry, 198, 424–427. Moore, G. E. (1903). Principia Ethica. Cambridge, UK: Cambridge University Press. Murgatroyd, C., & Spengler, D. (2012). Genetic variation in the epigenetic machinery and mental health. Current Psychiatry Reports, 1–12. Penke, L. (2010). Bridging the gap between modern evolutionary psychology and the study of individual differences. In D. M. Buss & P. H. Hawley (Eds.), The evolution of personality and individual differences (pp. 243–279). New York: Oxford University Press. Plomin, R., De Fries, J. C., McClearn, G. E., & McGuffin, P. (2008). Behavioral genetics. New York: Worth Purcell, S. (2002). Variance component models for gene–environment interaction in twin analysis. Twin Research, 5, 554–571. Risch, N., Herrell, R., Lehner, T., Liang, K. Y., Eaves, L., Hoh, J., et al. (2009). Interaction between the serotonin transporter gene (5-HTTLPR), stressful life events, and risk of depression: A meta-analysis. Journal of the American Medical Association, 301, 2462– 2471. Scarr, S., & McCartney, K. (1983). How people make their own environments: A theory of genotype–environment effects. Child Development, 54, 424–435. Schmalhausen, I. I. (1949). Factors of evolution: The theory of stabilizing selection. Philadelphia: Blakiston. Sears, D. O. (1986). College sophomores in the laboratory: Influences of a narrow data base on social psychology’s view of human nature. Journal of Personality and Social Psychology, 51, 515–530. Segal, N. L. (2000). Entwined lives: Twins and what they tell us about human behavior. New York: Plume.

Genetics of Social Behavior 223 Sell, A., Tooby, J., & Cosmides, L. (2009). Formidability and the logic of human anger. Proceedings of the National Academy of Sciences USA, 106, 15073–15078. Slatkin, M. (2009). Epigenetic inheritance and the missing heritability problem. Genetics, 182, 845–850. Trut, L. N. (1999). Early canid domestication: The fox farm experiment. American Scientist, 87, 160–169. Turkheimer, E. (2000). Three laws of behavioral genetics and what they mean. Current Directions in Psychological Science, 9, 160–164. Twain, M. (1882). The stolen white elephant. Boston: J. R. Osgood. Uher, R., & McGuffin, P. (2007). The moderation by the serotonin transporter gene of environmental adversity in the aetiology of mental illness: Review and methodological analysis. Molecular Psychiatry, 12, 1–16. Visscher, P. M., Medland, S. E., Ferreira, M. A., Morley, K. I., Zhu, G., Cornes, B. K., et al. (2006). Assumption-free estimation of heritability from genome-wide identity-bydescent sharing between full siblings. PLoS Genetics, 2, 316–325. Voracek, M., & Haubner, T. (2008). Twin-singleton differences in intelligence: A meta-analysis. Psychological Reports, 102, 951–962. Waddington, C. H. (1953). Genetic assimilation of an acquired character. Evolution, 7, 118– 126. Webbink, D., Posthuma, D., Boomsma, D. I., de Geus, E. J., & Visscher, P. M. (2008). Do twins have lower cognitive ability than singletons? Intelligence, 36, 539–547.

11 Evolutionary Theories Timothy Ketelaar

A

year before his death, Kurt Lewin taught a graduate course at MIT on theories in social psychology. Stanley Schacter, a student enrolled in the course, described it as “more philosophy of science than a review of different theories. Its purpose was to teach students what a theory is, how to test a theory appropriately, and what makes a theory useful” (Higgins, 2004, p. 138). In this light, Lewin’s (1951) assertion that “there is nothing so practical as a good theory” seems less like a simple call for an increased quantity of theorizing in social psychology, and more like a gentle admonition directed toward social psychologists, urging them to develop a better understanding of the role that theories can and should play in their field. More than a half century before Kurt Lewin encouraged his fellow social psychologists to ponder the role of theories in social psychology, William James expressed a more general concern about the lack of theorizing in scientific psychology writ large. James saw the emerging science of the mind as too much dominated by purely descriptive accounts of human mental activity. In reviewing the past several centuries of scholarship on human emotion, for example, William James (1890/2009, p. 308) worried that a focus on description rather than explanation had resulted in a psychological litera 224

Evolutionary Theories 225

ture that failed to provide “a central point of view, or a deductive or generative principle. They distinguish and refine and specify in infinitum without ever getting on to another logical level.” One interpretation of this call for psychology to get “on to another logical level” is that James was urging the young field of scientific psychology to identify a set of guiding theoretical assumptions akin to what scholars employed in more mature fields such as chemistry and physiology.1 Following in William James’s footsteps, Kurt Lewin (1935, 1951) argued a half century later that if social psychologists wanted to become sophisticated social scientists, they too would have to start behaving more like natural scientists such as physicists and chemists. Lewin’s influence on social psychology gave rise to two important legacies: One legacy inspired social psychologists to become ever more capable experimentalists, the other urged them to become more sophisticated theoreticians (Kruglanski, 2001). Jerome Singer (1987, p. 16), a student of Stanley Schacter, and thus an academic grandson of Lewin, noted: “The basic steps in conducting Lewinian based research were to set up an axiom system from which were derived theorems and correlates. Each derivation became a statement subject to experimental test.” Although Lewin’s legacy inspired a greater appreciation for sophisticated causal explanations, it appears that over the years social psychologists have focused more on developing their methodological sophistication (becoming more capable experimentalists) than their theoretical acumen (becoming more sophisticated theoreticians). Remarking on social psychology’s emphasis on experimentation over theorizing, Kruglanski (2001, p. 871) observed that “the field has generally eschewed broad theorizing and tended to limit its conceptualizations to relatively narrow ‘mid-range’ notions closely linked to the operational level of analysis.” In this light, evolutionary psychologists see evolutionary theorizing in social psychology not just as an opportunity to construct sophisticated explanations of social phenomena, but rather as a potential pathway to creating a mature science of psychology, one that offers a grand meta-theory “capable of integrating existing findings across different domains and guiding researchers toward new empirical hypotheses” (Simpson & Kenrick, 1997, p. ix; see also Buss, 1990; Schaller, Simpson, & Kenrick, 2006). The aim of the current chapter is to review the role that theories play in social psychology by exploring the role of one particular theoretical framework—evolutionary psychology—in accounting for social behavior. In the spirit of Kurt Lewin’s graduate course on theories in social psychology, this chapter focuses more on philosophy of science than on a review of different evolutionary theories in social psychology. Of primary interest are three Lewin-inspired questions: What is an evolutionary theory? How do you test an evolutionary theory? What makes evolutionary theories useful in social psychology?

226

BIOLOGICAL THEORIES

What Is an Evolutionary Theory?     The problem is not that a majority of researchers would say that theory     is irrelevant; the problem is that almost anything passes as a theory.      —Gerd Gigerenzer (1998, p. 195)

Perhaps the largest stumbling block in developing good theories in social psychology has been a basic confusion about just what constitutes a theory. Although a proliferation of explanatory accounts in social psychology over the past century have been given the label “theory,” social psychologists traditionally have received relatively little training in how to construct and evaluate theories. It would appear that most social psychologists learn about theories, not by receiving formal training on what constitutes a theory and how to construct one (i.e., by taking a graduate course with Kurt Lewin!), but rather by repeated exposure to a series of explanatory accounts that have been labeled as theories. Before describing what makes a theory “evolutionary,” it makes sense to first consider the more basic Lewin-inspired question of how to define a scientific theory per se.

A Theory Is a Level of Explanation By imploring philosophers of mind to behave more like natural scientists, William James urged early psychologists to construct explanations that connected empirical data to explicit theoretical principles. Consistent with William James’s recommendation, I offer the following definition of a theory: A theory is a middle-level explanation lying above the level of the empirical data (that the theory purports to explain) and below the level of higher-order assumptions (metatheory) from which the theory has been derived (see Figure 11.1). More specifically, a theory is an explanation that is both inductively consistent with a set of empirical facts and deductively consistent with a set of higher-order metatheoretical assumptions. According to this definition, for the label “theory” to be warranted, an explanation must not only invoke inductive and deductive inferences, but perhaps more important, it must also reference both: (1) empirical data and (2) higher-order assumptions (a metatheory). Although the definition of “theory” proposed here describes an essential role for deductive reasoning (see Figure 11.1), the current definition differs from the classic Lewin-inspired approach to theorizing observed in mainstream social psychology. Whereas the classic Lewin-inspired approach to theorizing emphasizes the theory-testing role of deductive reasoning (e.g., “Each derivation became a statement subject to experimental test”; in Singer, 1987, p. 16), the definition of theory proposed here also emphasizes the theory-constructing role of deduction. According to this definition, the label “theory” is justified if and only if an explanation satisfies two conditions: (1) it must be inductively connected to empirical facts, and (2) it must be deductively connected to a specific set of higher-order assumptions. This

evolutionary Theories

227

fIgURe 11.1 Theories as middle-level explanations sandwiched between facts and metatheoretical assumptions.

second point (concerning a deductive connection to a metatheory) is important because much of social psychology appears to subscribe to a Lewininspired view that almost any explanation that is inductively generated from data can be deemed worthy of the label “theory” if this explanation allows researchers to subsequently derive explicit statements that can be subjected to empirical test (falsification or verification). By contrast with this rather permissive use of the term “theory” found in mainstream social psychology, the definition of theory proposed here states that for the label “theory” to be warranted, an explanation must be shown to be deductively connected to a set of higher-order assumptions (a metatheory) that generated this explanation (see Figure 11.1). In other words, regardless of whether a particular explanation allows us to deduce statements that can be tested, an explanation might nonetheless be judged as “atheoretical” (i.e., not worthy of the label “theory”) if the explanation is not logically (deductively) connected to a specific set of higher-order assumptions (see Figure 11.1).

Inductive Elements of a Theory Induction is the process of generating nomothetic2 statements about the world from the observation of more particular states of the world. The claim that a theory must necessarily be inductively connected to empirical data implies two things. First, a scientific theory is not just any sort of inductive generalization; a theory—according to the currently proposed definition—is necessarily a generalization about empirical data (i.e., observable facts that can be verified by third-parties). Statements that are not grounded in empirical data may be interesting and important philosophical observations, but they are not worthy of the label “theory” according to the definition of a sci-

228

BIOLOGICAL THEORIES

entific theory proposed here (see Figure 11.1). Second, a theory corresponds to an attempt to provide a nomothetic statement about the world (e.g., “a theory of precipitation” from the science of meteorology), as opposed to an idiographic description of a specific state of affairs (e.g., the statement that “the ground was wet this morning”). By invoking an inductive style of reasoning that spans from specific statements (about empirical data) to a set of more general principles that purport to account for those data, this definition of theory necessarily entails that for an explanation to be worthy of the label “theory,” it must correspond to a nomothetic statement that is more general in scope than the facts that it purports to explain. It follows that merely descriptive statements—often referred to as virtus dormativa accounts3 (see Gigerenzer, 1998)—are poor candidates for theories. Thus, the statement “The ground is wet” is not a good candidate for a theory because this statement does not explicitly invoke a more general, nomothetic statement about the world. By contrast, the statement “Rain caused the ground to become wet this morning” would be a better candidate for a theory (explaining the wetness of the ground) because it is easier to see how this statement corresponds to a more general nomothetic statement that references general laws or principles (drawn from the science of meteorology). Similarly, the observation that “levels of xenophobia are correlated with pathogen prevalence in a particular sample of data” does not itself constitute a theory of xenophobia because it appears to merely provide an idiographic description of a particular sample of data. By contrast, the statement “psychological mechanisms that evolved to defend against pathogens are one cause of xenophobia in modern environments” would be a better candidate for a theory (of xenophobia) because this statement is more general in scope than a simple idiographic description of a particular set of facts that it purports to explain.

Deductive Elements of a Theory Deduction is the process of inferring a more specific statement about the world from a more general statement. Deduction is a form of logic that lies at the heart of the scientific method, not only insuring that scientific explanations are internally consistent (not self-contradicting), but also enabling scientists to evaluate their theories by testing whether the specific “theorems and correlates” derived from these theories stand up to the weight of empirical evidence. Yet, scientific theories should do more than simply allow Lewin-inspired social scientists to set up axiom systems from which theorems and correlates can be derived and subjected to experimental test. As stated earlier, according to the definition of theory proposed here, an explanation is not worthy of the label “theory” unless it also invokes a set of higher-order assumptions (a metatheory) from which it can be deduced. This is an important point because so much of what passes for “theory” in social psychology appears to be limited to inductive generalizations from robust patterns of data that are then labeled as “theories” simply because one can derive testable state-

Evolutionary Theories 229

ments from these generalizations. Personality theorist Raymond Cattell referred to such a process of inductively generating explanations and then deductively testing them, not as “theory testing” per se, but rather as the “inductive–hypothetico–deductive” method (Cattell, 1988). For many social scientists, this inductive–hypothetico–deductive strategy is simply a generic description of the scientific method. Yet, for legions of Lewin-inspired social scientists this is “theorizing,” a ritual whereby social scientists inductively construct explanatory accounts of data that they then label as “theories” (e.g., Fishbein & Azjen’s [1975] theory of reasoned action or Heider’s [1958] balance theory) from which they develop axiom systems of theorems and correlates that are subsequently translated into statements that are, in turn, subjected to experimental test. The end result, as anticipated by Cattell, is that these explanations (putative “theories”) are either eliminated or further refined in light of the results of these empirical tests. So far, so good, except that nothing about this entire process of deriving testable statements from inductively constructed explanations entails a logical connection to a set of higher-order theoretical assumptions. In other words, the inductive– hypothetico–deductive strategy employed by social psychologists does not require that their deductive reasoning involves reasoning from (1) a set of higher-order theoretical principles (a metatheory) to (2) a particular middlelevel explanation (a theory). Much of what is called theorizing in mainstream social psychology corresponds to a form of the inductive–hypothetico–deductive strategy that is limited to inductively reasoning from a body of data to a middle-level explanation (theory construction) and then reasoning back (deductively) from this theory to a set of statements (hypotheses and predictions) that are then tested empirically by collecting a new body of data (theory testing). Why is this a concern? This is a concern, one might argue, because it is the routine failure to employ deductive reasoning in theory construction in social psychology (as opposed to the routine use of deduction in theory testing) that may be responsible for the relatively poor rate of accumulation of knowledge in social psychology in comparison to the more mature sciences (see Meehl, 1978). By contrast, in the more mature sciences, such as biology, chemistry, and physics, an appeal to higher-order assumptions is so taken for granted in theory construction that it is rarely articulated; as when one observes that chemists do not generally propose theories of chemical phenomena that violate basic principles of elementary physics and biologists do not generally propose theories of biological phenomena that violate basic principles of chemistry (see Barkow, Cosmides, & Tooby, 1992). In other words, natural scientists in disciplines such as biology, chemistry, and physics do not restrict their use of deductive reasoning to theory testing in the form of deducing statements from inductively constructed middle-level explanations and then comparing these statements to another set of empirical data (see Figure 11.1). Instead, scholars operating in these more established scientific fields actually use deductive reasoning to construct their theories! It is in this sense that we might say that the explicit use of a metatheory in evolutionary social psychol-

230

BIOLOGICAL THEORIES

ogy results in middle-level theories that are constructed (partly) out of data (e.g., empirical observations) and partly out of metatheoretical assumptions (see Buss, 1990, 1995; Ketelaar & Ellis, 2000).

What Makes a Theory “Evolutionary?” By contrast with theorizing in mainstream social psychology, evolutionary psychologists are just as apt to use deductive reasoning in their theory construction as in their theory testing. One way in which deductive reasoning plays an important role in theory construction is by providing (deductive) constraints on the sorts of middle-explanations that one can generate when attempting to provide an account for a particular set of observations. This is not to say that evolutionary psychologists do not employ inductive reasoning when they construct theories (middle-level explanations) based upon empirical data (they do!). Rather, the point is that evolutionary psychologists also employ deductive reasoning (from meta-theory to middle-level explanation, see Figure 11.1) when they construct their theories. More specifically, the use of an evolutionary meta-theory plays an important role in narrowing the scope of evolutionary-psychological explanations to a delimited set of plausible a priori alternative hypotheses (see Ketelaar & Ellis, 2000; Ketelaar, 2002 for a fuller treatment of this issue). Through this process of deductive reasoning, an evolutionary meta-theory allows evolutionary psychologists to focus their efforts on generating middle-level explanations that entail psychological mechanisms that could have—in principle—evolved through natural and sexual selection. Evolutionary psychologists then focus their attention on testing hypotheses about these kinds of mechanisms, ignoring the larger set of hypotheses about psychological mechanisms that they could imagine but that—in principle—could not have evolved (and are thus unlikely to receive empirical support). By contrast, the tendency to restrict deductive reasoning to theory testing (at the expense of theory construction) results in a much different type of theorizing in mainstream social psychology. When one looks back at prominent “theories” in social psychology such as Fishbein and Ajzen’s (1975) theory of reasoned action or Heider’s (1958) balance theory, it is hard to avoid the conclusion that far from being explanations that have been deductively constructed from higher-order assumptions about the mind or human nature, such “theories” appear to be little more than labels for interesting patterns of empirical observations. As influential and important as these explanatory frameworks (labels such as “balance theory”) may be, they fail to appeal to an explicit set of deductive or general principles from which these explanatory systems have been deduced. This is not to say that some creative social scientist could not identify, post hoc, a set of more general assumptions about the world that could be used to deduce the “theory of reasoned action” or “balance theory.” Rather, the point is that these “theo-

Evolutionary Theories 231

ries” appear to be purely inductively generated nomothetic explanations, as opposed to being middle-level explanations that have been constructed (i.e., deduced) from an explicit set of higher-order metatheoretical assumptions about human psychology. This is an important claim because one suspects that the failure to distinguish between purely inductive “midrange” explanations and proper “theories” (defined in the current chapter as middle-level explanations that are both inductively connected to data and deductively connected [logically derived] from an explicit set of “general and deductive” principles; see Figure 11.1) may be the source of the perception that social psychology has progressed at a relatively slow rate compared to more mature sciences (Meehl, 1978). Rather than developing the sorts of grand theoretical frameworks that we observe in the more mature sciences, social psychologists appear to spend a disproportionate amount of time trapped in an inductive–hypothetico–deductive spiral endlessly testing a series of competing midrange explanations that have been inductively constructed from data. As several evolutionary social psychologists note: “What is ironic is that the general framework of such a grand theory—Darwin’s theory of evolution by natural selection—has been around for more than 130 years, yet until recently, it has been largely ignored or overlooked by most social psychologists” (Simpson & Kenrick, 1997, pp. ix–x). In sum, the phrase evolutionary theories should be used to refer to middle-level explanations that are deductively connected to a broader evolutionary metatheory used to construct these explanations (see Figure 11.1; see also Buss, 1990, 1995). For example, the behavioral immune system approach to xenophobia (see below) can be seen as an example of a middle-level evolutionary theory of xenophobia (see Schaller & Neuberg, 2011) but should not be confused with the broader evolutionary metatheory used to construct this theory of xenophobia. For most evolutionary psychologists, the broader evolutionary metatheory corresponds to a set of deductive and generative principles located in the adaptationist program in evolutionary biology (Ketelaar & Ellis, 2000). Although a complete description of the core assumptions of the adaptationist program in evolutionary biology is well beyond the scope of the current chapter, one can note that these higher-order assumptions about human biology and psychology are culled not only from Darwin’s (1859) theory of natural and sexual selection, but also from the modern synthesis of population genetics and evolutionary biology (see Barkow, Cosmides, & Tooby, 1992; Dennett, 1995; and Pinker, 1997, for a fuller discussion of the core assumptions of the evolutionary metatheory).

How Do You Test an Evolutionary Theory? If we define a theory as a type of explanation that entails both inductive and deductive reasoning, we can see right away that the deductive component of a theory readily lends itself to corroboration or falsification (Ellis & Kete-

232

BIOLOGICAL THEORIES

laar, 2000; Ketelaar & Ellis, 2000). This is an important realization and one that lies at the heart of the Popperian strategy of falsificationism, a strategy that is far too often the only arrow in the social psychologist’s quiver of philosophy of science strategies (see Gawronski & Bodenhausen, Chapter 1, this volume). Popper (1959) introduced the strategy of falsificationism as a deductive method of theory evaluation that proceeded by subjecting statements deduced from the theory to empirical test. The Lewin-inspired strategy of deriving theorems and correlates from one’s “theory” and then subjecting these derived statements to empirical test (see Singer, 1987) is entirely consistent with Popper’s method of falsification and Cattell’s inductive– hypothetico–deductive strategy (Cattell, 1988). Although the Popperian method of falsification is useful for evaluating the scientific status of specific statements (hypotheses and predictions) derived from middle-level theories (which may explain the appeal of this method to Lewin, Cattell, and many other social psychologists), philosophically minded psychologists have come to see falsificationism as an inefficient strategy for generating knowledge in human psychology and have argued that a Lakatosian philosophy of science provides a more accurate description of theory construction and evaluation in scientific psychology. (For a detailed treatment of the role of the Lakatosian philosophy of science in evolutionary psychology, see Ellis & Ketelaar, 2000; Ketelaar, 2002; Ketelaar & Ellis, 2000; see also Meehl, 1978, 1990; Newell, 1973, 1990, for a more general discussion.) Cognitive scientist Alan Newell (1990, p. 14) astutely observed: “We are not living in the world of Popper (Popper, 1959), as far as I’m concerned, we are living in the world of Lakatos (Lakatos, 1970). Working with theories is not like skeet shooting—where theories are lofted up and bang, they are shot down with a falsification bullet, and that’s the end of the story.” Lakatos’s (1970) philosophy of science emerged as a direct response to Popper’s (1959) emphasis on falsification. Rather than using the Popperian strategy of falsificationism to evaluate metatheories as false or not yet falsified, Lakatos (1970, 1974, 1978) argued that metatheories4 are more properly evaluated as progressive or degenerative based on the performance of the middle-level theories they generate (see Ketelaar & Ellis, 2000; cf. Gawronski & Bodenhausen, Chapter 1, this volume). According to the Lakatosian philosophy of science, the key scientific criteria for evaluating evolutionary psychology’s guiding metatheory is not whether its core assumptions are false or not yet falsified, but rather whether this metatheory leads to fruitful new discoveries, explanations, and avenues of research and how well the metatheory accommodates anomalies (see Ellis & Ketelaar, 2000; Ketelaar, 2002; Ketelaar & Ellis, 2000). A metatheory that uses its middle-level theories to (1) generate novel explanations/predictions and (2) digest apparent anomalies is viewed as a progressive metatheory. By contrast, a metatheory that utilizes its middle-level theories primarily to deal with anomalies and contributes relatively little new knowledge is viewed as a degenerative metatheory.

Evolutionary Theories 233

Are Evolutionary Theories Useful in Social Psychology? From a Lakatosian perspective, for one to conclude that evolutionary theories are useful in social psychology, one would essentially have to provide evidence that an evolutionary metatheory generates middle-level theories that (1) lead to novel insights about social psychological phenomena and (2) can account for anomalies (evidence that appears to run contrary to evolutionary conjectures) regarding social behavior.

Evidence That Evolutionary Theories in Social Psychology Generate Novel Insights One area of social psychology where evolutionary theorizing has generated novel insights is the study of xenophobia, defined as a dislike or fear of strangers or foreign peoples (Kirkpatrick, & Navarrete, 2006; Navarrete & Fessler, 2005; Schaller & Neuberg, 2012; Schaller & Park, 2011). As an example of xenophobia, consider the words of a young Englishwoman traveling through France to Geneva in 1817. Scribbled in her diary, she described the French villagers she encountered as “squalid with dirt, their countenances expressing everything that is disgusting and stupid” (in Mellor, 1989, p. 25). Lest one think that this depiction of ethnic derogation was simply due to this individual having a bad day, consider the following diary entry regarding her impressions of the German people, written just a few days later: Our companions in this voyage are the meanest class, smoked prodigiously, and were exceedingly disgusting. . . . The horrid and slimy faces of our companions in voyage; our only wish was to absolutely annihilate such uncleanly animals. . . . ‘Twere easier for God to make entirely new men than attempt to purify such monsters as these. (in Mellor, 1989, p. 25)

Although the invocation of disgust at the sight of foreigners is a common correlate of xenophobia (Schaller & Park, 2011), one might be surprised to learn that these private thoughts were penned by one of history’s most liberal-minded feminist scholars, Mary Shelley, the celebrated author of the 1818 Gothic novel Frankenstein: The Modern Prometheus.5 Recently, an evolutionary theory of xenophobia—the behavioral immune system approach—has offered novel insights into social prejudice, a topic that has been studied by social psychologists for more than half of a century (e.g., Allport, 1954; Sherif & Sherif, 1953; Tajfel & Turner, 1986). To explain some of the robust, systematic features of xenophobia, evolutionary psychologists have posited a behavioral immune system that evolved to serve as a first line of defense (ahead of the internal physiological immune system) in response to pathogens (Schaller & Neuberg, 2012; Schaller & Park, 2011; but see Hruschka & Heinrich, 2013, for an alternative view). There are at least two reasons why members of outgroups are

234

BIOLOGICAL THEORIES

often perceived to be vectors for increased risk of disease. First, it is well known that contact with outgroup members has historically been associated with increased exposure to exotic pathogens, pathogens that tend to be especially virulent when introduced to the local population (for reviews, see Diamond, 1997; Fincher & Thornhill, 2012; Schaller & Neuberg, 2012; Schaller & Park, 2011). Second, members of outgroups are less likely to be familiar with local norms pertaining to hygiene and food preparation, behavioral norms that “serve as barriers to pathogen transmission” (Schaller & Neuberg, 2012, p. 36). According to evolutionary social psychologists, the behavioral immune system evolved to facilitate avoidance of pathogens through a number of mechanisms including emotion systems that respond specifically to contagion threats (e.g., disgust) and perceptual systems designed to identify and avoid “people who appear especially likely to pose some risk of pathogen transmission” (Schaller & Park, 2011, p. 100). Central to the behavioral immune system view of xenophobia is the claim that perceptions of pathogens are biased toward false positives (erroneously inferring the presence of pathogens when there are none).6 To explain the interesting association between pathogen threat and increased disparagement of outgroup ideologies (e.g., pro-in-group bias), evolutionary psychologists have argued that these built-in biases toward false positives can take the form of a greater wariness and avoidance of individuals whose behavior signals that they are not members of the ingroup. By drawing attention to the assumption that ancestral humans would have recurrently faced the adaptive problem of dealing with especially virulent pathogens when coming into contact with outgroup members, evolutionary theories of xenophobia provide novel insights into the possible origins of the now welldocumented finding that xenophobic reactions are sometimes better conceptualized by the emotion of disgust rather than fear (Cottrell & Neuberg, 2005; Park, Schaller, & Crandall, 2007). By providing empirical evidence to support their assumptions about the conditions under which xenophobia is more accurately conceptualized as a disgust response than a fear response, evolutionary psychologists have provided novel insights into the proximate and ultimate functions of xenophobia (see Cottrell & Neuberg, 2005; Park, Schaller, & Crandall, 2007). But disentangling the emotional aspects of xenophobia is not the only novel insight that evolutionary social psychologists have brought to psychology’s understanding of this form of social prejudice. Evolutionary social psychologists have recently pointed out that another set of circumstances in which individuals are especially vulnerable to infection occurs during the first trimester of pregnancy when a woman’s body is naturally immunosuppressed. Given that the fetus’s susceptibility to pathogens and teratogens is heightened in the first trimester, it is not surprising to observe that women in the first trimester of pregnancy report greater disgust sensitivity and pregnancy sickness (Flaxman & Sherman, 2000; Profet, 1992). Consistent with the idea of a behavioral immune system, evolutionary social psychologists have found that women in their first trimester of pregnancy exhibit significantly

Evolutionary Theories 235

higher levels of xenophobia compared to women in later stages of pregnancy (Navarrete, Fessler, & Eng, 2007). These empirical findings constitute a novel insight into xenophobia not anticipated by over half of a century of research on social prejudice (Navarrete, Fessler, & Eng, 2007). Armed with these evolutionary insights linking weeks in pregnancy with intensity of outgroup negativity (Navarrete, Fessler, & Eng, 2007), one might ask whether Mary Shelley was pregnant at the time that she made her highly prejudicial entries in her diary. The answer to that question can be provided by considering the following fact: Shelley made these diary entries around August of 1814 (Mellor, 1989, p. 25). Seven months later, on February 22, 1815, Shelley gave birth, two months prematurely, to a baby girl fathered by her lover, poet Percy Shelley (see Mellor, 1989, p. 31). Mary Shelley would have been in the first trimester of her pregnancy when she penned these racist comments about the French and Germans.

Evidence That Evolutionary Theories in Social Psychology Can Successfully Digest Anomalies According to Lakatos, when a metatheoretical research program begins to contribute only marginally to the advancement of knowledge because it is employed primarily in dealing with anomalies, we say that this metatheoretical research program is degenerative rather than progressive (see Ellis & Ketelaar, 2002; Lakatos, 1978; see also Gawronski & Bodenhausen, Chapter 1, this volume). A good example of an early anomaly for evolutionary metatheory involved an argument introduced by Lord Kelvin. Kelvin argued that the laws of thermodynamics placed numerical constraints on the length of geologic time available for evolutionary forces to have operated. According to Kelvin, these constraints led to the conclusion that the age of the earth was too young to have enabled evolution by natural selection to have occurred in the manner specified by Darwin. Darwin, however, did not regard Lord Kelvin’s calculations as adequate grounds for rejecting his theory of natural selection. Instead, Darwin regarded these observations as an anomaly, which he (correctly) expected would be resolved by future research. In the early 20th century when the discovery of radiation (an internal source of heat) dramatically increased estimates of the age of the earth, these new estimates enabled Darwin’s theory to digest the apparent anomaly and turn it into positive evidence (see Ellis & Ketelaar, 2002; see also Lakatos, 1978, for a fuller discussion of the role of anomalies in evaluating metatheoretical research programs). Have evolutionary theories of social-psychological phenomena had similar success in digesting anomalies? One area of social psychology where evolutionary theorizing has successfully digested apparent anomalies is the study of interpersonal violence (see Ketelaar & Ellis, 2000). A basic higher-order assumption contained in evolutionary metatheory is that natural selection favors nepotism, the inclination to discriminate in favor of genetic relatives. Given identical levels of physical proximity and social interaction, parents should be much more

236

BIOLOGICAL THEORIES

inhibited against harming or killing their own biological children than against harming or killing stepchildren (Daly & Wilson, 1988). Yet, this core assumption of the evolutionary metatheory was directly called into question by criminologists who observed that the family is the single most common locus of all types of interpersonal violence. This “fact” led some criminologists to develop the mutual access hypothesis, which argued that “it cannot be surprising that more violence is directed against those with whom we are in more intimate contact. We are all within easy striking distance of our friends and spouses, for a goodly part of the time” (Goode, 1969, p. 941). The mutual access hypothesis specified a general set of psychological mechanisms underlying aggression (direct violence toward others who are around you most often and affect you most frequently and directly) that cuts across different types of social relationships. This suggests that children who are within easy striking distance of their parents are at the greatest risk for physical abuse, regardless of whether those children are step-relations or biological offspring. If this supposition were shown to be true (i.e., if the psychological mechanisms underlying family violence followed a general “easy striking distance” rule that applied equally across different genetic relations, and were not nepotistically biased), then such an observation would call into question a basic metatheoretical assumption of modern evolutionary theory. Such an observation would be an example of what Darwin’s bulldog, Thomas Huxley (1893), once referred to as “the great tragedy of Science—the slaying of a beautiful hypothesis by an ugly fact.” If the family was in fact the most frequent single locus of homicide (Gelles, 1979; Straus & Gelles, 1990), this would represent an anomaly for an evolutionary metatheory because—according to a core assumption of the evolutionary metatheory— individuals should be strongly inhibited against terminating their own fitness vehicles. However, evolutionary psychologists (Daly & Wilson, 1988) challenged this putatively anomalous “fact” by reevaluating the definition of family. It turned out that criminologists’ observations of greater familial homicide were based on a sociological definition of the family (a definition that included both genetically related and unrelated cohabitants). When Daly and Wilson examined homicide rates based on a biological definition of family, they found that cohabitants who were genetic relatives of the killer were more than 11 times less likely to be murdered compared to cohabitants who were nongenetic relatives of the killer, and that only 6.3% of all homicides occurred between genetic relatives. These new facts digested the apparent anomaly suggested by the mutual access hypothesis, calling into question the mutual access hypothesis and turning an apparent anomaly into positive evidence for evolutionary metatheory.

Conclusions In the spirit of Kurt Lewin’s graduate course on theories in social psychology, the current chapter focused on the philosophy of science rather than a

Evolutionary Theories 237

review of different evolutionary theories in social psychology. Of primary interest were three Lewin-inspired questions: (1) What is an evolutionary theory?; (2) How do you test an evolutionary theory?; and (3) What makes evolutionary theories useful in social psychology? In reviewing what constitutes an evolutionary theory, I began with the more basic Lewinian question of what a theory is by defining a theory as a type of explanation that is constructed (partly) out of data (e.g., empirical observations) and partly out of metatheoretical assumptions. I argued that what distinguishes a theory from other sorts of explanations is the simple fact that a theory references both empirical data and higher-order assumptions (a metatheory). I pointed out that by contrast with their colleagues in the more mature sciences, it would appear that theoreticians in social psychology tend to restrict their use of deductive reasoning to theory testing (i.e., the so-called inductive–hypothetico–deductive strategy), as when a Lewin-inspired social scientist subjects a particular statement derived from Heider’s (1958) balance theory to an empirical test in a 2 × 2 ANOVA design. By contrast with theorizing in mainstream social psychology, evolutionary psychologists are just as apt to use deductive reasoning in theory construction as in theory testing. Thus, evolutionary theories can be distinguished from nonevolutionary theories in social psychology by virtue of recognizing that evolutionary theories have an explicit deductive connection to the larger evolutionary metatheory consisting of a set of higher-order assumptions located in the adaptationist program in evolutionary biology (see Barkow, Cosmides, & Tooby, 1992; Ketelaar & Ellis, 2000). In addressing how one tests an evolutionary theory, I argued that the distinction between evolutionary theories and the evolutionary metatheory that generates these theories (see Figure 11.1) is important because philosophers of science have identified different criteria for constructing and evaluating these two different levels of explanation in science. I reviewed previous work (see Ketelaar & Ellis, 2000) arguing that a Lakatosian philosophy of science provides a better framework (than the Popperian program of falsification) for understanding how psychologists construct and evaluate the overarching metatheory employed by evolutionary social psychologists. Although middle-level evolutionary theories (such as the idea that an evolved behavioral immune system explains certain aspects of xenophobia in modern environments, see Schaller & Neuberg, 2011) can be evaluated rather directly through a Popperian process of attempting to falsify or corroborate predictions and hypotheses generated by these theories, metatheoretical research programs are not evaluated in terms of their ability to survive attempts at falsification. Instead, metatheories are more properly evaluated through a process of establishing that the metatheoretical research program displays evidence of progressivity rather than degenerativity (see Ketelaar & Ellis, 2000; Lakatos, 1970, 1974, 1978; see also Gawronski & Bodenhausen, Chapter 1, this volume). Simply put, predictions and hypotheses generated by middle-level evolutionary theories are judged by the same sorts of criteria (corroboration and falsification) that are used to evaluate statements derived

238

BIOLOGICAL THEORIES

from nonevolutionary theories such as Heider’s (1958) balance theory or Fishbein and Azjen’s (1975) theory of reasoned action. However, the overarching evolutionary metatheory used to construct middle-level evolutionary theories (such as the behavioral immune system account of xenophobia) is best evaluated using the tools that philosophers of science have deemed appropriate for evaluating grand unifying frameworks such as those found in more mature sciences such as biology, chemistry and physics. Simply put, from a Lakatosian perspective, one doesn’t wish to know whether a specific prediction constructed under the umbrella of the metatheory has been shown to be false. Instead, the Lakatosian wishes to know two things about evolutionary metatheory: (1) whether the middle-level theories constructed from the evolutionary metatheory leads to fruitful new discoveries, explanations, and avenues of research in social psychology and (2) how well this metatheory accommodates anomalies, observations that appear to run contrary to evolutionary explanations of social behavior (see Ellis & Ketelaar, 2000; Ketelaar & Ellis, 2000; Ketelaar, 2002). Finally, in examining whether evolutionary theories are useful in social psychology I reviewed several examples demonstrating that evolutionary metatheory appears to meet all of the requirements of a progressive research program as described by Lakatos (1978). In particular, I reviewed evidence that theories in evolutionary social psychology embody two important criteria of progressivity. First, evolutionary theories of xenophobia, for example, clearly provide novel insights into social behavior, despite the fact that topics such as xenophobia and social prejudice have been studied by social psychologists for over half a century. Second, I demonstrated that evolutionary theories of interpersonal violence have clearly succeeded not only in digesting apparent anomalies, but in turning these putative counterexamples into positive evidence supporting the use of an evolutionary metatheory in social psychology. In sum, evolutionary theories of social psychological phenomena show much promise not only for generating falsifiable predictions, but also in leading us to fruitful new discoveries, novel explanations, and promising avenues of research. Acknowledgments Portions of this chapter are based on Ketelaar and Ellis (2000) and Ellis and Ketelaar (2000). I would like to thank Martha R. Zella and the editors for valuable comments on an early draft of this chapter.

Notes 1. For William James (1892), the search for deductive and generative principles for scientific psychology, coupled with his formal training in physiology, led him to the core belief that mental states should be described in terms of physical causes lying in the central and peripheral nervous systems.

Evolutionary Theories 239 2. Nomothetic explanations are attempts to describe general scientific laws. Nomothetic explanations can be contrasted with idiographic explanations, explanations that are attempts to describe individual cases or events. 3. Virtus dormativa explanations are a form of tautology in which a phenomenon is explained in terms of the phenomenon itself, stated in somewhat different (often more technical) words (see Gigerenzer, 1998). An example would be the real world case of the death of a Hollywood actress in the 1990s that was reported in the press as being due to “sudden unexplained death in epilepsy (SUDEP)” which, despite the acronym, is just another way of stating the obvious fact that the actress had epilepsy and died suddenly for unknown reasons. 4. Lakatos (1970, 1974, 1978) did not actually use the term metatheory in his writing. Rather, Lakatos used the term hard core to refer to what we are calling the metatheoretical assumptions level of analysis depicted in Figure 11.1. 5. Shelley was the daughter of Mary Wollstonecraft, author of the (1792) liberal treatise A Vindication of the Rights of Woman, and her mother was described as “the most ardent advocate of her times for the education and development of female capacities” (Mellor, 1989, p. 1). 6. Proponents of “error management theory” (Haselton & Buss, 2000; Haselton & Nettle, 2006; Nesse, 2005) argue that perceptual and cognitive biases of this sort are not errors of reasoning, but rather adaptive biases because they ensure that the organism makes the less costly survival or reproductive error when confronted with a trade-off between making one of two types of incorrect inferences (false alarms vs. misses; see Schaller & Park, 2011).

References Allport, G. W. (1954). The nature of prejudice. Cambridge, MA: Addison-Wesley. Barkow, J. H., Cosmides, L., & Tooby, J. (Eds.). (1992). The adapted mind: Evolutionary psychology and the generation of culture. New York: Oxford University Press. Becker, E. (1973). The denial of death. New York: Free Press. Buss, D. M. (1990). Evolutionary social psychology: Prospects and pitfalls. Motivation and Emotion, 14, 265–286. Buss, D. M. (1995). Evolutionary psychology: A new paradigm for psychological science. Psychological Inquiry, 6, 1–30. Cattell, R. B. (1988). Psychological theory and scientific methodology. In R. B. Cattell & J. R. Nesselroade (Eds.), Handbook of multivariate experimental psychology (2nd ed.). New York: Plenum Press. Cottrell, C. A., & Neuberg, S. L. (2005). Different emotional reactions to different groups: A sociofunctional threat-based approach to “prejudice.” Journal of Personality and Social Psychology, 88, 770–789. Daly, M., & Wilson, M. (1988). Homicide. New York: deGruyter. Darwin, C. (1859). On the origin of species. London: Murray. Dennett, D. C. (1995). Darwin’s dangerous idea: Evolution and the meanings of life. New York: Simon & Schuster. Diamond, J. (1997). Guns, germs, and steel: The fates of human societies. New York: Norton. Ellis, B. J., & Ketelaar, T. (2000). On the natural selection of alternative models: Evaluation of explanations in evolutionary psychology. Psychological Inquiry, 11, 56–68. Fincher, C. L., & Thornhill, R. (2012). Parasite-stress promotes in-group assortative sociality: The cases of strong family ties and heightened religiosity. Behavioral and Brain Sciences, 35, 61–79.

240

BIOLOGICAL THEORIES

Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior: An introduction to theory and research. Reading, MA: Addison-Wesley. Flaxman, S. M., & Sherman, P. V. (2000). Morning sickness: Adaptive cause or nonadaptive consequence of embryo viability? American Naturalist, 172, 54–62. Gelles, R. J. (Ed.). (1979). Family violence. Beverly Hills, CA: Sage. Gigerenzer, G. (1998). Surrogates for theories. Theory and Psychology, 8(2), 195–204. Goode, W. (1969). Violence among intimates. In D. J. Mulvihill & M. M. Tumin (Eds.), Crimes of violence: Report to the National Commission on the causes and prevention of violence (Vol. 13, pp. 941–977). Washington, DC: U.S. Government Printing Office. Haselton, M. G., & Buss, D. M. (2000). Error management theory: A new perspective on biases in cross-sex mind reading. Journal of Personality and Social Psychology, 78, 81–91. Haselton, M. G., & Nettle, D. (2006). The paranoid optimist: An integrative evolutionary model of cognitive biases. Personality and Social Psychology Review, 10, 47–66. Heider, F. (1958). The psychology of interpersonal relations. New York: Wiley. Higgins, E. T. (2004). Making a theory useful: Lessons handed down. Personality and Social Psychology Review, 8, 138–145. Hruschka, D. J., & Heinrich, J. (2013). Institutions, parasites and the persistence of ingroup preferences. PLoS One, 8, E6342. Huxley, T. H. (1893). Collected essays; Vol. 8. Discourses biological and geological. London: Macmillan. James, W. (1892). A plea for psychology as a natural science. Philosophical Review, 1, 146– 153. James, W. (2009). Principles of psychology (Vol. 2). Scotts Valley, CA: IAP. (Original work published 1890) Ketelaar, T. (2002). The evaluation of competing approaches within human evolutionary psychology. In S. J. Scherer & F. Rauscher (Eds.), Evolutionary psychology: Alternative approaches. Norwell, MA: Kluwer Press. Ketelaar, T., & Ellis, B. J. (2000). Are evolutionary explanations unfalsifiable? Evolutionary psychology and the Lakatosian philosophy of science. Psychological Inquiry, 11, 1–21. Kirkpatrick, L. A., & Navarrete, C. D. (2006). Reports of my death anxiety have been greatly exaggerated: A critique of terror management theory from an evolutionary perspective. Psychological Inquiry, 17, 288–298. Kruglanski, A. W. (2001). That “vision thing”: The state of theory in social and personality psychology at the edge of the new millennium. Journal of Personality and Social Psychology, 80, 871–875. Kuhn, T. S. (1962). The structure of scientific revolutions. Chicago: University of Chicago Press. Lakatos, I. (1970). Falsificationism and the methodology of scientific research programmes. In I. Lakatos & A. Musgrave (Eds.), Criticism and the growth of knowledge (pp. 91–196). Cambridge, UK: Cambridge University Press. Lakatos, I. (1974). Popper on demarcation and induction. In P. A. Schilpp (Ed.), The philosophy of Karl Popper (pp. 241–273). La Salle, IL: Open Court. Lakatos, I. (1978). The methodology of scientific research programmes: Philosophical papers (Vol. 1). Cambridge, UK: Cambridge University Press. Lewin, K. (1935). A dynamic theory of personality: Selected papers by Kurt Lewin. New York: McGraw-Hill. Lewin, K. (1951). Field theory in social science: Selected theoretical papers. New York: Harper & Row. Meehl, P. (1978). Theoretical risks and tabular asterisks: Sir Karl, Sir Ronald, and the slow progress of soft psychology. Journal of Consulting and Clinical Psychology, 46, 806–834.

Evolutionary Theories 241 Meehl, P. E. (1990). Appraising and amending theories: The strategy of Lakatosian defense and two principles that warrant it. Psychological Inquiry, 1, 108–141. Mellor, A. K. (1989). Mary Shelley: Her life, her fiction, her monsters. New York: Routledge. Navarrete, C. D., & Fessler, D. M. T. (2005). Normative bias and adaptive challenges: A relational approach to coalitional psychology and a critique of terror management theory. Evolutionary Psychology, 3, 297–325 Navarrete, C. D., Fessler, D. M. T., & Eng, S. J. (2007). Elevated ethnocentrism in the first trimester of pregnancy. Evolution and Human Behavior, 28, 60–65. Nesse, R. M. (2005). Natural selection and the regulation of defenses: A signal detection analysis of the smoke detector principle. Evolution and Human Behavior, 26, 88–105. Newell, A. (1973). You can’t play 20 questions with nature and win: Projective comments on the papers in this symposium. In W. G. Chase (Ed.), Visual information processing (pp. 283–308). New York: Academic. Newell, A. (1990). Unified theories of cognition. Cambridge, MA: Harvard University Press. Park, J. H., Schaller, M., & Crandall, C. S. (2007). Pathogen-avoidance mechanisms and the stigmatization of obese people. Evolution and Human Behavior, 28, 410–414. Pinker, S. (1997). How the mind works. New York: Norton. Popper, K. R. (1959). The logic of scientific discovery. New York: Hutchison Education. Profet, M. (1992). Pregnancy sickness as adaptation: A deterrent to maternal ingestion of teratogens. In J. H. Barkow, L. Cosmides, & J. Tooby (Eds.), The adapted mind: Evolutionary psychology and the generation of culture. New York: Oxford University Press. Schaller, M., & Conway, L. G. (2000). The illusion of unfalsifiability and why it matters. Psychological Inquiry, 11, 49–52. Schaller, M., & Neuberg, S. L. (2012). Danger, disease, and the nature of prejudice(s), Advances in Experimental Social Psychology, 46, 1–54. Schaller, M., & Park, J. H. (2011). The behavioral immune system (and why it matters). Current Directions in Psychological Science, 20, 99–103. Schaller, M., Simpson, J. A., & Kenrick, D. T. (2006). Evolution and social psychology. New York: Psychology Press. Sherif, M., & Sherif, C. W. (1953). Groups in harmony and tension. New York: Harper & Row. Shelley, M. (1996). Frankenstein. New York: Norton. (Original work published 1818) Simpson, J. A., & Kenrick, D. T. (Eds.). (1997). Evolutionary social psychology. Mahwah, NJ: Erlbaum. Singer, J. E. (1987). The role of the mentor. In N. E. Grunberg, R. E. Nisbett, J. Rodin, & J. E. Singer (Eds.), A distinctive approach to psychological research: The influence of Stanley Schacter. Hillsdale, NJ: Erlbaum. Straus, M. A., & Gelles, R. J. (Eds.). (1990). Physical violence in American families. New Brunswick, NJ: Transaction. Tajfel, H., & Turner, J. C. (1986). The social identity theory of intergroup conflict. In S. Worchel & W. G. Austin (Eds.), Psychology of intergroup relations (pp. 7–24). Chicago: Nelson-Hall. Wollstonecraft, M. (2009). A vindication of the rights of woman (3rd ed.) (D. S. Lynch, Ed.). New York: Norton. (Original work published 1792)



Part IV Pragmatic Theories





12 Rational Actor Theories David Trafimow

I

magine the ideal case of a finite number of behaviors that could be performed in a particular situation, and the values (vi) of all possible positive and negative consequences of each behavior are known. In addition, the probability of occurrence (pi) is known for each possible consequence of each behavior. Finally, there is sufficient time and processing capacity available to compute the expected utility of each possible behavior. In that case, the rational way to maximize expected value would be to multiply the value of each consequence by its probability and sum over all products for each behavior: Expected value = ∑ni=1vi pi The behavior with the largest sum of products is the one that maximizes expected utility and therefore is the “rational” choice. It is instructive to keep this ideal case in mind when considering “rational actor” theories. According to such theories, how closely do people approach the ideal goal of maximizing expected utility given that people do not have unlimited processing capacity and that information about all relevant consequences and probabilities generally is not known? There are several ways to conceptualize rational thinking, given limitations on knowledge or cognitive capacity. One way is to contrast behaving on impulse versus undergoing some kind of reasoned process, involv 245

246 PRAGMATIC THEORIES

ing logically consistent thinking. In this case, the argument might be that, although the reasoned process doubtless would have flaws, it is nevertheless rational given existing limitations. Another way is to contrast behaving in a way that maximizes the likelihood of positive outcomes versus behaving in accordance with what a logician, mathematician, or philosopher might consider to be proper reasoning. In this case, it might be considered to be rational to commit fallacies in logic if committing those fallacies increases the probability of adaptive behavior. Consistent with this view of rationality, philosophers have noted that many fallacies in logic, though irrational from a deductive logic perspective, make sense when probability is taken into account. A third way to think about rationality is to contrast different goals people might have, keeping in mind that what seems irrational from the perspective of one goal might be much more rational from the perspective of another goal. Although logic can prescribe courses of action given a particular goal, logic cannot prescribe goals (Hollis, 1996). As an example, suppose a person has to choose between performing morally or in a way that maximizes financial gain; which choice is rational? A related issue pertains to the rationality of performing behaviors in order to increase knowledge. With increased knowledge comes the possibility of being able to make better choices, but gaining the knowledge might be costly, as Eve found out to her dismay as described in the book of Genesis. Arguably, there is no rational way to decide whether or not to attempt to increase one’s knowledge because there is no reasonable way to assess potential costs and benefits before that knowledge is obtained (Hollis, 1996). Given that there are many ways to think about rationality, it is not surprising that not all rational actor theories in psychology are the same. Some of them assume that people are poor scientists. Kelley (1973, p. 109) made a particularly famous statement of this perspective: The assumption is that the man in the street, the naive psychologist, uses a naïve version of the method used in science. Undoubtedly, his naive version is a poor replica of the scientific one—incomplete, subject to bias, ready to proceed on incomplete evidence, and so on. Nevertheless, it has certain general properties in common with the analysis of variance as we behavioral scientists use it.

Kelley’s dimensions, which he assumed that people used, refer to people, entities, and time. A complete description is beyond the scope of this chapter, but an example should provide the “flavor.” Suppose an observer sees a dog (entity) bite Joe (person). In addition, the observer knows that other dogs have not bitten Joe but that this particular dog also has bitten other people. In this case, it would be rational to assign causation to the dog. In contrast, if the dog has not bitten other people, whereas other dogs have bitten Joe, it would be more rational to assign causation to Joe. If neither Joe nor the dog

Rational Actor Theories 247

had previously been involved in previous biting cases, it would be rational to look for something special about the particular interaction to assign causation. Weiner et al. (1971) also proposed a rational theory of attribution where he assumed that causes vary on a stable–unstable and an internal–external dimension. If a person succeeds at a task, accompanied by an opinion of stability and internality on the part of the observer, that observer should attribute the success to the person’s ability. In contrast, the combination of stability and externality should induce causation to be attributed to the task being easy. The combination of instability and internality should result in the observer attributing causation to effort, whereas the combination of instability and externality should result in causation being attributed to luck. Weiner (2010) described how this simple rational theory was expanded in later articles by the addition of more dimensions, motivational factors, and personality variables. In contrast to Kelley and Weiner, Goldstein and Gigerenzer (Gigerenzer & Goldstein, 1996; Goldstein & Gigerenzer, 2002) proposed a very different rational actor theory. They argued that people have limited resources, processing time, and ability, and also tend to satisfice rather than find the absolute best option. Thus, rather than being poor scientists, people make decisions in a way shaped by evolution. Very rarely, in our evolutionary history, did people have complete information on which to base decisions. Consequently, people had to use decision-making strategies that used lack of information as well as the presence of information. The consequences of our evolutionary history are still with us today, as can be demonstrated by a simple example. Suppose a person is asked which of two cities has a larger population, and the person has heard of one city but not of the other. In that case, the person might use the lack of knowledge about one of the cities to infer that it likely is smaller than the city that the person recognizes. This would be a “rational” decision under the condition of limited knowledge. The argument is not that reasoning from lack of information always results in correct decisions, but rather, from a probabilistic perspective, such reasoning is beneficial more often than not. Possibly the most prominent of the rational actor theories, particularly with respect to the frequency with which it has been applied to address realworld problems, is the theory of reasoned action (TRA; Ajzen & Fishbein, 1980; Fishbein, 1980; Fishbein & Ajzen, 1975), and its descendant, the theory of planned behavior (TPB; Ajzen, 1988, 1991). Together, these theories exemplify the “reasoned action” approach to how people make decisions about how to behave (see Fishbein & Ajzen, 2010, for a review). Like the other rational actor theories, this approach does not assume that people always behave in an optimal manner, but it does assume that given what may or may not be good assumptions, people reason in a rational way to decide what to do. If the assumptions are poor, resulting behaviors might be suboptimal but are still “reasoned.”

248 PRAGMATIC THEORIES

The easiest way to understand the TRA is to work backwards, from the behavior to be explained to the constructs that are used to do so. According to the TRA, the proximal determinant of behavior is one’s behavioral intention, which, as its name suggests, is a person’s intention or motivation to perform the behavior. Thus, it then becomes necessary to ask, “What determines behavioral intention?” As will become clear presently, there is an “attitudinal” and “normative” pathway to behavioral intention (and then to behavior). The two proximal determinants of behavioral intention are attitude and subjective norm. Attitude is an evaluation of the behavior (how much does the actor like or dislike it?), whereas subjective norm refers to the actor’s opinion about whether most others who are important to her think she should perform it. The latter is a “subjective” norm because the issue is not what most important others actually think but rather what the actor thinks they think. Given that attitudes and subjective norms determine behavioral intentions, the next issue pertains to the constructs that determine attitudes and subjective norms, respectively. Attitude is determined by behavioral beliefs, which are beliefs about the likelihood of the positive and negative consequences that might arise from performing the behavior, and evaluations of how good or bad it would be if those consequences actually occurred. Subjective norms are determined by normative beliefs, which are beliefs about what specific important others think one should do, and motivations to comply, which refer to the extent to which the actor wishes to comply with each important other. There is a mathematical component to the TRA that makes it the closest of rational actor theories to the idealized case of maximizing expected utility that was described in the first paragraph of this chapter. Based on an assumption that there is a set of salient beliefs, the most famous TRA equation is Equation 12.1 below, indicating that attitude (A) equals the sum of behavioral belief (bi) and evaluation (ei) products. Note the similarity to the ideal notion of expected value described earlier. A = ∑ni=1bi ei

(12.1)

A similar equation, Equation 12.2 below, describes the formation of subjective norms. Specifically, subjective norm (SN) equals the sum of normative belief (ni) and motivation to comply (mi) products. SN = ∑ni=1ni mi

(12.2)

In summary, there is an attitudinal pathway to behavioral intention: ∑ni=1bi ei → A → BI. And there is a normative pathway: ∑ni=1ni mi → SN → BI. For any particular behavior, the attitudinal or normative pathway might be more important. Put more statistically, for any particular behavior, the atti-

Rational Actor Theories 249

tude might be more highly weighted than the subjective norm, or the reverse might be true. When the attitude is more highly weighted than the subjective norm, the behavior can be said to be under attitudinal control, whereas it can be said to be under normative control when the reverse is true. Finally, Ajzen (e.g., 1988) added a third pathway that features the notion of perceived behavioral control, which is an actor’s perception that the behavior is under his control. When Ajzen added this third pathway to the TRA, he renamed it the theory of planned behavior (TPB). However, the TRA is sufficient to bring out most of the philosophy of science issues of interest, and so we will focus on that.

Modeling, Description, and Explanation Like most rational actor theories, the TRA can be interpreted as describing how people ought to decide what to do; that is, if they were rational. An alternative, asserted by the TRA’s originators, is that the TRA describes what people actually do (so people are assumed to be “rational” or at least to make reasoned decisions). However, this last statement can be deceptive. Even Fishbein did not argue that people undergo complex mathematical calculations (e.g., as in Equation 12.1 and Equation 12.2) before deciding what to do. Rather, the argument is that people undergo a reasoned process when deciding what to do that can be modeled by Equations 12.1 and 12.2. To understand this idea, consider the analogy of a Major League outfielder judging where a fly ball will land, and running to the spot just in time to make a spectacular catch. It is unlikely in the extreme that the outfielder performed any of the complex calculations that describe ballistic phenomena. Nevertheless, an expert in these matters could derive mathematical equations that would model the outfielder’s behavior. In addition, these mathematical equations, or at least generalized versions of them, could be used to predict the behavior of the outfielder, with respect to future fly balls. In this sense, the argument would be that although the equations do not describe precisely the thoughts that go through the heads of outfielders, the behaviors outfielders perform conform to what the equations say. Suppose an outfielder failed to behave in accordance with the equations and consequently failed to catch the ball. Would we blame the outfielder or the equations? From the point of view of predicting behavior, the equations clearly failed. But from the point of view of saying what the outfielder ought to have done, the equations are successful and so the failure is due to the outfielder not conforming to the equations. This is a reoccurring theme with rational actor theories. When experimental findings are not in accord with the rational actor theory at hand, researchers often have changed the assertion to say that the theory is about what people ought to do, as opposed to what they really do. The TRA could be argued to be an exception because Fishbein and Ajzen

250 PRAGMATIC THEORIES

repeatedly claimed that the TRA is an explanation of actual (as opposed to ideal) human behavior. To some extent it is possible to support the explanation claim. After all, if behavior is determined by other constructs in the theory, doesn’t this mean that some variables cause behaviors, which constitutes an explanation of behavior? Let us attempt to consider the issues of explanation and description more precisely. According to the TRA, ∑ni=1bi ei causes attitude. And yet, as I pointed out earlier, it is unlikely that people actually calculate products and then sum them. Or consider Kelley’s theory, that people make attributions according to a poorly done three-dimensional analysis of variance. And yet it seems unlikely that people would perform such calculations. Is it possible for theories such as these to be true, even if people are unable to compute products or sums of products in their heads? Well, that depends. Suppose that experimental manipulations are performed to influence ∑ni=1bi ei in a TRA experiment, and suppose that attitudes really are influenced. Or suppose that information pertaining to Kelley’s dimensions is manipulated, and attributions really are influenced. By pretending the ideal case where there are no confounds, it would be clear that the putative causes really work. This does not mean that the causes work because people perform complex mathematical computations in their heads. It might just mean that people undergo some mental process that leads to results that would have been obtained if people really did perform complex mathematical computations. In this case, it would be possible to argue that rational actor theories explain behaviors up to a point, but that it is still unknown what the mental process is that seems so analogous to the performance of complex mathematical computations. If rational actor theories can be said to explain, in part, what people actually do, does this mean that they describe what people actually do? That depends on what we mean by “explain” and “describe.” At present, there is no philosophical consensus, though much effort has been devoted to these topics. The philosopher Nancy Cartwright (1983, 1999) has argued that no scientific theory actually describes the universe because the universe is too complicated. At best, theories might be said to describe what happens in idealized universes that are simpler than the real universe. It is possible to use Cartwright’s notion to connect explanation and description; an explanation can be argued to be a description of what happens in a nonreal (idealized) universe. However, many scientists believe that their theories actually do describe what happens in the real universe. Certainly, Fishbein (personal communication) would not have been content with having provided a description of what people do in an idealized universe. Complicated issues arise such as whether the usefulness of a theory depends on how close the idealized universe is to the real universe, whether we should care about idealized universes at all, and so on. There are no easy answers, and readers should reach their own judgments. I urge the reader to consider an interesting aspect of rational actor theories, which is that the obvious unlikelihood

Rational Actor Theories 251

that people perform complex mathematical computations in their heads is a possible reason why the topics of explanation and description seem to come up more often in connection with rational actor theories than in connection with other kinds of theories.

Definitions and Falsification The previous section was only a warm-up as the most important philosophical charge leveled against rational actor theories, especially the TRA, is that they are not falsifiable, and consequently not scientific (see Trafimow, 2009, for a review and refutation). Critics of rational actor theories (especially Greve, 2001) have used the TRA as an example of a definitional problem that plagues rational actor theories in general. Greve argued that the common definition of an action is something that is caused by an intention (involuntary reflexes, accidents, and so on do not count as “actions”). Therefore, it is tautological that when any action is observed, the cause was an intention. Because it is true, by definition, that intentions cause actions, it is a mistake to say that intentions explain actions. Furthermore, it is impossible to falsify a statement that is true by definition, which means that rational actor theories cannot be falsified. Although Greve tied the issues of definitions and falsification together, it is possible to separate them (e.g., Trafimow, 2013), and I will do so here. Let us consider the definitional issue first, in the context of Newton’s famous equation that force equals the product of mass and acceleration (f = ma). Both physicists and philosophers have noted that Newton never defined mass. Is this a problem for Newton? Suppose Newton had defined mass, how could he have done it? The most obvious way would have been for Newton to define mass in terms of another word. In that case, a critic could inquire about the definition of that second word. Newton could define the second word with a third word, but the critic would then inquire about the definition of the third word, and so on. A second solution would be for Newton to state that mass is defined as force divided by acceleration. But this would be circular, just as Greve argued the intention–action connection to be. Either way, then, Newton would have a problem. Either he would have an infinite series of definitions, or he would have circularity. Newton did not get into either of these problems because he treated mass as a primitive concept that was not defined (Trafimow & Rice, 2009). Returning to the TRA, it is worth noting that although Greve defined an action as something that is caused by an intention, Fishbein and Ajzen did not. Rather, they did not define action or behavior at all, though they did, of course, state that intentions cause behaviors. Because Fishbein and Ajzen did not define behavior, it is problematic to accuse them of perpetrating a tautology. Arguably, this defense actually places the TRA in a worse light because it now opens up the criticism that Fishbein and Ajzen did not define

252 PRAGMATIC THEORIES

an important term. However, this argument can be refuted by shouting “Newton!”—the idea being that definitions are always problematic because of the three unpalatable choices: infinity, circularity, or primitiveness. And if this issue is a problem for any theory, including Newton’s, there is little reason to single out rational actor theories, such as the TRA, for criticism on definitional grounds. There is an additional definitional point. Consider again that there always must be infinity, circularity, or primitiveness. As choosing infinity requires an infinite number of hours, pages, and so on, practicality forces the choice to really be between circularity and primitiveness. This choice is perhaps more a matter of form than of substance. Suppose that, rather than not defining mass, Newton had formally stated that mass is defined as force divided by acceleration (the circularity option), rather than that this is true because of algebraic manipulation of primitive concepts. Would it really have made any difference? Would physicists then have applied Greve’s argument to criticize Newton? I doubt it strongly! The way physicists thought about mass was dictated much more by how it was used in Newton’s laws, and in the experiments that were conducted, than by any definitional issues. And when Einstein used mass in a different way in his special theory of relativity, physicists set the theories against each other through empirical tests, and so physicists subsequently thought about mass in the context of the winning theory (Einstein’s). To see this concretely, recall that according to Newton, mass does not change as velocity changes but according to Einstein, mass increases as velocity increases. These differences are not spelled out by definitions but are implications of the whole theoretical contexts. Thus, physicists tend to argue about theories rather than about definitions. Psychologists might consider emulating physicists in this respect. Rather than arguing for or against rational actor theories on the basis of definitional issues, it would be more profitable to do so in terms of theory tests, which brings us to the falsifiability issue. The charge that rational actor theories are unfalsifiable is based on applying the writings of Sir Karl Popper (e.g., Popper, 1959, 1963, 1972; for a discussion, see Gawronski & Bodenhausen, Chapter 1, this volume), who argued that whether or not a theory is testable demarcates “scientific” from “nonscientific” theories. To be testable, the theory has to make predictions that could be wrong, the demonstrated wrongness of which would show the theory to be false. Theories that cannot be made to commit to predictions that could be wrong are not testable, not falsifiable, and not scientific. An oft-used example of an unfalsifiable theory is the creationist notion that God created the world. This theory seems to explain everything, which means that it could account for any possible finding, and so there is no way to falsify it, even in principle. Because the definition of words such as “rational,” “reasoned,” and so on are not clear, it is possible to argue that any finding that is obtained can be interpreted to be consistent with some conception of these words. If this is the case, so the argument might go, then it follows that rational actor theories are not falsifiable.

Rational Actor Theories 253

But there is a complication that was stated particularly clearly by Lakatos (1978). Lakatos commenced by pointing out the importance of auxiliary assumptions in science, which are assumptions that are not in the theory of concern but nevertheless are crucial in deriving theory-based predictions. As an example, consider that Newton made predictions about the future position of Mars, from his notions about motion and universal gravitation. However, Newton’s theorizing is about general principles and not about Mars. Clearly, to make specific predictions about Mars from a general theory, it is necessary to make assumptions that pertain specifically to Mars, such as what its current position is. Such an assumption is an auxiliary assumption, and it is not included in Newton’s general theorizing about motion or gravitation. The upshot is that predictions about future observations come from a combination of theory with auxiliary assumptions, and not from theory alone. Consequently, if the predicted observation does not work out, it could be that the theory or at least one auxiliary assumption is to blame. Well, then, if a failed observation can be blamed on the theory or on at least one auxiliary assumption, it follows that absolute theory disconfirmation is not possible. Rather, it is necessary to settle for some form of “reasonable” falsification (e.g., Trafimow, 2009). The necessity of auxiliary assumptions for predicting observations implies an important problem for Popper’s falsification program that psychologists have failed to see, in addition to the problem described above that absolute falsification is impossible. Specifically, if observation predictions come from a combination of theory and auxiliary assumptions, Popper’s demarcation between scientific theories that are falsifiable (in principle) and nonscientific theories that are not falsifiable cannot be maintained. To see why, consider how a critic would demonstrate that a theory is not falsifiable. The critic might combine the theory with a set of auxiliary assumptions and show that no testable prediction arises. But an aficionado of the theory could counter that perhaps a testable prediction would arise if the theory were combined with a different set of auxiliary assumptions. Even if the critic tried a hundred sets of auxiliary assumptions, an aficionado could counter that the critic simply has not yet found a sufficiently good set. Worse yet, it is easy to use the history of physics to find many examples of seemingly unfalsifiable ideas that eventually were falsified when sufficiently creative or advanced auxiliary assumptions became available. (This is one reason why philosophers talk about falsifiability “in principle” rather than falsifiability right now.) For example, it was once thought that notions about the chemical composition of the stars were unfalsifiable because there was no way to travel to a star to obtain a test tube of star material for chemical analysis. However, spectrographic analyses (determining chemical composition from light emitted) became available in the latter part of the 19th century, thereby rendering such tests easy to perform. Even the idea that God created the world is testable in principle. To see this, imagine that a person invents a new kind of prayer to induce God to

254 PRAGMATIC THEORIES

appear and answer questions truthfully. The prayer is used, God is asked about the creation of the world, and God declines to take responsibility, saying: “If I had created the world, I would have done a better job!” This would provide strong negative evidence against the theory that “God did it.” You might consider the invention of such a prayer to be unlikely, but that does not matter; in principle, such a prayer could be invented and the creation theory falsified. In general, then, it is very difficult to argue convincingly that any particular theory is unfalsifiable, in principle. With the philosophical underbrush cleared, let us consider again rational actor theories such as the TRA. Can these theories be made to make falsifiable predictions? In fact, the answer is “yes,” as long as one does not insist on absolute falsification that is impossible anyhow. Consider, for example, that Fishbein (1980) has argued against an affective route to behavioral intention. According to Fishbein, only the attitudinal route and the normative route are possible, and both are cognitive. And yet, Trafimow and Sheeran (1998) have shown that there is an affective route, thereby falsifying an important aspect of the TRA. Clearly, the fact that falsification already has occurred provides a strong argument that the TRA is reasonably falsifiable. Trafimow (2007, 2009) provides additional predictions of observations that demonstrate that the TRA is both falsifiable and, in some instances, falsified. What about other rational actor theories, can they be falsified? In fact, many already have been falsified. I mentioned earlier that experimental observations often are discordant with predictions made by rational actor theories in the attribution area (e.g., Kelley, 1973; Weiner et al., 1971) and that the reaction has been to change the claim. Rather than saying that these theories explain or describe what people actually do, they describe what people ought to do. Whether or not one agrees with the changed claim, it is obvious that the changed claim is an admission that the theories have been falsified, at least with respect to their original purpose of explaining or describing what people actually do. And if the theories have been falsified (according to reasonable rather than absolute falsification), then it follows that they must have been falsifiable. The foregoing discussion assumes that falsifiability matters, at least at the level of reasonable falsification. However, it also is possible to take a completely different tack and argue that whether the implications of a theory are surprising can be separated from whether that theory is falsifiable. To place this issue in Greve’s context, suppose Fishbein had defined a behavior as an action caused by an intention. Would that make Greve’s criticism that intentions cannot explain behaviors, because the causal connection is tautologically true, more valid? I would argue that this is not so. Trafimow (2013) pointed out that there are theories (especially measurement theories) that are based on definitions rather than on assumptions subject to empirical testing, and so they are tautologically true. For example, several authorities have noted that because classical true score theory conclusions

Rational Actor Theories 255

depend on the definitions of “true score” and “parallel tests,” and not on any assumptions that are subject to empirical testing, the theory is tautological, and thus true by definition. In this case, falsification is out of the question. And yet the contribution of measurement theories, such as classical true score theory, to psychology is undeniable. According to Trafimow, the error is in tying together whether the implications of a theory are surprising and whether the theory is falsifiable. Although most philosophers and psychologists have assumed that surprising implications imply more falsifiability, Trafimow pointed out that classical true score theory has surprising implications, explains why these implications are so, and yet is not falsifiable because of its tautological nature. Arguably, then, not only was Greve wrong in his falsification arguments, as described earlier, but he also makes an argument that would be flawed even if he were not wrong. Even if rational actor theories were fully guilty of the charge of being tautological (and therefore unfalsifiable), it does not follow that they are incapable of suggesting surprising implications, of not explaining anything, and of not being of value to psychology.

Unit Coherence There is an additional issue to consider in connection with rational actor theories, such as the TRA. Philosophers agree that good theories are internally coherent—elements of a good theory are not in contradiction. Because rational actor theories, such as the TRA, tend to be stated, at least in part, as mathematical equations, it might seem that internal coherence would be the last issue with which we need to be concerned. After all, mathematics would seem to be nothing if not logical! Contrary to what common sense might suggest, there is a certain type of internal coherence issue to which rational actor theories are particularly vulnerable (Trafimow, 2012a). To understand the issue, it is useful to take a short detour into Newtonian physics and, in particular, Newton’s famous equation relating force (f) to mass (m) and acceleration (a): f = ma. An oldfashioned unit of force is a dyne, which is defined as the amount of force needed to accelerate a mass of 1 gram by 1 centimeter per second squared: 1 dyne = 1[.(g × cm)/sec2)]. Consequently, if we just use units, we have Equation 12.3 below. × cm Force in dynes = gsec 2

(12.3)

The important thing to note about Equation 12.3, for present purposes, is that the units match on both sides of the equation. Suppose the units did not match. For example, suppose we were given Equation 12.4. Force in dynes = g × cm

(12.4)

256 PRAGMATIC THEORIES

Given Equations 3 and 4, we know immediately that something is wrong because combining them in Equation 5 results in a clear contradiction (unless there is exactly 1 second). g × cm sec2

≠ g × cm

(12.5)

What does the issue of units have to do with rational actor theories such as the TRA? To find out, consider again Equation 12.1, but keep units in mind. What are the units? Well, then, remembering that behavioral beliefs are beliefs about the likelihood of consequences, let’s use an abbreviation, bc. And remembering that there also are evaluations of how good or bad the consequences would be if they occurred, let’s use the abbreviation es. In units, then, Equation 12.1 could be rewritten as Equation 12.6 below. A = ∑(bci)(eci)

(12.6)

In turn, this implies that attitude units can be expressed as (bc)(ec). So what’s wrong with that? Let us recall that according to the TRA, attitudes are defined verbally as evaluations of behaviors. It is obvious that evaluations of behaviors and (bc)(ec) products are not the same thing. Thus, either the verbal definition of attitude is wrong or Equation 12.1 is wrong. Either way, the inconsistency renders the TRA as internally incoherent! That it is possible to demonstrate that the TRA is internally incoherent, without any data whatsoever, and by logical reasoning, is consistent with the overall theme of this book. There is a tendency in graduate education in psychology for professors and graduate students to get caught up in experimental details and to fail to realize that much can be accomplished by taking a philosophical approach and employing plain reason. Several decades of research have been performed to test the theory of reasoned action, but we see here that a simple unit analysis renders much of the research unnecessary. The TRA is internally incoherent, and that’s that! Worse yet, Trafimow (2012) showed that it is easy to perform similar unit analyses on additional TRA equations and find similar examples of incoherence. Trafimow has suggested that the problem can be remedied by adding constants with the necessary units to render the TRA to be unit coherent. Of course, such additions change the meanings of the TRA equations, and these would be subject to empirical testing. But what about other rational actor theories, are they vulnerable to unit coherence criticisms? In fact, many of them are not. For example, theories by Kelley (1973), Weiner, (2010; Weiner et al., 1971), and others are not stated in mathematical form. This can be both an advantage and a disadvantage. The advantage is obvious. Because there are no equations, a unit analysis such as the foregoing one cannot be used to demonstrate a lack of unit coherence. Consequently, nonmathematical rational actor theories are not vulnerable to criticisms on unit coherence grounds. The disadvantage, however, is that the

Rational Actor Theories 257

lack of mathematical expression renders these theories to be more vague; we might say that rather than being unit incoherent, they are unit a-coherent. In general, the less clear the theory, the greater the difficulty in finding its flaws. Many philosophers of science would prefer theories with clear flaws to theories that are unclear. Theories with clear flaws can be improved or replaced with better theories, thereby facilitating progress. In contrast, to the extent that a theory is unclear, it is more difficult to figure out whether it has flaws, what they are, and how to improve them, thereby slowing progress.

The Problem of Sufficiency To the modern reader, it seems obvious that people’s behaviors are not completely rational. Given this fact, the notion of a “rational actor” theory might seem silly on the face of it. And yet, as I alluded to earlier, it is possible to act “rationally” (or perhaps better words might be “reasonably” or “logically”) based on wrong assumptions. If a person performs a maladaptive behavior, based on good reasoning but on wrong assumptions, the process might be considered “reasoned.” But why might people make wrong assumptions? Rational actor theories are mostly silent on this issue. In the case of the TRA, for example, such issues are covered up in the “sufficiency assumption.” The idea is that there might be many causes of behavior other than the TRA constructs; however, their effects on behavior are mediated through TRA constructs. That is, many variables, whether justifiable or not, can influence behavioral beliefs, evaluations of them, normative beliefs, or motivations to comply. But from there, the process is completely reasoned as described by the TRA. Even the updated TPB assumes this, though it adds control beliefs as another distal construct that can act as a conduit for the influence of outside variables on behavior. The sufficiency assumption—that the effects of all variables work through TRA constructs—seems to account for the fact that people sometimes make poor or unreasonable decisions, while allowing researchers to retain the assumption that decisions nevertheless are reasoned. The sufficiency assumption also has led to a curious methodology for performing TRA research. Specifically, it has led to an almost exclusive reliance on hierarchical regression paradigms. To see why, consider that one way of interpreting the sufficiency assumption is that it renders all non-TRA variables as predictively unimportant. Even if a non-TRA variable has great causal power over behavior, the ability to predict behavior will not be enhanced by including it in the multiple regression equation. This is because the variance in behavior or behavioral intention that it accounts for already is accounted for by the TRA variables through which it transmits its influence on behavior. In fact, TRA researchers have repeatedly stated that the test for whether a new construct should be added to the theory is whether it accounts for unique variance in behaviors or behavioral intentions, above and beyond the

258 PRAGMATIC THEORIES

variance that can be accounted for by the TRA constructs. According to TPB researchers, perceived behavioral control passes this test, thereby justifying its inclusion in the newer TPB. Given that the test of the worthiness of new variables is their ability to account for unique variance in behaviors or behavioral intentions, it is perhaps not surprising that many researchers have attempted to show that other constructs are capable of accounting for unique variance in behaviors or behavioral intentions. And it is not particularly difficult to do this, if the criterion is accounting for a statistically significant percentage of unique variance in intentions or behaviors. This percentage could be as little as 1% if the number of participants is sufficiently high for the requisite statistical power, especially if the study is designed precisely for the purpose of showing off the new variable. In fact, Trafimow (2004) noted that this is the usual strategy for researchers to publish in the area. Despite the consistent use of the paradigm, its use arguably is contrary to good scientific practice. There is an empirical issue and there is a conceptual one. The empirical issue can be described quickly and easily. Although it is easy to account for a few percent of unique variance in behaviors or behavioral intentions with one’s favored new variable in the specific context designed for this, it is less easy to show that one’s favored variables account for a few percent unique variance across several contexts, including ones not specifically designed to show off the favored variable to best advantage. Without such a demonstration of the general importance of the favored variable, a concern for parsimony renders it unlikely that the new variable will be added to the TRA or TPB. Most scientists agree that theories should be as simple as possible and that the addition of extra variables should be avoided if unnecessary. Suppose the inclusion of a variable on Step 2 of a hierarchical regression analysis causes more variance to be accounted for than was accounted for in Step 1 where the variable was not included. How much of an increase in variance accounted for at Step 2, in how many study contexts, justifies inclusion of the new variable in the theory? There is a conceptual issue too that, though mostly ignored, is much more important than the empirical issue. Suppose that a TRA researcher wishes to predict behavioral intentions from attitudes, subjective norms, and a new touted variable. The researcher enters attitudes and subjective norms on Step 1, and enters the touted variable on Step 2, to see if it accounts for unique variance in behavioral intentions. Suppose that the touted variable accounts for 5% unique variance in behavioral intentions. What does this mean? Surprisingly, the 5% effect means something very different, depending on what the Step 1 result was. An easy way to see this is to convert “increase in variance accounted for” into “increase in percentage of successes” (Trafimow, 2004), which is much easier to understand. Suppose that under Treatment A, 60% of cancer patients were cured, whereas the addition of Treatment B increased this cure rate to 70%; the increase in percentage of successes would be 70% – 60% = 10%, and this is straightforward. The surprising mathematical fact of the matter is that the conversion of increase in

Rational Actor Theories 259

variance accounted for to increase in percentage of successes depends on the Step 1 correlation. This mathematical fact causes a conceptual dilemma for the researcher who wishes to use hierarchical regression to support a touted variable, and this includes the vast majority of TRA and TPB researchers. For example, suppose the Step 1 correlation is .10 (variance accounted for is 1%) and the touted variable provides an extra 5% more variance accounted for in behavioral intentions on Step 2. Note that the low Step 1 correlation suggests that the measures of the Step 1 variables likely were invalid. When translated into extra successes, the extra 5% translates into an increase of 7.5% in terms of increased successes. So is the touted variable really that good, or is it simply a matter of invalid measures of Step 1 variables leaving almost all of the variance in behavioral intentions left over to be accounted for by the touted variable? Or suppose that the Step 1 effect is large (e.g., r = .90). The good news is that it is more difficult to argue that the measures of the Step 1 variables are invalid. The bad news is that conversion to unique successes implies that the 5% increase in variance accounted for at Step 2 shrinks to 1.37% when converted to increase in percentage of successes. The examples illustrate two points. First, the same 5% increase in variance accounted for at Step 2 means very different things, depending on the Step 1 correlation; it means a large increase in percentage of successes on Step 2 when the Step 1 correlation is low, but it means a measly increase in percentage of successes on Step 2 when the Step 1 correlation is high. Second, the examples highlight that if the Step 1 correlation is low, the seemingly impressive increase in percentage of successes by the touted variable might be due to too much variance left over because the Step 1 measures were poor predictors of the criterion variable, whether due to invalidity or for some other reason. If the Step 1 correlation is high, a seemingly impressive unique variance accounted for percentage turns out to imply very little increase in actual successes. Thus, no matter what the results are, they fail to make a strong case for the value of the touted variable for uniquely predicting the criterion variable. The foregoing might seem surprising, but a more general perspective may render it less so. Consider what everyone learns in Introductory Psychology, that correlation does not necessarily mean causation. Multiple regression analysis is just a fancy way of dealing with correlations, and so it makes sense that there would be important problems with it for testing a causal theory, such as the TRA or TPB. Tests of whether ostensible determinants of behavioral intentions actually do determine them would be much stronger if researchers used experimental paradigms rather than resorting to fancy correlational paradigms. And because TRA variables are not difficult to manipulate, it arguably is inexcusable that TRA researchers rely so heavily on correlational paradigms. I hasten to add, however, that it is arguable to what extent this fault should be blamed on the TRA sufficiency assumption versus TRA researchers.

260 PRAGMATIC THEORIES

The foregoing might seem, at first observation, to be a completely methodological issue. However, methodological, theoretical, and philosophical issues tend to have implications for each other. Consider again that the sufficiency assumption, which is a theoretical issue, is largely responsible for the use of the methodology of hierarchical regression. Even if the issue of increase in variance accounted for were not problematic, is the implied philosophy of science sensible? Does it make sense to ignore potential distal causes of a phenomenon in the event that proximal variables completely account for the phenomenon of interest? I will leave it to the reader to imagine what the state of other sciences (e.g., physics, chemistry, biology, and so on) would be, if scientists in those areas had settled for accounting for variance with proximal variables and had not investigated distal variables. There is one last issue before proceeding to the next section. The sufficiency assumption is particularly associated with TRA and TPB research, and is generally not made in other rational actor theories. Consequently, experimental approaches have been used more in investigations of other rational actor theories than in TRA or TPB research. Another advantage of not making the sufficiency assumption, and a more important one from the perspective of this book, has to do with the search for distal variables and ultimate causes. Work by Goldstein and Gigerenzer (Gigerenzer & Goldstein, 1996; Goldstein & Gigerenzer, 2002) has employed evolutionary theory as an aid to investigate potential distal variables. Although it is arguable whether an evolutionary perspective is the best way to search for distal variables (see Ketelaar, Chapter 11, this volume, for a discussion), the attention paid to them is certainly a positive aspect of this work, at least compared to the TRA or TPB.

“Mattering” in a Correlational versus Experimental Sense The foregoing section focused on the sufficiency assumption, the problem with the heavy use of hierarchical regression analyses, a plea for true experiments, and the pitfalls of failing to look for distal variables. The present section continues but with the pretense that everything in the foregoing section can be ignored. The point of doing this is to illustrate that the connection between correlation and causation is even weaker than psychologists realize. Most psychologists understand that correlation does not mean causation, but I wish to demonstrate, in the present section, that the inability of a variable to account for unique variance in a dependent variable does not imply a lack of causation. This demonstration will take place in the context of the assumption that practically everyone makes, tacitly of course, that variables that fail to account for unique variance in the dependent variable (e.g., behavioral intentions or behaviors) do not matter. I will show that they can matter a lot. To begin, recall the implication of Equation 12.1 that several behavioral belief–evaluation pairs determine attitudes. But suppose it were demon-

Rational Actor Theories 261

strated that the prediction of attitudes from only a subset of these pairs was sufficient to account for all of the variance in attitudes that can be accounted for. Put another way, suppose that once, say three belief-evaluation pairs were included in a hierarchical regression equation, the inclusion of the rest of the pairs failed to increase the prediction of attitudes on the next step of the analysis. In that case, according to the typical way of thinking in psychology, or at least in TRA or TPB research, the conclusion would be that the rest of the belief-evaluation pairs are unimportant. In fact, van der Pligt and de Vries (1998a, 1998b) actually obtained this finding and came to the conclusion that only the belief–evaluation pairs that predict unique variance in attitudes are actually important; the rest do not matter. There are at least two problems with this sort of thinking, despite how widespread it is. First, the sufficiency assumption does not actually state that variables that fail to account for unique variance are unimportant; it merely states that if they have causal effects, these effects are mediated through TRA constructs. Therefore, even by strict TRA thinking, this is a conceptual error, despite the fact that TRA researchers make the error on a routine basis. Second, and more important, variables can share variance for a lot of reasons. Suppose, for example, that all belief–evaluation pairs influence attitudes, but that for whatever reason the belief–evaluation pairs become correlated with each other as time progresses. Trafimow (2009), for example, suggested that there is a general tendency for mental elements that are associated with each other to become increasingly compatible over time. It even is possible that the formation of an attitude has a feedback effect on the very belief–evaluation pairs from which it arose, causing the belief–evaluation pairs to become increasingly correlated with each other. Well, then, from a causation point of view, all of the belief–evaluation pairs would be important because all have a causal effect on attitudes. But from a hierarchical regression point of view, only one of the belief–evaluation pairs would be important because the others predict no unique variance in attitudes. We might even perform a thought experiment, where we imagine that we experimentally manipulate many beliefs and show that all of them have causal effects on attitudes, whereas, at the same time, we imagine that in a hierarchical regression analysis, only one belief (and its associated evaluation) is sufficient to account for all of the variance in attitudes that we can account for. Clearly, at least in the thought experiment, even the beliefs that fail to account for unique variance in attitudes are extremely important given that we pretended that manipulating them influences attitudes. Of course, one might counter that this thought experiment would be extremely unlikely to work if it were attempted in an actual laboratory, and so the thought experiment should not be taken too seriously. I have two responses to the notion that thought experiments should not be taken seriously if they seem unlikely to work if they were actually performed. First, let us recall the lesson we learned from the section on falsification, that seemingly unlikely auxiliary assumptions often have been discov-

262 PRAGMATIC THEORIES

ered (e.g., spectrographic analysis). Similarly, the fact that one might think that a particular result would not occur does not make it so. The road to Hades is paved with incorrect intuitions. Second, Trafimow et al. (2012) actually did perform the thought experiment (well, several experiments), and the results actually did come out in the way described. That is, Trafimow et al. showed that experimental manipulations of beliefs that failed to account for unique variance in attitudes nevertheless had strong unique effects on attitudes! This demonstration worked even when all of the relevant beliefs were simultaneously manipulated, meaning that the supposedly unimportant beliefs mattered, when manipulated, even on top of the manipulation of beliefs that accounted for all of the variance. Finally, Trafimow et al. also obtained evidence that the reason for the hierarchical regression null effects, at the same time as there were strong experimental effects, was because of a feedback effect from the attitudes that caused beliefs to become increasingly correlated with each other. Lest the reader get caught up in the complexities of this section, it is worth reiterating the main conceptual point. The usual assumption in research that employs correlational methods (e.g., hierarchical regression) is that a variable that fails to account for unique variance in the dependent variable is unimportant. On the contrary, it is quite possible for such a variable to have important and impressive causal effects even if it has no effect in the correlational context. It is a mistake not only to assume that correlation (or accounting for unique variance) means causation, but also to assume that a lack of ability to account for unique variance means a lack of unique causation. Much is tacitly, and often wrongly, assumed when one performs correlational analyses, and this becomes increasingly so as the analyses become increasingly complex.

Rationality An obvious way to criticize rational actor theories is to harp on the lack of a definition of “rational” on which we all can agree. By using the word “reasoned” rather than “rational,” the TRA does not avoid the problem because that word also is not defined. In the section discussing the definition of action as being caused by intention, we saw that definitions are always problematic because of the unpalatable choice between infinity, circularity, and primitiveness. Keeping this in mind, consider again the research I mentioned earlier that shows that there is an affective route as well as cognitive ones that lead to behavioral intentions and behaviors. I pointed out in the section on falsification that these data falsify the TRA claim of a lack of an affective route. But do the data falsify the notion that people’s behaviors are the result of a reasoned process? It is interesting that this might or might not be so depending on one’s way of conceptualizing the notion of a reasoned process. One way to think about it, that I believe is close to the way Fishbein thought about it (e.g., 1980), is that “reasoned” means “cognitive,” in which case the

Rational Actor Theories 263

data really would falsify the notion. However, there are other possibilities. To see this, imagine asking a computer whether it likes chocolate or vanilla ice cream better; the computer would have a difficult time providing a sensible answer. In contrast, being a human with genetic input from my parents who love chocolate, I can confidently answer that chocolate ice cream is better (and by a lot!). From my point of view, though possibly not from the computer’s point of view, it might be said to be rational for me to ask for chocolate rather than vanilla ice cream. The ice cream example points to the possibility that some sort of affective system might be a prerequisite for attaching values to consequences, which thereby allows the cognitive machinery specified by the TRA to run. If we accept that there has to be some way to attach values to consequences, and that this cannot be done strictly on the basis of logic, affect provides a way out of the dilemma. Nor is this conclusion restricted to the TRA, as most rational actor theories assume that the values of consequences matter. Consequently, it makes sense, and can be considered to be rational, that there should be an affective route to behavioral intentions or behaviors. A potential additional argument might be that a major benefit in having a cognitive system is that it provides rational means to satisfy affective demands. Clearly, the philosophical issues about what is rational or reasoned are complex. Whether it is inherently rational or irrational to believe in rational actor theories comes down, at least in part, to what one thinks constitutes rationality. Or to put it in terms of the example, is it rational to like chocolate ice cream or, assuming it is valued, to perform behaviors that have chocolate ice cream as a consequence?

Conclusion Rational actor theories are easily criticized. Do people really undergo the complex computations implied by the mathematical equations? If not, can these theories be said to either “describe” or “explain” behavior? If the conclusion is that these theories neither describe nor explain behavior, is it permissible to settle for predicting or modeling? Or is such settling akin to “punting on fourth down”? Rational actor theories also have been criticized on both definitional and falsification grounds. Here, however, we saw that these criticisms are based on an incomplete view of either definitions or auxiliary assumptions. When a more sophisticated view is taken, the rejection of rational actor theories on either of these grounds is not sensible. This is especially so given that some falsification already has been demonstrated. Rational actor theories fare less well on the criterion of internal consistency. We saw earlier that special attention to units, which TRA researchers fail to give, demonstrates that the proposed mathematical equations actually contradict TRA definitions of attitude and subjective norm. Either the TRA definitions are wrong or the TRA equations are. More generally, the TRA is

264 PRAGMATIC THEORIES

not unit coherent, which means that it fails the philosophy of science requirement that theories be internally coherent. Some other rational actor theories, which lack mathematical equations, are not subject to unit analysis. The advantage is that they cannot be demonstrated to be unit incoherent, but the disadvantage is that they are vague and can be described as unit a-coherent. Finally, the sufficiency notion—though there is nothing necessarily wrong with it as a theoretical assumption—has had the unfortunate consequence of leading researchers down the garden path of relying heavily on multiple regression analyses and other fancy correlational paradigms. Aside from the general problem with depending on correlations to draw causal conclusions or test causal assumptions, the conversion of unique variance accounted for to unique successes implies a dilemma. Regardless of whether the Step 1 correlation is a low or high number, conversion to percentage of successes illustrates that practically any result is potentially problematic. An additional but related problem is that researchers have tended to assume that “mattering” in the context of accounting for unique variance is the only kind of mattering that matters. It is quite possible for a variable to matter a lot in a causal sense, while still not accounting for unique variance. This is an extremely important conceptual error that unfortunately has dominated TRA research, though this is not as true of research pertaining to other rational actor theories. Many rational actor studies have been performed. Rather than providing a review, I have attempted to emphasize a few of the important philosophical and conceptual issues that are important for evaluating them. Had researchers taken a philosophical perspective, much of the research need not have been performed. My hope is that the graduate student or professor who wishes to conduct research will consider the larger philosophical and conceptual issues before embarking on what might be an expensive and timeconsuming excursion that leads nowhere. References Ajzen, I. (1988). Attitudes, personality, and behavior. Chicago: Dorsey. Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50, 179–211. Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behavior. Englewood Cliffs, NJ: Prentice-Hall. Cartwright, N. (1983). How the laws of physics lie. Oxford, UK: Clarendon Press. Cartwright, N. (1999). The dappled world: A study of the boundaries of science. Cambridge, UK: Cambridge University Press. Fishbein, M. (1980). Theory of reasoned action: Some applications and implications. In H. Howe & M. Page (Eds.), Nebraska Symposium on Motivation, 1979 (pp. 65–116). Lincoln: University of Nebraska Press. Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention and behavior: An introduction to theory and research. Reading, MA: Addison-Wesley. Fishbein, M., & Ajzen, I. (2010). Predicting and changing behavior: The reasoned action approach. New York: Psychology Press.

Rational Actor Theories 265 Gigerenzer, G., & Goldstein, D. G. (1996). Reasoning the fast and frugal way: Models of bounded rationality. Psychological Review, 104, 650–669. Goldstein, D. G., & Gigerenzer, G. (2002). Models of ecological rationality: The recognition heuristic. Psychological Review, 109, 75–90. Greve, W. (2001). Traps and gaps in action explanation: Theoretical problems of a psychology of human action. Psychology Review, 108, 435–451. Hollis, M. (1996). Reason in action: Essays in the philosophy of social science. Cambridge, UK: Cambridge University Press. Kelley, H. H. (1973). The process of causal attribution. American Psychologist, 28, 107–128. Lakatos, I. (1978). The methodology of scientific research programmes. Cambridge, UK: Cambridge University Press. Ogden, J. (2003). Some problems with social cognition models: A pragmatic and conceptual analysis. Health Psychology, 22, 424–428. Popper, K. R. (1959). The logic of scientific discovery. New York: Basic Books. Popper, K. R. (1963). Conjectures and refutations. London: Routledge. Popper, K. R. (1972). Objective knowledge. Oxford, UK: Oxford University Press. Smedslund, G. (2000). A pragmatic basis for judging models and theories in health psychology: The axiomatic method. Journal of Health Psychology, 5, 133–149. Trafimow, D. (2003). Hypothesis testing and theory evaluation at the boundaries: Surprising insights from Bayes’s theorem. Psychological Review, 110, 526–535. Trafimow, D. (2004). Problems with change in R2 as applied to theory of reasoned action research. British Journal of Social Psychology, 43, 515–530. Trafimow, D. (2007). Distinctions pertaining to Fishbein and Ajzen’s theory of reasoned action. In I. Ajzen & D. Albarracin (Eds.), Prediction and change of health behavior: Applying the reasoned action approach (pp. 23–42). Mahwah, NJ: Erlbaum. Trafimow, D. (2009). Reeder’s MIM as a special case of confluence theory. Psychological Inquiry, 20, 48–52. Trafimow, D. (2012a). The role of auxiliary assumptions for the validity of manipulations and measures. Theory and Psychology, 22, 486–498. Trafimow, D. (2012b). The role of mechanisms, integration and unification in science and psychology. Theory and Psychology, 22, 696–703. Trafimow, D. (2012). The concept of unit coherence and its application to psychology theories. Journal for the Theory of Social Behaviour, 42, 131–154. Trafimow, D. (2013). Are measurement theories falsifiable and should we care? Theory and Psychology, 23(3), 397–400. Trafimow, D., & Rice, S. (2009). What if social scientists had reviewed great scientific works of the past? Perspectives in Psychological Science, 4, 65–78. Trafimow, D., Rice, S., Hunt, G., List, B., Nanez, B., Rector, N., et al. (2012). It’s irrelevant, but it matters: Using confluence theory to predict the influence of beliefs on evaluations, attitudes and intentions. European Journal of Social Psychology, 42, 509–520. Trafimow, D., & Sheeran, P. (1998). Some tests of the distinction between cognitive and affective beliefs. Journal of Experimental Social Psychology, 34, 378–397. van der Pligt, J., & de Vries, N. K. (1998a). Belief importance in expectancy–value models of attitudes. Journal of Applied Social Psychology, 28, 1339–1354. van der Pligt, J., & de Vries, N. K. (1998b). Expectancy-value models of health behaviour: The role of salience and anticipated affect. Psychology and Health, 13, 298–305. Weiner, B. (2010). The development of an attribution-based theory of motivation: A history of ideas. Educational Psychologist, 45, 28–36. Weiner, B., Frieze, I. H., Kukla, A., Reed, L., Rest, S., & Rosenbaum, R. M. (1971). Perceiving the causes of success and failure. Morristown, NJ: General Learning Press.

13 Social Functionalism Philip E. Tetlock Katrina Fincher

Most social acts have to be understood in their setting, and lose meaning if isolated. No error in thinking about social facts is more serious than the failure to see their place and function.   —Solomon E. Asch (1952/1987, p. 61)

S

ocial functionalism is arguably less a theory than a point of view that sensitizes us to the interpersonal and institutional settings within which intrapsychic processes of long-standing social-psychological interest unfold. Once we get into the habit of looking at the world through a socialfunctionalist set of lenses, it is easy to spot social-functionalist disputes popping up across a variety of classic social-psychological research programs, from conformity studies that tried to disentangle normative from informational influence (do people go along with the group because they fear ruptured relationships and social censure or because they realize that there has often proven to be wisdom in the crowd?) to studies of person perception and causal attribution (when is the “fundamental attribution error” a sign that people are flawed intuitive scientists who overrely on simple heuristics and when is it a sign that people are tough intuitive prosecutors determined to minimize situational excuse wiggle room?) to studies of attitude change in forced compliance paradigms (when is it real attitude change driven by dis 266



Social Functionalism 267

sonance reduction and when is it the self-presentational posturing of intuitive politicians?). A listing of such debates could easily fill this chapter, each debate—in essence—being an explanatory turf war between social-functionalist and intrapsychic formulations that has stimulated empirical efforts to demarcate boundary conditions (Tetlock, 2002; Tetlock & Manstead, 1985). Our goal here is certainly not to argue that social functionalism can explain everything. As critics of functionalism have long known, it is all too easy to fall into the trap of finding a function for virtually any pattern of behavior (Boring, 1953; Boring, Watson, & Campbell, 1963). Functionalist explanations reduce to hollow labeling exercises without support from intrapsychic theories that clarify exactly how functionalist goals and mind-sets become engaged or disengaged in particular situations—and how people (consciously or unconsciously) choose among coping strategies in pursuit of these goals. We divide this how-to chapter on social-functionalist theorizing into three sections. The first section notes the variety of social-functionalist theories currently in circulation, with special emphasis on our favorite form of functionalism—the complementary triad of the intuitive politician, prosecutor, and theologian. The second section identifies key questions that any social-functionalist theory should address: Which interpersonal or societal goals are people trying to achieve—and why? When are these goals (and associated mind-sets) likely to be switched on or off? What types of behavioral, cognitive, and affect-regulation strategies do people deploy to achieve these goals? And how successful are they—both objectively and subjectively? The third section explores a serious shortcoming in much social-functionalist theorizing: the frequent failure to flesh out in detail the perceptual-cognitiveaffective mechanisms that make it possible for people to switch from one functionalist mind-set to another. Using the example of intuitive-prosecutor models of responsibility attribution and punitiveness, we examine some of the perceptual and cognitive mechanisms that make it possible for observers of norm violations to shift seemingly seamlessly from empathizing with the pain of a fellow human being to ignoring or sometimes or even relishing it.

Forms of Social Functionalism There is substantial evidence from cultural, developmental, social, and personality psychology that people have a deep evolutionarily rooted and neurologically hardwired drive to seek inclusion and to avoid exclusion from valued groups (Ainsworth, 1989; Barash, 1977; Baumeister & Leary, 1995; Hogan & Blickle, 2013; Leary, 1990). Social functionalism builds on the premise that human beings are indeed deeply social creatures. It takes as its starting point the idea that human beings have managed to adapt and continue to adapt to an extraordinary range of cultural-historical environments and poses the launching-pad question: How is this possible? How do people

268 PRAGMATIC THEORIES

need to be psychologically structured to survive and thrive in collectivities regulated by complex accountability relationships, norms, and values? Scholars have proposed a variety of candidate answers, including: psychodynamic theories that posit needs for elaborate defense mechanisms to cope with tensions between basic biological urges and the demands of the social order (Adorno, Frenkel-Brunswik, Levinson, & Sanford, 1950), to just– world and system–justification–style theories that posit a need to believe that we live in a nonarbitrary moral-political world and to endow that world with legitimacy (Jost, Kay, & Thorisdottir, 2009; Lerner & Lerner, 1981) to existentialist theories that posit a need to seek symbolic immortality via identification with the more transcendental aspects of the social order, such as the law, nationalism, and religiosity (Greenberg & Jonas, 2003). Tetlock (2002) proposed an integrative social-functionalist framework that includes three complementary functionalist metaphors: people as intuitive politicians, as intuitive prosecutors, and as intuitive theologians (see Table 13.1). Building on the work of eminent sociologists—Durkheim, Weber, Mead, Parsons—as well as recent advances in cross-cultural psychology (Fiske, 2004), the first order of theory-construction business was listing the psychological prerequisites of social order—in particular, the adroitness with which the vast majority of the population manages to shift from being targets of the accountability demands from others to being sources of accountability demands on others (see Aberle, Cohen, Davis, Levy, & Sutton, 1950). Tetlock’s triad of theories is designed to capture the most basic socialfunctional orientations that the vast majority of people, under the right activating conditions, adopt toward the social world. As objects of accountability pressures from others, people strive—like intuitive politicians—to establish, maintain, and enhance their social identities vis-à-vis significant TABLE 13.1.  The Triad of Functionalist Frameworks Motives

Adaptive challenge

Illustrative coping strategies

Intuitive politician

Social approval activated by evaluative audiences

Maintain or enhance social standing

Attitude shifting as a function of accountability pressures

Intuitive prosecutor

Retributive impulses activated by norm violations

Preserve social order

Punish norm violators in proportion to offenses

Intuitive theologian

Sacred value protection activated by threats to the sacrosanct

Social cohesion

Moral outrage at taboo trade-offs



Social Functionalism 269

constituencies in their lives. A key function of thought becomes internalized dialogue in which people try to anticipate objections to possible lines of action and to craft responses. As transmitters of accountability pressures onto others, people try—like intuitive prosecutors—to detect cheaters and free riders who seek the benefits but shirk the responsibilities of membership in the collective. A key function of thought now becomes anticipating and closing loopholes in accountability regimes that unscrupulous members of the collective might otherwise exploit. Finally, as beings capable of reflecting on the accountability regimes within which they live, love, and work, people are posited to be intuitive theologians who have an existential need to believe that the rules governing their social world are not just the arbitrary preferences of currently dominant interest groups but rather are anchored in sacred values that confer legitimacy on collective practices. A key function of thought becomes protecting sacred values from secular encroachments.

Core Requirements of Social-Functionalist Theories Since McDougall (1908) and Murray (1938) published their lengthy and seemingly rambling lists of needs, social functionalism has been a suspect theoretical enterprise in the eyes of many psychologists (Boring et al., 1963). And social functionalists still need to take special pains to show that they are not peddling thinly veiled tautologies. We propose five features of socialfunctionalist frameworks that would-be theorists should aspire to build into their own formulations—and that discerning consumers in the marketplace of ideas should assess in deciding how much epistemic credibility to confer on such formulations: 1.  Does the framework focus on fundamental adaptive challenges that arise whenever human beings are locked into complex patterns of interdependence? The quest for “fundamental” challenges could easily take us all the way back to the challenges facing our hominid ancestors on the savannah plains of Africa and the need for bonding as well as cheater-detection mechanisms (Cosmides, 1989). But social functionalism does not have to come in evolutionary-psychology form (see Ketelaar, Chapter 11, this volume). Sociocultural theorists are content to explore the role of the target class of behavior in helping people in modern societies survive and thrive in the here-and-now—and in promoting social solidarity and stability. Although social functionalism can be agnostic on the degree to which our social needs and coping strategies are products of natural and sexual selection that are now wired into our DNA, parsimony dictates at minimum that we should prefer to focus on those challenges likely to arise in a wide range of cultural-historical settings (Pepitone, 1976)—even if the forms of behavior and triggers of behavior are highly culture-specific (see Eom & Kim, Chapter 16, this volume). For example, sacred acts and rituals exist in

270 PRAGMATIC THEORIES

all societies even though what is deemed sacred can vary dramatically—as can official manifestations of moral outrage and moral cleansing in response to violations of the sacrosanct. 2.  Does the framework focus on a starting point that is conceptually rich enough to serve as a Lakatosian hard core (see Gawronski & Bodenhausen, Chapter 1, this volume) from which it is possible to launch competing middle-range theories that make testable predictions about the situational triggers and adaptive challenges that activate functionalist mind-sets, the exact goals people are trying to achieve, and the coping strategies that people use in pursuit of goals. The intuitive politician metaphor, for instance, can be a starting platform for tightly specified and falsifiable formulations that depict the intuitive politician variously as totally unscrupulous (prepared to do or say whatever it takes to please the audience or constituency of the moment) to totally conviction-driven (prepared to do whatever he or she thinks is ideologically correct, regardless of impact on the audience. These competing middle-range theories make different predictions (among other things) about whether the mere presence of an evaluative audience should be sufficient to change the attitudes people endorse. The intuitive prosecutor and intuitive theologian metaphors can play a similar role in theory construction. Intuitive prosecutors can range from ruthless retributive personalities (prone to reject all excuses) to determined defenders of procedural justice (preferring to let the guilty to go free rather than violate due process). These competing theories make competing predictions (among other things) about how open people are likely to be to extenuating conditions surrounding a norm violation. Figures 13.1, 13.2, and 13.3 illustrate in schematic form the logical progression from a crude undifferentiated hard-core metaphor, such as the politician, prosecutor, or theologian, to increasingly precise testable middle-range theories that take varying stands on how people implement the functional imperatives of the hard core. The figures also underscore how hard it is to falsify a functionalist hard core. Confronted by dissonant evidence, defenders of the research-programmatic hard core have a multilayered, protective belt of defenses sketched in Figure 13.4. Potential lines of defense include challenging the degree to which the operational definitions of key constructs adequately capture those constructs; challenging the tightness of the logical links between the middle-range theories and the hypotheses “derived” from them; and, after repeated efforts to rescue a particular middle-range theory have failed, deciding to abandon that theory but simply shift to a different middle-range theory equally consistent with the hard-core premises of the research program. Table 13.1 sketches the three core questions that, at minimum, any social-functionalist theory should address: core motives, adaptive challenges for activating those motives, and coping strategies for achieving desired end states. The answers to these questions provide the key conceptual ingredi-



Social Functionalism 271

FIGURE 13.1.  Functionalist logic of the intuitive politician. The figure illustrates the gap between hard-core assumptions and empirical data. The protective belt gives defenders of the research program many interpretive options before concluding that the hard core itself needs to be revised (e.g., criticizing the operational definitions, the logical links between predictions and the middle-range theory, and the middle-range theory itself). .

FIGURE 13.2.  Functionalist logic of the intuitive prosecutor. The figure brackets the three layers of the protective belt of the intuitive prosecutor research program: the capacity to deflect dissonant data by criticizing operational definitions, the logical links between predictions and theories, and the theories themselves.

272 PRAGMATIC THEORIES

FIGURE 13.3.  Functionalist logic of the intuitive theologian. The figure brackets the three layers of the protective belt of the intuitive theologian research program: the capacity to deflect dissonant data by criticizing operational definitions, the logical links between predictions and theories, and the theories themselves.

ents for cooking up middle-range theories of varying flavors but consistent with the hard-core theme of the research program. Defenders of functionalist hard cores always have the option of positing individual difference boundary conditions on when one middle-range theory or another applies. For instance, within the intuitive politician program, it seems reasonable to acknowledge that the highly flexible variant of the intuitive politician better fits people who score high on certain types of personality measures (e.g., Snyder’s, 1974, high self-monitors), whereas the conviction-constrained politician better fits low self-monitors who are more concerned with self-verification and affirmation than they are with accommodating social pressures of the moment. One could construct parallel arguments linking existing personality variables and typologies to the partly competing and partly complementary middle-range theories of the intuitive prosecutor or theologian (see Cervone, Caldwell, & Mayer, Chapter 8, this volume). 3.  Do researchers working within the framework recognize the need to strike the right epistemic balance between commitment to developing middle-range theories consistent with the hard core and willingness to acknowledge when these efforts have run out of steam? Given the variety of middle-range theories that can be derived from each social-functionalist hard core, it becomes virtually impossible to falsify the hard core itself. If one type of intuitive politician/prosecutor/theologian



D 1

Social Functionalism 273

O 1

Hypotheses

Middle-range theory

Hard core

Protective belt

FIGURE 13.4.  Illustration of the increasingly tenuous links between concepts and data within the Lakatosian framework. As one moves toward the most abstract concept, the hard core, the links fade.

(or any other functionalist theory) middle-range theory does not fit, determined defenders of the research program can always spin off another. Social functionalists thus confront a delicate balancing act here: They need to be enthusiastic enough about the functionalist hard core to develop and test middle-range theories, but they also need to recognize when perseverance shades into dogmatism. For instance, what would it take to convince me that people are not intuitive politicians who respond to accountability manipulations to manage impressions, but rather the behavior in question is best captured by a nonsocial form of functionalism (people as intuitive scientists or statisticians) or by a type of theory that is purely process focused and seemingly does not rely on any functionalist assumptions (e.g., associationist models of priming). 4.  Has the framework proven, in McGuire’s (1983) phrase, to be “heuristically provocative”? The proof of the pudding of any research program is its capacity to generate novel and useful discoveries. Middle-range theories can be successful in one of two basic ways—and theories that can claim neither are prime candidates for abandonment. An empirically progressive middle-range social-functionalist theory should (1) stimulate discoveries of empirical boundary conditions on well-established effects that would not have been recognized if theorizing had been confined to mainstream intrapsychic frameworks. For instance, we might discover that an effect only holds up under conditions of anonymity or public scrutiny—or only when people think the researchers possess methods for detecting deception; (2) stimulate discoveries of normative boundary conditions that lead us to reconsider classifications of earlier discovered effects as adaptive or maladaptive. For instance, is the dilution effect (Nisbett, Zukier, & Lemley, 1981) a sign that people are flawed intuitive statisticians who too quickly lose confidence in the predictive power of diagnostic cues when the experimenter also provides nondiagnostic cues or a sign that people are attentive conversationalists who rely on usually reliable conversational norms to decode social interaction (Tetlock, Lerner, & Boettger, 1996).

274 PRAGMATIC THEORIES

5. Does the framework address the vexing problem of integrating conflicting psychosocial functions? An old saying—adopted by cybernetic learning theorists—is that we never have had enough until we have had more than enough. It is easy for people to overshoot in the pursuit of one or another functionalist objective: the cloyingly ingratiating intuitive politician, the relentless intuitive prosecutor, or the jihadist intuitive theologian. The more interesting observation, however, may not be that overshooting happens but that it does not happen all the time. People often seem to come up with cleverly subtle ways of reconciling the seemingly irreconcilable. Compare, for instance, the absolutist-sounding priorities of the intuitive theologian (pay any price and bear any burden to defend sacred values) and the pragmatic mind-set of the intuitive economist (in a world of finite resources, trade-offs are inevitable—and everything has a shadow, if not explicit price). Intuitive theologians don’t want to admit that they attach dollar valuations to human life, so they feel a need to invent a host of ways of obfuscating the fact that they have done just that (e.g., Tetlock, Kristel, Elson, Green, & Lerner, 2000). Remarkably little is yet known about when the pursuit of one selfregulatory objective has negative externalities that impede other objectives. In political theory, negative externalities offer the classic justification for empowering government to regulate the conduct of its citizens (to paraphrase John Stuart Mill, your freedom to swing your arm ends where my nose begins). Likewise, in psychological theory, negative intrapsychic externalities have offered the classic justification—since Freud—for some system of mental governance or executive control to adjudicate conflicts between compartmentalized functions. This fourth requirement has proven a serious blindspot in our initial social-functionalist theorizing—to which we turn in the final section of this chapter. 6. Does the framework tackle the thorny issue of how successful or unsuccessful people are in achieving their functional objectives? Defining success raises complex conceptual issues: success in whose eyes? A purely subjective definition would be: “I am successful whenever I truly believe myself to be successful.” An objective definition of success would require a measurable impact in the external world. For instance, I am successful whenever audiences perceive me the way I want to be perceived or whenever I have been seen by others as having played a key role in upholding the normative order. Well-specified functionalist theories should take stands on when subjective and objective metrics of consummating functional goals are likely to converge and diverge, a test that virtually all currently fail. Functionalist theories should also allow for the possibility that people are not always aware of their goals or of the full social impact of their actions. Merton’s (1949/1968) analysis of the latent and manifest functions of social practices provides a useful reminder of how often mismatches can occur between the functions we think we are advancing and those we are advanc-



Social Functionalism 275

ing. Manifest functions serve ends that people expect and intend, whereas latent functions serve ends that are neither expected nor intended. The manifest function of a rain dance may be to produce rain, whereas the latent function may be to reinforce group identity by encouraging group members to engage in a common activity. A promising underdeveloped area for socialfunctionalist theorizing concerns the potential connections between the latent–manifest distinction and the implicit–explicit distinction in early-21stcentury attitude theory. Explicit attitude measures should be fully up to the task of assessing manifest functions, but implicit measures may be essential for tapping into the latent associations between, say, rain dances and social solidarity (Fincher & Tetlock, 2013a). Functionalist theorists should make more efforts to distinguish conceptually and operationally between the latent and manifest functions of social conduct. When does tearing off the manifest mask disable the latent function as well? And when are people quick simply to don another mask?

Filling a Common Gap in Social-Functionalist Theorizing In this section, we use social-functionalist work on intuitive-prosecutor middle-range theories to illustrate what we see as the most serious deficiency in current social-functionalist theorizing. Understanding punitiveness requires middle-range theorizing that bridges individual and societal levels of analysis. Putting aside whether authoritarian-personality theorists were right that punitiveness is just masked aggression (a dispute over manifest vs. latent functions), most people in everyday life insist there is—holding the physical acts of coercion constant—a real, not illusory, difference between being arrested and kidnapped—or between being murdered and executed. Unlike personal vendettas, punishment is subject to institutional constraints and rules. “Punishment” refers to penalties imposed in a ritualized fashion by legitimate third parties, and is so determined and enacted by human beings working within complex institutional structures. That said, there is also a good deal of evidence that people as intuitive prosecutors want to punish wrongdoers for more than the officially approved (in Western societies) reasons of specific and general deterrence. Most of us seem to harbor a deep-rooted preference to inflict pain on wrongdoers commensurate with the harm they have done to individual victims as well as to society as a whole (Carlsmith, Darley, & Robinson, 2002; Darley, Carlsmith, & Robinson, 2000). A social-functionalist analysis of the intuitive prosecutor therefore needs to grapple with the interplay between intrapsychic retributivist drives and institutional constraints. Tetlock et al.’s (2007) model of the intuitive prosecutor posits that people seek to defend social orders that they endow with legitimacy. The model builds on the evolutionary-psychological notion that there is a moralistic streak in human nature that predisposes people to value, as an end in itself,

276 PRAGMATIC THEORIES

the punishment of norm violators. The possibility of punishment increases cooperation and enables the emergence of stable cooperative equilibria in public goods games (Fowler, Johnson, & Smirnov, 2005)—crucial outcomes for small-group survival in states of nature. And despite the personal costs, people in an extraordinary range of societies often proactively seek to defend the social order against norm violators (Henrich et al., 2006; Wu et al., 2009). The three key assumptions of the model (from which hypotheses are derived) are: 1. The fairness postulate: Most people see themselves as fair-minded, anchor this self-image in the adherence to shared norms of fair play, and are roused to retributive wrath when others display contempt for these norms (Lerner, 1980; Miller & Vidmar, 1981). 2. The bias postulate: As creatures of bounded rationality with imperfect cognitive self-control, people often fall prey to judgmental biases that cause them either to overweight or underweight relevant criteria (Gilovich, Griffin, & Kahneman, 2002). 3. The self-correction postulate: When people catch themselves straying from their own private standards of good judgment, they try to correct themselves (and self-correction is easiest in repeated-measures settings that facilitate monitoring the cues they are using) (Petty, Wegener, & White, 1998). Blending these assumptions, the fair-but-biased-yet-correctible model (FBC; Tetlock, 2002; Tetlock et al., 2007) hypothesizes that observers shift into the prosecutorial mind-set to the degree they have been induced to believe, or were dispositionally predisposed to believe, that: (1) norm violations are widespread; (2) violations are intentional; (3) violators are flaunting their contempt for shared values; (4) violators are routinely escaping punishment; (5) the social order is legitimate; and (6) the norm violations offend shared moral values. Observers disengage from the mind-set when reassured that (1) norm violations have been punished by legitimate representatives of the collectivity; and (2) the goals of retribution, incapacitation, and general deterrence have been satisfied. Grounding these ideas in specific social settings forces us to think much harder than we normally would about boundary conditions. For instance, the model resolves the tension between the procedural-justice and punitiveness postulates by drawing on intrapsychic theories of motivated reasoning (Kunda, 1990) and positing that people will bend rules to reach desired conclusions (favoring one party over another) only when they feel they can do so by generating convincing justifications that they have not done so (linking to work on attributional ambiguity). Naked discrimination and nastiness is apparently unacceptable in most cultures wealthy enough to fund psychological research programs.



Social Functionalism 277

This notion helps to explain, among other things, the reasonably well-established phenomena of hyperpunitiveness and hypopunitiveness (Fincher & Tetlock, 2013a). Hyperpunitiveness arises, for example, when people declare that they have an in-principle aversion to corporal punishment of norm violators but they assign certain categories of norm violators to violent prisons as long as they can attribute their decisions to other factors (a phenomenon that it was possible to demonstrate by linking social-functionalist theory to an attributional ambiguity paradigm—Fincher & Tetlock, 2013a). An example of hypopunitiveness arises when people declare that they support basic principles of procedural justice but they selectively apply these principles to favor plaintiffs or defendants with whom they sympathize. We suggest that the tension between societal rules and individual casespecific information is often resolved by strategically creating attributional ambiguity that allows individuals to act in line with their punitiveness and forgiveness drives without directly violating impartiality (Fincher, Skitka, & Tetlock, 2013). Social-functionalist theorizing on intuitive prosecutors is far from complete, however. We know little about the cognitive processes, the “transistors” (to invoke a physical science metaphor) that enable efficient switching between mind-sets. For example, work on the intuitive prosecutor has almost completely ignored the deep response conflict between empathy and retributivism. On the one hand, a large body of both behavioral and neurological evidence suggests that we are social creatures who find it painful to see fellow human beings in pain—and are often willing to make substantial sacrifices to help them. On the other hand, people in the prosecutorial mindset become insensitive to the pain inflicted on norm violators—or perhaps even relish the pain. The question arises: How can people shift so readily into a harm–norm–violators mind-set, instead of being paralyzed by countervailing empathic responses or moral dissonance? To perform their norm-enforcement duties, intuitive prosecutors need a switching mechanism that allows them to move seamlessly between empathic and punitive mind-sets. We offer three examples of such bridging operations (the list is far from exhaustive but illustrates the types of careful conceptual and empirical work that needs to be done). 1.  It is one thing to document hypopunitiveness or hyperpunitiveness and quite another to clarify whether people genuinely believe that they are being faithful to their formally declared principles of justice (true dissonance reduction) or whether people are engaging in strategic impression management designed to conceal moral hypocrisy and thus protect their social identities in the eyes of various audiences. These competing interpretations can be tested using well-established techniques designed to convince research participants that the researchers have the wherewithal to detect their true attitudes (Fincher & Tetlock, 2013a). The true dissonance reducers should feel no need to back down from their prior stances.

278 PRAGMATIC THEORIES

2. One could argue that the distinction between self-deception and other-deception in punitiveness judgments does not matter much if the two are behaviorally equivalent. Linking social functionalism to another intrapsychic formulation, learning theory (Rescorla, 1988), raises the possibility, however, that the two are not behaviorally equivalent; it is more difficult for people to adjust their decisions in response to policy feedback when, as dissonance theory implies, they have so thoroughly rationalized their choices as to be unaware of the true drivers of those choices. Fincher and Tetlock (2013a) show that intuitive prosecutors who are also dissonance reducers are at great risk of slipping into punitiveness traps in which they fail to throttle back punitiveness when they are confronted with evidence of false-positive convictions of the innocent; instead they become all the more punitive when they are confronted with evidence of false-negative acquittals of the guilty. 3.  As noted earlier, social-functionalist theories of the intuitive prosecutor ultimately have to answer the question: How can people shift from being empathic intuitive politicians, eager to bond with their fellow citizens and ready to feel their pain, into punitive intuitive prosecutors? One candidate intrapsychic mechanism is that, once people classify an individual as a threat to the social order (the Durkheimian collective conscience), they automatically stop processing the person as an object or they perceptually “dehumanize” the target individual—at least to some degree. Presumably, the degree of dehumanization would be a function of the intensity of the retributive feelings activated by the norm violation. Objectification and dehumanization are much more likely for serial child killers than for shoplifters—but it may even occur to some degree for the shoplifter (a twinge of satisfaction at seeing a “cheater” or “thief” being led away in handcuffs for at least a brief incarceration timeout). The crucial bridging task is to specify exactly how varying degrees of objectification and dehumanization occur—and how to tap into these cognitive-affective transformations of our reactions to the target individual. Here, social-functionalist theorizing needs to become more psychologically nuanced than the earlier Tetlock formulations. Sadists aside, few people welcome the thought of an innocent citizen being subjected to the status-degradation ritual of being seized, shackled, and jailed (Goffman, 1959/2002). And few people welcome the thought of an actual shoplifter being seized by the owner of the store (with the implicit threat of lethal force if resisted) and incarcerated (that would smack of vigilante justice and feel more like a kidnapping than an arrest). It is only when a legitimate authority takes custody of a known norm violator that the functionalist preconditions for activating some degree of objectification and dehumanization of the target individual are satisfied. The hardest work, however, lies ahead. The social-functionalist theorist now needs an intrapsychic model that specifies not only when objectification



Social Functionalism 279

and dehumanization unfold (once functional activating conditions are satisfied), but also how this cognitive-affective transformation can be conceptualized and operationalized. We suspected, correctly, that few observers would be willing to acknowledge in a self-report survey (even if they had introspective access) that they view someone being arrested or even convicted as closer to an animal or physical object than as a fellow human being with conscious feelings deserving of respect. Here the hard bridge-building work arises. Social functionalists need to connect their frameworks with more intrapsychic frameworks for detecting far-from-fully-conscious cognitive-affective shifts of how we see each other. The key link proved to be to neuroscience and perceptual work on how people process information about the human face. In a series of studies, Fincher and Tetlock (2013b) have found evidence that dehumanization or objectification alters a fundamental perceptual process: the well-replicated facial-inversion effect, an effect that will require a digression. Neurocognitive researchers have long suspected that there is something special about the human face: brain-damaged patients show specific face recognition impairment (prosopagnosia); newborns preferentially orient toward stimuli with face-like first-order relations; babies react to facial distress; and certain regions of the brain are dedicated to facial pattern recognition. It turns out that specialized facial processing is particularly vulnerable to orientation effects. Many studies have observed a face-inversion effect: 180-degree rotations of faces impair recognition much more than 180-degree rotations of comparably complex objects—an effect often seen as support for the view that human beings process each other’s faces configurally (as wholes), not on a feature-by-feature basis. An inhibition of the face-inversion effect would thus suggest that individuals are perceptually no longer processing the face as human and are instead relying on other types of processing—the face is perceptually dehumanized. Fincher and Tetlock (2013b) repeatedly find that participants are more likely to show reduced face-inversion effects for the faces of norm violators— and that those who show reduced face-inversion effects are more supportive of imprisoning, beating, and even killing norm violators and members of threatening outgroups. This perceptual dehumanization suggests that socialfunctionalist mechanisms of norm enforcement—which enable cooperative equilibria inside groups—are deeply internalized in the individual psyche.

Concluding Thoughts Like many social functionalists, we sympathize with a recurring critique of experimental social psychology, one that has popped up over the last 60 years, from Solomon Asch (1952) to Paul Rozin (2001, 2006, 2009), with numerous variations in between (Gergen, 1973; Pepitone, 1976). In this view, social psychologists have skipped over the descriptive naturalistic phase

280 PRAGMATIC THEORIES

of research and moved too quickly into controlled laboratory paradigms. Systematic description is foundational in science. As Paul Rozin carefully documents, in biology both The Origin of Species (Darwin, 1859/2009) and The Molecular Structure of Nucleic Acids (Crick & Watson, 1954) are largely descriptive. The description of the adaptive radiation of animals and plants provided a basis for a theory of evolution. The descriptive studies of crossspecies regularities in nucleotides were key clues to discovering the molecular structure of DNA. In our view, psychology, like biology, should not be embarrassed to build on naturalistic observations. It is no accident that all advanced societies have specialized institutions that regulate and define ground rules for conducting political competition, exercising prosecutorial discretion and expressing theological commitments to transcendental values. Given how many times these institutional forms have been invented and reinvented— over thousands of years and across six continents—it strikes us as likely that these institutions are essential to life in complex collectivities. And it strikes us as a good idea to use these clues—offered in abundance by our colleagues in historical sociology, cultural anthropology, and comparative politics—in designing psychological theories of people as social beings. Social psychology is an interstitial field, wedged between the biological and social sciences, and social psychologists should feel as free to borrow concepts and findings from our macro colleagues as we do from our micro colleagues. References Aberle, D. F., Cohen, A. K., Davis, A. K., Levy, M. J., & Sutton, F. X. (1950). The functional prerequisites of a society. Ethics, 60, 100–111. Adorno, T. W., Frenkel-Brunswik, E., Levinson, D. J., & Sanford, R. N. (1950). The authoritarian personality. Oxford, UK: Harpers. Ainsworth, M. S. (1989). Attachments beyond infancy. American Psychologist, 44, 709–716. Asch, S. E. (1952). Social psychology. New York: Oxford University Press. Barash, D. P. (1977). Sociobiology and behavior. New York: Elsevier. Baumeister, R. F., & Leary, M. R. (1995). The need to belong: Desire for interpersonal attachments as a fundamental human motivation. Psychological Bulletin, 117, 497–529. Boring, E. G. (1953). A history of introspection. Psychological Bulletin, 50, 169–189. Boring, E. G., Watson, R. I., & Campbell, D. T. (1963). History, psychology, and science: Selected papers. Oxford, UK: Wiley. Carlsmith, K. M., Darley, J. M., & Robinson, P. H. (2002). Why do we punish? Deterrence and just deserts as motives for punishment. Journal of Personality and Social Psychology, 83, 284–299. Cosmides, L. (1989). The logic of social exchange: Has natural selection shaped how humans reason? Studies with the Wason selection task. Cognition, 31, 187–276. Crick, F. H., & Watson, J. D. (1954). The complementary structure of deoxyribonucleic acid. Proceedings of the Royal Society of London. Series A. Mathematical and Physical Sciences, 223, 80–96. Darley, J. M., Carlsmith, K. M., & Robinson, P. H. (2000). Incapacitation and just deserts as motives for punishment. Law and Human Behavior, 24, 659–683.



Social Functionalism 281

Darwin, C. (1991). On the origin of species by means of natural selection, 1859. London: Murray. Fincher, K. M., Skitka, L. J., & Tetlock, P. E. (2013). The dark side of procedural justice: Its power to mask both hyper- or hypo-punitiveness. Manuscript in preparation. Fincher, K. M., & Tetlock, P. E. (2013a). Brutality under cover of ambiguity: Covert retributivism and punitiveness traps. Manuscript in preparation. Fincher, K. M., & Tetlock, P. E. (2013b). Perceptual dehumanization of faces is activated by norm violations—and facilitates norm enforcement. Manuscript in preparation. Fiske, A. P. (2004). Four modes of constituting relationships: Consubstantial assimilation; space, magnitude, time and force; concrete procedures; abstract symbolism. In N. Haslam, (Ed.), Relational models theory: A contemporary overview (pp. 61–146). Mahwah, NJ: Erlbaum. Fowler, J. H., Johnson, T., & Smirnov, O. (2005). Egalitarian motive and altruistic punishment. Nature, 433. Gergen, K. J. (1973). Social psychology as history. Journal of Personality and Social Psychology, 26, 309. Gilovich, T., Griffin, D., & Kahneman, D. (Eds.) (2002). Heuristics and biases: The psychology of intuitive judgment. New York: Cambridge University Press. Goffman, E. (2002). The presentation of self in everyday life. New York: Garden City. (Original work published 1959) Greenberg, J., & Jonas, E. (2003). Psychological motives and political orientation—the left, the right, and the rigid: Comment on Jost et al. Psychological Bulletin, 129, 376–382. Henrich, J., McElreath, R., Barr, A., Ensminger, J., Barrett, C., Bolyanatz, A., et al. (2006). Costly punishment across human societies. Science, 312, 1767–1770. Hogan, R., & Blickle, G. (2013). Socioanalytic theory. In N. D. Christiansen & R. P. Tett (Eds.), Handbook of personality at work (pp. 53–70). New York: Routledge. Jost, J. T., Kay, A. C., & Thorisdottir, H. (2009). Social and psychological bases of ideology and system justification. New York: Oxford University Press. Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108, 480–498. Leary, M. R. (1990). Responses to social exclusion: Social anxiety, jealousy, loneliness, depression, and low self-esteem. Journal of Social and Clinical Psychology, 9, 221–229. Lerner, M. J. (1980). The belief in a just world: A fundamental delusion. New York: Plenum Press. Lerner, M. J., & Lerner, S. C. (Eds.). (1981). The justice motive in social behavior: Adapting to times of scarcity and change. New York: Plenum Press. McDougall, W. (1908). An introduction to social psychology. London: Methuen. McGuire, W. J. (1983). A contextualist theory of knowledge: Its implications for innovation and reform in psychological research. Advances in Experimental Social Psychology, 16, 1–47. Merton, R. K. (1968). Social theory and social structure. Glencoe, IL: Free Press. (Original work published 1949) Miller, D. T., & Vidmar, N. (1981). The social psychology of punishment reactions. Justice Motive in Social Behavior, 145, 172. Murray, H. A. (1938). Explorations in personality. Oxford, UK: Oxford University Press. Nisbett, R. E., Zukier, H., & Lemley, R. E. (1981). The dilution effect: Nondiagnostic information weakens the implications of diagnostic information. Cognitive Psychology, 13, 248–277. Pepitone, A. (1976). Toward a normative and comparative biocultural social psychology. Journal of Personality and Social Psychology, 34, 641. Petty, R. E., Wegener, D. T., & White, P. H. (1998). Flexible correction processes in social judgment: Implications for persuasion. Social Cognition, 16, 93–113.

282 PRAGMATIC THEORIES Rescorla, R. A. (1988). Pavlovian conditioning: It’s not what you think it is. American Psychologist, 43, 151–160. Rozin, P. (2001). Social psychology and science: Some lessons from Solomon Asch. Personality and Social Psychology Review, 5, 2–14. Rozin, P. (2006). Domain denigration and process preference in academic psychology. Perspectives on Psychological Science, 1, 365–376. Rozin, P. (2009). What kind of empirical research should we publish, fund, and reward? A different perspective. Perspectives on Psychological Science, 4, 435–439. Snyder, M. (1974). Self-monitoring of expressive behavior. Journal of Personality and Social Psychology, 30, 526–537. Tetlock, P. E. (2002). Social functionalist frameworks for judgment and choice: Intuitive politicians, theologians, and prosecutors. Psychological Review, 109, 451–471. Tetlock, P. E., Kristel, O. V., Elson, S. B., Green, M. C., & Lerner, J. S. (2000). The psychology of the unthinkable: Taboo trade-offs, forbidden base rates, and heretical counterfactuals. Journal of Personality and Social Psychology, 78, 853–870. Tetlock, P. E., Lerner, J. S., & Boettger, R. (1996). The dilution effect: Judgmental bias, conversational convention, or a bit of both? European Journal of Social Psychology, 26, 915–934. Tetlock, P. E., & Manstead, A. S. (1985). Impression management versus intrapsychic explanations in social psychology: A useful dichotomy? Psychological Review, 92, 59–77. Tetlock, P. E., Visser, P. S., Singh, R., Polifroni, M., Scott, A., Elson, S. B., et al. (2007). People as intuitive prosecutors: The impact of social-control goals on attributions of responsibility. Journal of Experimental Social Psychology, 43, 195–209. Wu, J. J., Zhang, B. Y., Zhou, Z. X., He, Q. Q., Zheng, X. D., Cressman, R., et al. (2009). Costly punishment does not always increase cooperation. Proceedings of the National Academy of Sciences USA, 106, 17448–17451.

14 Socially Situated Cognition Gün R. Semin Margarida V. Garrido

W

e live in a dynamically changing environment. In order to navigate such an environment successfully, we have to adapt to the constantly changing demands that we face. While our adaptive responses constitute one side of the coin of how we navigate our complex social and physical worlds, the other side of the coin is that we actively structure our physical and social environment to reduce its complexity and release cognitive resources. We achieve this in part by relying on others who provide us with scaffolds, given their specialized knowledge and competencies. However, we do not only rely only on others to reduce the complexity of our social and physical environment; we also download knowledge structures to the environment as in the case of street names. Thus, we reduce the complexity of our navigational tasks by relying on distributed knowledge, namely, knowledge distributed across people and objects, which we use as resources. Navigating our environment is not a limitless engagement. The biological constitution of our bodies puts limits on how we structure and process the dynamic reality surrounding us. In other words, our knowledge is embodied. An important aspect of this adaptive negotiation processes is the emergent nature of social cognition, namely, the situatedness of social cognition. The above overview encapsulates the essential elements of situated cognition, which suggests that cognitive processes emerge from adaptive senso 283

284 PRAGMATIC THEORIES

rimotor interactions with a dynamically changing social and physical environment and are grounded by the constraints of the human body and the environment (see Semin, Garrido, & Farias, 2014; Semin, Garrido, & Palma, 2012, 2013; Semin & Smith, 2002, 2013; Smith & Semin, 2004). This newly emerging dynamic perspective is referred to as socially situated cognition (SSC). The “situated cognition movement” started evolving in the 1990s as a reaction to the decontextualized view of human functioning. It should be noted at the onset that the SSC approach does not constitute a systematic theoretical framework. The approach advances a set of general principles that cut across numerous scientific disciplines and holds the promise of a unified perspective on human functioning (Robbins & Aydede, 2009). The origins of this approach go back to the mid-19th century (see William James, Lev Vygotsky, Frederic Bartlett, Margaret Mead, and John Dewey), but these origins were not pursued with any vigor as the second half of the last century saw the flourishing of the cognitive revolution which made the “mind” and “cognitive processes” the central research foci. The emerging fascination with computers narrowed the focus on human functioning even more to the isolated processing and representation of information. This remarkably narrow focus would concentrate scientific attention on processes within the narrow confines of the cranium at the expense of a richer and more realistic vision of cognition as contextually embedded, situated, and embodied action.

Overview We begin this chapter by drawing attention to the guiding framework of the SSC approach, which contrasts starkly with traditional approaches to human cognitive functioning, namely, the level of analysis that a SSC perspective introduces. The level of analysis in SSC is not at the phenomenon level but presents a framework about human functioning in general and incorporates a considerably broader perspective than is possible with a focus on the individual alone. Understanding human cognitive functioning means taking into consideration individuals in relationships in socially and physically connected environments that are in constant change. Within this umbrella framework there are specific metaphors or pillars of SSC as we refer to them. These pillars are emphases that are singled out in order to draw attention to neglected aspects of human functioning that have to be incorporated into theory and grounded with empirical tools. We list and summarize four important metaphors (for more details, see Semin et al., 2012, 2013; Semin & Smith, 2013). The first metaphor is the reconceptualization of the “social.” Traditionally, the object of analysis in social cognition involved concepts that had a social content (e.g., trait terms) rather than a “nonsocial” one (e.g., number, color). However, “social” is about two or more individuals in an interdepen-



Socially Situated Cognition 285

dent relationship and the products of this interdependence. It is not about detached concepts. Moreover, the social is about how a bridge between individuals is established rather than about static concepts and their representation. An important corollary of the SSC’s level of analysis is how the social is grounded. Notably, we refer to the issue as grounded rather than defined. Definitions are conceptual delineations, whereas grounding, as used here, draws attention to the processes by which the social is constituted. It is in that sense that we argue that social cognition is biologically grounded. We then turn to the three other guiding metaphors or pillars of SSC. The first is that cognition is grounded by sensorimotor experiences. This is the well-known embodiment argument, broadly defined by the idea that thinking is not driven exclusively by symbols but rather by multimodal states that activate the sensory-motor and affective systems. In short, embodiment posits that “the body” plays a central role in shaping the contents of the mind. The next metaphor is that cognition is for action. The final metaphor that we shall present is that cognition is supplemented by the aid of tools and the exploitation of others. This last metaphor or pillar of SSC states simply that cognition is distributed. The final section draws out the implications of this general framework and its guiding metaphors for explanation and theory construction as well as the construction of the experimental environment in which phenomena are investigated. One of the main lessons that can be derived from this framework is that informed progress in theory and research cannot be achieved by focusing on one of the metaphors alone, such as “cognition is embodied.” The SSC perspective is a package whereby the general level of analysis and the specific pillars of SSC have to be considered as a whole, rather than emphasizing the one or the other pillar and ignoring the implications of the rest. These points are illustrated in the concluding section where the evolution of two research areas are discussed— namely, facial mimicry of emotional expressions, and arm movement and affective compatibility effects.

The Conceptual Umbrella: Level of Analysis of Human Functioning The individual constituted a center stone for mainstream psychology, much as the pre-16th-century world viewed the earth as the orbital center of all celestial bodies. Similarly, the standard view of explaining human functioning (still) relies on the decontextualized individual as the unit of analysis. Moreover, it is assumed that explaining human functioning at the individual level will pave the way to grasping a more complex level, namely, emergent behavior. This view is clearly articulated, for instance, in classic arguments for social cognition (e.g., Devine, Hamilton, & Ostrom, 1994). Thus, the individual constituted the level of analysis, with the added corollary that cognition could be analyzed and understood as symbolic mental structures and processes along with a set of combinatorial “rules” (e.g., Fodor, 1980; Smith, 1998).

286 PRAGMATIC THEORIES

In the heliocentric view, the earth is in the context of the solar system with the sun at the center and the planets orbiting around it. Importantly, this view contrasted with the pre-16th-century geocentric view, which, as noted above, put the earth in the center of the solar system, by introducing a context within which the earth was placed, namely, the solar system. Analogically, the level of analysis introduced by the situated cognition perspective places cognitive phenomena as emergent products in the situated context of social interaction in a physical environment. Thus, the analysis of emergent phenomena is to be understood as being driven by the higher level of organization—that is, a situated context that shapes how the specific parts are interlinked and not vice versa—that is, how the individual or aspects of individual functioning shape the situated context. The essential idea here is that the higher level of organization that emerges as a composite enables us to understand the way the parts are functionally related to each other. The reverse is not possible. An analysis of the parts alone without knowing the emergent composite that defines the unique relationships of the parts does not advance us. The situated view of cognition contrasts with the traditional individual- and representation-centered focus in mainstream psychology. It means that the cognitive phenomena that require explanation, as well as the processes driving them, are the products of social interaction in specific contexts. Consequently, cognition can only be captured within a broader framework, or rather at a higher level of organization. But this means that the explanation of cognition is at a different level of analysis, rather than one that is locked in the cranial vault. (Semin et al., 2012, p. 638)

The contrasting views typifying the geocentric and the heliocentric perspectives can be simply put as follows: Focus on the individual and explain human functioning in terms of individual “invariants” versus understanding human functioning as situated and contextualized. Recent research developments attest to the gradual emergence of the situated view, providing a fuller account of specific phenomena. Essentially, the individualcentered view was concerned with identifying invariants of psychological functioning, namely, specific behaviors and their invariant meanings across contexts. More recently, there has been an increasing acknowledgment that behaviors are situated and their meanings vary according to the context within which they are manifested. This transition from the cross-situational and invariant vision of behavior to the socially situated perspective has been in the ascendance recently, and we will provide examples of this transition in the concluding section. One example that we will detail is how early work regarding the mimicking of facial movements relied on the assumption that behavioral mimicry relied on an automatic, matched motor response. What you saw (perception) was what you mimicked (motor action). This invariance assumption, shared by different fields in psychology (e.g., “affective



Socially Situated Cognition 287

compatibility” that has relied on a presumed invariant relationship between action and valence) has come under questioning in recent years (e.g., Hess & Fischer, 2013, in the case of emotional mimicry). This questioning across a range of behavioral phenomena resembles the transition from a geocentric point of view to the heliocentric one initiated by the Copernican revolution. The shift to a more situated and contextually driven understanding of phenomena does not necessarily proffer more accurate predictions of behavioral phenomena and human functioning, much like the Copernican revolution over the Ptolemaic geocentric view, as Kuhn (1962) argued. Nevertheless, it begins a more accurate characterization of how we navigate a dynamically changing social environment and pulls together some of the diverse parameters of this navigation process. In the following, we briefly summarize the four pillars of socially situated cognition that we have articulated recently (see Semin et al., 2012, 2013; Semin & Smith, 2013). It should be emphasized that the pillars of socially situated cognition outlined below are but specific emphases on aspects of human behavior that have been neglected in the practice of normal science in psychology. Thus, for instance, the embodiment argument, which states that the biological constitution of our bodies puts limits on how we structure and process the dynamic reality surrounding us, points to one specific issue of relevance in understanding human cognition. It is simply the argument that conceptualizing human cognition as exclusively representational, and thus symbolic, ignores the fact that cognition is also bodily grounded and not merely symbolic. While this emphasis and redirection of scientific attention that has recently captured the scholarly imagination is undoubtedly important, the research and argumentation under the exclusively embodiment banner (e.g., Barsalou, 1999, 2008) actually ignores the level of analysis umbrella advanced by the SSC perspective. Most (if not all) embodiment research is centered on the individual, and this seems to be a recurrent problem: that is, to analyze specific phenomena from an individual perspective and not a contextually situated perspective. This cautionary remark about one of the four pillars of SSC applies equally well to the other pillars. While the idea of a multipillar, holistic embodied/situational system is intuitively powerful in the abstract, the interactive nature of the components may run the risk of making testing it a highly challenging enterprise. In the last section we present some illustrations of how such research agenda can progress.

The Pillars of SSC: Metaphors to Guide Theory and Research The Biological Basis of Sociality In the classic work on social cognition, there was and is no difference between the processes driving cognition and social cognition. The difference was merely in the object of inquiry. In the one case, the objects of inquiry were not social, and in the other case they were social, such as traits. This

288 PRAGMATIC THEORIES

division based on the nature of the object of inquiry avoided a consideration of what it means to be social as well as the epistemic implications of such an analysis. The discovery of the mirror neuron system (see Iacoboni, 2009; Rizzolatti & Craighero, 2004) highlighted the biologically distributed nature of knowledge, driven by the architecture of the human perceptuomotor system designed for the reproduction of movements of conspecifics in a privileged way (see Buccino et al., 2004; Ferrari, Tramacere, Simpson, & Iriki, 2013). Critically, mirror neurons reveal congruence between observed and executed action with respect to the goals and to the means to achieve that goal. Moreover, this research brought to the fore the notion that there is a different ontological status of social knowledge compared to knowledge about the world in general. Social knowledge is underpinned by an isomorphism in mapping movements as a consequence of synchronization (Semin, 2007) processes. This points to the biologically “pre-grounded” nature of social knowledge in contrast to our knowledge of the object world. The latter knowledge is acquired through sensorimotor processes that are shaped by goal-directed interaction with it. Social cognition has therefore an entirely different epistemic status than cognition about the nonsocial world and has a generative neural basis that allows access to others’ states, movements, and actions. This capability is contextually adapted and activated (Caggiano et al., 2011, 2012; Fogassi et al., 2005; Yamazaki et al., 2010); it is also one that we carry out multimodally, namely, with all our senses (cf. Semin, 2007; Semin & de Groot, 2013). Thus, social cognition is specialized for a distinctive class of stimuli: humans in interaction, and not about trees or cars (Semin & Cacioppo, 2008, 2009). The biologically distributed nature of knowledge gives mutual access to and constitutes the foundations of communication (see Semin, 2007, for details). The take-away message of this section is that our bodies are involved in the way we represent others. In the next section we turn to a discussion of the role of the body in human mental functioning. The view is bolstered by an exponential growth in supportive evidence revealing that mental contents are often grounded in bodily experiences and that models that rely on symbolic, abstract, and amodal representations are not sufficient to account for such observations.

Embodiment A set of relatively invariable conditions (e.g., ecological, existential, material), including the architecture of our body, constrains our experience of the world and our psychological functioning in general. It is within these constraints that our actions and interactions take place, shaping the knowledge that we form, allowing us to navigate our social and physical world. Notably, our actions and interactions are sensorimotor experiences contributing to the embodied nature of human functioning. Thus, the meaning of an object or a person does not merely consist of some abstract set of features but is shaped



Socially Situated Cognition 289

largely by the nature of the actions or interactions with an object or a person (e.g., Gibson, 1966). These ideas are not new. Motor theories of perception, such as William James’s account of “ideomotor action” (1890/1950) or Jean Piaget’s developmental psychology, according to which cognitive abilities grew out from sensorimotor abilities, were predicated on precisely such a conception of knowledge. Another example is J. J. Gibson’s (e.g., 1966) ecological psychology. More recently, Clark’s work (e.g., 2008), Brooks’s contributions to robotics (e.g., 1999), and Prinz’s (e.g., 1984) common coding theory, according to which a shared representation or common code for perception and action exist, are well-worked-out examples of “embodiment.” What marks embodied approaches is their fundamental opposition to amodal approaches that conceptualize psychological functioning in terms of a closed loop of symbols or an internal model of the world. The extensive research on embodiment has shown that the body is not an output device for the computational processes on which cognition rests in representational models. Embodiment played a central role in early work in social psychology (Cacioppo, Priester, & Berntson, 1993; Strack, Martin, & Stepper, 1988; Valins, 1966; Wells & Petty, 1980), although the impressive and creative research that emerged from the 1960s onward did not have a common term identifying their embodied nature. This literature is extensively reviewed elsewhere (Semin et al., 2012, 2013; Semin, Garrido, & Farias, 2014). Recent research in social psychology and cognitive psychology has produced a wealth of information about the embodiment of concrete (e.g., Barsalou, 1999, 2008; Borghi, Glenberg, & Kaschak, 2004; Zwaan & Yaxley, 2003) and abstract concepts such as affection (e.g., Casasanto & Dijkstra, 2010; Crawford, 2009; Lakens, Semin, & Foroni, 2012; Meier & Robinson, 2004; Palma, Garrido, & Semin, 2011), power (e.g., Lakens, Semin, & Foroni, 2011; Schubert, 2005), time (e.g., Blom & Semin, 2013; Boroditsky, 2000; Lakens, Semin, & Garrido, 2011), or even politics (e.g., Farias, Garrido, & Semin, 2013; van Elk, van Schie, & Bekkering, 2010; see Semin et al., 2012, 2013, 2014, for reviews of both the relevant research literature and the theoretical controversies). The explosive growth of embodied cognition has taken place across different disciplinary perspectives ranging from cognitive psychology to social psychology, and from computational linguistics to neuropsychology and neurophysiology. The embodiment perspective is not without its critics. Some argue that the capacity to acquire semantic content that goes beyond perceptual experience reflects a fundamental design feature of human minds and that the human conceptual system is characterized by a representational division of labor in which modal and amodal representations handle different aspects of our concepts (Dove, 2009, 2011). Others claim that the evidence provided in support of embodied representations is also compatible with amodal representations (given certain auxiliary assumptions (e.g., Mahon & Caramazza, 2008); accordingly, they advance an alternative model suggesting that the core of a concept is amodal or symbolic and that sensorimotor information is an

290 PRAGMATIC THEORIES

embellishment that “colors conceptual processing, enriches it, and provides it with a relational context” (Mahon & Caramazza, 2008, p. 10). A similar proposal (language and situated simulation—LASS; Barsalou, Santos, Simmons, & Wilson, 2008) suggests that concepts are represented in terms of both linguistic representations and sensorimotor simulations. Their joint operation contributes to the representation of concepts, with the former, the linguistic system, assumed to induce superficial processing and the latter, deeper conceptual processing via the simulation system. According to Machery (2007), the critical question to be investigated should not revolve around whether concepts are modal or amodal, but rather the question should be: “To what extent do we use reenacted perceptual representations in cognition and to what extent do we use amodal representations?” (Machery, 2007, p. 42). Indeed, recent research (e.g., Farias et al., 2013) shows that the structure of a concept is likely to be reproduced in different modalities by which a concept can be represented, including its symbolic representation. The claim that an opposition exists between symbolic representational and modalityspecific representations is, in our view, misleading at best. We believe rather that representations of concepts are multimodal and inseparably interwoven with their linguistic representations.

Cognition Is for Adaptive Action This important pillar of SSC is highly relevant for appreciating the dynamic function of cognition as adaptive action in a constantly changing social and physical world. The socially situated perspective rejects the passive representational view of cognition and argues that the primary function of cognition is the control of adaptive action. This emphasis on cognition as adaptive action is detached from the standard representational or information processing paradigm of social cognition involving the construction and manipulation of inner representations that have no bearing on real interaction in and with the world. Modeling cognition in terms of abstract, detached symbolic representations has meant treating mental representations as invariant, timeless, and largely immune to contextual influences. Accordingly, enduring abstract mental structures are supposed to play a central role in attaining cognitive economy (e.g., Taylor, 1981). Relatedly, cognitive representations (e.g., attitudes and stereotypes) were assumed to be temporally invariant and relatively immune to contextual changes (e.g., Hamilton & Trolier, 1986; Snyder, 1981). Contrast this perspective with the dynamic view that cognition emerges in the interaction with a constantly changing social and physical environment (e.g., Semin & Smith, 2002). The assumption that mental representations resist contextual factors makes it very difficult to reconcile it with the fact that cognition needs to be adaptive to situational requirements. Obviously, offline cognition (Wilson, 2002) is important in a variety of situations involving, for instance, forward planning. But even this kind of cognition is situated in that it is a contextually simulated mental activity.



Socially Situated Cognition 291

Indeed, there is considerable research showing situational influences on cognitive processes. These studies have revealed the adaptive nature of cognition by highlighting the context sensitivity of mental “knowledge structures” and processes. For example, attitudes have been shown to be responsive to a multitude of contextual effects (e.g., Schwarz & Sudman, 1992), and stereotypes display considerable malleability in the face of changing contexts (e.g., Dovidio, Brigham, Johnson, & Gaertner, 1996). To illustrate, extensive research has long established that the accessibility of specific exemplars or group members affects category and subtype descriptions (e.g., Smith & Zárate, 1992) as well as central tendency and variability judgments about the group as a whole (Garcia-Marques & Mackie, 2001). Different members of a group can also apparently make stereotypes differentially accessible (e.g., Macrae, Mitchell, & Pendry, 2002). Stereotypes are sensitive to subtle contextual cues (e.g., Wittenbrink, Judd, & Park, 2001) and context stability (Garcia-Marques, Santos, & Mackie, 2006). Finally, recent research has also started to document the effects of the physical features of the environment on social-cognitive processes. For example, Williams and Bargh (2008) have shown that receiving information about a hypothetical person while holding a cup of warm or cold coffee led to distinct personality impressions of this hypothetical person on the warm–cold dimension. Similarly, Ijzerman and Semin (2009) observed that participants in a warmer room (relative to a colder room) reported higher social proximity to a target person. On the other hand, social exclusion situations led people to feel colder (Zhong & Leonardelli, 2008). Other physical features of the environment such as warm temperature, pleasant smell, or closer physical distance also triggered more positive impressions about a target or uninvolved others (see Semin & Garrido, 2012). Evidence for the adaptive function of cognition is illustrated not only by the sensitiveness and responsiveness of cognitive structures and processes to contextual circumstances but also by research highlighting the functions of those cognitive representations and processes—namely, as guides for action rather than internal states locked in the cranial vault. The action-oriented nature of representations is further underlined by research demonstrating how situated social motives (e.g., Pickett, Gardner, & Knowles, 2004; Sinclair & Kunda, 1999) and relationships (e.g., Neuberg & Fiske, 1987) with others shape mental representations and guide psychological, communicative, and behavioral processes.

Cognition Is Distributed The information that is contained in our physical and social environment shapes our actions and helps us navigate the world mentally and physically. We know where to turn left to find the location of a new shop because we can see the street names and numbers. We know that we have to stop when we are driving if we see a red light. We know who to ask how to best get to the train station in a new town. The artifacts (e.g., street signs) and experts (e.g.,

292 PRAGMATIC THEORIES

policeman) serve as external devices to orient ourselves (cf. Caporael, 1997). Thus, knowledge (cognition) is distributed across different tools or devices as well as different people, and these sources provide scaffolds for cognitive activity (see Clark, 1997; Hutchins, 1995). Indeed, as we use tools such as paper and pencil to perform a small calculation, we also structure our environment to aid others and ourselves, such as having different colored bins for the disposal of different wastes. We also make idiosyncratic use of spontaneous mnemonic devices by leaving an empty milk bottle near the front door in the evening so as to not forget to buy milk the next day (Kirsh, 1995). We also extend our cognitive processes beyond the individual by making use of knowledge distributed across other people and engaging them in the construction of interindividual mental representations. This is illustrated by Edwin Hutchins’s (1995) study of how to navigate a large vessel, where people draw on each other’s expertise and establish a type of knowledge that supersedes a single individual’s capabilities. The concept of transactive memory systems proposed by Dan Wegner and his colleagues (e.g., Wegner, 1986; Wegner, Erber, & Raymond, 1991) constitutes another example of such social scaffolding. According to this view and the research conducted using this perspective, individuals in close relationships develop a distributed memory system, such that they divide responsibility for the encoding, storage, and retrieval of information from different domains. Thus, individuals in close relationships jointly remember information better than do individuals who do not share a systematic relationship (e.g., Andersson & Rönnberg, 1995; Dixon & Gould, 1996; Hollingshead, 1998; Wegner, 1986; Wegner et al., 1991). Distributed cognition is a pervasive aspect of our daily social lives (e.g., Levine, Rensick, & Higgins, 1993) and regardless of whether the information is socially (e.g., Garcia-Marques, Garrido, Hamilton, & Ferreira, 2012; Garrido, Garcia-Marques, & Hamilton, 2012a, 2012b; Garrido, Garcia-Marques, Hamilton, & Ferreira, 2006, 2012; for a review, see Rajaran & Pereira-Pasarin, 2010) or physically distributed in terms of artifacts, they provide distinct cues that shape and determine the processes of our cognitive activities. Moreover, social and material scaffolds also serve the function of releasing cognitive space as well as extending one’s capabilities. Notably, such scaffolds constitute pervasive and integral aspects of our cognitive environment (see Palma, Garrido, & Semin, 2014) and often constitute unrealized influences of our cognitive activities demanding much more careful attention in our research.

Implications and Conclusions: From Invariant to Situated Behavior Meanings Research over a recent span of 50 to 60 years has witnessed a gradual transition from demonstrating and explaining the very same psychological phenomena in terms of their invariant regularities to contextualizing these



Socially Situated Cognition 293

phenomena due to their increasing malleability, as demonstrated with different experimental manipulations and conditions. The gradual emergence of a situated understanding of psychological phenomena from an original invariant view of the very same phenomena indicates a shift in paradigmatic thinking—much like Thomas Kuhn’s analysis (1962). In the concluding section of this chapter, we will illustrate this transition with two specific examples. Both examples have their origins in early research in sensorimotor bases of knowledge. The first example comes from the classic work on the relationship between valence and arm flexion and extension (e.g., Cacioppo et al., 1993; Priester, Cacioppo, & Petty, 1996). The second example is emotional mimicry, which is often regarded as a specific instantiation of behavioral or motor mimicry (e.g., Chartrand & Bargh, 1999; but see Hess & Fischer, 2013), namely, the imitation of the facial emotional display of another person.

Research on Affective Compatibility Early research that has led to diverse complementary research paradigms, and currently classified under titles such as affective compatibility, has relied on a presumed invariant relationship between action and valence. The first reported study was by Solarz (1960), who showed that participants were faster pulling cards with positive words toward themselves and pushing those with negative words than the reverse. In a highly cited study, Chen and Bargh (1999) showed that when participants classified positive words with an arm flexion movement as positive and used an arm extension movement to classify negative words as negative, they were faster than when doing the reverse classification movements. These studies suggested an invariant relationship between valence and arm flexion and extension. Indeed, research by Cacioppo et al. (1993) and Priester et al. (1996) mirrored these findings by showing that when people were exposed to novel stimuli they judged these novel stimuli as positive or negative if they were performing either isometric arm flexion or extension respectively. The message from this research was framed in terms of an invariant relationship between the type of muscle movement and evaluation (see also Neumann, Förster, & Strack, 2003). In this view, arm flexion is invariantly related to approach and arm extension to an avoidance orientation. These orientations are assumed to have motivational implications: Approach is positive and avoidance is negative. A logical extension of this type of theorizing is that “positive” or “negative” arm movements in association with a neutral stimulus lead to the neutral stimulus acquiring the corresponding valence (e.g., Cacioppo et al., 1993). Underlying this research is the assumption that arm flexion is part of a behavioral approach system. This is either innate or acquired through repeated associative linking (e.g., Neumann et al., 2003) and has become an integral part of an approach system. Similarly, arm extension is assumed to have become part of an avoidance system. Thus, associating a neutral stimulus with arm flexion or exten-

294 PRAGMATIC THEORIES

sion means that it acquires positive or negative valence. This association is mirrored by stimuli that are valenced. Valenced stimuli activate specific muscle movements, thus speeding up or slowing down their classification (e.g., Chen & Bargh, 1999). In the interim, the research literature has started to draw a more differentiated picture of this bidirectional link. The emerging view is that the meanings of the motor movements are construed online. It has been shown that the way the movement is mapped in space determines the motivational meaning of the movement. Thus, flexion can be approach and extension can be avoidance. However, if the mapping system is reversed, then the movements can acquire precisely the opposite meanings (e.g., Eder & Rothermund, 2008; Lavender & Hommel, 2007; Markman & Brendl, 2005; Rotteveel & Phaf, 2004). For instance, Markman and Brendl (2005) have shown that compatibility effects rely on how people represent their selves in space and not on their physical location. Thus, instead of understanding embodied phenomena as invariantly anchored, one needs to examine the complex interplay between perceptual and motor representations as well as how people represent their selves in space. Thus, the meaning of flexion and extension depends on the situation.

Research on Mimicry of Emotional Expressions There is a considerable amount of research on emotional mimicry: notably, the imitation of the emotional expressions of others. Early research on mimicry was guided by an invariance assumption, which stated that the facial display of a specific emotion is spontaneously copied in an unintended manner and occurs even when emotional faces are presented subliminally (Sonnby-Borgström, 2002) or when participants are instructed to suppress imitation of the facial expression they see (Dimberg, Thunberg, & Grunedal, 2002). Hess and Fischer (2013) refer to this phenomenon as the “Matched Motor Hypothesis.” In fact, two invariance assumptions drove this research. The first one was the matched motor hypothesis, which stated that there was an automatic, unintentional synchronization of emotional expression between the stimulus and the perceiver’s face. The stimulus was usually a photo, or sometimes a video (Dimberg, Thunberg, & Elmehed, 2000; Dimberg et al., 2002; Hess & Blairy, 2001; Sato, Fujimura, & Suzuki, 2008). The matching process was usually measured by electromyography (EMG; e.g., Dimberg et al., 2000). The second invariance assumption was largely due to Ekman’s argument that there are culturally universal facial expressions of emotions (Ekman, 1972; Ekman & Friesen, 1975). Over the years, the invariance assumptions have been called into question, in two ways. The first is an increasing number of studies illustrating the context sensitivity of mimicry. The second is research questioning whether mimicry of discrete emotions is a tenable assumption in the first place. The



Socially Situated Cognition 295

way a facial emotion is perceived is arguably dependent on situated multimodal contexts affecting how a face is perceived and what type of emotion is expressed (e.g., Barrett, Mesquita, & Gendron, 2011; Gendron, Lindquist, Barsalou, & Barrett, 2012; Hess & Fischer, 2013). Neither invariance assumption has stood the test of time, and much like the research on affective compatibility, has been shown to be subject to situated variation as a function of the contexts in which they acquire differing meanings. Firstly, different analyses (e.g., Hess & Fischer, 2013) have revealed that it is difficult, if not impossible, to sustain the matching hypothesis that speaks to EMG data that is assumed to support the matching of specific facial expressions with specific emotional states. These data can speak to valence in general but not to the specific emotion that is expressed in a stimulus. More importantly, situated context has been found to be an important determinant of whether or not any mimicry will occur. For instance, Van der Schalk and colleagues (2011) reported that when members of the outgroup expressed fear, then participants reacted with contempt, thereby not mimicking outgroup members’ fear. Similarly, Herrera, Bourgois, and Hess (1998) reported more smiling to outgroup fear, which was a function of negative attitude toward the outgroup. Indeed, considerable evidence exists that perceivers’ judgments of facial actions are influenced by descriptions of the social situation (e.g., Carroll & Russell, 1996). Indeed, such judgments are influenced by context, which includes voices, body postures, and visual scenes (e.g., Aviezer, Hassin, Bentin, & Trope, 2008; Righart & de Gelder, 2008; for a review, see de Gelder et al., 2006).

Closing Remarks The translation of a multipillar, holistic embodied/situational system into a feasible research agenda is one of the main challenges of the socially situated perspective. The claim that a research program should take each pillar seriously in formulating theoretical assumptions and designing research may be powerful but not necessarily feasible. Whereas all pillars need to be considered for a theoretically fully informed and integrated manner, progress can be and is being made by focusing explanatory and empirical efforts on particular pillars/subcomponents. Notably, the socially situated cognition perspective comes as a complete package and has to be applied as such. An exclusive focus on only one of its pillars (e.g., embodiment, distributed cognition) has potential pitfalls. The examples of affective compatibility or mimicry that we have outlined above illustrate the problems that beset our understanding of human functioning when we focus on one aspect of behavior alone at the expense of its situated and contextualized nature. Indeed, the direction in which these research areas (i.e., affective compatibility or mimicry) have evolved over time attests to a realization of this shortcoming and the increasing sensitivity to the situ-

296 PRAGMATIC THEORIES

ated nature of both phenomena, which has brought improved understanding and theorizing in these fields. As we mentioned earlier, SSC is not a theory. It consists of a set of assumptions or metaphors that are highlighted to redress the imbalance of the current conceptualization of human functioning. It is therefore not possible to say that SSC has predictive power. The metaphors underlying SSC as a whole are not testable. However, certain guiding metaphors, such as embodiment, have generated an enormous amount of research illustrating how bodily influences affect psychological functioning (cf. Semin & Smith, 2008). The bodily influences include examinations of facial expressions (Niedenthal, Mermillod, Maringer, & Hess, 2010; Strack, Martin, & Stepper, 1988), dynamic motor actions (Förster & Strack, 1997, 1998), and static postures (Stepper & Strack, 1993), as well as investigations of the influence of symbolic systems (language) on bodily action (e.g., such as facial expressions see Foroni & Semin, 2009) and neural activity (Pulvermüller, 2005). Unequivocally, these findings support the notion that mental contents are often grounded in bodily experiences and go beyond mere symbolic, abstract, and amodal representations. In short, they show that the mind needs the body to function. Although the volume of converging evidence is impressive, there is nevertheless an increasing awareness of the need for a more precise understanding of what the integrative notion of embodiment entails. Moreover, a further important step—showing how it is related to the other metaphors of SSC—has yet to be developed in a systematic way. In short, there is a lack of integration between the different metaphors driving SSC. In summary, the transition from an elementary approach and decontextualized symbolic cognition to a markedly contextual, dynamic, and systemic approach has its dangers. As Clancey has already mentioned (1997), one of the difficulties in articulating a situated view of cognition has been and continues to be an approach that invites a view of science that is embroiled with cultural relativism (Bruner, 1990; Slezak, 1989a). A view of cognition as infinitely malleable, distributed, and responsive to the physical and social context lacks predictive power. On the other hand, narrowing down the diverse constraints on human cognition, bodily, contextually, socially, and culturally as a systemic whole is an invitation toward an integrated science that is likely to pave the road to a higher level of analysis and explanation. References Andersson, J., & Rönnberg, J. (1995). Recall suffers from collaboration: Joint recall effects of friendship and task complexity. Applied Cognitive Psychology, 9, 199–211. Aviezer, H., Hassin, R., Bentin, S., & Trope, Y. (2008). Putting facial expressions into context. In N. Ambady & J. Skowronski (Eds.), First impressions (pp. 255–288). New York: Guilford Press.



Socially Situated Cognition 297

Barrett, L. F., Mesquita, B., & Gendron, M. (2011). Context in emotion perception. Current Directions in Psychological Science, 20, 286–290. Barsalou, L. W. (1999). Perceptual symbol systems. Behavioral and Brain Sciences, 22, 577– 660. Barsalou, L. W. (2008). Grounded cognition. Annual Review of Psychology, 59, 617–645. Barsalou, L. W., Santos, A., Simmons, W. K., & Wilson, C. D. (2008). Language and simulation in conceptual processing. In M. de Vega, A. M. Glenberg, & A. C. Graesser (Eds.), Symbols, embodiment, and meaning (pp. 245–283). Oxford, UK: Oxford University Press. Blom, S., & Semin, G. R. (2013). Moving events in time: Time referent hand–arm movements influence perceived temporal distance to past events. Journal of Experimental Psychology: General, 142, 319–322. Borghi, A. M., Glenberg, A. M., & Kaschak, M. P. (2004). Putting words in perspective. Memory and Cognition, 32, 863–873. Boroditsky, L. (2000). Metaphoric structuring: Understanding time through spatial metaphors. Cognition, 75, 1–28. Brooks, R. A. (1999). Cambrian intelligence. Cambridge, MA: MIT Press. Bruner, J. (1990). Acts of meaning. Cambridge, MA: Harvard University Press. Buccino, G., Lui, F., Canessa, N., Patteri, I., Lagravinese, G., Benuzzi, F., et al. (2004). Neural circuits involved in the recognition of actions performed by non-conspecifics: An fMRI study. Journal of Cognitive Neuroscience, 16, 114–126. Cacioppo, J. T., Priester, J. R., & Berntson, G. G. (1993). Rudimentary determinants of attitudes: II. Arm flexion and extension have differential effects on attitudes. Journal of Personality and Social Psychology, 65, 5–17. Caggiano, V., Fogassi, L., Rizzolatti, G., Casile, A., Giese, M. A., & Their, P. (2012). Mirror neurons encode the subjective value of an observed action. Proceeding of the National Academy of Sciences, USA, 109, 11848–11853. Caggiano, V., Fogassi, L., Rizzolatti, G., Pomper, J. K., Thier, P., Giese, M. A., et al. (2011). View-based encoding of actions in mirror neurons of area f5 in macaque premotor cortex. Current Biology, 21, 144–148. Caporael, L. (1997). The evolution of truly social cognition: The core configurations model. Personality and Social Psychology Review, 1, 276–298. Carroll, J. M., & Russell, J. A. (1996). Do facial expressions signal specific emotions? Judging emotion from the face in context. Journal of Personality and Social Psychology, 70, 205–218. Casasanto, D., & Dijkstra, K. (2010). Motor action and emotional memory. Cognition, 115, 179–185. Chartrand, T. L., & Bargh, J. A. (1999). The chameleon effect: The perception–behavior link and social interaction. Journal of Personality and Social Psychology, 76, 893–910. Chen, M., & Bargh, J. A. (1999). Consequences of automatic evaluation: Immediate behavioral predispositions to approach or avoid the stimulus. Personality and Social Psychology Bulletin, 25, 215–224. Clancey, W. J. (1997). Situated cognition: On human knowledge and computer representations. New York: Cambridge University Press. Clark, A. (1997). Being there: Putting brain, body and world together again. Cambridge, MA: MIT Press. Clark, A. (2008). Supersizing the mind. Embodiment, action and cognitive extension. Oxford, UK: Oxford University Press. Crawford, L. E. (2009). Conceptual metaphors of affect. Emotion Review, 1, 129–139. de Gelder, B., Meeren, H. K. M., Righart, R., Van Den Stock, J., Van De Riet, W. A. C., &

298 PRAGMATIC THEORIES Tamietto, M. (2006). Beyond the face: Exploring rapid influences of context on face processing. Progress in Brain Research, 155, 37–48. Devine, P. G., Hamilton, D. L., & Ostrom, T. M. (1994). Social cognition: Impact on social psychology. Orlando, FL: Academic Press. Dimberg, U., Thunberg, M., & Elmehed, K. (2000). Unconscious facial reactions to emotional facial expressions. Psychological Science, 11, 86–89. Dimberg, U., Thunberg, M., & Grunedal, S. (2002). Facial reactions to emotional stimuli: Automatically controlled emotional responses. Cognition and Emotion, 16, 449–472. Dixon, R., & Gould, O. (1996). Adults telling and retelling stories collaboratively. In P. B. Baltes & U. M. Staudinger (Eds.), Interactive minds: Life-span perspectives on the social foundation of cognition (pp. 221–241). New York: Cambridge University Press. Dove, G. (2009). Beyond perceptual symbols: A call for representational pluralism. Cognition, 110, 412–431. Dove, G. (2011). On the need for embodied and dis-embodied cognition. Frontiers in Psychology, 1, 1–13. Dovidio, J. F., Brigham, J. C., Johnson, B. T., & Gaertner, S. L. (1996). Stereotyping, prejudice, and discrimination: Another look. In C. N. Macrae, C. Stangor, & M. Hewstone (Eds.), Stereotypes and stereotyping (pp. 276–319). New York: Guilford Press. Eder, A. B., & Rothermund, K. (2008). When do motor behaviors (mis)match affective stimuli? An evaluative coding view of approach and avoidance reactions. Journal of Experimental Psychology: General, 137, 262–281. Ekman, P. (1972). Universals and cultural differences in facial expressions of emotion. In J. Cole (Ed.), Nebraska Symposium on Motivation (Vol. 19, pp. 207–283). Lincoln: University of Nebraska Press. Ekman, P., & Friesen, W. V. (1975). Unmasking the face: A guide to recognizing emotions from facial clues. Englewood Cliffs, NJ: Prentice Hall. Farias, A. R., Garrido, M. V., & Semin, G. R. (2013). Converging modalities ground abstract categories: The case of politics. PLoS ONE, 8(4), e60971. Ferrari, P. F., Tramacere, A., Simpson, E. A., & Iriki, A. (2013). Mirror neurons through the lens of epigenetics. Trends in Cognitive Sciences, 17, 450–457. Fodor, J. A. (1980). Methodological solipsism considered as a research strategy in cognitive psychology. Behavioral and Brain Sciences, 3, 63–109. Fogassi, L., Ferrari, P. F., Gesierich, B., Rozzi, S., Chersi, F., & Rizzolatti, G. (2005). Parietal lobe: From action organization to intention understanding. Science, 308, 662–667. Foroni, F., & Semin, G. R. (2009). Language that puts you in touch with your bodily feelings: The multimodal responsiveness of affective expressions. Psychological Science, 20, 974–980. Förster, J., & Strack, F. (1997). The influence of motor actions on retrieval of valenced information: A motor congruence effect. Perceptual and Motor Skills, 85, 1419–1427. Förster, J., & Strack, F. (1998). Motor actions in retrieval of valenced information: II. Boundary conditions for motor congruence effects. Perceptual and Motor Skills, 86, 1423–1426. Garcia-Marques, L., Garrido, M. V., Hamilton, D. L., & Ferreira, M. (2012). Effects of correspondence between encoding and retrieval organization in social memory. Journal of Experimental Social Psychology, 48, 200–206. Garcia-Marques, L., & Mackie, D. M. (2001). Not all stereotype-incongruent information is created equal: The impact of sample variability on stereotype change. Group Processes and Intergroup Relations, 4, 5–20. Garcia-Marques, L., Santos, A. S., & Mackie, D. M. (2006). Stereotypes: Static abstractions or dynamic knowledge structures? Journal of Personality and Social Psychology, 91, 814–831. Garrido, M. V., Garcia-Marques, L., & Hamilton, D. L. (2012a). Hard to recall but easy



Socially Situated Cognition 299

to judge: Retrieval strategies in social information processing. Social Cognition, 30, 57–71. Garrido, M. V., Garcia-Marques, L., & Hamilton, D. L. (2012b). Enhancing the comparability between part-list cueing and collaborative recall: A gradual part-list cueing paradigm. Experimental Psychology, 59, 199–205. Garrido, M. V., Garcia-Marques, L., Hamilton, D., & Ferreira, M. (2006). Collaborative impression formation and recall of impression-relevant information. In R. Sun & N. Miyake (Eds.), Proceedings of the 28th annual conference of Cognitive Science Society (pp. 1358–1363). Mahwah, NJ: Erlbaum. Garrido, M. V., Garcia-Marques, L., Hamilton, D. L., & Ferreira, M. (2012). Person memory: A matter of non-collaboration? In N. Payette & B. Hardy-Vallée (Ed.), Connected minds: Cognition and interaction in the social world (pp. 134–145). Newcastle upon Tyne, UK: Cambridge Scholars. Gendron, M., Lindquist, K. A., Barsalou, L. W., & Barrett, L. F. (2012). Emotion words shape emotion percepts. Emotion, 12, 314–325. Gibson, J. J. (1966). The senses considered as perceptual systems. Boston: Houghton Mifflin. Hamilton, D. L., & Trolier, T. K. (1986). Stereotypes and stereotyping: An overview of the cognitive approach. In J. Dovidio & S. L. Gaertner (Eds.), Prejudice, discrimination, and racism (pp. 127–163). New York: Academic Press. Herrera, P., Bourgois, P., & Hess, U. (1998, September). Counter mimicry effects as a function of racial attitudes. Paper presented at the 38th annual meeting of the Society for Psychophysiological Research, Denver, CO. Hess, U., & Blairy, S. (2001). Facial mimicry and emotional contagion to dynamic emotional facial expressions and their influence on decoding accuracy. International Journal of Psychophysiology, 40, 129–141. Hess, U., & Fischer, A. (2013). Emotional mimicry as social regulation. Personality and Social Psychology Review, 17, 142–157. Hollingshead, A. B. (1998). Retrieval processes in transactive memory systems. Journal of Personality and Social Psychology, 74, 659–671. Hutchins, E. (1995). Cognition in the wild. Cambridge, MA: MIT Press. Iacoboni, M. (2009). Imitation, empathy, and mirror neurons. Annual Review of Psychology, 60, 653–670. Ijzerman, H., & Semin, G. R. (2009). The thermometer of social relations: Mapping social proximity on temperature. Psychological Science, 20, 1214–1220. James, W. (1890). The principles of psychology. (Reprinted 1950). New York: Dover. Kirsh, D. (1995). The intelligent use of space. Artificial Intelligence, 73, 31–68. Kuhn, T. S. (1962). The structure of scientific revolutions. Chicago: University of Chicago Press. Lakens, D., Semin, G. R., & Foroni, F. (2011). Why your highness needs the people: Comparing the absolute and relative representation of power in vertical space. Social Psychology, 43, 205–215. Lakens, L., Semin, G. R., & Foroni, F. (2012). But for the bad, there would not be good. Conceptual opposition vs. mere association: Metaphoric grounding of valence. Journal of Experimental Psychology: General, 141, 584–594. Lakens, L., Semin, G. R., & Garrido, M. V. (2011). The sound of time: Cross-modal convergence in the spatial structuring of time. Consciousness and Cognition, 20, 437–443. Lavender, T., & Hommel, B. (2007). Affect and action: Towards an event-coding account. Cognition and Emotion, 21, 1270–1296. Levine, J., Rensick, L., & Higgins, E. (1993). Social foundations of cognition. Annual Review of Psychology, 44, 588–612.

300 PRAGMATIC THEORIES Machery, E. (2007). Concept empiricism: A methodological critique. Cognition, 104, 19–46. Macrae, C. N., Mitchell, J. P., & Pendry, L. F. (2002). What’s in a forename? Cue familiarity and stereotypical thinking. Journal of Personality and Social Psychology, 38, 186–193. Mahon, B. Z., & Caramazza, A. (2008). A critical look at the embodied cognition hypothesis and a new proposal for groundal conceptual content. Journal of Physiology, 102, 59–70. Markman, A. B., & Brendl, C. M. (2005). Constraining theories of embodied cognition. Psychological Science, 16, 6–10. Meier, B. P., & Robinson, M. D. (2004). Why the sunny side is up. Psychological Science, 15, 243–247. Neuberg, S. L., & Fiske, S. T. (1987). Motivational influences on impression formation: Outcome dependency, accuracy-driven attention, and individuating processes. Journal of Personality and Social Psychology, 53, 431–444. Neumann, R., Förster, J., & Strack, F. (2003). Motor compatibility: The bidirectional link between behavior and evaluation. In J. Musch & K. C. Klauer (Eds.), The psychology of evaluation: Affective processes in cognition and emotion (pp. 371–391). Mahwah, NJ: Erlbaum. Niedenthal, P. M., Mermillod, M., Maringer, M., & Hess, U. (2010). The simulation of smiles (SIMS) model: Embodied simulation and the meaning of facial expression. Behavioral and Brain Sciences, 33, 417–480. Palma, T. A., Garrido, M. V., & Semin, G. R. (2011). Grounding person memory in space: Does spatial anchoring of behaviors improve recall? European Journal of Social Psychology, 41, 275–280. Palma, T. A., Garrido, M. V., & Semin, G. R. (2014). Situating person memory: The role of the visual context on memory for behavioral information. Journal of Experimental Social Psychology, 52, 32–43. Pickett, C. L., Gardner, W. L., & Knowles, M. (2004) Getting a cue: The need to belong and enhanced sensitivity to social cues. Personality and Social Psychology Bulletin, 30, 1095–1107. Priester, J. R., Cacioppo, J. T., & Petty, R. E. (1996). The influence of motor processes on attitudes toward novel versus familiar semantic stimuli. Personality and Social Psychology Bulletin, 22, 442–447. Prinz, W. (1984). Modes of linkage between perception and action. In W. Prinz & A.-F. Sanders (Eds.), Cognition and motor processes (pp. 185–193). Berlin: Springer. Pulvermüller, F. (2005). Brain mechanisms linking language and action. Nature Reviews Neuroscience, 6, 576–582. Rajaran, S., & Pereira-Pasarin, L. P. (2010). Collaborative memory: Cognitive research and theory. Perspectives on Psychological Science, 5, 649–663. Righart, R., & de Gelder, B. (2008). Recognition of facial expressions is influenced by emotional scene gist. Cognitive, Affective, and Behavioral Neuroscience, 8, 264–272. Rizzolatti, G., & Craighero, L. (2004). The mirror–neuron system. Annual Review of Neuroscience, 27, 169–192. Robbins, P., & Aydede, M. (Eds.). (2009). The Cambridge handbook of situated cognition. Cambridge, UK: Cambridge University Press. Rotteveel, M., & Phaf, R. H. (2004). Automatic affective evaluation does not automatically predispose for arm flexion and extension. Emotion, 4, 156–172. Sato, W., Fujimura, T., & Suzuki, N. (2008). Enhanced facial EMG activity in response to dynamic facial expressions. International Journal of Psychophysiology, 70, 70–74. Schubert, T. W. (2005). Your highness: Vertical positions as perceptual symbols of power. Journal of Personality and Social Psychology, 89, 1–21.



Socially Situated Cognition 301

Schwarz, N., & Sudman, S. (Eds.). (1992). Context effects in social and psychological research. New York: Springer Verlag. Semin, G. R. (2007). Grounding communication: Synchrony. In A. Kruglanski & E. T. Higgins (Eds.), Social psychology: Handbook of basic principles (2nd ed., pp. 630–649). New York: Guilford Press. Semin, G. R., & Cacioppo, J. T. (2008). Grounding social cognition: Synchronization, entrainment, and coordination. In G. R. Semin & E. R. Smith (Eds.), Embodied grounding: Social, cognitive, affective, and neuroscientific approaches (pp. 119–147). New York: Cambridge University Press. Semin, G. R., & Cacioppo, J. T. (2009). From embodied representation to co-regulation. In J. A. Pineda (Ed.), Mirror neuron systems: The role of mirroring processes in social cognition (pp. 107–120). Totowa, NJ: Humana Press. Semin, G. R., & de Groot, J. H. B. (2013). The chemical bases of human sociality. Trends in Cognitive Sciences, 17, 427–429. Semin, G. R., & Garrido, M. V. (2012). A systemic approach to impression formation: From verbal to multimodal processes. In J. Forgas, K. Fiedler, & C. Sedikides (Eds.), Social thinking and interpersonal behavior (pp. 81–100). New York: Psychology Press. Semin, G. R., Garrido, M. V., & Farias, A. R. (2014). How many processes does it take to ground a concept? In J. W. Sherman, B. Gawronski, & Y. Trope (Eds.), Dual process theories of the social mind (pp. 542–559). New York: Guilford Press. Semin, G. R., Garrido, M. V., & Palma, T. A. (2012). Socially situated cognition: Recasting social cognition as an emergent phenomenon. In S. Fiske & N. Macrae (Eds.), Sage handbook of social cognition (pp. 143–169). Seven Oaks, CA: Sage. Semin, G. R., Garrido, M. V., & Palma, T. A. (2013). Interfacing body, mind, the physical, and social world: Socially situated cognition. In D. E. Carlston (Ed.), The Oxford handbook of social cognition (pp. 637–655). New York: Oxford University Press. Semin, G. R., & Smith, E. R. (2002). Interfaces of social psychology with situated and embodied cognition. Cognitive Systems Research, 3, 385–396. Semin, G. R., & Smith, E. R. (2008). Embodied grounding: Social, cognitive, affective, and neuroscientific approaches. New York: Cambridge University Press. Semin, G. R., & Smith, E. (2013). Socially situated cognition in perspective. Social Cognition, 31, 125–146. Sinclair, L., & Kunda, Z. (1999). Reactions to a black professional: Motivated inhibition and activation of conflicting stereotypes. Journal of Personality and Social Psychology, 77, 885–904. Slezak, P. (1989a). How NOT to naturalize the theory of action. In P. Slezak & W. R. Albury (Eds.), Computers, brains and minds: Essays in cognitive science. Dordrecht, The Netherlands: Kluwer. Slezak, P. (1989b). Man NOT a subject for science? In W. R. Albury & P. Slezak (Eds.), Dimensions of cognitive science. Kensington Studies, Centre for Cognitive Science, University of New South Wales. Smith, E. R. (1998). Mental representation and memory. In D. Gilbert, S. Fiske, & G. Lindzey (Eds.), Handbook of social psychology (4th ed., Vol. 1, pp. 391–445). New York: McGraw-Hill. Smith, E. R., & Semin, G. R. (2004). Socially situated cognition: Cognition in its social context. Advances in Experimental Social Psychology, 36, 53–117. Smith, E. R., & Zárate, M. A. (1992). Exemplar based model of social judgment. Psychological Review, 99, 3–21. Snyder, M. (1981). On the self-perpetuating nature of social stereotypes. In D. L. Hamilton (Ed.), Cognitive processes in stereotyping and intergroup behavior (pp. 183–212). Hillsdale, NJ: Erlbaum.

302 PRAGMATIC THEORIES Solarz, A. K. (1960). Latency of instrumental responses as a function of compatibility with the meaning of eliciting verbal signs. Journal of Experimental Psychology, 59, 239–245. Sonnby-Borgström, M. (2002). Automatic mimicry reactions as related to differences in emotional empathy. Scandinavian Journal of Psychology, 43, 433–443. Stepper, S., & Strack, F. (1993). Proprioceptive determinants of emotional and nonemotional feelings. Journal of Personality and Social Psychology, 64, 211–220. Strack, F., Martin, L. L., & Stepper, S. (1988). Inhibiting and facilitating conditions of the human smile: A nonobtrusive test of the facial feedback hypothesis. Journal of Personality and Social Psychology, 54, 768–777. Taylor, S. E. (1981). A categorization approach to stereotyping. In D. L. Hamilton (Ed.), Cognitive processes in stereotyping and intergroup behavior (pp. 83–114). Hillsdale, NJ: Erlbaum. Valins, S. (1966). Cognitive effects of false heart-rate feedback. Journal of Personality and Social Psychology, 4, 400–408. Van der Schalk, J., Fischer, A. H., Doosje, B. J., Wigboldus, D., Hawk, S. T., Hess, U., et al. (2011). Congruent and incongruent responses to emotional displays of ingroup and outgroup. Emotion, 11, 286–298. van Elk, M., van Schie, H. T., & Bekkering, H. (2010). From left to right: Processing acronyms referring to names of political parties activates spatial associations. Quarterly Journal of Experimental Psychology, 63, 22202–22219. Wegner, D. M. (1986). Transactive memory: A contemporary analysis of the group mind. In B. Mullen & G. R. Goethals (Eds.), Theories of group behavior (pp. 185–208). New York: Springer-Verlag. Wegner, D. M., Erber, R., & Raymond, P. (1991). Transactive memory in close relationships. Journal of Personality and Social Psychology, 61, 923–929. Wells, G. L., & Petty, R. E. (1980). The effects of head movement on persuasion: Compatibility and incompatibility of responses. Basic and Applied Social Psychology, 1, 219–230. Williams, L. E., & Bargh, J. A. (2008). Experiencing physical warmth promotes interpersonal warmth. Science, 322, 606–607. Wilson, M. (2002). Six views of embodied cognition. Psychonomic Bulletin and Review, 9, 625–636. Wittenbrink, W., Judd, C. M., & Park, B. (2001). Spontaneous prejudice in context: Variability in automatically activated attitudes. Journal of Personality and Social Psychology, 81, 815–827. Yamazaki, Y., Yokochi, H., Tanaka, M., Okanoya, K., & Iriki, A. (2010). Potential role of monkey inferior parietal neurons coding action semantic equivalences as precursors of parts of speech. Social Neuroscience, 5, 105–117. Zhong, C. B., & Leonardelli, G. J. (2008). Cold and lonely: Does social exclusion literally feel cold? Psychological Science, 19, 838–842. Zwaan, R. A., & Yaxley, R. H. (2003). Spatial iconicity affects semantic-relatedness judgments. Psychonomic Bulletin and Review, 10, 954–958.

Part V Social Theories





15 Interdependence Theory and Related Theories Harry T. Reis Ximena B. Arriaga

Analogous to contemporary physics—where the relations between particles are as meaningful as the particles themselves—in interdependence theory, between-person relations are as meaningful as the individuals themselves.    —Caryl Rusbult and Paul Van Lange     (2008, pp.  2049–2050)

S

andor and Ellis work on the salesforce of a large medical supply company. They must decide whether to cooperate or compete with each other as they pursue customers. Shannon and Pat having been dating for about 2 months. As their relationship develops, they each consider whether to deepen their commitment or preserve their independence. Eamon and Kyoko, proud parents of a newborn baby girl, must allocate responsibility for middle-of-the-night feedings. Who will get up and who will sleep? Players and owners from a professional sports league are engaged in negotiations for a new contract to cover the coming season. The season will take place only if they come to an agreement, yet each side feels that the other is not negotiating in good faith. Melissa and Patrick are considering a divorce. Melissa is employed in a high-paying job, whereas Patrick has been a stayat-home dad and has few prospects for finding a good job. Does each person 305

306

SOCIAL THEORIES

persist in trying to save the marriage or give up? Three college friends agree to form a start-up Internet business upon graduating. It will take several years of long hours and living on the cheap to determine whether the company will be profitable, but it can succeed only if all of them persevere and forgo more secure alternative jobs. Do they remain committed or do they take the first opportunity to cash out? A patient resists a much-needed treatment plan, prompting his physician to suggest less effective alternatives that will elicit greater adherence. How will the physician and patient negotiate a treatment plan that will help the patient while avoiding wasted effort on the part of the physician? Each of these examples describes an interdependence dilemma: The outcomes that each person will receive depend not only on what both of them do but also on the structure of the situation they face. Interdependence theory was first proposed by Thibaut and Kelley (1959) to help explain how people represent and consider situations of interdependence with respect to choosing among potential courses of action. The theory also considers how cognition and emotion reflect people’s analysis of these interdependent situations, as well as the role of individual differences in people’s response to various situational contingencies. Most importantly, interdependence theory is unique among social-psychological theories in addressing questions about how interacting people influence each other’s preferences, motives, and actions. Interdependence theory is therefore one of social psychology’s most far-reaching and comprehensive theoretical statements. It is also arguably the field’s most social theory. We see interdependence theory as a quintessentially social-psychological theory because its analysis of situations examines their interpersonal and interactional components. Whereas social psychology has justifiably championed itself as the science of situations (Ross & Nisbett, 1991), to a large extent it has emphasized the impersonal features of situations. In life, however, the distinctive properties of situations often (and perhaps more often than not) reflect their interpersonal aspects: who one is with, and who affects or is affected by one’s behavior (Kelley, 2000; Reis, 2008; Reis & Holmes, 2012). In fact, the social dimension of situations was the driving interest of many of the field’s earliest thinkers, such as Ross (1908), Lewin (1943), and Allport (1960), leading Kelley to assert that “the proper study of social psychology is the study of interaction and its immediate determinants and consequences” (2000, p. 11). Interdependence theory therefore has great potential to refocus the field’s attention on what is truly social about situations and in turn elucidate how situations influence social behavior. Such a refocus seems to us likely to enhance the explanatory power of social psychology. Thorngate (1975) opined that because no theory could be simultaneously general, accurate, and simple (in the sense of parsimonious), all theories must sacrifice one of these. In the case of interdependence theory, the odd factor out is simplicity. Interdependence theory is based on a highly abstract analysis of the interpersonal dimensions of situations and, as might be expected,

Interdependence Theory and Related Theories 307

is every bit as complex as the multifaceted interpersonal world that it seeks to represent. Indeed, novices to the theory are sometimes astounded by the complexity of the constructs needed to represent judgments that ordinary people make easily. Interdependence theory postulates are often portrayed through the tool of payoff matrices, borrowed from early game theorists such as Luce and Raiffa (1957). Although these matrices provide a degree of logical precision that is essential to a deep understanding of the theory, they sometimes deter casual readers from appreciating the clarity and richness of its analysis and predictions. In this chapter, we therefore eschew matrix representations in favor of a verbal analysis of the theory’s main concepts, assumptions, and goals. Consistent with the theme of this volume, our chapter does not directly review interdependence theory. Readers interested in a more forthright account of the theory might consult any of several excellent sources. Interdependence theory was first proposed by John Thibaut and Harold Kelley in their seminal volume, The Social Psychology of Groups (1959). That volume reflected the authors’ goal of capturing Kurt Lewin’s idea, which had not been fleshed out in field theory, that interdependence was “the essence of a group” (Lewin, 1948, p. 84). The theory was later revised by Kelley and Thibaut (1978) and then applied to dyadic relationships by Kelley (1979). More recently, Kelley et al. (2003) used interdependence theory concepts to describe and analyze 20 of the most common interpersonal situations. Excellent reviews can be found in Rusbult and Arriaga (1997), Rusbult and Van Lange (1996), and Van Lange and Rusbult (2012). By focusing on interdependence theory, we limit our discussion to those theories that use or derive from the theoretical framework first articulated by Thibaut and Kelley (1959). We have little doubt that this theory offers the field’s most prototypic, best articulated, and most influential theory of interdependence processes. Our discussion is based both on their original work and its subsequent elaborations, as well as on other theoretical models that build directly on this approach. Of course, many other social-psychological theories describe ways in which people are influenced by the presence or behavior of others, which are surely examples of, literally, “interdependence.” However, not all theories about the influence of others make use of specific interdependence theory processes or mechanisms, so we focus our analysis on interdependence theory itself and models that most directly reflect that framework. This chapter begins by reviewing certain key metatheoretical assumptions that define interdependence theory. These assumptions should help readers comprehend the basic principles behind interdependence theory’s analysis of interaction and interdependence, as well as its place within the family of social-psychological theories. We next discuss key structural features of interdependence theory as well as its explanatory and predictive value, followed by a review of several of the theory’s more noteworthy advantages and disadvantages, an accounting that is admittedly selective

308

SOCIAL THEORIES

and to some extent biased by our own perspective. We hope that readers will come away from this chapter with an appreciation for the unique contribution that interdependence theory offers to theorizing in social and personality psychology.

The Metatheoretical Framework of Interdependence Theory Interdependence theory begins with two key assumptions. The first assumption is that the most important features of situations are found in their interpersonal core. This emphasis, which derives from Lewin’s (1943) account of the E term in his famous formulation, B = f(P, E)—namely, that behavior is a function of the person and the environment—is consistent with widespread recognition of the centrality of social relations to human activity and wellbeing (Reis, Collins, & Berscheid, 2000). It also converges with contemporary interest in the adaptive significance of interpersonal activity and connections in human evolution (Buss & Kenrick, 1998; see also Ketelaar, Chapter 11, this volume). The second assumption is that the fundamental ground for an analysis of social situations is the study of social interactions, which are best understood by considering people’s interdependence with respect to outcomes— namely, the nature and extent to which co-acting individuals or groups are dependent on each other in order to attain desired outcomes. This assumption follows from the functional analysis inherent in Thibaut and Kelley’s (1959) original work: that groups exist to help people solve problems more effectively than they could do alone. In other words, because “socially significant behavior will not be repeated unless it is reinforced, rewarded in some way” (Thibaut & Kelley, 1959, p. 5), analysis of the manner in which groups go about solving problems taking into account the facts of their interdependence represents a useful way to comprehend social behavior. This is accomplished by examining the abstract properties of outcome interdependence in the various situations that groups (or, more typically, dyads) encounter. A marked difference between this approach and many other socialpsychological theories is evident in these assumptions. Whereas other theories often begin in the mind of the individual, with his or her needs, goals, or mental representations, interdependence theory starts with the objective situation itself—how two or more individuals (or two groups) are linked interdependently with respect to possible actions and desired outcomes. Thus interpersonal interactions, rather than individuals, become the fundamental units of analysis. This origin has considerable implications for the way in which motives, cognitive processes, and individual differences are conceptualized, as discussed below. Interdependence theory characterizes interpersonal situations according

Interdependence Theory and Related Theories 309

to six properties that define patterns of interdependence, each of which has clear implications for the choices interacting individuals make: the degree to which individuals’ interests correspond or conflict; the extent to which outcomes depend on one’s own actions or the actions of others; whether co-actors have mutual or asymmetric power over each other’s outcomes; whether the task requires coordination or exchange to produce a desired result; whether the situation involves interaction over time; and whether or not substantial uncertainty exists about the likely result of particular actions. These properties can be combined to portray common or prototypical situations, akin to a periodic table of situations (Kelley et al., 2003). Novel-seeming situations may not be experienced as completely novel if a person has had past experience with comparable situations in the same category. Certain of these patterns correspond to the distinctive norms of social exchange that are proposed by some theorists to define prototypical categories of social relations (e.g., Bugental, 2000; Clark & Mills, 2012; Fiske, 1992). It is conceivable that future analyses will identify different dimensions for describing situations, an eventuality that most interdependence theorists would welcome. To be consistent with the theory’s conceptual perspective, these alternatives should be derived from the objective properties of situations and have logical and discernible implications for interaction.

What Does Interdependence Theory Mean by “Self-Interest”? Interdependence theory was first proposed during the heyday of reinforcement theories. Thibaut and Kelley (1959) were clearly influenced by these theories, as well as by Homans’s (1950) attempt to describe social exchange processes in reinforcement terms. Thus, as mentioned above, interdependence theory assumes that people seek rewarding, noncostly solutions to the choices they encounter in social interaction. This principle has led some observers to erroneously conclude that the primary drive behind interdependence theory is outcome maximization or, in the extreme case, that it sanctions selfishness (e.g., Wallach & Wallach, 1983). To so conclude seriously mistakes the theory’s fundamental tenets. As Rusbult and Van Lange (1996) explain, the “self-interest” to which interdependence theory refers is not immediate and narrow but rather long term and broad. Because the ability of situations to provide rewards and costs depends on the degree to which important needs, values, and goals are fulfilled, broader considerations come into play. These broader considerations address the complexity and diversity of human needs, values, and goals, and therefore may include, for example, contributing to the well-being of another person, delaying gratification to further a long-term goal, following socially appropriate rules of conduct, or acting in a manner that is consistent with personal moral standards. Any of these can (and often do) run counter to immediate self-interest, although of course they benefit the self in the long

310

SOCIAL THEORIES

run. It is in this latter sense that interdependence theory conceptualizes selfinterest (see Deutsch, Chapter 7, this volume, for additional discussion). In fact, interdependence theory is unique among motivational approaches in clearly specifying a process by which broader considerations alter more immediate (or automatic) self-interest. To wit, the theory distinguishes the given situation—the direct and immediate outcomes that would accrue, were individuals to ignore the consequences of their actions over time or for others—from the effective situation—a reconceptualized set of outcomes that takes into account the immediate or long-term impact of one’s actions on the self and others. Just how individuals transition from the given to the effective matrix, a process known as transformation of motivation, is beyond the scope of this chapter (see Rusbult & Van Lange, 1996, for a thorough account). It is important to recognize, however, that this transformation—this taking into account of temporal and interpersonal considerations—points to another assumption of interdependence theory: that people recognize the nature of their interdependence with others and the impact that their behavioral choices may have for the well-being of others. For example, a husband may choose to spend his free time running errands for his wife rather than watching a football game, knowing that by lessening her burden, she will be happier and more committed to their relationship, which likely benefits both of them over time. Even strangers consider the interpersonal effects of their behavior (e.g., by helping strangers in an emergency or by practicing mundane helpful behaviors such as opening the door for others) and exhibit stable individual differences in how they approach interdependent situations (Van Lange, De Bruin, Otten, & Joireman, 1997). The self-interest of interdependence theory, in other words, is a broader form of self-interest that takes metacognition and temporal factors into account. By metacognition, we mean that before deciding how to act, people must infer how interaction partners perceive the situation and are likely to feel about each of the possible alternatives that one might pursue. Although not grounded in interdependence theory, Kenny’s (1994) social relations model provides conceptual and analytic tools for examining these cognitions in some detail—for example, in distinguishing self-perception from metaperception (understanding how others see oneself), and in distinguishing dispositional attributes from relationship-specific attributes (i.e., how generally likeable a person is versus how much one relationship partner likes the other, over and above general likeability). In its interest in metacognition, interdependence theory overlaps to some extent with work on intersubjectivity: shared meaning systems constructed by interacting parties during their interaction (Ickes & Gonzalez, 1994). A related misconception is that by describing rewards and costs associated with situations, interdependence theory assumes that people make rational choices among available behavioral alternatives (see Trafimow, Chapter 12, this volume, for a discussion of rationality in social-psychological

Interdependence Theory and Related Theories 311

theories). The logic of the transformation process, explained in the previous paragraph, indicates otherwise. The properties of the given situation may indicate what a “rational” actor should do (i.e., considering only oneself), but people often transform those options (i.e., considering oneself in relation to others) for idiosyncratic, nonobvious, perhaps even dysfunctional reasons. In other words, transformations of motivation may be based on any of the many “nonrational” or construal processes that have been documented in social-psychological research (e.g., self-esteem maintenance, increased competitiveness of groups relative to dyads). In fact, a key insight of the interdependence approach is that an individual’s deviation from behavioral options strongly suggested by the given situation provides unambiguous evidence of the operation of personal motivational factors. For example, a spiteful response to an ordinary question suggests something about the responder’s motives or circumstances, precisely because the more common and situationally appropriate neutral answer would be associated with better outcomes (e.g., avoiding unnecessary conflict).1

The Analysis of Situations Begins with Their Objective Properties A key and somewhat misunderstood assumption of interdependence theory is its insistence that the analysis of situations begin with the objective properties of those situations. On its face, this assumption seems to contradict social psychology’s traditional adherence to the principle of construal, namely, that causal analysis should focus on the personal and subjective meaning of situations to the individuals involved (e.g., Ross & Nisbett, 1991; see also Kashima, Chapter 3, this volume). To be sure, construals are an important part of interdependence theory. Nevertheless, interdependence theory conceptualizes construals as the product of the individual’s mental processing (i.e., transformational activity) of the situation as it actually exists—in other words, “what the individual makes of the situation.” To properly understand situation construals, therefore, theoretical analysis must begin with the objective properties of those situations, to which individuals respond. Consider the following example. The theory of psychological reactance stipulates that when people’s behavioral freedom is threatened, they will take steps to reassert that freedom, either behaviorally or psychologically (Brehm, 1966). This might be investigated by having an experimenter tell subjects that they must carry out some action in an experiment, in contrast to being given a choice to do the same thing. Interdependence theory characterizes this scenario as a situation in which participants have the option of protesting against or acceding to a domineering experimenter. The analysis thus starts with the situation as it is objectively defined—that there exists an authority figure who has some degree of power over the subject’s outcomes (perhaps their experimental credits). The subject must then respond to that situation by choosing whether to rebel or to submit, a choice that

312

SOCIAL THEORIES

depends on whether he or she construes the experimenter’s demand as a threat to be resisted or alternatively as a danger that requires acquiescence. This construal, in turn, is likely to derive from something in the individual’s mind—perhaps a personality predisposition or a momentary feeling. Rather than illustrating “the power of the situation,” then, as scenarios such as this one is commonly labeled, interdependence theory sees this example as demonstrating “the power of what the person makes of the situation” (Kelley et al., 2003, p. 7). This is not an arcane distinction. Social-psychological research often attributes causality to the properties of a particular situation; indeed, some characterize the field’s raison d’etre in this way (Kashima, Chapter 3, this volume). But if one views construals as the key causal agent, how are we to distinguish the objective properties that led to a certain construal from the motives, attitudes, and dispositions that shape individual interpretations (Reis, 2008)? Interdependence theory proposes that we begin causal analysis one step back from the individual’s construal, by instead defining the situation according to its objective properties. Of course, outside of the carefully controlled conditions of the experimental laboratory, it may not always be possible to do so accurately; among several reasons, independent observers may not always be aware of all relevant contingencies in a particular interaction. Nevertheless, by explicitly separating situation and construal, a less ambiguous causal analysis is made possible (for a more detailed discussion, see Reis & Holmes, 2012). Murray and Holmes’s risk regulation model (Murray & Holmes, 2009, 2011) follows directly from this analysis. Committed relationships, they argue, are fraught with risk: By committing, partners sacrifice autonomy and make themselves vulnerable to exploitation and harm. The inherent risk of commitment can be mitigated by motivated reinterpretation that not only minimizes the potential threat of such situations but also even increases— through confidence in the relationship—one’s trust and commitment (Murray & Holmes, 2009). By distinguishing the situation (dependence) from the construal, it becomes possible to clearly identify the underlying motivational process. Rusbult’s model of commitment processes (Rusbult, Agnew, & Arriaga, 2012) relies on a similar analysis. When confronted with the same conflicts of interest, highly committed partners differ markedly from their less committed counterparts in what they make of the situation. Highly committed partners are more likely to construe the situation as an opportunity to support the relationship rather than as a threat to its continuance. Thus they are more likely to respond to a partner’s destructive behavior by inhibiting the urge to retaliate and responding constructively instead, more likely to sacrifice preferred activities to appease a partner, and less likely to escalate negative conflict behaviors. Committed partners thus reconstrue threatening situations in a manner that bolsters rather than hinders their relationship. The emphasis on distinguishing situations from construals is also evident

Interdependence Theory and Related Theories 313

in interpersonal-process models that differentiate partner behaviors from the perception of those behaviors. For example, Reis and Clark’s (2013) model of responsiveness posits that perceptions of a partner’s responsiveness reflect the joint (and empirically separable) influence of person factors—the various traits, motives, values, preferences, beliefs, feelings about a particular other, and other attributes that differentiate one person from another—and the partner’s actual behavior. In this model, as in the risk regulation model, the partner’s behavior provides a potentially ambiguous stimulus, which the perceiver transforms according to the motives that are momentarily most accessible. The conceptual benefits of distinguishing the objective properties of situations from construals are discussed further below.

How Are Individual Differences Conceptualized? Individual differences are commonly represented in terms of the consistent patterns of behavior that a person displays across situations—for example, tendencies to worry or to be socially outgoing in diverse circumstances. Interdependence theory takes a different approach, building on the idea of affordance—that situations provide opportunities for the impact of person factors to be revealed in behavior. The analysis starts with the given situation—as discussed above, an objective description of what sorts of behavior are possible in that situation. Person factors then determine the individual’s perception of and response to those objective properties (Rusbult & Van Lange, 1996). In this way of theorizing, then, person factors are considered to be nested within situations: situations afford but individuals choose. Although this conceptualization differs from the traditional view that defines personality in terms of cross-situational consistency, it is consistent with other models that similarly point to situation-based variability as the essence of personality. For example, Mischel and Shoda’s (1995) cognitiveaffective processing systems (CAPS) model proposes that situations activate distinctive patterns of cognition and affect that in turn lead to behavior. These patterns constitute behavioral “signatures”: predictable patterns of response to critical features of situations. Consistency in personality is thereby not defined cross-situationally but rather by “distinctive and stable patterns of situation–behavior relations” (Mischel, Shoda, & Mendoza-Denton, 2002, p. 50)—in other words, as a series of “if X exists, then response Y is more likely” contingencies. These distinctive patterns can be identified by examining the degree to which within-person variability in behavior is captured by the properties of situations (Fleeson & Noftle, 2008). Interdependence theory takes the idea of affordance considerably further than the CAPS model does in theorizing that the range of person factors that can be revealed in a particular situation is limited by the properties of that situation. In the language of affordance, a given situation affords only certain responses and not others. This is because it is logically implausible to infer the operation of a person factor when the situation compels a particular

314

SOCIAL THEORIES

behavior or when a situation does not permit it to occur. For example, most people would not attribute Rafael’s sad mood to a gloomy personality if they knew that his grandmother had just died, nor would they infer altruistic intent from Rachel’s kindly act if they were aware that she stood to gain a great deal from that act. The logic of affordance in interdependence theory further assumes that behavioral choices reflect a person’s preference among the various possibilities afforded by a particular situation. Most situations afford multiple behavioral opportunities: An interpersonal challenge is an opportunity to be defiant or back down; a friend in need is an occasion to assist, exploit, or ignore; and a job that needs to be done is a chance to divide labor, delegate, or do it oneself. By considering the various logically relevant alternatives in particular situations, perceivers can deduce the dispositions, motives, or goals that may have shaped the individual’s choice. Another way to describe this process is to say that person factors represent tendencies to psychologically transform situations by emphasizing certain of their afforded possibilities over others. For instance, in the example of the job that needs to be done, a self-reliant person would be someone who turns this somewhat ambiguous situation into a more individualistic one by attaching particular value to the opportunity to go it alone. (Holmes, 2002, referred to these tendencies as valuation rules; see also Van Lange et al.’s, 1997, theorizing about social value orientations.) On the other hand, a person who prizes egalitarianism might instead turn the situation into a more cooperative one. (The analysis of transformational activity borrows from the Lewinian [1943] concept of restructuring the field.) It can also be seen from this example that ambiguous situations—sometimes called weak situations (Snyder & Ickes, 1985)—are better able to reveal the impact of person factors than are so-called strong situations, in which behavior is normatively mandated by the requirements of the situation. For example, an officer who confronts an ambiguous situation in which a suspect reaches in his coat pocket will make a split-second decision about whether the suspect is reaching for a gun or something else, a decision that reveals the impact of the officer’s personal beliefs or stereotypes precisely because the situation is ambiguous (see Reis & Holmes, 2012, for additional discussion of situational ambiguity). In this light, interdependence theory’s stance in the long-standing person–situation debate is readily apparent: Individual differences are inextricably intertwined with situations. The impact of persons can only be understood as figure against the ground of situations. Although this may not be news—in 2006, David Funder proclaimed, “Nowadays, everybody is an interactionist” (Funder, 2006, p. 22)—we would argue that interdependence theory goes further than other interactionist (Person × Situation) theories in proposing a particular, theoretically integrated functional form of how person factors and situation factors interact (see Cervone, Caldwell, & Mayer, Chapter 8, this volume). This conceptualization has provided a springboard for many specific programs of research, as well as for the detailed analyses

Interdependence Theory and Related Theories 315

of 20 prototypical situations and their characteristic person factors found in the Atlas of Interpersonal Situations (Kelley et al., 2003).

Structural Features of Interdependence Theory Interdependence theory has several key structural features, each with implications for the type of data needed to test relevant hypotheses. Because dyadic interaction (actual or implied) involves two individuals, interdependence theory structurally models the responses and experiences of both individuals by using an outcome matrix to represent their behavioral options (Kelley, 1991). A conceptually similar structural model is the actor–partner interdependence model (APIM), which is used to control for dependencies in dyadic research, a statistical necessity when interdependence exists, and to quantify the influence of each partner to an interaction or relationship (Kenny, 1996; Kenny & Ledermann, 2010). Therefore, in both approaches, the most relevant data are obtained from two interaction (or relationship) partners, although a single individual’s data may also be useful if they adequately capture the real or implied influence of others. Group research has likewise made use of outcome matrices to model interdependence that also can be captured by generalizing APIM analyses to group intergroup contexts (e.g., an ingroup–outgroup interdependence model; Kenny & Kashy, 2014). The APIM specifies that in a dyad, each partner’s actions can be affected by his or her own attributes or prior behavior, the partner’s attributes or behavior, or a combination of the two. As Wickham and Knee (2012) explain, different combinations of these three types of effects (described by Kenny & Ledermann, 2010) correspond to different patterns of outcome interdependence. For example, actor effects with no partner effects in the APIM would correspond to bilateral actor control in interdependence theory, whereas equal actor and partner effects would correspond to mutual joint control. If partner effects vary from one partner to the other (e.g., if husbands influence wives more than wives influence husbands), then control is asymmetric. Wickham and Knee show how the matrix representations of interdependence theory may be used to calculate different patterns of influence in the APIM (Kenny & Ledermann, 2010). Thus, insofar as the influence of situations on outcomes is concerned, APIM and interdependence theory provide conceptually analogous representations of the ways in which partners influence each other’s outcomes. Structurally, interdependence theory goes beyond the description of such influence. The assumptions described earlier in this chapter are indicative of a fundamental structural distinction in interdependence theory, differentiating among (1) situational features as reflected in the given situation; (2) inthe-moment, motivational, and cognitive transformation processes guided by person factors that are activated in the given situation, as described ear-

316

SOCIAL THEORIES

lier (e.g., general personality attributes, specific tendencies with the current partner, and immediate motives or valuation rules); and (3) overt behavior. Research has demonstrated the validity of conceptualizing the given situation as a structural factor separate from the person factors that shape interpretations of a situation. For example, Yovetich and Rusbult (1994) found that with limited time to respond, people tend to follow the selffocused, immediate preferences suggested by the given situation, rather than the transformed (effective situation) preferences that take into account the impact of one’s actions on others. They concluded that given preferences do indeed exist and that straying from them requires mental elaboration that is not feasible in a limited response window. Finkel and Campbell (2001) took another approach, reasoning that transformational activity is more cognitively demanding than self-focused reactions. They demonstrated that participants who were depleted of cognitive resources were less likely to consider other-oriented responses. These studies show that interaction is not directly caused by the given situation, nor is it invariant across situations, guided solely by person factors. Rather, when people are able and motivated to consider their interaction goals, behavior strays from the given situation to incorporate social or interpersonal factors. Transformational activity is particularly likely to tax cognitive resources when a task is novel, as it was in these studies. Many social interactions are not novel, of course. They occur in ongoing relationships or in situations that are comparable to past experiences, so that certain responses are likely to be well learned and perhaps even habitual. In such situations, responses can become automated, presenting little or no cognitive demand, and yet they still reflect the transformational processes that occurred when the responses were originally formed (Arriaga, 2013; cf. Fazio, 1990). As Rusbult and Van Lange (1996) and Murray and Holmes (2011) theorize, with repeated, ongoing interaction, transformations, once deliberate, can become habitual and automatic (see also Deutsch, Chapter 7, this volume). The opening paragraph of this chapter described several situations that appear to involve deliberate transformations. But the same examples could easily provoke automatic responses that reflect past transformational activity. When Shannon and Pat consider whether to deepen their commitment, each may respond automatically, in ways that reflect past experiences in situations of vulnerability; Pat, an anxiously attached person, may be eager to deepen their commitment, whereas Shannon, an avoidantly attached person, may reflexively withdraw. The new parents, Eamon and Kyoko, may have automated responses to middle-of-the-night feedings that are based on their prior experiences negotiating other conflicts of interest. And the physician and patient may be reminded of interactions with other patients and other providers, respectively, during which they formed attitudes that are activated automatically and that direct their response in the current interaction.

Interdependence Theory and Related Theories 317

Other Structural Considerations in Interdependence Theory Assessments that conceptually differentiate the given situation from transformational activity often combine varied levels of analysis. The given situation defines an immediate pattern of interdependence shaped by physical characteristics (e.g., an individual’s physical capabilities or disabilities, spatial arrangements), universal person characteristics (i.e., meeting basic needs and avoiding aversive experiences, such as getting food or sleep, satiating the need for connection with others, avoiding pain), or the nature of the task (e.g., a task can only be accomplished successfully if people coordinate their movements). The transformational activity that mediates between a given situation and behavior is similarly shaped by factors that reside at different levels of analysis, such as physical (e.g., biological temperament, limitations in self-regulatory strength, excessive stress), psychological (e.g., past experiences, individual cognitive and affective tendencies), and sociocultural characteristics (e.g., socially or culturally specific norms, influence of network members or peers, religious mores). As such, interdependence theory provides breadth by integrating factors residing at different levels of analysis that are often kept separate in social-psychological theories (De Houwer & Moors, Chapter 2, this volume). Another key structural feature of interdependence theory is the clear distinction it makes between satisfaction and dependence. Satisfaction reflects comparison of current outcomes with expectations based on prior experiences (known as the comparison level, or CL). Dependence, or the willingness to remain in a relationship or group, reveals comparison of current outcomes with the most desirable alternative perceived to be available (known as the CL-alt). This important distinction led Rusbult to propose the investment model of commitment (see Rusbult et al., 2012, for a review). Laypersons and clinicians alike had commonly assumed that relationship stability depended on satisfaction, despite the fact that many unhappy relationships persist, and many happy relationships end. As a direct legacy of her training in interdependence theory, Rusbult recognized that dependence (needing a particular relationship to attain desired outcomes) is distinct from whether or not one is satisfied with a relationship. Thus, she theorized that people may stay in unhappy relationships for structural reasons, in particular a lack of alternatives (e.g., having no way of earning a living except relying on a partner) and high investments (e.g., having put too much into a relationship that would be lost if the relationship ended). This proposition has provided valuable insights into many practical problems, such as why women stay in abusive relationships (Rusbult & Martz, 1995), why businesses often exploit workers who have few options (Rusbult, Insko, Lin, & Smith, 1990), and why individuals persist in costly business relationships (the sunken cost effect; Rusbult & Farrell, 1983). A further structural consideration is that because, as described above, situations vary in the specific responses that they afford, situations also vary

318

SOCIAL THEORIES

in their degree of informativeness about persons. Thus, certain situations are considered diagnostic about other persons because they are particularly likely to reveal their personal valuation rules (Holmes & Rempel, 1989). For example, Rusbult’s analysis of accommodative dilemmas (Rusbult, Verette, Whitney, Slovik, & Lipkus, 1991) highlights responses to conflictof-interest situations, as do Murray and Holmes’s (2009) analysis of trust and commitment and Simpson’s (2007) description of relationship strain tests. Conflicts of interest are diagnostic of relational valuation rules because they reveal whether partners are inclined to prioritize the other person (or the relationship). The same may be said of groups. For example, as their outcomes become increasingly noncorrespondent, interacting groups become more competitive than interacting individuals, presumably because group members anticipate more self-interest in the other group than shared interest (Schopler et al., 2001). Yet another structural consideration underlying the predictions of interdependence theory is that many relevant processes unfold over time and involve different levels of awareness. Suppose one hypothesized that perceiving (and reporting on) a partner’s selfish behavior is negatively correlated with relationship satisfaction. This rather obvious claim, which might be substantiated straightforwardly in a one-time survey, may mask a more nuanced process that unfolds over time. Murray and colleagues (Murray, Holmes, & Pinkus, 2010) found that when newlyweds were exposed to more high-risk situations (e.g., conflicts of interest involving high unilateral dependence), four years later they exhibited more negative implicit attitudes but not more negative explicit reports of satisfaction or love. These results suggest that the impact of situations may not be immediately apparent, may take time to develop, and may operate, at least initially, beyond one’s awareness. Temporal considerations may also be evident in partners’ adaptations to their interaction experiences. Consider the temporal impact of interaction processes on personality change (Kelley, 1983). For example, individuals who must interact repeatedly in inherently competitive situations, facing only competitive response options over time, may develop automatic tendencies in similar future situations. In interdependence theory terms, it is not repeated experiences with a particular situation per se that influences individual differences, but rather repeated experiences with a specific configuration of abstract features of situations. A final example of an adaptation process that unfolds over time occurs when the affect generated by specific interactions directs situation selection in future interactions (Kelley et al., 2003). Interactions that cause strong affect do not go unnoticed. They are encoded adaptively, so that similar situations can be elicited or avoided in the future. For example, individuals whose interaction history has led them to develop an avoidant attachment style steer clear of interactions that suggest intimacy, whereas individuals who have developed an anxious attachment style shun potentially threatening interactions.

Interdependence Theory and Related Theories 319

In sum, interdependence theory encapsulates processes that occur at levels of analysis that vary in specificity and that link together over time. Indeed, interdependence theory provides impressive explanatory breadth precisely because it accounts for many important phenomena over extended time.

Explanatory and Predictive Value of Interdependence Theory Epistemologists sometimes describe a trade-off between explanatory breadth and predictive power, such that some theories explain much and yet offer little in the way of specific predictions, while other theories offer the opposite (see Gawronski & Bodenhausen, Chapter 1, this volume). Interdependence theory is clearly and intentionally a theory that emphasizes explanatory breadth; after all, its concepts and analytic framework are well suited to explaining most any phenomenon that depends on social interaction and interpersonal influence. To our knowledge, no “critical tests” of interdependence theory have been conducted or published, and it is difficult to imagine an empirical study that would disconfirm the theory. Such a study would have to show that people do not repeat rewarding interactions, that they do not attend to the influence of others on their outcomes nor their own impact on others, or that the objective properties of social situations have no impact on behavior. These are, of course, unlikely. Nevertheless, interdependence theory suggests that the above-mentioned trade-off between explanatory breadth and predictive power is a misleading dichotomy. In addition to its explanatory breadth, interdependence theory has clear predictive power in at least two important senses. First, in using the language of payoff matrices and transformations of motivation, it offers predictions about the behaviors people will choose. In particular instances, these values may be difficult to ascertain, but this is a measurement problem, not an absence of predictive power. Second, and more generatively, by virtue of its comprehensive, multilevel conceptualization, interdependence theory points to constructs and principles that would not have been imagined from narrower theoretical approaches, opening the door to novel theoretical insights. Nowhere is this influence more evident than in the impact of interdependence theory on the study of interpersonal processes. In the close relationships literature, analyses of dyadic data have become synonymous with adopting an interdependence theory framework (Wickham & Knee, 2012).2 But the claim that people influence each other—although profound and not always adopted in study designs—alone does not yield specific insights about the causes and consequences of interaction. Interdependence theory’s application to dyadic relationships highlights more specific, testable propositions, including (but not limited to) the manner in which people: solve interpersonal problems concerning their relative influence across diverse types

320

SOCIAL THEORIES

of interactions (e.g., in professional or romantic couples, or among peers); infer traits and motives about others who can affect a person’s outcomes; establish habitual or automated patterns of interaction in couples over time; develop relationship-specific attitudes that may operate beyond awareness; form stable expectations about relationships (i.e., the CL) and the next-best alternative relationship (i.e., CL-alt) and compare these comparison standards to current outcomes; and clarify the relational grounding of personality. (Arriaga, 2013, explains how these topics fit into an integrated model of couple interaction.) Research on each of these topics utilizes constructs not currently available in other theoretical approaches. Research on group processes also has been influenced by interdependence theory. For example, as mentioned above, the discontinuity effect indicates that two interacting groups are more likely to compete in a mixedmotive situation than are two interacting individuals confronting an identical situation (Schopler et al., 2001). Interdependence theory describes specific situational features that intensify or dampen the competitive orientation that groups adopt—that is, their tendency to transform a given situation from one that allows for cooperation into one that seems to necessitate competition. In this way, many of the processes that characterize dyads also characterize groups.3 For example, just as dyads may increase their closeness by becoming more interdependent, so too do highly interdependent groups have a stronger group identity (i.e., greater entitativity) than groups in which members are relatively more independent (Gaertner, Iuzzini, Witt, & Oriña, 2006). By identifying necessary constructs, an explanatory theory like interdependence theory gains predictive power. To illustrate this principle, we discuss the example of how individuals draw meaning from interpersonal interactions. Reinforcement theories suggest that if an interaction is rewarding, it will be repeated. But the evaluation process is not that simple. Interactions are evaluated on (at least) two levels: namely, in terms of their direct and symbolic outcomes (Kelley, 1979). Whereas direct outcomes are exactly what they suggest—how something feels in the moment without being modified by higher-order inferential processes—symbolic outcomes rely on attributional activity to detect broader implications of a partner’s actions. An act of kindness may yield an immediate positive outcome, but it becomes an even more positive experience when it comes from a person who typically is rude to others. Knowing that one is the unique recipient of the other’s kind behavior suggests that he or she has invoked a relationship-enhancing transformation specific to oneself, which in turn increases the symbolic value derived from the interaction. Actions that convey personalized meaning are evaluated differently than actions that are generalized to others (e.g., speeddaters appreciate being liked uniquely more than being liked by others who like everyone; Eastwick, Finkel, Mochon, & Ariely, 2007). It is in this sense that person factors become socially situated, as mentioned above (Semin & Garrido, Chapter 14, this volume). Similarly, small acts of compassion, like making a partner’s favorite meal when he or she is having a bad day, not only provide immediate positive outcomes for the partner (i.e., eating

Interdependence Theory and Related Theories 321

a favorite meal) but also have symbolic value in conveying the underlying sense of caring that led to extra effort on behalf of the partner. This analysis of symbolic meaning suggests a prediction that is unique to interdependence theory: Varying the situation (strong vs. weak, as described above, or in terms of specific features, such as making it personally costly to benefit a partner) will vary how much diagnostic information the situation conveys about another person’s motives and, thus, how much symbolic value might to be gained from an interaction. Programs of research on trust and perceived partner responsiveness—constructs that have been enormously influential in the literature on interpersonal relations (e.g., Holmes & Rempel, 1989; see Reis & Clark, 2013; Simpson, 2007, for reviews)—have been profoundly influenced by this idea. This chapter has described several phenomena that in all likelihood would not have been examined, and various specific predictions that would not have been offered, without the general theoretical framework provided by interdependence theory. They illustrate ideas that have been enormously generative. Rusbult and Van Lange (2008) discuss additional predictions that are uniquely derived from interdependence theory, such as predictions about the interpersonal nature of goal pursuits and regulatory fit (Righetti, Finkenauer, & Rusbult, 2011), and the importance of maintaining error-free channels of communication in mixed-motive situations, as commonly occur in negotiations between parties whose interests are not aligned (Tazelaar, Van Lange, & Ouwerkerk, 2004).

Advantages and Disadvantages of Interdependence Theory In our account of interdependence theory, we have already noted several compelling advantages. In this section, we review these advantages in a more integrated fashion, fittingly enough, because we see interdependence theory as a remarkably integrative theory. This quality is evident both as metatheory and in the theory’s particulars. Metatheoretically, interdependence theory and its offshoots adopt elements from several theoretical traditions that seemingly involve inconsistent or contradictory philosophical assumptions. One is structuralism: Interaction processes can be understood in terms of specific structural components. A second is functionalism: Interactions reflect responses that “make sense” or provide what is perceived to be the most beneficial (or least aversive) experience (see also Tetlock & Fincher, Chapter 13, this volume). A third is behaviorism: Interaction involves observable behavior, and much of what creates a rewarding or aversive experience follows from what can be observed. A fourth is the social-cognitive tradition: The mental states of each interaction partner—mental states with developmental and experiential origins—figure prominently in interpreting interpersonal situations and responding to them (see Gawronski & Bodenhausen, Chapter 4, this volume). A fifth is the gestalt tradition: The influence of situations is

322

SOCIAL THEORIES

best understood in terms of “what the person makes of the situation,” rather than as objective attributes that operate uniformly across persons. At the level of particulars, most interdependence theorists would probably agree that the focus on interaction best differentiates interdependence theory from other social-psychological theories. To be sure, social interaction is prominent in many existing theories. Most of them, however, tend to focus either on the contributions of individuals to interaction or on interaction processes per se. Interdependence theory, in contrast, forces theorists to fully integrate these levels of analysis: It is at once a theory of how human needs and goals create the situations in which people interact, of how situations facilitate certain patterns of interaction while discouraging others, and of how individual differences are shaped by the interactional possibilities that situations afford. Interdependence theory is furthermore a deeply social theory: The situations it studies concern the manner in which people coordinate their actions with the actions of others. By representing these situations in abstract terms, the theory allows an analysis of social interaction with considerable generalizability. Another advantage of interdependence theory is the clear distinction it makes between the objective environment and the person’s construal of that environment (see also Kashima, Chapter 3, this volume). Many theories conflate these two, or at least fail to specify how one relates to the other. In contrast, interdependence theory is explicit about the process of transforming situational contingencies into goal-directed action. This is another point in which the theory’s deeply social nature can be seen: Goals are defined broadly rather than individualistically, encompassing, when it is suitable to do so, actions that take into account the likely responses of others. For example, Rusbult’s model of behavioral affirmation (sometimes called the Michelangelo Phenomenon; Rusbult, Finkel, & Kumashiro, 2009) describes how partner support contributes to, and no-support hinders, people’s progress in attaining personal goals. By incorporating thinking about the actual or anticipated response of others, interdependence theory is one of the few social-psychological theories to explicitly incorporate metacognition (Epley, 2008). A somewhat more general advantage is the role that interdependence theory may play in social psychology’s elusive quest for a taxonomy of situations, akin to the periodic table in chemistry or the Big Five in personality theory. Much has been written about the field’s lack of a taxonomy of situations, despite numerous efforts to provide one (see Reis & Holmes, 2012, for a review). Most such endeavors have proceeded bottom-up, from lay accounts of social activity. Interdependence theory in contrast takes a topdown approach, beginning with an abstract analysis of properties likely to influence patterns of interdependence and then generating the most plausible prototypes (Kelley et al., 2003). If these efforts bear fruit, it will be possible to theorize more systematically about the varieties of human social interaction.

Interdependence Theory and Related Theories 323

Like any theory, interdependence theory has shortcomings. Prominent among them is the inherent ambiguity of specifying in any given case the values of relevant variables. For example, although the theory requires analysts to determine the objective properties of the given situation, the fact that observers will have experience with such situations may confound distinctions between its objective properties and common construals (which reflect person factors as well as situational features). Similarly, because many different transformation rules can produce the same outcome, it may prove difficult to identify the particular transformation that a person is using, as well as the underlying motive (i.e., person factor or social norm) for that transformation. Such ambiguity suggests an emphasis on explanatory breadth rather than predictive power, but in our view that conclusion should not be taken too far. Real-life interactions are messy, often simultaneously involving multiple motives and circumstances that can be deductively ambiguous. In a highly structured lab study, it may be feasible to identify behavioral options that are not mediated through the transformation process. In fact, stranger interactions in novel settings are often guided more by situational factors than by broader considerations; the absence of a history together and an expectation of future interaction typically render specific person factors relatively immaterial. In contrast, interaction with relationship partners reflects current, past, or anticipated social considerations, making the given situation less prominent and current person factors more prominent (Arriaga, 2013). Different dyads may react to the same situation in different ways because of their own unique history and their habitual interactions. Moreover, the same person may react differently to the same situation with a different partner. The given situation thus becomes more difficult to identify. Another limitation is that interdependence theory has little to say a priori about which of the person factors afforded in a particular situation are most likely to be influential in transforming interpersonal situations. Nearly anything that can affect an individual’s interpretation of and response to an interdependent situation in pursuit of outcomes can serve as a person factor: traits, attitudes, values, habits, preferences, feelings, defenses, roles, even momentary goals. Although this degree of inclusiveness follows from the theory’s explanatory aims, it highlights a weakness insofar as interdependence theory might serve as a model of individual differences or of their origins or functional significance. A final shortcoming is that, even though this theory is highly generative of research, few studies directly test propositions of interdependence theory. Indeed, evidence for the theory’s propositions often draws on existing research conducted for other purposes (see, for example, the analysis of prototypical situations in Kelley et al., 2003). To some, this may suggest that interdependence theory is more of a general framework and descriptive language than a specific theory, a conclusion with which we do not concur, as discussed earlier. A rather different explanation is that newcomers

324

SOCIAL THEORIES

may be daunted by the complexity of the theory, mentioned at the beginning of this chapter. We find this circumstance somewhat ironic, given people’s “intuitive ability to recognize situations in their own lives without any formal training in interdependence analysis and moreover, usually without the ability to articulate the abstract properties that underlie such recognition” (Kelley et al., 2003, p. 7). It is to be hoped that this shortcoming will soon become irrelevant.

Conclusion Since its first appearance in 1959, interdependence theory has had a somewhat mixed experience in the history of social-psychological theories. Never one of the field’s most popular theories, over time it has not only endured but grown in scope and influence. This influence can be seen in research and theory, particularly in the related theoretical models, mentioned throughout this chapter, born from Thibaut and Kelley’s formulations. This growth and influence likely reflect the theory’s quintessentially social nature, as we have elaborated in this chapter. Many important psychological phenomena occur in a social context that directs, moderates, and frames important processes. To scholars looking for social explanations of these behaviors, we encourage them to capitalize on the breadth and depth that an interdependence analysis already offers, rather than search for new theories or explanatory concepts. Its roots in past theoretical traditions, its utility for understanding crucially important interpersonal processes, and its promise for framing novel research questions make interdependence theory an exceptional and satisfying theory. Notes 1. Readers may be interested to learn that this sort of analysis led Kelley to propose his analysis-of-variance model of attribution theory as a sidelight to his work on interdependence theory. That is, observers infer causality by considering which of the plausible causes of a behavior covary with its occurrence. 2. In fact, Ellen Berscheid once pointed out to one of us that although Thibaut and Kelley’s initial work (1959) was entitled The Social Psychology of Groups, most of the examples in that marvelously generative book concerned close dyadic relationships. 3. This may follow from Thibaut and Kelley’s (1959) thorough training in the Lewinian tradition. Lewin wrote that “marriage is a group situation, and, as such, shows the general characteristics of group life,” explaining that marital interaction follows the general properties of “the relation between an individual and his group” (1948, p. 84). However, the group in this instance is very small, is central to the person’s values, desires, and goals, and is characterized by “the least social distance” (1948, p. 88) between its members.

Interdependence Theory and Related Theories 325

References Allport, G. W. (1960). Personality and social encounter. Boston: Beacon Press. Arriaga, X. B. (2013). An interdependence theory analysis of close relationships. In J. A. Simpson & L. Campbell (Eds.), The Oxford handbook of close relationships (pp. 39–65). Oxford, UK: Oxford University Press. Brehm, J. W. (1966). A theory of psychological reactance. New York: Academic Press. Bugental, D. B. (2000). Acquisition of the algorithms of social life: A domain-based approach. Psychological Bulletin, 126, 187–219. Buss, D. M., & Kenrick, D. T. (1998). Evolutionary social psychology. In D. T. Gilbert, S. T. Fiske, & G. Lindzey (Eds.), The handbook of social psychology (4th ed., Vol. 2, pp. 982– 1026). Boston: McGraw-Hill. Clark, M. S., & Mills, J. (2012). A theory of communal (and exchange) relationships. In P. A. M. Van Lange, A. Kruglanski, & E. T. Higgins (Eds.), Handbook of theories of social psychology (pp. 232–250). Thousand Oaks, CA: Sage. Eastwick, P. W., Finkel, E. J., Mochon, D., & Ariely, D. (2007). Selective versus unselective romantic desire: Not all reciprocity is created equal. Psychological Science, 18, 317–319. Epley, N. (2008). Solving the (real) other minds problem. Social and Personality Psychology Compass, 2, 1455–1474. Fazio, R. H. (1990). Multiple processes by which attitudes guide behavior: The MODE model as an integrative framework. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 23, pp. 75–109). New York: Academic Press. Finkel, E. J., & Campbell, W. (2001). Self-control and accommodation in close relationships: An interdependence analysis. Journal of Personality and Social Psychology, 81, 263–277. Fiske, A. (1992). The four elementary forms of sociality: Framework for a unified theory of social relations. Psychological Review, 90, 669–723. Fleeson, W., & Noftle, E. E. (2008). Where does personality have its influence? A supermatrix of consistency concepts. Journal of Personality, 76, 1355–1386. Funder, D. C. (2006). Towards a resolution of the personality triad: Persons, situations, and behavior. Journal of Research in Personality, 40, 21–34. Gaertner, L., Iuzzini, J., Witt, M., & Oriña, M. (2006). Us without them: Evidence for an intragroup origin of positive in-group regard. Journal of Personality and Social Psychology, 90, 426–439. Holmes, J. G. (2002). Interpersonal expectations as the building blocks of social cognition: An interdependence theory analysis. Personal Relationships, 9, 1–26. Holmes, J. G., & Rempel, J. K. (1989). Trust in close relationships. In C. Hendrick (Ed.), Review of personality and social psychology (Vol. 10, pp. 187–220). London: Sage. Homans, G. C. (1950). The human group. New York: Harcourt, Brace. Ickes, W., & Gonzalez, R. (1994). “Social” cognition and social cognition: From the subjective to the intersubjective. Small Group Research, 25, 294–315. Kelley, H. H. (1979). Personal relationships: Their structures and processes. Hillsdale, NJ: Erlbaum. Kelley, H. H. (1983). The situational origins of human tendencies: A further reason for the formal analysis of structure. Personality and Social Psychology Bulletin, 9, 8–30. Kelley, H. H. (1991). Lewin, situations, and interdependence. Journal of Social Issues, 47, 211–233. Kelley, H. H. (2000). The proper study of social psychology. Social Psychology Quarterly, 63, 3–15. Kelley, H. H., Holmes, J. G., Kerr, N. L., Reis, H. T., Rusbult, C. E., & Van Lange, P. A. M. (2003). An Atlas of interpersonal situations. Cambridge, UK: Cambridge University Press.

326

SOCIAL THEORIES

Kelley, H. H., & Thibaut, J. W. (1978). Interpersonal relations: A theory of interdependence. New York: Wiley. Kenny, D. A. (1994). Interpersonal perception: A social relations analysis. New York: Guilford Press. Kenny, D. A. (1996). Models of non-independence in dyadic data. Journal of Social and Personal Relationships, 13, 279–294. Kenny, D. A., & Kashy, D. A. (2014). The design and analysis of data from dyads and groups. In H. T. Reis & C. M. Judd (Eds.), Research methods in social and personality psychology (2nd ed., pp. 589–607). New York: Cambridge University Press. Kenny, D. A., & Ledermann, T. (2010). Detecting, measuring, and testing dyadic patterns in the actor-partner interdependence model. Journal of Family Psychology, 24, 359–366. Lewin, K. (1943). Defining the “field at a given time.” Psychological Review, 50, 292–310. Lewin, K. (1948). The background of conflict in marriage. In K. Lewin (Ed.), Resolving social conflicts (pp. 84–102). New York: Harper. Luce, R. D., & Raiffa, H. (1957). Games and decisions: Introduction and critical survey. New York: Wiley. Mischel, W., & Shoda, Y. (1995). A cognitive-affective system theory of personality: Reconceptualizing situations, dispositions, dynamics, and invariance in personality structure. Psychological Review, 80, 252–283. Mischel, W., Shoda, Y., & Mendoza-Denton, R. (2002). Situation-behavior profiles as a locus of consistency in personality. Current Directions in Psychological Science, 11, 50–54. Murray, S. L., & Holmes, J. G. (2009). The architecture of interdependent minds: A motivation-management theory of mutual responsiveness. Psychological Review, 116, 908–928. Murray, S. L., & Holmes, J. G. (2011). Interdependent minds: The dynamics of close relationships. New York: Guilford Press. Murray, S. L., Holmes, J. G., & Pinkus, R. T. (2010). A smart unconscious? Procedural origins of automatic partner attitudes in marriage. Journal of Experimental Social Psychology, 46, 650–696. Reis, H. T. (2008). Reinvigorating the concept of situation in social psychology. Personality and Social Psychology Review, 12, 311–329. Reis, H. T., & Clark, M. (2013). Responsiveness. In J. A. Simpson & L. Campbell (Eds.), The Oxford handbook of close relationships (pp. 400–424). Oxford, UK: Oxford University Press. Reis, H. T., Collins, W., & Berscheid, E. (2000). The relationship context of human behavior and development. Psychological Bulletin, 126, 844–872. Reis, H. T., & Holmes, J. G. (2012). Perspectives on the situation. In K. Deaux & M. Snyder (Eds.), The Oxford handbook of personality and social psychology (pp. 64–92). New York: Oxford University Press. Righetti, F., Finkenauer, C., & Rusbult, C. (2011). The benefits of interpersonal regulatory fit for individual goal pursuit. Journal of Personality and Social Psychology, 101, 720–736. Ross, E. A. (1908). Social psychology. New York: Macmillan. Ross, L., & Nisbett, R. E. (1991). The person and the situation: Perspectives of social psychology. New York: McGraw-Hill. Rusbult, C. E., Agnew, C. R., & Arriaga, X. B. (2012). The investment model. In P. A. M. Van Lange, A. W. Kruglanski, & E. T. Higgins (Eds.), Handbook of theories of social psychology (Vol. 2, pp. 218–231). Thousand Oaks, CA: Sage. Rusbult, C. E., & Arriaga, X. B. (1997). Interdependence processes in close relationships. In S. Duck (Ed.), Handbook of personal relationships (2nd ed., pp. 221–250). Chichester, UK: Wiley. Rusbult, C. E., & Farrell, D. (1983). A longitudinal test of the investment model: The impact

Interdependence Theory and Related Theories 327 on job satisfaction, job commitment, and turnover of variations in rewards, costs, alternatives, and investments. Journal of Applied Psychology, 68, 429–438. Rusbult, C. E., Finkel, E. J., & Kumashiro, M. (2009). The Michelangelo phenomenon. Current Directions in Psychological Science, 18, 305–309. Rusbult, C. E., Insko, C. A., Lin, Y. H. W., & Smith, W. J. (1990). Social motives underlying rational selective exploitation: The impact of instrumental versus social-emotional allocator orientation on the distribution of rewards in groups. Journal of Applied Social Psychology, 20, 984–1025. Rusbult, C. E., & Martz, J. M. (1995). Remaining in an abusive relationship: An investment model analysis of nonvoluntary dependence. Personality and Social Psychology Bulletin, 21, 558–571. Rusbult, C. E., & Van Lange, P. A. M. (1996). Interdependence processes. In E. T. Higgins & A. W. Kruglanski (Eds.), Social psychology: Handbook of basic principles (pp. 564–596). New York: Guilford Press. Rusbult, C. E., & Van Lange, P. A. M. (2008). Why we need interdependence theory. Social and Personality Psychology Compass, 2, 2049–2070. Rusbult, C. E., Verette, J., Whitney, G. A., Slovik, L. F., & Lipkus, I. (1991). Accommodation processes in close relationships: Theory and preliminary empirical evidence. Journal of Personality and Social Psychology, 60, 53–78. Schopler, J., Insko, C. A., Wieselquist, J., Pemberton, M., Witcher, B., Kozar, R., et al. (2001). When groups are more competitive than individuals: The domain of the discontinuity effect. Journal of Personality and Social Psychology, 80, 632–644. Simpson, J. A. (2007). Psychological foundations of trust. Current Directions in Psychological Science, 16, 264–268. Snyder, M., & Ickes, W. (1985). Personality and social behavior. In G. Lindzey & E. Aronson, Handbook of social psychology (3rd ed., Vol. 2, pp. 883–948). New York: Random House. Tazelaar, M. A., Van Lange, P. M., & Ouwerkerk, J. W. (2004). How to cope with “noise” in social dilemmas: The benefits of communication. Journal of Personality and Social Psychology, 87, 845–859. Thibaut, J. W., & Kelley, H. H. (1959). The social psychology of groups. New York: Wiley. Thorngate, W. (1975). Possible limits on a science of social behavior. In K. Gergen, R. Aboud, & G. Jahoda (Eds.), Social psychology in transition (pp. 121–139). New York: Plenum Press. Van Lange, P. M., De Bruin, E. N., Otten, W., & Joireman, J. A. (1997). Development of prosocial, individualistic, and competitive orientations: Theory and preliminary evidence. Journal of Personality and Social Psychology, 73, 733–746. Van Lange, P. M., & Rusbult, C. E. (2012). Interdependence theory. In P. M. Van Lange, A. W. Kruglanski, & E. Higgins (Eds.), Handbook of theories of social psychology (Vol. 2, pp. 251–272). Thousand Oaks, CA: Sage Wallach, M. A., & Wallach, L. (1983). Psychology’s sanction for selfishness: The error of egoism in theory and therapy. New York: Freeman. Wickham, R. E., & Knee, R. C. (2012). Where theory and method converge: Interdependence theory and the actor–partner interdependence model. Personality and Social Psychology Review, 16, 375–393. Yovetich, N. A., & Rusbult, C. E. (1994). Accommodative behavior in close relationships: Exploring transformation of motivation. Journal of Experimental Social Psychology, 30, 138–164.

16 Cultural Psychological Theory Kimin Eom Heejung S. Kim

Cultural psychology aims to develop a principle of intentionality by which culturally constituted realities and reality-constituting psyches continually and continuously make each other up, perturbing and disturbing each other, interpenetrating each other’s identity, reciprocally conditioning each other’s existence.      —Richard Shweder (1995, p. 71)

C

ultural psychology has revived the original intention of the cognitive revolution in which psychologists aimed to bring meaning to the study of the mind (Bruner, 1990). In contrast to much of psychological research that has been devoted to discovering “pure” context-free psychological mechanisms, the basic assumption of cultural psychology is that the human psyche cannot exist independently of its sociocultural contexts, and therefore, the study of human actions must consider the contexts in which these actions take place (Shweder, 1995). From the beginning, cultural psychology has aimed to understand the mutual influence between psyche and cultural contexts. According to the framework of mutual constitution (e.g., Fiske, Kitayama, Markus, & Nisbett, 1998), the human psyche is regarded as a product as well as a producer of culture; psychological tendencies are not only shaped by culture but also shape cultural realities. Using this general framework, cultural psychological research has flourished over the last couple of decades, 328

Cultural Psychological Theory 329

providing ample empirical evidence for the idea that culture is an inseparable aspect of human experiences, and thus, a central element to consider in understanding human behaviors. In the present chapter, we review cultural psychology as a field through the lens of the mutual constitution framework. We do so by highlighting some of the field’s notable findings and methodological approaches, and also their strengths and limitations in order to locate empirical efforts within the larger framework of mutual constitution. We also evaluate the basic assumptions prevalent in the field to discuss the explanatory breadth and predictive power of the framework. We first describe the core ideas of the mutual constitution framework and present several empirical approaches to test the ideas put forth by the framework. Next, we briefly summarize a couple of middle-range theories developed from the general framework of mutual constitution and discuss more general issues regarding theory development in cultural psychology. We then identify and evaluate a few basic relatively unquestioned assumptions in cultural psychological research, and sample empirical approaches that address these questions. Finally, we conclude with a brief discussion of future directions and challenges of cultural psychology as a field.

Human Minds as Products and Producers of Culture The framework of the mutual constitution explains how human psychological processes, such as cognitive, emotional, motivational, behavioral, and biological processes, are shaped by individuals’ participation in their cultural worlds that are replete with ideas, values, practices, institutions, and artifacts as shown in different panels in Figure 16.1. Through this participation in specific cultural worlds, individuals adopt particular ways of being and become functional cultural members. The individuals who incorporate certain cultural models into their psyche in turn act according to these models, creating, maintaining, and altering cultural realities that shape their psychology. This cycle of mutual constitution suggests that the human psyche is at the same time a cultural product and a cultural producer. It is important to note that the idea of mutual constitution between culture and psyche is intended as a broad theoretical framework, rather than a specific theory. It aims to provide a way to conceptually organize and to simultaneously consider numerous and divergent aspects of human lives that comprise culture. Thus, the explanantia of the framework are quite broad and inclusive, as they aim to explain both the shaping of psychological processes and the construction of culture. It considers practically all products of human minds, including both things that are external (e.g., documents and texts) and internal (e.g., emotions, and motivation), and the processes that connect the external and internal (e.g., socialization and creation of cultural products) (see arrows in Figure 16.1).

330

SOCIAL THEORIES Production of human minds (e.g., Socialization, Acculturation, Cultural Priming)

Socioculturally, historically rooted ideas and values  Ontologies  Ideologies  Values  Assumptions

Institutions and cultural products  Education  Politics  Law  Religion  Media

Daily situations and episodes in local worlds

Psychological tendencies

Daily choice Decision-making Social interaction Patterns of close relationship

 Perception  Cognition  Emotion  Motivation

Biological tendencies  Genetics  Neural  processes  Physiology

Construction of social realities (e.g., creation of cultural products, co-construction of social situations)

FIGURE 16.1.  Mutual constitution of psychology and culture.

The strength of the framework is that it inherently incorporates multiple levels of analysis and that it explicitly addresses the processes that link these different levels. Given its focus on mutuality, the framework also affords a great deal of flexibility in theoretical and methodological development, and consequently, a wide range of empirical evidence has accumulated. The majority of studies in cultural psychology to date have focused on the process of culture influencing psychological and behavioral processes of individuals (upper arrow in Figure 16.1), probably in part because psychology generally concerns itself with the question of where human psychological tendencies come from. In this process, culture may be conceptualized as a set of shared beliefs and values that are made cognitively salient and accessible through social practices and interactions (Oyserman, Coon, & Kemmelmeier, 2002), or as a system in which meanings, practices, and mental processes and responses are loosely organized and often causally connected (D’Andrade, 2001; Kitayama, 2002). Given the inherent complexity of the concept, culture is unlikely to be fully captured by any one operationalization. Thus, the operationalization of culture in empirical efforts varies a great deal, and it inevitably relies on the use of proxies of culture. The most common way of operationalizing culture is to use existing social categories within which values, practices, and behavioral norms are shared, such as nationality (e.g., Heine et al., 2001; Kim, 2002; Kitayama, Mesquita, & Karasawa, 2006), religious affiliation (e.g., Cohen, 2009; Tsai, Miao, & Seppala, 2007), social class (e.g., Kraus, Piff, & Keltner, 2011; Snibbe & Markus, 2005; Stephens, Markus, & Townsend, 2007; Varnum, Na, Murata, & Kitayama, 2012), or region within a nation (e.g., Cohen,

Cultural Psychological Theory 331

Nisbett, Bowdle, & Schwarz, 1996; Nisbett, 1993). There are other forms of operationalization that are more psychological in nature, such as individual differences measured by value scales and questionnaires (e.g., Lee, Aaker, & Gardner, 2000), and cognitive priming of key cultural concepts (e.g., Lee et al., 2000; Oyserman & Lee, 2008) and cultural icons (e.g., Hong, Morris, Chiu, & Benet-Martínez, 2000). Utilizing at least one or a combination of multiple proxy variables to capture culture, studies show the fundamental influence of culture on how people think, feel, and behave, both within the mind and with other people. In spite of the dominant research paradigm examining processes of culture that influence the psyche in cultural psychology, some notable studies also have provided the other half of the question: how the human psyche influences culture (lower arrow in Figure 16.1). Such efforts are exemplified by studies on cultural products (see Morling & Lamoreaux, 2008, for a metaanalytic review). These studies investigate how cultural products—tangible objects produced by members of a specific culture—are created by the synergy among the intentions of individuals, and thus, reflect the cultural values and norms within their society. In so doing, cultural product research conceptualizes the human psyche as a producer of cultural realities, not just as a product of cultures. Studies consider many different types of cultural products, such as advertisements (Kim & Markus, 1999), church websites (Sasaki & Kim, 2011), children’s books (Tsai, Louie, Chen, & Uchida, 2007), school textbooks (Imada, 2012), paintings (Masuda, Gonzalez, Kwan, & Nisbett, 2008), and online web pages (Wang, Masuda, Ito, & Rashid, 2012) as products of social representation (Moscovici, 1984). Research findings show that these cultural products reflect the cultural values, norms, and psychological characteristics of their creators. For example, the contents of cultural products created in Western cultures tend to be more individualistic (e.g., valuing independence, uniqueness, and high-intensity positive affect) and less collectivistic (e.g., less valuing interdependence, conformity, and low-intensity positive affect) than products created in Eastern or Latin American cultures (Morling & Lamoreaux, 2008). The framework of the mutual constitution of culture and psyche is intended to encourage simultaneous consideration of multiple levels of analysis. Thus, full appreciation of the model requires considering these levels simultaneously in one program of research with the goal of seeking a cultural thread that is present across different aspects of sociocultural environments and psychological tendencies (e.g., Kim & Markus, 1999; Sasaki & Kim, 2011; Snibbe & Markus, 2005). For instance, the article by Kim and Markus (1999) provided an initial example of using the multimethod research paradigm. Contrasting East Asian cultures, where conformity is relatively valued, to American culture, where uniqueness is relatively valued, this article showed that this core cultural valuation of being different from others or being like others is consistently found across different levels of analysis from basic

332

SOCIAL THEORIES

preference judgment (Studies 1 and 2) to choice in a social interaction (Study 3), to the dominant themes found in advertisements (Study 4). Taken together, cultural psychology has accumulated an impressive body of literature over the last couple of decades. These studies have tackled different aspects of the big question of how culture and human psychology make each other up, and their results facilitate the understanding of a broad range of psychological phenomena, informing both the possibilities and limitations of human psychology. In reviewing the field broadly, in the next section we will more specifically discuss theoretical contributions made by cultural psychology by focusing on specific theories developed from the perspective of mutual constitution.

Middle-Range Theories of the Mutual Constitution Framework As noted earlier, the idea of the mutual constitution between culture and psychology is an overarching framework that encompasses all components of culture and psychology in order to inspire middle-range theories—less general, lower-level theories with more specific predictions and hypotheses than grand theoretical frameworks (see Merton, 1968, for a more detailed concept of middle-range theory). Our reference is solely to middle-range theories that are relevant to the mutual constitution framework and that focus on reciprocal maintenance processes between culture and psychology, rather than all theories formulated in cultural psychology. Numerous theories have been developed to elucidate the origins of cultural differences, such as the question of why some cultures become more individualistic and other cultures become more collectivistic. These theories attend to various factors ranging from biological factors (e.g., gene–culture coevolution theory: see Chiao & Blizinsky, 2010, Cavalli-Sforza & Feldman, 1981, and Lumsden & Wilson, 1981; the pathogen prevalence hypothesis: see Fincher, Thornhill, Murray, & Schaller, 2008) to socioecological factors (e.g., voluntary settlement hypothesis: see Kitayama, Ishii, Imada, Takemura, & Ramaswamy, 2006; residential mobility hypothesis: see Oishi, 2010) as potential environmental or biological pressures to develop specific cultural values and ideas. For example, the voluntary settlement hypothesis proposes that life circumstances in frontier regions (e.g., California during 19th century or the Hokkaido region in Japan)—such as survival threats, low population density, and high residential mobility—may be the origin of the cultural values of individualism (Kitayama et al., 2006). These theories offer valuable theoretical insights, with compelling evidence to understand how cultural differences develop to begin with (see Oishi & Graham, 2010, for a socioecological perspective review). Although these are theories based on the cultural psychological perspective and are areas of very active research, our focus in the present chapter is on theories developed to understand how

Cultural Psychological Theory 333

culture, once established, influences individual psychology—notably, collective constructionist theory and affect valuation theory—and we also briefly discuss the issue of middle-range theory building in the field.

Collective Constructionist Theory One specific theory developed directly from the perspective of mutual constitution is the collective constructionist theory (Kitayama, Markus, Matsumoto, & Norasakkunkit, 1997). The theory posits that the co-creation processes between culture and minds occur via everyday situations that are collectively experienced in specific cultural contexts. More specifically, the theory proposes that daily situations are culturally constructed realities, and thus systematically vary from one culture to another. Individuals who subscribe to shared cultural values and assumptions collectively produce daily situations that are consistent with their cultural values and assumptions. Those situations in turn function as mechanisms of promotion and maintenance of a particular set of psychological tendencies. Thus, it is a theory that focuses on the mutual-shaping processes between daily situations and psychological tendencies from the inclusive mutual constitution framework (i.e., the interrelation between the third and fourth panels from the left in Figure 16.1). The specific methodology developed to substantiate the collective constructionist theory is the situational sampling method (e.g., Kitayama et al., 1997; Morling, Kitayama, & Miyamoto, 2002; Uskul, Cross, Sunbay, GercekSwing, & Ataca, 2012). In this research method, by asking participants from different cultural backgrounds to describe certain situations (e.g., situations affecting self-esteem), researchers can analyze how certain situations are defined and constructed in different cultures and how individuals respond to those situations. Researchers typically find that situations produced by different cultural groups have subtle characteristics that reflect dominant psychological tendencies in their respective cultural contexts, even though participants are given an identical prompt. These findings elegantly indicate the process of mutual constitution: Even in situations that apparently serve similar functions (e.g., self-enhancing situations), situations in different cultures present subtle differences that reflect the important values and assumptions shared in their culture. In turn, engaging in these different situations fosters corresponding psychological characteristics (e.g., being in American situations is more effective in promoting self-enhancement, whereas being in Japanese situations is more effective in promoting self-criticism) (see Kitayama et al., 1997, for a detailed discussion).

Affect Valuation Theory Affect valuation theory (Tsai, 2007; Tsai, Knutson, & Fung, 2006) proposes that ideal affect—affective states that people want to feel—is influenced by

334

SOCIAL THEORIES

culture, and consequently, the ideal affect in a specific culture becomes a goal for its cultural members to pursue. More specifically, the theory builds on three main assumptions: (1) ideal affect differs from actual affect, (2) cultural factors shape ideal affect, whereas temperamental factors shape actual affect, and (3) discrepancies between ideal and actual affect lead to moodproducing behaviors in order to reduce the discrepancies. The research findings from affect valuation theory provide empirical evidence on how cultural ideals play a role in mutual constitution processes in shaping not only actual psychological tendencies but also consequences of living up to cultural ideals or failing to do so. When cultures have different ideals (Tsai, Knutson, & Fung, 2006), these culturally varied ideals are manifested in cultural realities such as social situations (e.g., interpersonal interaction patterns; Tsai, Levenson, & McCoy, 2006) and artifacts (e.g., books; Tsai, Louie, et al., 2007). Thus, culturally different ideals motivate individuals to engage more actively in behaviors that help them approach their cultural ideals (e.g., Gobster & Delgado, 1992); failure to get close to cultural ideals may have a negative impact on well-being (Tsai et al., 2006). Although the theory focuses specifically on ideals of affective states, at a broader level, it has significant implications for understanding the processes through which cultural ideals shape and influence a wide range of psychological tendencies beyond affective processes (see Na, Choi, & Sul, 2013, for a related point on culturally valued cognitive styles). Reviewing the field reminds us that despite the ample empirical findings in cultural psychology during the last decades, only a few middle-range theories have been developed from the mutual constitution framework. Rather than articulating overarching theories that explain processes of cultural influence, a majority of studies have focused on testing whether given psychological phenomena (e.g., self-enhancement, cognitive dissonance) are culturally varied. And for their theoretical foundation, researchers generally have relied on the taxonomical organization of cultural systems (e.g., independent vs. interdependent self-construal, analytic vs. holistic mode of thinking). Of course, these approaches have satisfied one goal of cultural psychology, that is to document culturally varied ways of being, generating numerous important predictions and supportive findings about cultural variation in human psychology. These trends, however, at the same time have led the field of cultural psychology to be in a somewhat paradoxical state in which numerous testable predictions are made with few middlerange theories. The value of middle-range theories is their ability to generate specific predictions and hypotheses. Of course, the mutual constitution framework has been a useful tool allowing researchers to systematically consider incredibly complex cultural systems and offering a great explanatory breadth. However, given its broadness and relative lack of middle-range theories, the framework has relatively limited predictive power, and the field still has relatively limited knowledge about how culture shapes psychology and

Cultural Psychological Theory 335

how psychology makes up culture. We thus underscore the importance of developing middle-range theories that could allow researchers to formulate specific, testable, and falsifiable predictions. One of the reasons for the relative lack of theories in the field lies in some of the basic assumptions of the framework that are widely shared without much explicit reflection. As a dominant theoretical framework in cultural psychology, the framework of mutual constitution brought its own set of implicit assumptions and empirical routines into the field. Thus, we now direct our focus to and evaluate basic assumptions in cultural psychology.

Dominant Assumptions in Cultural Psychology and How to Question Them In this section, we outline several basic assumptions commonly shared in cultural psychological research. Some of these assumptions are core aspects of the field, but others are implicitly shared assumptions that are sustained by habitual omission of explicit empirical efforts to address the questions. We will discuss three particular issues: causal understanding of cultural influence, cultural changes and variation within culture, and intersectionality of different layers of cultural influences. In addition to outlining the issues, we will describe a few existing approaches that exemplify the muchneeded effort to address these specific concerns.

Consideration of Causality The key aspect of the framework of mutual constitution is its emphasis on the mutuality of the influences. It recognizes that the causal influences are by nature bi-directional. This position allows great explanatory flexibility in how culture and human minds create and shape each other, but at the same time, it could pose the danger of overinterpreting causality. Whenever shared threads are found across different levels of analysis, it is easy to assume that the observed tendencies exist because of mutual shaping between culture and human psyche. Moreover, coupled with the fact that actual experimental treatment of “culture” is virtually impossible, psychological investigation of the exact processes through which culture shapes the psyche and the processes through which the psyche makes up culture poses a challenge. One useful way to examine the question of causality is use of cultural priming methods (Oyserman & Lee, 2008). The cultural priming studies complement other cultural psychological findings that rely on culture as a measured variable. Moreover, these studies reveal a pathway through which cultural influences occur. Research using cultural priming may be grouped into at least two types. One set of studies focuses on activation of cultural frames via cultural symbols, taking the dynamic constructivist approach

336

SOCIAL THEORIES

(Benet-Martínez, Leu, Lee, & Morris, 2002; Hong et al., 2000; Zhang, Morris, Cheng, & Yap, 2013) or languages (Ji, Zhang, & Nisbett, 2004). These cultural icons (and/or languages) activate a generalized set of cultural frames rather than specific concepts. Studies show that when a certain cultural frame is activated, individuals generally act in a way that is more prevalent in that cultural context (but also see Cheng, Lee, & Benet-Martínez, 2006, for the moderating role of bicultural identity). These studies allow the study of causal processes in which being in a specific cultural context triggers a set of associated behaviors among people who have already acquired the specific cultural knowledge. Another set of studies focuses on direct activation of particular key conceptual elements of a cultural dimension of interest, such as individualistic and collectivistic values and beliefs (see Oyserman & Lee, 2008, for a metaanalytic review). This method aims to uncover the role of a specific aspect of culture that is theorized to be a cause of the observed cultural difference. One example is the pronoun-circling task in which participants are instructed to search and circle the first-person singular pronouns (e.g., I, me, or mine) or plural pronouns (e.g., we, us, or ours) in order to activate individualistic or collectivistic mind-sets, respectively (e.g., Brewer & Gardner, 1996; Gardner, Gabriel, & Lee, 1999). These cultural priming studies successfully demonstrate how stored cultural minds are activated by situational cues, and these stored thoughts, in turn, shape social judgments and behaviors. It is also important to note, however, that these studies focus primarily on the cognitive aspects of culture. Culture is an inclusive concept by nature that cannot be reduced to a psychological schema or a knowledge structure (Miller, 1999). Activated thoughts are elements of cultural systems, not the culture per se. Consequently, it does not allow investigations of how the cultural schemas are internalized into individuals’ minds through participation in specific cultural worlds (e.g., institutions and social interactions as shown in the panels in Figure 16.1). Thus, the cultural priming research effectively captures relatively proximal processes of cultural influence (i.e., cultural schemas’ influence on psychological tendencies), but not necessarily their links with other cultural factors at different levels. For more inclusive ways to uncover the processes through which culture shapes psychology, we argue that the field will need to direct its empirical efforts to the processes of enculturation and acculturation. Culturally shared meanings and perspectives enter the minds of individuals through socialization from birth, through parenting and teaching, as well as through engagement in social practices and interactions with other cultural members. Research suggests that at birth, infants in all cultures are quite similar, and as they get older and psychologically mature, expected cultural differences emerge (e.g., Miller, 1984). Moreover, even fully grown individuals go through psychological changes, and these changes occur especially when one’s cultural context is changed through immigration. Research in accul-

Cultural Psychological Theory 337

turation shows that immigrants adopt the psychological pattern of their new culture through their exposure to and their engagement with the culture (e.g., De Leersnyder, Mesquita, & Kim, 2011). Uncovering these processes would also inform the question about how culture causally shapes human minds.

Consideration of Change and Variation within Cultures The primary focus of the mutual constitution framework is on maintenance and perpetuation of a cultural system. Therefore, the framework is effective in explaining the cultural consistency across different levels of analysis as well as across historical time frames, but it is limited in explaining and predicting specific processes of cultural change and in offering a systematic understanding of variations in the degree to which culture shapes the human mind. We argue that this limitation comes from two implicit assumptions of the framework. First, although theoretical discussion acknowledges culture as a malleable and dynamic system, the framework would predict a set of discernible core cultural values and world views to be consistently present across different levels of analysis and time (e.g., see Kim & Markus, 1999, for consistency across levels of analysis; and see Nisbett, Peng, Choi, & Norenzayan, 2001, for historical consistency). However, the fact that culture can change, and is always changing, is a far more important fact than what has been empirically acknowledged to date. Change in one level of the cultural system brings subsequent change in other interlocked levels, so that significantly different cultural realities are formed and revised. For example, innovative technological developments (e.g., the Internet) leading to different social interaction patterns and/or influential and exceptional thinkers (e.g., Mandela, Darwin) bringing new ideas and values into a society can trigger huge cultural changes. These dynamic processes of cultural change have been investigated relatively little in cultural psychology thus far. Second, although it is acknowledged that how and why people behave, think, and feel vary within a cultural context in the abstract, the extent to which individuals are influenced by cultural engagement is assumed to be relatively unvaried. At the least, the possibility of individual variation in the degree to which people are impacted by culture has not been fully incorporated. In typical research paradigms in cultural psychology that compare psychological characteristics among cultural groups (or among conditions), researchers tend to focus on the mean levels of each culture rather than on individual variations within culture. This paradigm has been used to provide powerful, contrasting characteristics between cultural groups, but may lead researchers to overlook an important question of whether and how cultural influence can be manifested differently among individuals. More recently, some theoretical advances have been made allowing sys-

338

SOCIAL THEORIES

tematic considerations of cultural influence in conjunction with individual differences. One approach comes from the gene–culture interaction model (Kim et al., 2010a, 2010b, 2011; see Kim & Sasaki, 2014, for a review). This model suggests that the degree to which individuals demonstrate culturally prototypical psychological tendencies may vary depending on whether or not individuals carry genotypes associated with greater sensitivity to environmental input. Indeed, studies show that individuals with a more socioemotionally susceptible genotype (e.g., GG genotype of OXTR rs53576) engage more in culturally fostered behaviors (e.g., emotional coping in the United States or emotion suppression in Korea) than those with a less susceptible genotype (e.g., AA genotype of OXTR rs53576) (for details, see Kim et al., 2010a, 2011). These findings show that the degree to which individuals are susceptible to cultural influences may vary and that some of this variation may be shaped by genes. Another approach that examines within-culture variation is the CuPS (culture × person × situation) model (Leung & Cohen, 2011). This model proposes that the ways that the same type of person behaves in particular situations can vary between cultures. For example, Leung and Cohen (2011) showed that those who endorse the value of honor (person) behaved in opposite ways in a situation in which they received a small favor from others (e.g., receiving candies) (situation) according to their cultural backgrounds (culture). Among people from honor cultures (e.g., American Southerners), those who endorse the value of honor are more likely to return a favor to the others who offered a small gift, whereas among people from nonhonor cultures (e.g., Northerners), the same types of people who endorse the value of honor more are less likely to reciprocate a favor. These findings highlight the ideas that individuals may vary in the degree to which they are influenced by cultural norms and that considering this individual difference may help to explain variation in behaviors and responses in specific situations within culture. Conversely, they show that even the same individual difference factors (e.g., genetic factors) do not necessarily predict the same behaviors in different cultural contexts. Thus, these new approaches that integrate between- and within-culture variation provide important frameworks for a more complete understanding of the relationship between culture and psychology.

Consideration of Intersectionality Another important but relatively understudied question in cultural psychology is intersectionality (Crenshaw, 1989; see Cole, 2009, for review). The concept of intersectionality highlights the notion that individuals are often at the intersections of multiple social categories (e.g., race, sex, social status, etc.), and consequently, these simultaneously experienced multiple social categories lead to specific psychological outcomes that cannot be explained by the sum of the effects of those categories. For example, African American

Cultural Psychological Theory 339

females may experience discrimination specifically as Black women that is different from the sum of race and sex discrimination. In cultural psychology, reflecting psychologists’ general preference for relatively simple models, researchers have tended to investigate the effects of a single cultural category or value dimension of interest, such as nationality, religion, or social class, controlling or collapsing across other social/ cultural categories. This is, of course, a necessary approach to scientifically abstract, coherent, and comprehensible patterns (Kim, Sherman, & Taylor, 2009). Yet, it is also important to recognize that one’s cultural experiences are interactively shaped by multiple cultural groups to which one concurrently belongs. For instance, a Christian Asian American would have the unique cultural experience of being at the intersection of his or her religious culture and ethnic culture, which cannot be reduced to the effect of either cultural group. Thus, neglecting the issue of intersectionality would limit understanding of the relationship between culture and psychology, and might even result in biased understanding of any one social/cultural category (Cole, 2009). Recent findings suggest that considering the intersectionality between multiple social/cultural categories may allow researchers to conduct more complicated analyses of relationships between different cultural identities and psychology. For example, Gobel and Kim (2014) investigated national culture and social class in combination, showing that social background signaling behaviors of the high social class vary between national groups. Specifically, they found that in high power distance cultures such as France, high social-class individuals nonverbally signal dominance more than low social-class individuals. In contrast, in low power distance cultures such as the United States, there is no significant difference in dominance signaling between high and low social-class individuals. Also, Grossmann et al. (2012), considering the interaction between age and nationality, showed that Japanese and Americans differ in terms of the change in reasoning styles according to age. Wise reasoning (e.g., consideration of multiple perspectives, flexibility, etc.) increases with older age among Americans, whereas there was no significant association between age and wise reasoning among Japanese. These findings demonstrate that the psychological influences that come from belonging to these different social categories are not necessarily additive. Thus, it would be important to develop specific theories and accumulate empirical findings about how multiple social/cultural categories are simultaneously and interactively associated with psychological outcomes.

Future Challenges in Cultural Psychology In this chapter, in addition to a very brief review of empirical advances in cultural psychology, we reviewed a couple of significant middle-range theories in the field, underscoring the importance of developing more cul-

340

SOCIAL THEORIES

tural psychological theories. Finally, we presented a critical assessment of a few implicit assumptions shared within the field. Reflecting on the field of cultural psychology makes it clear that the mutual constitution framework has allowed cultural psychologists to begin to understand and investigate the interplay between culture and human minds in systematic ways. The main strength of the framework is that it offers great explanatory breadth and reasonably good predictive power at least regarding how psychology and behaviors would manifest themselves in different cultural contexts. The framework allows consideration of cultural environments as inclusive systems and at the same time permits the generation of testable predictions. However, most of the empirical successes have taken place in documenting cultural differences based on a taxonomical understanding of the world, and it has not been as successful in creating theories indigenous to the field of cultural psychology. Moreover, the survey of the field as a whole brings forward the basic assumptions of the framework and underscores the importance of evaluating them. The next set of challenges for cultural psychology lies in generating new theories from a cultural psychological perspective in order to stimulate the next wave of research. The field started with and focused on nationallevel cultural comparisons for the first 10 years or so and since then has been successfully branching out in terms of the topics of research and forms of culture to be investigated over the next 10 years. An impressive number of findings have accumulated showing cultural divergence in psychological phenomena. Perhaps it is now time to place greater focus on the process of mutual influence between culture and psychology. It is important to remember that documentation of culturally varied ways of being is only one of the missions of cultural psychology. Another mission is to understand specific processes of how humans become cultural beings and how cultural systems are made up by human minds. The field needs more formal theories to address this goal. During the past few decades, cultural psychology has made a remarkable set of discoveries showing that humans are products of culture and that humans in turn act as co-creators of culture. The current review presents small slices of cultural psychologists’ endeavors in order to examine the core assumption that launched the field. These efforts collectively testify to the importance of contextualizing the human psyche within the continual cycle of culturally constituted realities and reality-constituting psyches making each other up, as well as underscore the potential for future discoveries. Acknowledgments Preparation of this chapter was supported by a Fulbright Graduate Study Award to Kimin Eom and National Science Foundation Grant No. BCS-1124552 to Heejung S. Kim.

Cultural Psychological Theory 341

References Benet-Martínez, V., Leu, J., Lee, F., & Morris, M. W. (2002). Negotiating biculturalism: Cultural frame switching in biculturals with oppositional versus compatible cultural identities. Journal of Cross-Cultural Psychology, 33, 492–516. Brewer, M. B., & Gardner, W. (1996). Who is this “we”? Levels of collective identity and self representations. Journal of Personality and Social Psychology, 71, 83–93. Bruner, J. S. (1990). Acts of meaning. Cambridge, MA: Harvard University Press. Cavalli-Sforza, L., & Feldman, M. (1981). Cultural transmission and evolution: A quantitative approach. Princeton, NJ: Princeton University Press. Cheng, C. Y., Lee, F., & Benet-Martínez, V. (2006). Assimilation and contrast effects in cultural frame switching bicultural identity integration and valence of cultural cues. Journal of Cross-Cultural Psychology, 37, 742–760. Chiao, J. Y., & Blizinsky, K. D. (2010). Culture–gene coevolution of individualism–collectivism and the serotonin transporter gene. Proceedings of the Royal Society B: Biological Sciences, 277, 529–537. Cohen, A. B. (2009). Many forms of culture. American Psychologist, 64, 194–204. Cohen, D., Nisbett, R. E., Bowdle, B. F., & Schwarz, N. (1996). Insult, aggression, and the southern culture of honor: An “experimental ethnography.” Journal of Personality and Social Psychology, 70, 945–960. Cole, E. R. (2009). Intersectionality and research in psychology. American Psychologist, 64, 170–180. Crenshaw, K. (1989). Demarginalizing the intersection of race and sex: A black feminist critique of antidiscrimination doctrine, feminist theory and antiracist politics. University of Chicago Legal Forum, 138–167. D’Andrade, R. (2001). A cognitivist’s view of the units debate in cultural anthropology. Cross-Cultural Research, 35, 242–257. De Leersnyder, J., Mesquita, B., & Kim, H. S. (2011). Where do my emotions belong? A study of immigrants’ emotional acculturation. Personality and Social Psychology Bulletin, 37, 451–463. Fincher, C. L., Thornhill, R., Murray, D. R., & Schaller, M. (2008). Pathogen prevalence predicts human cross-cultural variability in individualism/collectivism. Proceedings of the Royal Society B: Biological Sciences, 275, 1279–1285. Fiske, A., Kitayama, S., Markus, H. R., & Nisbett, R. E. (1998). The cultural matrix of social psychology. In D. Gilbert, S. Fiske, & G. Lindzey (Eds.), The handbook of social psychology (Vol. 2, 4th ed., pp. 915–981). San Francisco: McGraw-Hill. Gardner, W., Gabriel, S., & Lee, A. (1999). “I” value freedom but “we” value relationships: Self-construal priming mirrors cultural differences in judgment. Psychological Science, 10, 321–326. Gobel, M. S., & Kim, H. S. (2014). The visibility of the hierarchical self: Cultural differences in signaling social rank. Manuscript in preparation. Gobster, P. H., & Delgado, A. (1992). Ethnicity and recreation use in Chicago’s Lincoln Park: In-park user survey findings. In P. Gobster (Ed.), Managing urban and high-use recreation settings (General Technical Report NC-163, pp. 75–81). Washington, DC: U.S. Department of Agriculture. Grossmann, I., Karasawa, M., Izumi, S., Na, J., Varnum, M. E. W., Kitayama, S., et al. (2012). Aging and wisdom: Culture matters. Psychological Science, 23, 1059–1066. Heine, S. J., Kitayama, S., Lehman, D. R., Takata, T., Ide, E., Leung, C., et al. (2001). Divergent consequences of success and failure in Japan and North America: An investigation of self-improving motivations and malleable selves. Journal of Personality and Social Psychology, 81, 599–615.

342

SOCIAL THEORIES

Hong, Y., Morris, M. W., Chiu, C., & Benet-Martinez, V. (2000). Multicultural minds: A dynamic constructivist approach to culture and cognition. American Psychologist, 55, 709–720. Imada, T. (2012). Cultural narratives of individualism and collectivism: A content analysis of textbook stories in the United States and Japan. Journal of Cross-Cultural Psychology, 43, 576–591. Ji, L. J., Zhang, Z., & Nisbett, R. E. (2004). Is it culture, or is it language? Examination of language effects in cross-cultural research on categorization. Journal of Personality and Social Psychology, 87, 57–65. Kim, H. S. (2002). We talk, therefore we think? A cultural analysis of the effect of talking on thinking. Journal of Personality and Social Psychology, 83, 828–842. Kim, H., & Markus, H. R. (1999). Deviance or uniqueness, harmony or conformity? A cultural analysis. Journal of Personality and Social Psychology, 77, 785–800. Kim, H. S., & Sasaki, J. Y. (2014). Cultural neuroscience: Biology of the mind in cultural context. Annual Review of Psychology, 64, 487–514. Kim, H. S., Sherman, D. K., Mojaverian, T., Sasaki, J. Y., Park, J., Suh, E. M., et al. (2011). Gene–culture interaction: Oxytocin receptor polymorphism (OXTR) and emotion regulation. Social Psychological and Personality Science, 2, 665–672. Kim, H. S., Sherman, D. K., Sasaki, J. Y., Xu, J., Chu, T. Q., Ryu, C., et al. (2010a). Culture, distress and oxytocin receptor polymorphism (OXTR) interact to influence emotional support seeking. Proceedings of the National Academy of Sciences, 107, 15717–15721. Kim, H. S., Sherman, D. K., & Taylor, S. E. (2009). The irony of cultural psychology research. American Psychologist, 64, 564–565. Kim, H. S., Sherman, D. K., Taylor, S. E., Sasaki, J. Y., Chu, T. Q., Ryu, C., et al. (2010b). Culture, the serotonin receptor polymorphism (5-HTR1A), and locus of attention. Social Cognitive and Affective Neuroscience, 5, 212–218. Kitayama, S. (2002). Culture and basic psychological processes—Toward a system view of culture: Comment on Oyserman et al. (2002). Psychological Bulletin, 128, 89–96. Kitayama, S., Ishii, K., Imada, T., Takemura, K., & Ramaswamy, J. (2006). Voluntary settlement and the spirit of independence: Evidence from Japan’s “Northern frontier.” Journal of Personality and Social Psychology, 91, 369–384. Kitayama, S., Markus, H. R., Matsumoto, H., & Norasakkunkit, V. (1997). Individual and collective processes in the construction of the self: Self-enhancement in the United States and self-criticism in Japan. Journal of Personality and Social Psychology, 72, 1245– 1267. Kitayama, S., Mesquita, B., & Karasawa, M. (2006). Cultural affordances and emotional experience: Socially engaging and disengaging emotions in Japan and the United States. Journal of Personality and Social Psychology, 91, 890–903. Kraus, M. W., Piff, P. K., & Keltner, D. (2011). Social class as culture: The convergence of resources and rank in the social realm, Current Directions in Psychological Science, 20, 246–250. Lee, A. Y., Aaker, J. L., & Gardner, W. L. (2000). The pleasures and pains of distinct selfconstrual: The role of interdependence in regulatory focus. Journal of Personality and Social Psychology, 78, 1122–1134. Leung, A. K. Y., & Cohen, D. (2011). Within- and between-culture variation: Individual differences and the cultural logics of honor, face, and dignity cultures. Journal of Personality and Social Psychology, 100, 507–526. Lumsden, C. J., & Wilson, E. O. (1981). Genes, mind, and culture. Cambridge, MA: Harvard University Press. Masuda, T., Gonzalez, R., Kwan, L., & Nisbett, R. E. (2008). Culture and aesthetic prefer-

Cultural Psychological Theory 343 ence: Comparing the attention to context of East Asians and European Americans. Personality and Social Psychology Bulletin, 34, 1260–1275. Merton, R. K. (1968). Social theory and social structure. New York: Free Press. Miller, J. G. (1984). Culture and the development of everyday social explanation. Journal of Personality and Social Psychology, 46, 961–978. Miller, J. G. (1999). Cultural psychology: Implications for basic psychological theory. Psychological Science, 10, 85–91. Morling, B., Kitayama, S., & Miyamoto, Y. (2002). Cultural practices emphasize influence in the United States and adjustment in Japan. Personality and Social Psychology Bulletin, 28, 311–323. Morling, B., & Lamoreaux, M. (2008). Measuring culture outside the head: A meta-analysis of individualism—collectivism in cultural products. Personality and Social Psychology Review, 12, 199–221. Moscovici, S. (1984). The phenomena of social representation. In R. M. Farr & S. Moscovici (Eds.), Social representation (pp. 3–69). Cambridge, UK: Cambridge University Press. Na, J., Choi, I., & Sul, S. (2013). I like you because you think in the “right” way: Culture and ideal thinking. Social Cognition, 31, 390–404. Nisbett, R. E. (1993). Violence and U. S. regional culture. American Psychologist, 48, 441–449. Nisbett, R. E., Peng, K., Choi, I., & Norenzayan, A. (2001). Culture and systems of thought: Holistic versus analytic cognition. Psychological Review, 108, 291–310. Oishi, S. (2010). The psychology of residential mobility implications for the self, social relationships, and well-being. Perspectives on Psychological Science, 5, 5–21. Oishi, S., & Graham, J. (2010). Social ecology: Lost and found in psychological science. Perspectives on Psychological Science, 5, 356–377. Oyserman, D., Coon, H. M., & Kemmelmeier, M. (2002). Rethinking individualism and collectivism: Evaluation of theoretical assumptions and meta-analysis. Psychological Bulletin, 128, 3–72. Oyserman, D., & Lee, S. W. S. (2008). Does culture influence what and how we think? Effects of priming individualism and collectivism. Psychological Bulletin, 134, 311–342. Sasaki, J., & Kim, H. S. (2011). At the intersection of culture and religion: A cultural analysis of religion’s implications for secondary control and social affiliation. Journal of Personality and Social Psychology, 101, 401–414. Shweder, R. (1995). Cultural psychology: What is it? In N. R. Goldberger & J. B. Veroff (Eds.), The culture and psychology reader (pp. 41–86). New York: New York University Press. Snibbe, A. C., & Markus, H. R. (2005). You can’t always get what you want: Social class, agency, and choice. Journal of Personality and Social Psychology, 88, 703–720. Stephens, N. M., Markus, H. R., & Townsend, S. S. M. (2007). Choice as an act of meaning: The case of social class. Journal of Personality and Social Psychology, 93, 814–830. Tsai, J. L. (2007). Ideal affect: Cultural causes and behavioral consequences. Perspectives on Psychological Science, 2, 242–259. Tsai, J. L., Knutson, B. K., & Fung, H. H. (2006). Cultural variation in affect valuation. Journal of Personality and Social Psychology, 90, 288–307. Tsai, J. L., Levenson, R., & McCoy, K. (2006). Cultural and temperamental variation in emotional response. Emotion, 6, 484–497. Tsai, J. L., Louie, J. Y., Chen, E. E., & Uchida, Y. (2007). Learning what feelings to desire: Socialization of ideal affect through children’s storybooks. Personality and Social Psychology Bulletin, 33, 17–30. Tsai, J. L., Miao, F., & Seppala, E. (2007). Good feelings in Christianity and Buddhism: Religious differences in ideal affect. Personality and Social Psychology Bulletin, 33, 409–421. Uskul, A. K., Cross, S. E., Sunbay, Z., Gercek-Swing, B., & Ataca, B. (2012). Honor bound:

344

SOCIAL THEORIES

The cultural construction of honor in Turkey and the Northern United States. Journal of Cross-Cultural Psychology, 43, 1131–1151. Varnum, M. E. W., Na, J., Murata, A., & Kitayama, S. (2012). Social class differences in N400 indicate differences in spontaneous trait inference. Journal of Experimental Psychology: General, 141, 518–526. Wang, H., Masuda, T., Ito, K., & Rashid, M. (2012). How much information? East Asian and North American cultural products and information search performance. Personality and Social Psychology Bulletin, 38, 1539–1551. Zhang, S., Morris, M. W., Cheng, C., & Yap, A. J. (2013). Heritage-culture images disrupt immigrants’ second-language processing through triggering first-language interference. Proceedings of the National Academy of Sciences, USA, 110, 11272–11277.

Part VI Formal Theories





17 Computer Simulation Klaus Fiedler Florian L. Kutzner

W

hen naturally existing systems are too complex and dynamic to be assessed and explained in all detail, simplifying simulation models can nevertheless help researchers to understand the system and to develop testable hypotheses about its behavior. During the last decades of social-psychological research, simulation models have been developed and tested for such diverse phenomena as mating behavior (Kalick & Hamilton, 1986), majority and minority influences (Nowak, Szamrej, & Latané, 1990), group decisions (Penrod & Hastie, 1980) and multiagent interaction (Sun, 2006), peace and conflict resolution (Nowak, Deutsch, Bartkowski, & Solomon, 2010), stereotype formation and distribution (Kunda & Thagard, 1996), language acquisition (Elman, 1993), person memory (Hastie, 1988), and judgment and decision making (Dougherty, Gettys, & Ogden, 1999).

Why Computer Simulation? While the diverse computer models that have been proposed for simulation research rely on different mathematical and structural notations, they all share a number of desirable properties that benefit scientific progress and the development of overarching theories (Kruglanski, 2004). As sum 347

348

FORMAL THEORIES

marized by Hastie and Stasser (2000), simulation models (1) force researchers to create an internally consistent theory that can be implemented on a computer; (2) motivate researchers to provide complete accounts that leave fewer assumptions implicit than when working with traditional methods; and (3) afford efficient deductive tools for deriving new theoretical predictions that often go way beyond common sense. Further useful aspects of simulation approaches might be added. They (4) enhance the transparency of theoretical models, the precision of which is no longer restricted by ordinary language. They (5) detach the implications of a model from the theoretician’s subjective opinion. And last but not least, they (6) mimic quasiempirical evidence where the understanding and lining of pale and abstract theories are low. Let a “classical” example illustrate these desirable properties of simulation research. Using Latané’s (1981) social impact theory to simulate the consequences of mutual interactions among all majority and minority members of a social system, Nowak et al. (1990) devised a simplified but meaningful computer model. A relatively large number of individuals (1600) holding one of two opposite attitudinal positions (coded as 1 vs. –1) were arranged in a two-dimensional (40 × 40) matrix, which determined the distance dij between one individual i and all other individuals j. The numbers of other individuals holding the same attitude and the opponent attitude, Ns and No, were varied across simulation runs, and the power or strength of attitudes, pj, varied between individuals. In each simulated interaction cycle, the persuasive impact PIi on individual i’s attitude of all others holding opposite attitudes was assumed to be PIi = No1/2⋅[Σ(pj,/d2ij)/No], while the supportive impact SIi of individuals holding the same attitude (including i) was SIi = Ns1/2⋅[Σ(p,/d2i)/Ns]. The model thereby implements social impact theory’s key assumption that social influence is a multiplicative function of the number (N), power (p), and distance (d) of interaction partners. The simulation algorithm would let individual i’s attitude switch whenever PIi was stronger than SIi and remain unchanged otherwise. After a series of (typically a few dozens of) such interaction cycles, in which the net effect of social impact on all individual attitudes was recomputed accordingly, an equilibrium was found; there were no longer any attitude changes. Several meaningful results emerged regularly across numerous simulation runs. Consistent with empirical research on group dynamics, the repeated operation of social impact led to attitude polarization; attitudes became more extreme and less moderate than at the outset. Whereas minority and majority attitudes were randomly distributed at the beginning, the resulting patterns in the 40 × 40 matrix reflected clear-cut clustering effects. Minority attitudes persisted only in compact subgroups or neighborhoods. While many minority members had been “swallowed by” the majority, reflecting an overall conformity effect, only highly coherent minorities (sharing strong congruent attitudes) survived, typically in marginal regions of the grid. Altogether, these intriguing results demonstrated nicely how macro-

Computer Simulation 349

level phenomena like segregation and polarization emerge as a result of the repeated operation of simple individual-level rules. All aforementioned assets of simulation research are nicely evident in this seminal study. Researchers were forced to articulate a precise theory of social influence; they had to commit themselves to a specific implementation of social impact theory rather than leaving many assumptions implicit; and their simulation model actually turned out to account for existing group phenomena as well as to suggest new implications for future research (e.g., regarding the conditions for minority influence). Moreover, the Nowak et al. (1990) study highlights the ability of simulation methods to render highly complex processes (involving 1,600 individuals) amenable to quasi-empirical inquiry. This is not to say that a useful and enlightening simulation model must provide a full account of the real phenomenon it is meant to elucidate. Most simulation studies provide nothing but existence proofs that one causal mechanism is sufficient to produce a certain effect. It cannot exclude the possibility that in reality the same effect can also be brought about by other causal influences or may reflect multiple causes. Thus, the Nowak et al. simulation model demonstrates that under idealized conditions, when no other influences are at work, the repeated operation of pairwise social impact will lead to typical patterns of attitude change. However, it does not exclude many other influences on attitude change that may complicate the prediction of attitude change in real life.

Distinctive Features of Different Types of Simulation Approaches Simulation models emphasize the role of structural principles, as distinguished from concrete phenomena. The same structural principle of repeated pairwise interaction in a social network that Nowak et al. (1990) applied to social impact on public opinions might be applied to the simulation of the distribution of, say, distinct myths or social stereotypes through interpersonal communication. For example, the acquisition and maintenance of the belief that physically handicapped people are humorless might similarly depend on the variables of social impact theory, namely, number, power, and distance of people’s interaction partners. The same simulation model can thus be used to explain a variety of structurally similar phenomena; they are rarely confined to explaining singular phenomena. Because our illustration relied on the simulation of the consequences of multiple interactions between social agents, it may be considered an example of agent-based modeling, which can be distinguished from variable-based modeling (cf. Smith & Conrey, 2007). The aim of agent-based modeling—the more focused topic of Chapter 19 of this volume—is to explain aggregate phenomena (at group, institutional, or societal level) as an emergent bottom-

350

FORMAL THEORIES

up result of reiterating interactions between multiple individual agents. The purpose of variable-based models, in contrast, is to examine the causal and statistical relationships that hold between the variables of theoretical interest. Note, however, that the distinction is not exclusive. Depending on the researcher’s style of theorizing, the Nowak et al. (1990) simulation can be framed either as an emergent outcome of the repeated interactions between multiple agents or in terms of a causal model of the joint (multiplicative) impact of three distinct variables, numerosity, power, and distance. Similarly, Axelrod’s (1980) famous demonstration of the success of tit for tat1 among competing strategies in the prisoner’s dilemma game speaks to agents who apply tit for tat, or to tit for tat as a variable. A common denominator of all these examples is that behavior is conceived as an adaptive interplay of individual and environmental influences. A noteworthy distinction in simulation research, therefore, is between environmental models that explicitly represent the individual as part of an overarching ecological context and purely intrapsychic models that are confined to cognitive, affective, or motivational processes within the individual. Other distinctions refer to the structure of the simulation algorithm and the interpretation of underlying units of analysis. With regard to the elementary units, one can distinguish between symbolic models, using meaningful entities as units (such as stimulus objects, attributes, roles, or names) and subsymbolic models that represent these entities in terms of more primitive elements (such as binary vector patterns). Regarding the learning algorithm, one can distinguish error-reducing recurrent feedback models that permanently change their operation as a function of feedback about the accuracy of the model’s predictions from unidirectional feedforward models that simply aggregate a growing amount of stimulus input.

Insights Gained from Simulation Studies in Social Psychology This chapter seeks not only to introduce some prominent simulation approaches but also, and primarily, to demonstrate the aforementioned assets of simulation models: their transparency, their explanatory and predictive power. We point out how simulation studies have fertilized and inspired new empirical research in several areas, providing alternative explanations of old findings and innovative implications for future research. While any simulation algorithm must be internally coherent to be implemented on a computer, the simulation-based explanations and implications may be tested and falsified empirically. From a sketch of some notable debates between different simulation approaches we finally try to derive conclusions about the crucial catalysts of simulation success: What aspects of real systems have simulation studies revealed to be most essential for the explanation of social behavior? The

Computer Simulation 351

remainder of this chapter is ordered from simple to more complex simulation approaches. Starting from simple extrapolation models of the cumulative effect of a multiply repeated operation, we move on to simulations of cumulative reward from instrumental behavior, to ecologically constrained sampling processes, and then to comparative simulations of the impact of small versus large amounts of information.

Illustrating the Accumulation of Tiny Influences A widely shared rule says that simulation should be restricted to research questions that cannot be solved analytically. Sometimes, however, simulations are also applied to analytically obvious facts just because empirically working researchers can be more easily persuaded by concrete and vivid results than by abstract and pallid arguments. For example, although the formula to compute compound interest from interest is well known, a graph that depicts the development of cumulative interest over time provides more impressive evidence. Likewise, it is easy to see analytically that unanimous group decisions are unlikely to be correct if the individual group members’ accuracy is imperfect. Assuming a constant individual correctness chance of pI = .8, the likelihood that a five-person group is unanimously correct is only pG = .85 = .33. Nevertheless, a quasi-empirical demonstration of this truism in the context of a refined simulation study can be very informative and compelling (cf. Stasser, 1988). Therefore, a respectable purpose of simulation research is to render incontestable rules more vivid and intelligible. One such incontestable rule is statistical regression (Galton, 1877), suggesting that it is sometimes advisable to engage in anticyclic behavior. To illustrate, Hubert (1999) conducted a retrospective simulation study of different investment strategies on the stock market. He compared the long-term returns that would have been earned over several years if one had consistently invested in high-performing versus low-performing shares. An anticyclic strategy that exploit the regressive nature of the stock market—that is, the tendency of low indices to rise and high indices to drop—proved to be clearly superior. That is, ironically, investing in shares that recently outperformed the market may lead to less profit than investing in recently less successful shares (Hubert, 1999). Note that the purpose of simulation is not to model real investors’ behavior that will hardly invest only in high- or low-performing shares over a longer period. The purpose is rather to understand the causal impact of a clearly defined, pure strategy under idealized conditions. Simulation studies that only visualize and illustrate the cumulative results of well-understood statistical or ecological laws should not be devalued as scientifically insignificant or trivial. They actually afford a powerful method in high-esteem disciplines, such as economy or evolution science. In biological evolution, one can simulate the long-term cumulative effect of tiny reproduction or survival advantages, as a function of distinct assump-

352

FORMAL THEORIES

tions about possible changes in climate, nutrition or mobility. Such simulation models of the evolution process may lead to new theoretical insights, such as the insight that long-term success is not merely a function of finding an equilibrium or optimal solution, but is also contingent on variability in a portfolio (of partially suboptimal) solutions. To understand the key role of variability, imagine some rodent species whose dark brown skin gives it a survival advantage, because dark animals are hard to detect for their predators. Albino mutants of the same rodent are more easily detected and caught and therefore decrease the overall survival rate of the species. Nevertheless, a simulation model that extrapolates assumptions about reproduction rate, likelihood of falling prey to predators, and ecological changes may figure out distinct conditions under which a mixture of dark and (suboptimal) albino phenotypes will benefit survival (e.g., when white albinos help the species to overcome extended winters in which the landscape is covered with snow). In biology, this survival advantage is called adaptive coin-flipping (Cooper & Kaplan, 1982). Waiting for an experiment of nature on adaptive coin-flipping would clearly exceed any researchers’ lifetime. Computer simulation, however, offers a helpful surrogate.

Illustrating the Impact of Reward: Instrumental Learning of Attitudes In his seminal work on reinforcement, Herrnstein (1961) postulated a matching rule, saying that organisms tend to respond to a stimulus at the same rate as they are reinforced by that stimulus. Thus, when the winning rates of two options A and B in a binary choice lottery are pwin(A) = .75 and pwin(B) = .25, respectively, reward can be maximized by always choosing A with pchoose(A) = 1.00, whereas probability matching means to afford choosing A only with pchoose(A) = .75. Probability matching was already a topic in Bandura’s (1965) early work on reinforcement in social learning. The take-home message from this research is that it pays off in the long run to invest into exploration at the expense of the optimal exploitation of the best strategy (Schul, Mayo, Burnstein, & Yahalom, 2007). Nevertheless, the rate of approaching or choosing a stimulus decreases stochastically with the rate of negative reinforcement or losses associated with that stimulus. This basic “law of effect” (Thorndike, 1927) had a strong impact on recent simulation modeling on the learning of attitudes in probabilistic environments.

Attitude Learning A pertinent simulation approach by Eiser, Fazio, Stafford, and Prescott (2003) is best explained by the experimental work it inspired (Fazio, Eiser, & Shook, 2004). Like an animal whose survival depends on finding food providing sufficient energy, the participants’ task was to identify and col-

Computer Simulation 353

lect high-energy beans. The environment consisted of an array of 100 beans representing all combinations of 10 increasingly longish shapes and 10 densities of speckles. Whenever a bean was chosen and consumed, participants would receive positive or negative feedback about high versus low energy level. Because the available energy was limited and exhausted over time, learning to select positive and avoid negative beans was of utmost importance for survival. In the critical contingent-feedback condition, feedback was only received for chosen and consumed beans; nothing could be learned about the energy value of nonchosen beans. An efficient and almost necessary strategy in such a learning environment is to avoid negative beans, in accordance with the law of effect. A radical avoidance strategy, however, will produce a typical negativity bias in attitude learning. Whenever a bean (or its “neighbors” with similar shape and freckle) is associated with negative value, correctly or erroneously, it will be avoided so that negative initial attitudes cannot be corrected. Learning about positive attitude targets, in contrast, will continue. Unwarranted positive attitudes can thus be corrected, whereas unwarranted negative attitudes become irreversible. Crucially, though, this avoidance of negative stimuli, and its carryover to neighboring stimuli, should only occur in a contingent-feedback schedule. If choosing and consuming the beans is not a precondition for feedback learning, the asymmetry disappears. This intriguing framework of attitude learning, which has obvious implications for the persistence of prejudice and negative stereotypes, was the focus of two different simulation models that emerged roughly at the same time in different literatures. These two models can be used to illustrate the difference of symbolic versus subsymbolic or connectionist models. Let us first consider Denrell’s (2005) symbolic implementation of this experience-sampling approach, which figures prominently in the recent decision-making literature. This model explains attitude learning by two simple and straightforward assumptions: First, every valenced (positive or negative) experience vt+1 at time t + 1 with a target bean X changes the old attitude xt at time t by the weighted-average rule xt+1 = (1 – b)⋅xt + b)⋅vt+1. The parameter b determines the weight given to the new experience vt+1. Second, for every update of the attitude x, the probability p of sampling the target bean increases or decreases after positive or negative experience vt+1, respectively, according to the logistic function p = ec+Sx / 1 + ec+Sx. If the parameter S is relatively high (e.g., S = 3), information search is strongly dependent on the current evaluation or attitude. If b is also high (e.g., b = .5), attitudes change quickly with new hedonic experience. Both factors together will soon lead to the radical avoidance of negative beans, the “hot-stove effect” (Denrell & March, 2001). This is the crucial mediator underlying the predicted negativity bias. Note that this prediction can be tested empirically in behavioral research. Variants of the same hedonic-sampling rule offer new explanations of social influence (conceived as the joint experience of two

354

FORMAL THEORIES

people exposed to overlapping samples), or conformity (Denrell & Le Mens, 2007). As illustrated in the left-hand part of Figure 17.1, the attitude targets (e.g., bean X) and attribute meanings (e.g., negative valence) themselves provide the representational units in a symbolic model. As a function of the frequency of the joint experience of X and negative (vs. positive) valence, the corresponding connection weights in the network will change and reflect the outcome of evaluative learning. Let us now consider, for comparison, the subsymbolic model suggested by Eiser et al. (2003) for the Beanfest phenomenon. Rather than using an algebraic equation for the association of attitude targets and valence, this model assumes that targets and attributes are represented as distributed patterns of subsymbolic features, offering an intrinsic account of similarity-based stimulus generalization. The model consists of several layers (see right-hand part of Figure 17.1). In the input layer, each presentation of a stimulus bean is represented as a pattern of elementary units, each of which can take on an activation value between 0 and 1. The overlap of the vector patterns defines the similarity between the respective beans in terms of shape and speckle. Each unit of the input layer is connected with each of three units in a hidden layer that may be thought of as a conceptual transformation of the perceptual representation in the input layer, depending on the positive or negative connection weights that determine the direction and degree to which input unit activation increases or decreases the activation of the units in the hidden layer. The three hidden units are then connected to a single-output unit supposed to reflect positive versus negative evaluation and thus to determine the decision to choose or not to choose a bean. When choosing a bean, a positive or negative experience is then fed back into the system, and the weights connecting the units are adjusted accordingly. The weights of connections supporting a successful decision increase, whereas the weights of the connections working against a successful decision will decrease. An updating in the opposite direction takes place after a wrong decision. A noteworthy distinctive feature of such a connectionist model is that it offers a natural explanation for stimulus generalization. One of the most intriguing results of Fazio et al.’s experimental test of the model was that negative attitudes carried over to neighboring beans of similar shape and speckle. No auxiliary assumptions had to be introduced to account for this carryover, which is naturally built into the subsymbolic representation of the shape and speckle attributes defining the beans’ identity. This may be considered an advantage over the Denrell (2005) model, which is, however, more parsimonious, being able to account for asymmetric attitude learning in terms of two simple and straightforward assumptions. One should thus refrain from asking which of the two models is superior; they apparently fulfill different fertilizing functions for research and theorizing on attitude learning. The precision of such algorithmic models renders theoretical reasoning

Computer Simulation 355 Positive valence

Negative valence Output layer Weighted connections Hidden layer

Weighted connections Weighted connections W

X

Y

Target nodes

Input layer

Z Targets represented by patterns of elementary units

FIGURE 17.1.  Illustration of a symbolic (left-hand side) and a subsymbolic (righthand side) connectionist network.

sharp and transparent. A phenomenon is not only re-labeled in terms of sonorous nominalizations (e.g., referring to the “negativity bias” or the “dominance of avoidance behavior”). Rather, an algorithm produces the phenomenon when implemented on a computer. Such a simulation approach fosters new insights and enlightening conclusions. Thus, the models advocated by Denrell (2005) and Eiser et al. (2003) both demonstrate, unequivocally, that attitude learning can be biased without any biased cognitive or motivational process, just as a cumulative consequence of the law of effect. No sentiment or deeply anchored prejudice is required to produce these effects. Regardless of how closely these simulation models really describe the process of attitude learning in reality, they provide an incontestable proof of principle that biased attitudes are possible without biased processes. At the same time, the precision and clarity of simulation-based theorizing instigate critical reasoning about limiting conditions. An open question that suggests itself from a psychological point of view is whether attitude learning must inevitably produce a negativity bias. Under what conditions might the process be more balanced, or might attitudes even exhibit a positivity bias? One answer emphasized by Fazio et al. (2004) is that only feedback-contingent sampling leads to biased learning, that is, only learning guided by exploitation (i.e., consumption of the beans) rather than mere exploration goals (Hills & Hertwig, 2010). Neither the simulation studies nor the aligned experiments led to any negativity bias when learning was detached from the need to choose and eat a bean. Another answer lies in the algorithm’s parameters. When the parameters (S and b) used by Denrell (2005) decrease so that the sampling of information about negative stimuli will not be truncated abruptly and completely, then there is still a chance to correct for inappropriate negative impressions and even allow another process to induce a reversal.

356

FORMAL THEORIES

Evaluative Learning Experiments and simulations in a simulated classroom paradigm (Fiedler, Walther, Freytag, & Plessner, 2002; Fiedler, Woellert, Tauber, & Hess, 2013) highlight another reason why the negativity bias may reverse into the opposite, a positivity bias. In this paradigm, participants take the role of a teacher who asks knowledge questions to the students of a school class represented on the computer screen. Their task is to provide accurate and fair judgments of “their” students’ performance, based on the information gathered about all individual students’ proportions of correct and incorrect responses. Although this looks like a purely epistemic (exploration) task, it actually has a hedonic component. Teachers prefer to ask more questions from smart (with a high parameter for the proportion of correct responses) than from weaker students (with a low parameter), presumably because they identify with “their” class and because smart students may reflect good teaching. In any case, empirical evidence shows that subsequent judgments of the students’ performance reflect a distinct positivity bias. With increasing numbers of observations, judgments become generally more extreme: Evaluations of positive students become even more positive, and evaluations of negative students become even more negative. However, crucially, because larger samples are drawn from positive students, this polarization effect is more pronounced for positive than for negative evaluations. Note that a simulation model that originally predicted a negativity bias inspired new empirical research resulting in a positivity bias. Why does the hedonic preference for desirable information produce a negativity bias in the Beanfest but a positivity bias in the simulated school class? A first sketch of the underlying simulation model (BIAS; Fiedler, 1996) helps to understand this reversal. In this framework, the stimulus attributes are also represented as distributive vector patterns. For a simple illustration, let every positive (p) or negative (n) experience with a student (i.e., every correct and incorrect response, respectively) be represented as a pattern of 12 binary features (cf. Figure 17.2). Correct responses are encoded as noisy copies of an ideal pattern (P) of positive valence, in which two randomly selected elements have been inverted. Incorrect responses are noisy copies of an ideal pattern (N) of negative valence. Assuming that evaluative learning can be conceived as an averaging or summation process (Anderson, 1965), aggregating across all student responses (i.e., all matrix columns in Figure 17.2) yields aggregate patterns (A8 and A18) that resemble the more prevalent valence pattern (i.e., the ideal pattern P of positive valence). The two parts of the matrix, on the left and on the right of Figure 17.2, represent the first eight and the second eight observations of the same student with a high ability parameter of p(correct) = .75. Indeed, 75% of the columns in the entire matrix are noisy copies of the ideal positivity pattern. They are noisy in that two elements of the ideal pattern have been randomly inverted. Because these positive columns (denoted “p”) are clearly the majority, the

Computer Simulation 357

Observation:

Ideal patterns 1

4 0 0 4 2 4 0 2

2

3

4

….

4 4 2 2 0 0 4 2 =

16 = i SRi

R

+ -

+ -

+ + +

+ -

+ + -

+ -

+

+ +

+ -

+ -

+ +

+ + -

+ + + -

+ -

+ -

+ + -

+ -

Valence segment

+ + + + + + -

+ -

+ + + + + + +

+ + + + + + -

+ + +

+ + + + + -

+ + + + +

+ + + + +

+ + + + + +

+ + -

+ + + + + +

+ + + + +

+ + + + +

+ + + + -

+ + + -

+ + + + + -

Outcome:

p n p p n p p p

Cross product:

Target segment

8

9 ….

A8

16 0 12 8 -8 12 8 -8

A16

18 -2 -2 2 -2 22 22 -10

p n p p n p p p rA8P = .20 after n=8

+ + + + + + -

+ +

P N rA16P = .59 after n=16

FIGURE 17.2. Graphical illustration of BIAS (Brunswikian induction algorithm for social cognition; Fiedler, 1996). Positive (p) and negative (n) stimulus behaviors observed in a target person named Ruth appear as column vectors in a matrix. Each stimulus is represented as a pattern of binary elements. As all behaviors refer to Ruth, the target segment (first 4 elements) always resembles the ideal pattern R of Ruth. The valence segment (last 8 elements) resembles the ideal pattern of either positive or negative valence, P or N. The aggregate patterns A8 and A16 obtained after 8 or 16 observations, respectively, are the sums of all columns weighted by SRi, the similarity (dot-product) of each observed pattern i with the Ruth pattern R. Because most behaviors are positive (6:2), the sign pattern of both aggregates resembles P rather than N. However, as the number of observations increases from 8 to 16, the aggregate’s correlation increases from rA8P = +.20 to rA16P = +.59.

sign patterns of the aggregate traces (A8 and A16, reflecting weighted sums of all matrix columns, weighted by SRi as explained below) also correlate substantially with the ideal pattern (P) of positivity. However, crucially, this correlation (i.e., the simulated measure of valence) is more pronounced after 16 observations (see A16) than after only eight observations (see A8), simply due to increasing sample size. Thus, rather than depolarizing unwarrantedly strong positive impressions, extended sampling from positive sources can serve to polarize positive impressions. Closer inspection of the simulation algorithms helps to understand the conditions under which attitude learning accentuates positive rather than negative impressions. On one hand, when attitude targets are human entities

358

FORMAL THEORIES

(i.e., students, rather than beans), a few recent negative observations will not cause a dramatic truncation of all information sampling. A teacher cannot stop sampling from low-performing students, as confirmed empirically by Fiedler et al. (2013). Technically speaking, the impact of overly high parameters S (radical truncation) and b (strong recency effect) has been eliminated in the BIAS account of teachers’ evaluative learning. On the other hand, an essential property of BIAS is noise. If all stimulus items were perfect replicas of positive or negative valence, both matrix sums in Figure 17.2 would bear perfect (noiseless) correlations with ideal positivity. Extended sampling from positive targets (i.e., the law of effect) can only be expected to produce a polarization effect when there is noise in the system. Noise or error variance is a precondition for an aggregation advantage with increasing sample size. This principle is at the heart of many simulation findings (Kutzner, Vogel, Freytag, & Fiedler, 2011), as the studies reviewed in the next section will demonstrate.

Regression and Polarization: Mere Thinking, Outgroup Homogeneity, and Self versus Others The aggregation effect is the key to understanding an entire class of models that resemble BIAS in many respects. In Hintzman’s (1986) MINERVA model (see also Smith, 1991), individual stimulus observations are also encoded in a matrix representation as in Figure 17.2. Each stimulus pattern (i.e., each matrix column) consists of several segments supposed to represent different stimulus aspects (e.g., a person’s name, her academic achievement, group membership, etc.). For example, to simulate a judgment of Ruth (e.g., of the achievement of a person named Ruth), the name pattern R would be used as a judgment prompt and the similarity SRi of this prompt to the name segments of all memory traces i (columns) in the matrix would be computed (i.e., the dot product of the respective patterns leaving a single similarity value). In BIAS, the entries in each trace would then be multiplied or weighted by the trace’s similarity to the Ruth segment or SRi. MINERVA uses S3Ri as weights to increase the differential impact of Ruth-like and non-Ruthlike entries. In any case, the vector resulting from the horizontal sum of this weighted matrix would finally be correlated, in the segment of the attribute to be judged, with the ideal type of positive valence or achievement (P). Models of this type produce polarization with an increasing amount of information. As the number of matrix columns (representing noisy manifestations of Ruth’s achievement) increases (e.g., from 8 to 16 in Figure 17.2), the weighted aggregate will more and more approximate the prototype (i.e., Ruth’s actual or average achievement). Cancellation of noise (or unreliability of specific observations) is the major factor underlying this aggregation effect. Specific parameters are of lesser importance; aggregation is obtained regardless of whether the matrix columns are weighted by the raw simi-

Computer Simulation 359

larities SRi (as in BIAS) or by similarities raised to the power of 3 (S3Ri as in MINERVA). In any case, with increasing sample size, unsystematic error (i.e., the overestimation of Ruth’s achievement in some observations and the underestimation in others) is canceled off and latent structures become more and more visible. Through aggregation, simulation models can explain why mere thinking about an attitude target (e.g., generating an increasing number of thoughts about football) can render the initial attitude more extreme (Tesser, 1978), or why extended group discussion (e.g., generating many reasons for a defendant’s guilt) serves to polarize an initially most prevalent group opinion (Myers & Lamm, 1976). The aggregation principle also implies that an increasing amount of information renders a larger variety of target attributes visible, thus resulting in more differentiated judgments of well-known than unfamiliar targets. For example, the failure to develop a differentiated picture of an outgroup (i.e., outgroup homogeneity) may simply reflect the paucity of available information (Linville, Fischer, & Salovey, 1989) and explain why outgroup homogeneity is reduced for relatively large outgroups (Mullen & Hu, 1989). The same argument applies to asymmetric judgments of the self versus others. Computer simulations by Fiedler, Kemmelmeier, and Freytag (1999) using the BIAS algorithm showed that in the absence of any cognitive or motivational bias, the assumption that we know more about ourselves than about others can account for a differentiated pattern of judgments. When judging the self and others on uni-polar scales with respect to both antonyms of trait dimensions (e.g., extraverted and introverted), self-knowledge is so rich that people rate themselves to be high on both antonyms. Due to aggregation, the sum of both uni-polar ratings is regularly higher for the self then for others. In contrast, on bi-polar scales, when people have to choose between either extravert or introvert, judgments are more extreme for others than for the self. This is because impoverished knowledge about others (i.e., small number of matrix columns) facilitates simplifying one-sided judgments when rich knowledge entails conflicting evidence. Empirical findings corroborate these insights gained from computer simulations (Sande, Goethals, & Radloff, 1988). While the existence of noise in the empirical world constitutes a truism—because empirical correlations are inevitably imperfect—simulation models help researchers to understand the breadth of the implications.

Theoretical Insights Gained from Simulation For simulation research to produce new insights and theoretical progress in social psychology, it is important to identify the specific properties of an algorithm that generates an effect. In the preceding sections, we have seen that periodic accumulation of a tiny influence can have a huge effect, that negativity in attitude learning may reflect a radical sampling truncation

360

FORMAL THEORIES

effect, and that aggregation over noisy information can account for various judgment biases. One should, however, refrain from equating a successful computer algorithm with a psychological process per se. Different models can often simulate the same phenomenon, especially when they are sensitive to the same crucial property. Aggregation is a property shared by various algorithms, and it is thus no surprise that, say, outgroup homogeneity (i.e., the readiness to generalize judgments across outgroup members) or ingroup favoritism (i.e., the tendency to provide positive ingroup judgments) can be simulated by error-reducing recurrent PDP models2 with many feedback loops (Vanhoomissen & van Overwalle, 2010), by aggregation models such as BIAS (Fiedler, 1996), or by exemplar-based memory models (Linville et al., 1989).

Simulation Models Informing Major Insights on Illusory Correlations In the social-psychological research reviewed so far, simulation models have at best played a newcomer or outsider role, providing nice illustrations of core processes and translating theories into computer programs. The present section is devoted to one paradigm in which simulation models have come to play a more advanced theoretical role, helping researchers to answer central theoretical questions, for which there is no direct empirical evidence. This paradigm, called illusory correlations, is again concerned with the unequal aggregation of information about large groups (majorities) and small groups (minorities). Given the same high rate (e.g., 70%) of positive behaviors observed in a unknown majority and in a unknown minority, but a larger absolute number of observations about the majority (e.g., 21 out of 30) than about the minority (e.g., 7 out of 10), observers will nevertheless acquire a more positive impression of the majority than of the minority. This memorable finding, which was first discovered by Hamilton and Gifford (1976) and later replicated in dozens of experiments (cf. Mullen & Johnson, 1990), implies a basic advantage of majorities over minorities (in a world in which positive behavior is the norm). Unequal aggregation is sufficient to explain the illusory correlations. Given the same positivity rate in both groups, but clearly more trials to learn the preponderance of positive behavior in the majority, illusory correlations reflect fully normal learning rules. In PDP models, where connection weights between the memory nodes are zero at the beginning (i.e., prior to any information about the groups), the connections will be gradually changed in a way that the output nodes reflect the predominant positivity. In an exemplar-based aggregation model, too, the matrix aggregate will also gradually change from a neutral pattern to a positive pattern. This aggregation effect increases with sample size. Let us first consider the impressive list of theoretical insights from simulation studies that had a strong impact on research and theorizing

Computer Simulation 361

on illusory correlations (ICs). First, successful IC simulations demonstrate unequivocally that for the illusion to occur it is not necessary to assume any cognitive or motivational bias or selective attention given to negative behavior or to minorities. In all pertinent simulations, every stimulus was given the same weight and attention. Second, it is particularly not necessary to assume a memory advantage for the rarest class of behaviors (i.e., negative minority behaviors), as suggested to be essential for the illusion (Hamilton & Gifford, 1976). Third, if the illusion reflects a normal aggregation effect, it must be less restricted and more robust than originally expected (Hamilton & Sherman, 1989). Illusory correlations are not restricted to group impressions rather than person impressions, or to learning under memory instructions, and they do not disappear under impression formation instructions. Fourth, analogous illusions are obtained with completely different tasks, provided there is a difference in sample size. For instance, the illusion occurs in social hypothesis testing, when there is the same high rate of confirming evidence for two hypotheses, but more information is sampled for one than for the other hypothesis. The illusion even occurs in operant conditioning paradigms where one signal is encountered more frequently than another. And last but not least, the illusion also occurs when there are more than two targets. For instance, when a teacher has to assess the achievement of many students in a school class, the same high accuracy rate will lead to better evaluations of students with large samples of observations. All of these theoretically strong and innovative predictions have received ample support from empirical research inspired by simulation models (e.g., Fiedler, Freytag, & Unkelbach, 2007; Fiedler, Hemmeter, & Hofmann, 1984; Fiedler, Russer, & Gramm, 1993; Fiedler, Walther, & Nickel, 1999; Klauer & Meiser, 2000; Kutzner, Freytag, Vogel, & Fiedler, 2008; Meiser, 2003; Meiser & Hewstone, 2001, 2006). However, while all these findings derived from simulation models testify to the robustness and the generality of illusory correlations, closer inspection of the difference between different simulation models has also led to clarifying open questions and refined controversies. For example, an important practical question is whether illusory correlations persist in the face of abundant evidence. Without the simulation models at hand, researchers would now be thrown back to a verbal debate about the implications of models stated in ordinary language. While recurrent PDP models (e.g., Van Rooy, Van Overwalle, Vanhoomissen, Labiouse, & French, 2003) emphasize an implication of the error-reducing delta rule in feedback learning (McClelland & Rumelhart, 1988), the BIAS (Fiedler, 1996) or MINERVA models (Smith, 1991) point to the impact of noise on the aggregation effect. Similar to Rescorla and Wagner’s (1972) learning model, the delta rule predicts that the learning effect on every trial depends on the difference between the model’s prediction and the obtained feedback. Once learning reaches an asymptote, so that delta approximates zero, first for the majority and later also for the minority, the difference should disappear. Therefore, delta-based feedback learning models suggest that illusory cor-

362

ForMal TheorieS

relations should decrease and eventually disappear when the stimulus series becomes very long. In contrast, aggregation models would emphasize the role of noise at any time. Learning is never perfect because part of the memorized information is always forgotten. Even when the number of stimuli increases to several hundred, forgetting, fatigue, and erroneous inferences should cause noise and, according to this notion, illusory correlations should persist even for very large samples. By simulating the full pattern of predictions from both models, it is possible to test whether these plausible predictions are actually a consequence of the mechanisms employed in the models. Simulating an illusory correlation effect across 320 observations—10 times more than in previous experiments— Kutzner et al. (2011) first tested the PDP model used by Van Rooy et al. (2003). As is evident from Figure 17.3, a marked IC effect appeared after the first 16 observations (i.e., 9 majority positive, 3 majority negative, 3 minority positive, and 1 minority negative), reflecting a large difference in the activation of the positive and negative valence nodes for the majority and a smaller one (though notably in the same direction) for the minority. Also as expected, the illusion gradually converges to zero after many observations. Note also in passing that the PDP simulation predicts a transitory reversal (i.e., higher evaluation of the minority) after an intermediate number of trials—a prediction that probably would not have been made in the absence of formal simulations. Analogous simulations of the aggregation model (MINERVA) sug-

fIgURe 17.3. Activation levels in the valence nodes (frequent and infrequent) resulting from majority and minority prompts, as a function of the number of observations, or trials. Results are averaged over 250 simulated runs of a PDP network with an error-reducing delta algorithm (Van Rooy et al., 2003). Group evaluations = difference in activation levels for the frequent minus the infrequent valence. The bottom line connecting triangles reflects the illusory correlation (IC) effect (i.e., the difference of the majority minus minority evaluation). Learning rate = proportion of prediction error used to change connection weights (i.e., speed of learning).

computer Simulation

363

gested by Smith (1991) also yielded IC effects. As is apparent from Figure 17.4, the illusion remained similarly strong across the extended learning process, even after 320 trials. Moreover, the crucial role of noise was supported. When noise was set to zero (i.e., when no matrix elements were randomly changed), the correlation of both groups with the ideal of positivity was perfect (see the uppermost lines), thus eliminating the IC effect. An experiment with 320 trials lends direct support to noise as a main causal ingredient. However, not only did participants continue to express illusory correlations after many trials, but experiments also confirmed that illusory correlations largely disappear when noise is eliminated, either by presenting the information in tabular form (Hamilton, Dugan, & Trolier, 1985) or by always repeating the same positive or negative stimulus (“nasty” vs. “nice”; Kutzner & Fiedler, 2014). Thus, the illusory correlation effect is a prime example of how simulation models inspire new research, are empirically tested, and help to articulate and resolve theoretical debates about an intriguing social-psychological phenomenon.

Simulation Models Illustrating the Advantages of Small Samples Simulation research is not only helpful to point out and explain existing empirical findings. It also helps researchers to figure out the limits of empirical phenomena and to anticipate curious reversals. This also applies to the

fIgURe 17.4. Simulated group evaluations, averaged over 1000 simulated runs of a distributed memory model for illusory correlations (Smith, 1991). Evaluations are plotted as a function of the number of observations and the amount of noise in memory. Noise = proportion of binary elements in the columns to be randomly changed. Illusory correlation (IC) effect = difference in majority minus minority evaluation.

364

FORMAL THEORIES

seeming advantage of large amounts of information in learning and inductive judgment; sometimes it is of advantage to start small or to restrict the capacity of the cognitive system (Elman, 1993; Newport, 1990). This perplexing “less is more” phenomenon was explained by Elman (1993) in a connectionist model tailored to explain the amazing phenomenon that language acquisition unfolds optimally when young children’s working memory capacity is still restricted. Because natural language is often irregular and noisy, rather than deterministic, a complete input of correct language samples is less effective than the kind of degraded input that results from limited capacity. A similar point was made in a seminal paper by Kareev (2000), based on a Monte Carlo simulation of statistical sampling processes. When a substantial correlation between two variables actually exists in the world (e.g., r = .50), then the distribution of sample correlations drawn from such a world will be left-skewed; most samples will indicate correlations stronger than r. Because the skewness is maximal for small samples sizes of n = 7 ± 2, the simulation suggests an intriguing theoretical reason for the size of human working memory. Evolution may have equipped Homo sapiens with a working memory of roughly 7 chunks because this “window size” (Kareev, 1995a, 1995b) renders environmental correlations maximally detectable. In a more general version of the same sampling model, Fiedler and Kareev (2006) simulated the implications of the truism that the dispersion of sample estimates increases with decreasing sample size—the inverse of the empirical law of large numbers. Assume that the true population difference, or contingency, ∆ = p(+|A) – p(+|B) between the proportions of positive attributes observed in two targets A and B (consumer products or job applicants) is D = .20. That is, in the universe, when sample size is very large, A is 20% more likely to be positive than B. However, with limited sample size n, the observed sample difference dD = .20 will often exceed the true value, especially when n is small. As a consequence, when the decision threshold is high (i.e., when a decision in favor of A is only made when the observed difference is very strong (e.g., when dD = .20 > .50), the very unreliability of small samples may facilitate decisions. Simulations show that when the threshold is higher than the true difference D, small samples may enable more correct decisions than larger samples (under conditions specified by the model) because they are more likely to exaggerate the true difference. These and other simulations of small-sample advantages (Hertwig & Pleskac, 2010; Schooler & Hertwig, 2005) offer interesting accounts for the oftennoted accuracy of judgments and decisions based on intuition, gut feeling, and “thin slices” of expressive behavior (Ambady & Rosenthal, 1993).

Concluding Remarks Throughout this chapter, we have treated simulation as a methodological approach that has the potential to further our understanding of socialpsychological phenomena beyond the level at which behavior can be

Computer Simulation 365

observed and manipulated directly. The purpose was neither to provide an exhaustive review of all the relevant simulation literature nor to present a systematic introduction to the mathematical underpinnings of simulation models. The purpose was, rather, to illuminate selected aspects of simulation research that have actually had an impact on current social psychology. With this goal in mind, we covered such diverse topics as social influence, attitude learning, intergroup relations, and the impact of sample size on judgment and decision making. We placed an emphasis on illusory correlations as a paradigm that has become a stronghold of social-psychological simulation research. Within this paradigm, simulation approaches have led to particularly deep reflection and to fruitful debates about the causal conditions of illusory correlations that could then be tested and cross-validated experimentally. Despite the heuristic value of the simulation models we have reviewed, and despite their advantage in theoretical precision and transparency, it should have also become apparent that simulation models should not be mistaken as unique explanations of behavioral phenomena. Neither the specific cognitive metaphor that underlies a simulation model—whether it resembles an exemplar-based or a feature abstraction model (Barsalou, 1990)—nor the specific parameters used in the simulation algorithms appear to be crucial for understanding the simulation models’ success. Rather, the major lessons to be gained from simulation studies refer to the explanatory value of such structural principles as periodic accumulation of tiny errors, aggregation, hedonic information search, or the impact of noise in an uncertain, probabilistic world. These theoretical constructs are often located in the interface of the human mind and the stimulus environment, clearly distinct from the intrapsychic constructs that have so long dominated the empirical research traditions in social psychology.

Caveats and Limitations There are obvious limitations to the explanatory and predictive value of simulation models. Any implementable simulation program inevitably provides an idealized, perfect causal model of the phenomenon it produces. The underlying algorithm not only entails a commitment to one particular causal mechanism but also a number of parameter settings and process assumptions that afford potentially critical moderators or boundary conditions. For example, a distributed memory model entails assumptions about the number of elements reflecting a memory trace, how to measure similarities between traces, and so on. Modifications in any of these “silent” assumptions may alter the simulated results and their theoretical and practical implications. More technically, when allowed to vary from one implementation to another, these silent assumptions constitute degrees of freedom that allow a model to fit empirical data—regardless of the causal mechanism central to a model. Thus, even an impressive fit of model and data is not infallible evidence for the proposed causal process. In fact, too many degrees of freedom in a simu-

366

FORMAL THEORIES

lation model will not only undermine its ability to indicate core mechanisms, but once specified they will also undermine the model’s ability to predict novel data due to “overfitting” arbitrary variation (Gigerenzer & Brighton, 2009). Nevertheless, simulation models identify causal conditions that are sufficient to produce types of effects, even when they do not necessarily apply to all manifestations of the same effect type. In this regard, they hardly differ from all other scientific models, as it is always possible that an empirical result originates from multiple causes. To illustrate ,we return to the opening example, the famous simulation of social influence (Nowak et al., 1990). The model illustrates that the joint operation of numerosity, strength, and distance on changing individual agents’ attitudes is sufficient to produce attitude polarization and segregation. This, however, does not mean that true attitude change of the conversion type (Moscovici, 1980) is necessary for polarization and segregation to occur. In a recent simulation model, Brown (2014) assumes quite the opposite, namely, that agents are motivated to conform, both to their own and their neighbors’ attitudes, without ever changing their attitude. Allowing agents to move (e.g., spatially) to reconcile possible self-neighbor differences produces the same pattern: segregation and polarization of (observed) attitudes. Thus, slightly different assumptions, either conversion or compliance, can both predict an observed pattern, though with completely different deep meanings of the psychological processes. As a matter of rule, then, fully specified computer models of one causal mechanism can never fully explain and predict real phenomena that are almost always blends of multiple causal influences. They also never come with the guarantee that all elements of the computer model are at work in the real-world analogue of the simulated idealized world. Overfitting (of specific parameters and unwarranted model elements) is always a problem, and multicausation always complicates comprehensive explanations and predictions of real phenomena. However, crucially, the theorist’s ambitious task of understanding the causal and conditional structures of behavioral phenomena can be greatly facilitated through transparent and complete simulations of idealized causal mechanisms. Acknowledgments The research underlying the present chapter was supported by a Koselleck grant from the Deutsche Forschungsgemeinschaft (No. Fi 294/23-1).

Notes 1. “Tit for tat” means to mirror another player’s behavior on the next trial, that is, to cooperate when the opponent has cooperated and to defect whenever the opponent has defected.

Computer Simulation 367 2. Parallel distributed processing models (PDPs) represent the most prominent class of connectionist approaches (McClelland & Rumelhart, 1988).

References Ambady, N., & Rosenthal, R. (1993). Half a minute: Predicting teacher evaluations from thin slices of nonverbal behavior and physical attractiveness. Journal of Personality and Social Psychology, 64, 431–441. Anderson, N. H. (1965). Averaging versus adding as a stimulus-combination rule in impression formation. Journal of Experimental Psychology, 70, 394–400. Axelrod, R. (1980). More effective choice in the prisoner’s dilemma. Journal of Conflict Resolution, 24, 379–403. Bandura, A. (1965). Influence of models’ reinforcement contingencies on the acquisition of imitative responses. Journal of Personality and Social Psychology, 1, 589–595. Barsalou, L. W. (1990). On the indistinguishability of exemplar memory and abstraction in category representation. In T. K. Srull & R. R. Wyer (Eds.), Content and process specificity in the effects of prior experiences (pp. 61–88). Hillsdale, NJ: Erlbaum. Brown, G. D. A. (2014). An agent-based model of social norm effects and polarization. Unpublished manuscript, University of Warwick, Cokentry, UK. Cooper, W. S., & Kaplan, R. (1982). Adaptive “coin-flipping”: A decision-theoretic examination of natural selection for random individual variation. Journal of Theoretical Biology, 94, 135–151. Denrell, J. (2005). Why most people disapprove of me: Experience sampling in impression formation. Psychological Review, 112, 951–978. Denrell, J., & Le Mens, G. (2007). Interdependent sampling and social influence. Psychological Review, 114, 398–422. Denrell, J., & March, J. G. (2001). Adaptation as information restriction: The hot stove effect. Organization Science, 12, 523–538. Dougherty, M., Gettys, C., & Ogden, E. (1999). MINERVA-DM: A memory processes model for judgments of likelihood. Psychological Review, 106, 180–209. Eiser, J., Fazio, R. H., Stafford, T., & Prescott, T. J. (2003). Connectionist simulation of attitude learning: Asymmetries in the acquisition of positive and negative evaluations. Personality and Social Psychology Bulletin, 29, 1221–1235. Elman, J. L. (1993). Learning and development in neural networks: The importance of starting small. Cognition, 48, 71–99. Fazio, R., Eiser, J., & Shook, N. (2004). Attitude formation through exploration: Valence asymmetries. Journal of Personality and Social Psychology, 87, 293–311. Fiedler, K. (1996). Explaining and simulating judgment biases as an aggregation phenomenon in probabilistic, multiple-cue environments. Psychological Review, 103, 193–214. Fiedler, K., Freytag, P., & Unkelbach, C. (2007). Pseudocontingencies in a simulated classroom. Journal of Personality and Social Psychology, 92, 665–677. Fiedler, K., Hemmeter, U., & Hofmann, C. (1984). On the origin of illusory correlations. European Journal of Social Psychology, 14, 191–201. Fiedler, K., & Kareev, Y. (2006). Does decision quality (always) increase with the size of information samples? Some vicissitudes in applying the law of large numbers. Journal of Experimental Psychology: Learning, Memory, and Cognition, 32, 883–903. Fiedler, K., Kemmelmeier, M., & Freytag, P. (1999). Explaining asymmetric intergroup judgments through differential aggregation: Computer simulations and some new evidence. European Review of Social Psychology, 10, 1–40. Fiedler, K., Russer, S., & Gramm, K. (1993). Illusory correlations and memory performance. Journal of Experimental Social Psychology, 29, 111–136.

368

FORMAL THEORIES

Fiedler, K., Walther, E., Freytag, P., & Plessner, H. (2002). Judgment biases in a simulated classroom—a cognitive-environmental approach. Organizational Behavior and Human Decision Processes, 88, 527–561. Fiedler, K., Walther, E., & Nickel, S. (1999). The auto-verification of social hypotheses: Stereotyping and the power of sample size. Journal of Personality and Social Psychology, 77, 5–18. Fiedler, K., Woellert, F., Tauber, B., & Hess, P. (2013). Applying sampling theories to attitude learning in a virtual school class environment. Organizational Behavior and Human Decision Processes, 122(2), 222–231. Galton, F. (1877). Typical laws of heredity. Nature, 15, 492–495. Gigerenzer, G., & Brighton, H. (2009). Homo heuristicus: Why biased minds make better inferences. Topics in Cognitive Science, 1, 107–143. Hamilton, D. L., Dugan, P. M., & Trolier, T. K. (1985). The formation of stereotypic beliefs: Further evidence for distinctiveness-based illusory correlations. Journal of Personality and Social Psychology, 48, 5–17. Hamilton, D., & Gifford, R. (1976). Illusory correlation in interpersonal perception: A cognitive basis of stereotypic judgments. Journal of Experimental Social Psychology, 12, 392–407. Hamilton, D. L., & Sherman, S. J. (1989). Illusory correlations: Implications for stereotype theory and research. In D. Bar-Tal, C. F. Graumann, A. W. Kruglanski, & W. Stroebe (Eds.), Stereotype and prejudice: Changing conceptions (pp. 59–82). New York: Springer. Hastie, R. (1988). A computer simulation model of person memory. Journal of Experimental Social Psychology, 24, 423–447. Hastie, R., & Stasser, G. (2000). Computer simulation methods for social psychology. In H. T. Reis & C. M. Judd (Eds.), Handbook of research methods in social and personality psychology (pp. 85–114). New York: Cambridge University Press. Herrnstein, R. J. (1961). Relative and absolute strength of response as a function of frequency of reinforcement. Journal of the Experimental Analysis of Behavior, 4, 267–272. Hertwig, R., & Pleskac, T. J. (2010). Decisions from experience: Why small samples? Cognition, 115, 225–237. Hills, T. T., & Hertwig, R. (2010). Information search in decisions from experience: Do our patterns of sampling foreshadow our decisions? Psychological Science, 21, 1787–1792. Hintzman, D. (1986). “Schema abstraction” in a multiple-trace memory model. Psychological Review, 93, 411–428. Hubert, M. (1999). The misuse of past-performance data. In L. E. Likson & R. A. Geist (Eds.), The psychology of investing (pp. 147–157). New York: Wiley. Kalick, S., & Hamilton, T. E. (1986). The matching hypothesis reexamined. Journal of Personality and Social Psychology, 51, 673–682. Kareev, Y. (1995a). Positive bias in the perception of covariation. Psychological Review, 102, 490–502. Kareev, Y. (1995b). Through a narrow window: Working memory capacity and the detection of covariation. Cognition, 56, 263–269. Kareev, Y. (2000). Seven (indeed, plus or minus two) and the detection of correlations. Psychological Review, 107, 397–403. Klauer, K., & Meiser, T. (2000). A source-monitoring analysis of illusory correlations. Personality and Social Psychology Bulletin, 26, 1074–1093. Kruglanski, A. W. (2004). The quest for the gist: On challenges of going abstract in social and personality psychology. Personality and Social Psychology Review, 8, 156–163. Kunda, Z., & Thagard, P. (1996). Forming impressions from stereotypes, traits, and behaviors: A parallel-constraint-satisfaction theory. Psychological Review, 103, 284–308. Kutzner, F. L. W., & Fiedler, K. (2014). No correlation, no evidence or attention shift in category

Computer Simulation 369 learning: Different mechanisms behind illusory correlations and the inverse base-rate effect. Manuscript submitted for publication. Kutzner, F., Freytag, P., Vogel, T., & Fiedler, K. (2008). Base-rate neglect as a function of base rates in probabilistic contingency learning. Journal of the Experimental Analysis of Behavior, 90, 23–32. Kutzner, F., Vogel, T., Freytag, P., & Fiedler, K. (2011). A robust classic: Illusory correlations are maintained under extended operant learning. Experimental Psychology, 58, 443–453. Latané, B. (1981). The psychology of social impact. American Psychologist, 36, 343. Linville, P. W., Fischer, G. W., & Salovey, P. (1989). Perceived distributions of the characteristics of in-group and out-group members: Empirical evidence and a computer simulation. Journal of Personality and Social Psychology, 57, 165–188. McClelland, J., & Rumelhart, D. (1988). Computational models of cognition and perception: Explorations in parallel distributed processing: A handbook of models, programs, and exercises. Cambridge, MA: MIT Press. Meiser, T. (2003). Effects of processing strategy on episodic memory and contingency learning in group stereotype formation. Social Cognition, 21, 121–156. Meiser, T., & Hewstone, M. (2001). Crossed categorization effects on the formation of illusory correlations. European Journal of Social Psychology, 31, 443–466. Meiser, T., & Hewstone, M. (2006). Illusory and spurious correlations: Distinct phenomena or joint outcomes of exemplar-based category learning? European Journal of Social Psychology, 36, 315–336. Moscovici, S. (1980). Toward a theory of conversion behavior. Advances in Experimental Social Psychology, 13, 209–239. Mullen, B., & Johnson, C. (1990). Distinctiveness-based illusory correlations and stereotyping: A meta-analytic integration. British Journal of Social Psychology, 29, 11–27. Myers, D. G., & Lamm, H. (1976). The group polarization phenomenon. Psychological Bulletin, 83, 602–627. Newport, E. L. (1990). Maturational constraints on language learning. Cognitive Science, 14, 11–28. Nowak, A., Deutsch, M., Bartkowski, W., & Solomon, S. (2010). From crude law to civil relations: The dynamics and potential resolution of intractable conflict. Peace and Conflict: Journal of Peace Psychology, 16, 189–209. Nowak, A., Szamrej, J., & Latané, B. (1990). From private attitude to public opinion: A dynamic theory of social impact. Psychological Review, 97, 362–376. Penrod, S., & Hastie, R. (1980). A computer simulation of jury decision making. Psychological Review, 87, 133–159. Rescorla, R., & Wagner, A. (1972). A theory of pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In A. Black & W. Prokasy (Eds.), Classical conditioning: II. Current research and theory (pp. 64–99). New York: Appleton-Century-Crofts. Sande, G. N., Goethals, G. R., & Radloff, C. E. (1988). Perceiving one’s own traits and others’: The multifaceted self. Journal of Personality and Social Psychology, 54, 13–20. Schooler, L., & Hertwig, R. (2005). How forgetting aids heuristic inference. Psychological Review, 112, 610–628. Schul, Y., Mayo, R., Burnstein, E., & Yahalom, N. (2007). How people cope with uncertainty due to chance or deception. Journal of Experimental Social Psychology, 43, 91–103. Smith, E. (1991). Illusory correlation in a simulated exemplar-based memory. Journal of Experimental Social Psychology, 27, 107–123. Smith, E. R., & Conrey, F. R. (2007). Agent-based modeling: A new approach for theory building in social psychology. Personality and Social Psychology Review, 11, 1–18.

370

FORMAL THEORIES

Stasser, G. (1988). Computer simulation as a research tool: The DISCUSS model of group decision making. Journal of Experimental Social Psychology, 24, 393–422. Sun, R. (2006). Cognition and multi-agent interaction: From cognitive modeling to social simulation. New York: Cambridge University Press. Tesser, A. (1978). Self-generated attitude change. Advances in Experimental Social Psychology, 11, 289–338. Thorndike, E. L. (1927). The law of effect. American Journal of Psychology, 39, 212–222. Vanhoomissen, T., & Van Overwalle, F. (2010). Me or not me as source of ingroup favoritism and outgroup derogation: A connectionist perspective. Social Cognition, 28, 84–109. Van Rooy, D., Van Overwalle, F., Vanhoomissen, T., Labiouse, C., & French, R. (2003). A recurrent connectionist model of group biases. Psychological Review, 110, 536–563.

18 Mathematical Modeling Karl Christoph Klauer

Varieties of Mathematical Models Mathematical models serve many functions in psychological research. They (1) underlie statistical analyses, (2) serve as measurement models, and (3) provide formal theories at the algorithmic or computational level of theory formulation. Almost any statistical analysis is based on a mathematical model or implies one. In statistical analyses, the mathematical model specifies the distribution of observed variables. This enables one to estimate distribution parameters of interest (such as a correlation between two variables) and to test for significance of observed effects and relationships (such as testing whether or not a correlation is significantly different from zero). A prominent example is the general linear model treated in most textbooks of statistics. These models are typically general-purpose models that are not adapted to the particular psychological processes under study. Mathematical models as measurement models aim at providing dependent variables based on some principled transformation of the raw data that are better operationalizations of theoretical constructs of interest than ad-hoc transformations and indices derived from the data. A prominent example is the signal-detection model that is regularly used in social-psychological 371

372

FORMAL THEORIES

research (e.g., Correll, Judd, Park, & Wittenbrink, 2002; Greenwald, Draine, & Abrams, 1996). Measurement models usually imply at least a rudimentary understanding of the processes that lead to the observed data and provide corrections for the influence of secondary and currently uninteresting processes. For example, signal detection measure d’ provides a measure of sensitivity or performance in binary decisions that is assumed to partial out response bias in favor of one of the two decision options. In order to achieve this, assumptions about how response bias affects the decisions need to be made. Thus, signal detection in its simplest form assumes a decision process based on a one-dimensional strength-of-evidence axis. A response criterion is placed on this axis, splitting the axis into two regions associated with the two different decision options. The decision maker is believed to condense the information in the stimulus display to be judged into a single numerical value on that axis. The region into which this value falls then determines which decision option is chosen as response. Measurement models thereby fall between general-purpose statistical models such as the general linear model and formal theories of mental processes specified at the algorithmic level as considered in this chapter. In this chapter, I focus on the use of mathematical models as full-fledged psychological theories. Theories can be stated at different levels according to Marr’s (1982) classification of levels of theoretical analysis (De Houwer & Moors, Chapter 2, this volume). The computational level specifies the function relating input (stimuli) and output (decision, responses, behavioral tendencies, etc.) that the mind computes in the domain under scrutiny. This is the level of theoretical analysis typically involved in normative models. Normative models specify the functional relationships that should be computed to guide behavior according to an appropriate normative framework. The resulting computational-level theory claims that people have the competence to perform, or mimic, the normatively appropriate computations and that they indeed use the normative prescriptions to guide their choices and behaviors. Computational theories do not address the algorithmic level, that is, the mechanisms by which the computations are performed or mimicked by the cognitive system, although they constrain the range and nature of candidate mechanisms (Anderson, 1990). Mathematical models that explicitly specify and characterize the underlying processes and their interactions provide analyses at the algorithmic level. Before considering computational and algorithmic models, let us briefly contrast mathematical models as formal statements of psychological theories and verbal theories.

Verbal Theories and Mathematical Models Verbal theories have several advantages that are especially helpful at an early stage of theory development. At an early stage, our understanding



Mathematical Modeling 373

of the psychological phenomena under study may be vague and limited: There may be only a sparse empirical database to build on, focused on a few key effects, and our ideas about the psychological mechanisms causing the effects may still be vague and preliminary. Verbal theories are easy to communicate, and there are appropriate linguistic devices for demarcating the areas and degree of vagueness. Eliminating these areas of vagueness and replacing them by explicit and specific assumptions may even be counterproductive in early work because there may be many possibilities for specifying vague aspects, with different consequences for the empirical predictions and little basis on which to choose among the possibilities. In such situations, it may be reasonable to remain deliberately silent initially about how to fill in vague aspects of the theory and to await further data collection in order to be able to specify empty slots on the basis of better and more information. The alternative would be to come up with a detailed, yet somewhat arbitrary, set of assumptions and to run the risk of an early and perhaps stifling refutation of the developing theory because of false auxiliary assumptions. The downside of these advantages is that it is often difficult to judge a vague verbal theory on important criteria for the evaluation of theories. Theories should be logically consistent, that is, free of internal contradictions, and predictions derived from theories should be based on a logically valid chain of arguments with core assumptions and auxiliary assumptions involved in the chain formulated openly and explicitly. In particular, there should be neither gaps nor reasoning fallacies in the deductive chain (Gawronski & Bodenhausen, Chapter 1, this volume). Relative to verbal theories, perhaps the greatest advantage of mathematical models is that it is very difficult to fail on these criteria in developing a mathematical model. This is because the mathematical tools involved in specifying the model and in applying it to generate artificial data, to estimate model parameters from observed data, and to fit observed data incorporate the principles and power of deductive logic and make it very likely that logically inconsistent assumptions built into the theory would be detected at an early stage. Applying the model in these ways also necessitates the explicit statements of all assumptions and premises in the process and ensures a deductive chain from these premises to model predictions that is free of gaps and reasoning fallacies. Hintzman (1991) sees this as one of the major contributions of mathematical models, based on a list of common limitations of human verbal reasoning and examples of verbal theories containing misconceptions and mutually inconsistent assumptions (see also Gawronski & Bodenhausen, Chapter 1, this volume, for a discussion of common reasoning fallacies in psychological theorizing). Let us illustrate these and additional issues by means of two examples: a mathematical model of pseudocontingencies formulated at the computational level of theoretical analysis, and a mathematical model of illusory correlations formulated at the algorithmic level of theoretical analysis.

374

FORMAL THEORIES

Computational Models: A Normative Model of Pseudocontingencies Stereotypes are often conceived as perceived contingencies between social groups and certain behaviors or traits (Tajfel, 1969). In an influential approach, stereotype formation has accordingly been studied as a special case of contingency learning (Hamilton & Gifford, 1976), with an emphasis on the illusory or erroneous perception of contingencies between group membership and other variables. In particular, a number of different situations have been identified in which contingencies are perceived that do not accurately reflect the information provided and that lead to stereotyped perception of the groups in question, including the illusory-correlation paradigm (Hamilton & Gifford, 1976), ecological correlations (Hammdon, 1973), Simpson’s paradox (Simpson, 1951), and related cases (e.g., Meiser & Hewstone, 2004; Schaller, 1994). Fiedler and colleagues (Fiedler, 2000; Fiedler & Freytag, 2004; Fiedler, Freytag, & Meiser, 2009; Fiedler, Freytag, & Unkelbach, 2007) have recently emphasized the role of sampling in this context. Sampling is important inasmuch as the information given to perceivers in these paradigms most often consists of a number of statements about members of the groups in question that represents at best a sample of the group members’ behaviors or traits rather than an exhaustive list of a hypothetical population of the different group members’ behaviors and traits. In this view, perception of the information in samples, and inferences drawn therefrom, are subject to a number of biases. In particular, Fiedler and colleagues demonstrated a pervasive cognitive bias that they termed the pseudocontingency (PC) illusion (Fiedler & Freytag, 2004; Fiedler et al., 2007, 2009). It is a phenomenon that occurs in contingency assessment and related inferences. As argued by Fiedler (2000), Fiedler and Freytag (2004), and Fiedler et al. (2007), the PC phenomenon reflects a broadly applied cognitive algorithm, qualitatively different from genuine contingency assessment, that may account for illusory correlations (Hamilton & Gifford, 1976), so-called ecological correlations (Hammdon, 1973), contingency perception in Simpson’s (1951) paradox, and related cases. Meiser and Hewstone (2004) have argued that the notion of pseudocontingencies provides a better account of cognitive processes in stereotype formation than its competitors. A PC illusion arises when the distributions of two attributes in a group are skewed. When the skew for both attributes points in the same direction, in that most values on both attributes are high or most values on both attributes are low, the two attributes appear to be positively correlated. If the skew is in opposite directions, with mostly high values on one attribute and mostly low values on the other, the resulting PC will mimic a negative relationship. As implied by the prefix “pseudo”, however, this base-rate information does



Mathematical Modeling 375 not determine the sign and the size of the actual correlation. (Fiedler, Freytag, & Unkelbach, 2007, pp. 665–666)

To see this, consider Table 18.1, the following contingency table taken from Fiedler and Freytag’s (2004) Experiment 1. It cross-tabulates two attributes A and B taking values 0 and 1. In all, 48 cases were observed, and the correlation phi between A and B in the sample is zero. When observers are aware of only the marginals shown in the table, a PC illusion is given, if observers behave as if a positive correlation was in force. Thus, when asked to reconstruct the joint frequencies in the cells of the table or when asked to predict Attribute B on the basis of given values of Attribute A for 48 new cases, their estimates might result in the frequencies shown in Table 18.2 with phi > 0: Fiedler et al. (2009) provide the following verbal account of PC: First, they argue that base-rate driven response tendencies often make sense. For example, “in the absence of other information, maximizing the rate of correct decisions means invariantly predicting the more prevalent Y value. To detect changes in the environment, though, organisms often engage in probability matching rather than maximizing, predicting the more likely outcome Yfrequent at the same rate as it occurs, p(Yfrequent). In any case, a response bias toward the more likely criterion outcome would be functional” (Fiedler et al., 2009, p. 198). Having argued for base-rate-driven response tendencies in one dimension by means of the above and additional arguments, they go on to explain PC: Granting this rationale for base-rate-driven response tendencies in one dimension, y, the algorithm can be easily extended to cover PCs between two dimensions, y and x. Thus, given skewed base rates in both y and x (e.g., a high predator rate and a high rate of dark places), the animal should acquire two simultaneous expectancies or response sets, for predicting many predators and for predicting many dark places in territory Ek. Especially when other ecologies predict fewer predators and fewer dark places, Yfrequent should be aligned with Xfrequent and Yinfrequent with Xinfrequent in any memory representation that is organized by categories or ecologies. Even when the criterion rate Yfrequent does not vary from Xfrequent to Xinfrequent

TABLE 18.1. Contingency Table for Two Attributes A and B, phi = .00 Attribute B Attribute A

0

1

Marginals

0

27

 9

  36

1

 9

 3

  12

Marginals

36

12

376

FORMAL THEORIES

TABLE 18.2. Frequency Estimates Exhibiting PC, phi = .22 Attribute B Attribute A

0

1

Marginals

0

29

 7

  36

1

 7

 5

  12

Marginals

36

12

(i.e., when the contingency is zero or even reversed), the alignment of separately learned response tendencies for Y and X can be expected to produce a PC effect for any of the following reasons: (a) The aligned association of frequent versus infrequent levels on both variables may induce a propositional inference that the two variables are correlated, which can then be used for guessing under uncertainty. This variant could be termed the syllogistic-reasoning account. (b) The more prevalent level of one variable, Xfrequent, may activate or prime ecological information about Ek more effectively, prompting more reliable information on p(y|Ek), than the less prevalent value, Xinfrequent, which is less typical for Ek. This variant suggests why connectionist learning models (Dougherty, Gettys, & Ogden, 1999; Fiedler, 1996) can easily account for PC inferences. (c) Given two uncorrelated but skewed distributions, the number of matching cases can still be larger than the number of nonmatching cases. If the cognitive concept of a contingency is represented in terms of an average match value (cf. White, 2001), this would imply that psychologically—though not statistically—PCs do imply contingencies. (Fiedler et al., 2009, pp. 198–199)

Thus, PC may arise along different routes: via the organization of memory, via syllogistic reasoning, by differences in the reliability of estimating conditional probabilities for frequent versus infrequent categories, and by a psychological representation of contingency via average match value. One underlying assumption as stated in the above quote is that PCs do not imply contingencies statistically, more precisely that the inference from skewed base rates to an underlying contingency is not statistically warranted, and hence that the perceived contingencies are really pseudocontingencies. This assumption calls for a normative analysis of the PC phenomenon via a suitable mathematical model at the computational level of analysis. Remember that participants in PC research are usually asked to make an inference from given marginals to data that were not seen or that had not been encoded. In the above example, they are asked to estimate the (unobserved or not encoded) joint frequencies in the interior of the contingency table (Tables 18.1 and 18.2) on the basis of the marginal frequencies. Because the joint frequencies are not determined by the marginals, there



Mathematical Modeling 377

are a number of different contingency tables such as Tables 18.1 and 18.2 that are consistent with the information on the marginals that participants have. In a sense, participants have to pick one of these in responding. But how should people go about making such an inference to an unseen (or not encoded) set of joint frequencies? Bayesian statistical inference (e.g., Gelman, Carlin, Stern, & Rubin, 2004) provides a formal, normative framework describing how given data should inform beliefs about unobserved and latent quantities such as correlations and how this leads to predictions for unseen or new data. In modeling this task, we assume that the frequencies stem from a probability distribution with probabilities pij for each combination of level i of Attribute A and level j of Attribute B summing to 1 across combinations (see Table 18.3). Bayesian statistical inference works by specifying subjective beliefs about the underlying probabilities and by modifying those beliefs according to data that are observed (e.g., Gelman et al., 2004). The analysis has three building blocks: 1. a prior distribution P(pij) of the probabilities pij, 2. a posterior distribution P(pij | y) of the probabilities pij given the data y, and 3. a posterior predictive distribution. Prior distribution and posterior distribution are distributions of the probabilities pij. They describe the beliefs that a respondent holds regarding these probabilities before and after, respectively, seeing the data y. The data y (e.g., the marginals of a contingency table) are used to update the prior distribution, resulting in a modified set of beliefs that take the data into account, expressed by the posterior distribution P(pij | y). Based on the posterior distribution, new data y′ can be extrapolated or predicted from y by means of the so-called posterior predictive distribution, and these new data are the model predictions. Assume that before seeing the marginals, our participant experiences great uncertainty about the pij generating the information he or she is about to see. We express this uncertainty by choosing a uniform distribution as

TABLE 18.3. Notation for Probabilities p and Frequencies y Attribute B Attribute A

0

1

Marginals

Attribute B 0

1

Marginals

0

p00

p00

  p0.

y00

y00

  y0.

1

p10

p11

  p1.

y10

y11

  y1.

Marginals

p.0

p.1

y.0

y.1

378

FORMAL THEORIES

prior distribution. In other and rough words, prior to seeing the data, the participant believes that every conceivable set of pijs is as likely as every other set. Under the uniform prior, a prior distribution for the phi coefficient is implied; its density is shown in Figure 18.1; the figure is based on 10,000 simulated pij drawn from the uniform prior. As can be seen, the distribution is uni-modal, centered on zero, so that positive and negative phi coefficients are a priori equally likely. Furthermore, it has a huge standard deviation, expressing the high uncertainty about the phi coefficient in the absence of data. Upon seeing the data, the prior distribution is updated according to the laws of probability and Bayes’s theorem in particular. For example, upon learning that the marginals for Attributes A and B have the skewed values shown in Tables 18.1 and 18.2, the posterior distribution of the pij can be determined. It implies a posterior distribution for the phi coefficient as shown in Figure 18.2. As can be seen in Figure 18.2, values of phi smaller than –.33 are very unlikely under the posterior distribution, but almost all other phi values are about equally likely. In all, the posterior distribution favors positive phi values; for example, the posterior probability that phi is larger than zero is .745. The final step is to extrapolate from the marginals to the unseen joint frequencies or to predict joint frequencies of a new sample of, let us say, 48 values. This is achieved by means of the predictive posterior distribution.

FIGURE 18.1.  Prior distribution of phi coefficient.



Mathematical Modeling 379

FIGURE 18.2.  Posterior distribution of phi coefficient given yi. and y.j from Table 18.1 or 18.2.

The predicted data is a sample of joint frequencies yij summing up to 48 drawn from that distribution. I drew 10,000 such samples, and the resulting mean frequencies are shown in Table 18.4. Note that the marginals in Table 18.4 exhibit some regression to the mean in that they are a little less unbalanced than the given marginals. The average phi across the 10,000 samples was phi = .31 with a very large standard deviation of .40, reflecting the remaining uncertainty in the underlying correlation. Nevertheless, for a random sample of 20 draws from the posterior predictive distribution, corresponding to the responses from 20 participants in the prediction task, the mean phi correlation was significantly larger than zero, M = .26, SD = .32, t(19) = 3.60, p < .01. In other words, the normative analysis based on a mathematical model at the computational level of theoretical analysis shows that the inference

TABLE 18.4.  Mean Predicted Frequencies Attribute B Attribute A

0

1

Marginals

0

28.55

 6.50

 35.05

1

 6.50

 6.45

 12.95

Marginals

35.05

12.95

380

FORMAL THEORIES

from skewed marginals to an underlying contingency—albeit fraught with great uncertainty and large confidence intervals—is statistically warranted. Marginals do contain valid, if weak, statistical information on underlying contingencies. The PC phenomenon is neither an illusion in contingency perception nor a bias in statistical judgments; it is rational to show the PC effect. This example of a mathematical model serves to illustrate additional points. Thus, it is easy to use the model to generate precise predictions for new tasks and situations. For example, the posterior distribution of the phi coefficient as seen in Figure 18.2 is diffuse. This means that although the marginals convey information on the underlying contingency, it is only weak information. This in turn implies that the prior distribution will have a large weight in determining the shape of the posterior distribution. As a consequence, PC should be fragile and in particular easily erased or reversed by prior information about the contingency between the two attributes in question. In other words, it should be very difficult to alter preexisting beliefs about the contingency of the two attributes, stemming, for example, from existing stereotypes, by providing only the information about skewed marginals. Perhaps more impressively, the present theory can be readily applied to new tasks and situations to generate predictions that would be difficult to make a priori based on only a verbal account such as the ones exemplified above for PC. For example, one might ask whether PC is dependent on whether the information about the base rates of Attribute A and Attribute B is obtained from one and the same sample, or whether PC also arises when one sample is used to learn about the base rate of Attribute A, and a new sample to learn about the base rate of Attribute B, thereby eliminating even implicit bivariate observations. Similarly, one might ask whether pseudocontingencies should be seen when information about only one cell of the contingency table is provided, for example, when participants are told or learn that 40% of the cases have value 0 on both Attribute A and B, which implies neither skewed marginals nor a contingency with necessity. The present theory predicts that PC occurs in all of these cases.1 Normative models play a role in various subfields of social-psychological research. For example, Baron (2012) discusses normative models in the field of judgment and decision making such as the famous expected-utility model. The famous “heuristics and biases” program relies on normative models as baseline: Biases or fallacies such as the conjunction fallacy or the base-rate fallacy are defined as departures from normative prescriptions. Although containing a number of non-normative features and principles, central parts of McGuire’s probabilogical model are normative (Kruglanski & Stroebe, 2005). As a final example of a normative model, let us mention Campbell’s (1963) elegant model of the attitude–behavior link recently extended by Kaiser, Byrka, and Hartig (2010).



Mathematical Modeling 381

Algorithmic Models: A Multinomial Processing-Tree Model of Illusory Correlations If the above computational model of pseudocontingencies should receive further empirical support, a natural next step would be to specify how participants actually perform or approximate the complex and extensive computations postulated by the Bayesian normative analysis. In other words, it would be desirable to point to a few fast and frugal processes of mental simulation or statistical sampling that are plausibly within participants’ capabilities and mimic the kind of Bayesian computation underlying the normative analysis. This would lead to a theory at the algorithmic level of theoretical analysis. I have sketched some ideas in a technical report (Klauer, 2008). Many mathematical models are couched at the algorithmic level of theoretical analysis. That is, they incorporate explicit assumptions about the stages and processes intervening between stimuli and observable responses. Ulrich (2009) has argued that one of the major functions of mathematical models is to uncover unobservable cognitive mechanisms. For example, a large and useful family of models, the multinomial processing-tree models, consists of models of categorical data. These models address responses that can be classified into a set of exhaustive and mutually exclusive categories (e.g., into false and correct responses) and model the probabilities with which the different response categories occur. They do this on the basis of the frequencies with which the different response categories are observed. Consider as an example a multinomial model for illusory correlations (Klauer & Meiser, 2000). Illusory correlations are a special case of PC. Participants learn about positive and negative behaviors of two groups, labeled A and B, with skewed frequency distributions so that there are more Group A than Group B behaviors and more positive than negative behaviors, but group membership and valence of the behavior are crossed orthogonally. For example, the frequencies of the different kinds of behaviors might be as in Table 18.1, with Attribute A encoding group membership (with level 0 and 1 corresponding to Group A and B, respectively) and Attribute B encoding valence (with level 0 and 1 corresponding to positive and negative behaviors, respectively). Despite the absence of a contingency in the behavior lists, participants perceive and express a contingency such that Group A is associated with more positive behaviors than Group B. For example, when asked to assign the previously seen behaviors to groups, perceivers attribute proportionally more positive behaviors to Group A than to Group B (as in Table 18.4). Klauer and Meiser (2000) proposed a mathematical model of this assignment task that is a member of the family of multinomial processing-tree models. Specifying a mathematical model at the algorithmic level requires an intuition of the most important processes involved in the task and of their

382

FORMAL THEORIES

interaction at least as far as they affect the observable data, the assignment frequencies in the present case. Specifically, Klauer and Meiser (2000) argued that assignments could be memory-based or based on guessing in the absence of memory. They further distinguished between memory for the individual behaviors (i.e., recognizing a to-be-assigned behavior as one that was previously seen or not) and memory for its group origin. The memory and guessing processes were assumed to jointly determine the assignment frequencies. In multinomial processing-tree models, the processes are arranged in stages. Each process defines a decision or choice point that can result in two or more mental states as output, which are taken as starting point for further processes. Alternatively, a completed process may lead to an observable response directly. For example, in Klauer and Meiser’s (2000) model of illusory correlations in the assignment task, the first process to occur given a to-be-assigned behavior is probing for recognition of the behavior. That is, perceivers access their memory to answer the question whether or not the behavior is recognized as having been presented earlier. This can give rise to two mental states, one in which the behavior is recognized and another one in which there is no memory for the behavior. In the case of recognition, the next process would be probing for source memory, that is, perceivers more specifically probe their memory for a trace of the group origin of the behavior. In the absence of recognition, the response is based on guessing processes, which are engaged next. The model characterizes the sequential processing across mental stages to an observable response characterizing the contribution of each process in terms of the probabilities with which the mental states associated with it are reached. In building this model, it becomes clear via mathematical analysis of model identifiability that the data typically collected in the assignment task are too sparse to allow one to disentangle these different processes. Klauer and Meiser (2000) therefore augmented the task by adding new behaviors that were never seen at the testing stage. Figure 18.3 gives a graphical overview of the model. Fitting the model to the data means choosing numerical values for the model parameters so that the probabilities for the different response categories computed from these parameter values match the observed relative frequencies of the different categories as closely as possible. For categorical data, the statistical machinery is sufficiently well developed to allow one to do this using standard software and to assess whether the model fits the data to a degree that is consistent with the level of random noise inherent in the data or, alternatively, whether there are significant deviations of the data from the model (Klauer, Stahl, & Voss, 2012). If the model fits the data, this means that the ensemble of model assumptions is consistent with the observed data. Getting goodness-of-fit information is another important advantage of mathematical models over and above verbal theories. Having a model that statistically fits the data implies (1) that



Mathematical Modeling 383 Group origin (G) of behavior:

Old behavior

New behavior

FIGURE 18.3. Graphical depiction of the multinomial processing-tree model for the (augmented) assignment task in the illusory-correlation paradigm. Behaviors are shown to the left, responses in rectangles to the right. Given a behavior, a perceiver can respond that he or she does not remember the behavior as having been seen before (response N for “new”) or that it is an old one that was seen before and stemmed from group A (response A) or B (respond B). Given an old behavior from group G (G = A or B), perceivers either recognize the item as old (with probability DG) or not (with probability 1 –DG). In the case of recognition, they either have memory for the behavior’s group origin (with probability dG), leading to a correct response (i.e., G = A or B as the case may be), or they do not have such memory (with probability 1 – dG). In this latter case, participants resort to guessing A or B, and the probability that they guess A is given by probability a. If they do not remember having seen the item in the first place (with probability 1 – DG), they first guess the behavior’s old versus new status, guessing old with probability b. A new behavior is detected as new with probability DN and not detected with probability 1 –DN, in which case guessing processes determine the response as before.

the ensemble of model assumptions is internally free of contradictions and (2) consistent with the entire pattern of data at the level of detail at which the data were recorded and fitted. That is, we then have a theory that accounts not only for one or two crucial effects in the responses, but also for the entire and often complex pattern of responses. In contrast, a verbal theory is often focused on accounting for one or two effects of crucial theoretical interest, whereas it is unclear whether the theory is consistent with the entire pattern

384

FORMAL THEORIES

of observed data. That is, it is often difficult to answer the question for a verbal theory whether a set of plausible auxiliary assumptions exists that— when added to the core assumptions of the theory—would allow one to account not only for the central effects but for the entire pattern of observed data. Moreover, having identified and estimated the contribution of the major processes involved in generating the response pattern, we can now judge which of these processes is responsible for the effects of crucial theoretical interest. For example, in the illusory correlation data, the effect of interest is that the majority Group A is assigned proportionally more positive behaviors than the minority Group B. This could come about via different kinds of memory effects such as through especially good memory for the doubly distinctive negative Group B behaviors as postulated in one prominent verbal account (Hamilton & Sherman, 1990) or, alternatively, through an inference from the skewed marginals (there are more Group A than Group B behaviors, and there are more positive than negative behaviors) to an underlying contingency between group membership and behavior valence as in PC. The perceived contingency might then guide guessing processes if the information retrieved from memory does not suffice to make a memory-based assignment. In this case, the effect should be mapped on guessing parameter a such that positive behaviors would be associated with a larger value of a, the tendency to guess Group A, than negative behaviors. In contrast, if the effect is based on especially good memory for the doubly distinctive (negative Group B) behaviors, then one would expect memory parameters DB and/or dB for negative behaviors to be especially large compared to the other memory parameters. These different hypotheses thus predict different patterns of parameter values for the model parameters that can be tested using traditional significance tests. In other words, the complete characterization of the assignment paradigm in terms of underlying processes allows one to decompose observed effects of theoretical interest in terms of their latent causes. This illustrates another central function of mathematical models: allowing one to make inferences as to latent unobservable theoretical constructs from the observed data (Ulrich, 2009). Many of our central theoretical concepts cannot be directly observed. There is no perceptual correlate of things such as associations, memory traces, and mental control. Instead, such concepts have to be justified and investigated in terms of their effects on observable responses. Observable responses in turn are typically multiply determined: They are the outcome of a sequence or cascade of processes typically involving perceptual processes, memory processes, interpretative and construal processes, and responsemonitoring processes. The ability of mathematical models to disentangle the contribution of these different processes that shape the data underlies their ability to make latent constructs more amenable to empirical scrutiny. This issue is closely tied to the issue of model validation, of which statis-



Mathematical Modeling 385

tical fit is only one step. It is just as important to collect more specific evidence in support of the assumption that the central latent processes incorporated in the model are indeed adequately captured by the model. One way to do so is to implement experimental manipulations targeting one of the processes in question (such as memory for the behaviors in the above example) that should not affect the other processes (such as source memory and guessing) and to see whether the manipulation selectively affects the parameter estimates for only the manipulated process (Klauer et al., 2012). Such demonstrations of selective influence go beyond model fit in that they further raise one’s confidence in the particular substantive psychological interpretation of the process and associated parameter in question (e.g., as capturing memory for the behaviors in the example). Finally, multinomial processing-tree models also illustrate that mathematical models at the algorithmic level of theoretical analysis can operate at different degrees of specification. For example, the typical multinomial model deals with response frequencies and leaves temporal aspects of the involved processes out of consideration. Thus, it is typically not necessary to specify the temporal characteristics of the involved processes beyond the temporal order of processing stages incorporated in the tree representation. If the response latencies are also recorded, however, we could subdivide response categories into fast and slow responses of each category or aim at modeling the response latency data directly via a suitable mathematical model such as a diffusion model (e.g., Klauer, Voss, Schmitz, & Teige-Mocigemba, 2007). In a similar vein, we may consider response frequencies for each participant or item or participant–item combination separately to take into account possible differences between participants, items, or their combinations that can compromise the simpler analyses of response frequencies aggregated across participants and items (e.g., Klauer, 2006, 2010). In recent years, multinomial processing-tree models have been proposed as algorithmic models for many different paradigms of interest for social psychology such as for Taylor, Fiske, Etcoff, and Ruderman’s (1978) “Who said what?” paradigm (Klauer & Wegener, 1988); for the implicit association test (Conrey, Sherman, Gawronski, Hugenberg, & Groom, 2005; Meissner & Rothermund, 2013). An overview of the models is provided by Klauer et al. (2012). A review of applications of the process dissociation model, a simple multinomial processing-tree model, can be found in Payne and Bishara (2009). In addition, algorithmic mathematical models outside the family of multinomial processing-tree models have been applied to social-psychological questions (e.g., Smith & Zaraté, 1992).

Summary Taken together, mathematical models serve a number of important functions in psychological research. I focused here on their use as formal theories

386

FORMAL THEORIES

of psychological phenomena couched either at the computational level of theoretical analysis as in normative models or at the algorithmic level as in multinomial processing-tree models. I illustrated what I see as their major advantages. One major advantage is that mathematical modeling employs a constrained formal language and associated mathematical and statistical machinery that build on and incorporate the principles and power of deductive logic. This makes it difficult for mathematical models to fall short of a number of important criteria that any theory should meet. Thus, the mathematical formulation encourages exactness and making explicit all assumptions that go into the theory. It is also much more difficult than in verbal theories to build in inadvertently inconsistent or mutually contradictory assumptions. Such internal conflicts will typically become quickly apparent in attempts to apply the faulty model to real data using standard tools of parameter estimation and data simulation that bring the power of deductive logic to bear on the model premises. As already mentioned, the strong links with deductive logic also ensure that the deductive chains from the model assumptions to predictions are free of gaps (i.e., that they are in fact valid deductions) and reasoning fallacies. Furthermore, successful fitting of a model to a set of data ensures not only that the theory specified by the model accounts for a few key effects of central theoretical interest, but is consistent with the entire pattern of data at the level of detail at which the data are modeled. This is obviously a stronger outcome than the successful prediction of a few key effects in the data pattern as often achieved by verbal theories. Here, the question often remains open as to whether the theory can or cannot be augmented by plausible auxiliary assumptions to account for all of the data. Moreover, successful model fit is a kind of existence proof in that it shows that the ensemble of model assumptions can in principle account for the data. This is often interesting in actively assessing the possibility of alternative accounts. For example, tasks currently explained by means of dual-process models in terms of two qualitatively different processes (e.g., the shooter task; Correll et al., 2002) might also be successfully modeled by means of a mathematical diffusion model that postulates only one homogeneous underlying process of evidence accumulation (Klauer, 2014). If so, the modeling exercise would reveal that a single-process alternative model provides a viable alternative account of the data. Similarly, the normative model of PC sketched above reveals that there is a viable alternative account of PC as rational statistical inference. Relatedly, a theory that is stated as a formal mathematical model can be systematically explored in terms of the range of data that it can account for in principle, leading to quantifications of its testability, parsimony, and complexity. This facilitates comparing theories cast as mathematical models in terms of these criteria (e.g., Klauer & Kellen, 2011) that take a central role



Mathematical Modeling 387

in philosophy-of-science treatments of theory evaluation and comparison (Gawronski & Bodenhausen, Chapter 1, this volume). As we have seen, mathematical models differ in scope in several respects. The normative model of the PC phenomenon is relatively broad in scope inasmuch as it can be applied to many different tasks and paradigms in that context. For example, it is easy in principle to generalize the model to contingency tables with more than two attributes as in Simpson’s (1951) paradigm, to attributes with more than two levels, and to different assignment, prediction, and extrapolation tasks in use in PC research. Moreover, the normative model is able to make many new and as yet untested predictions in this area, as exemplified in this chapter. Other models are more limited in scope. For example, a given multinomial processing-tree model is usually tailor-made for one experimental paradigm such as the assignment task in illusory-correlation research. And it is usually limited to accounting for the frequencies of different responses, leaving aside their temporal characteristics. Finally, mathematical models are particularly well suited to uncover unobservable mechanisms and to make unobservable theoretical constructs more amenable to empirical study. They do so because in modeling a given behavior we need to incorporate the processes and associated constructs that are the most influential ones in shaping the behavior under study, and in order to be able to achieve a reasonable fit to the data, we have to specify how they interact in generating the behavior in question. This in turn allows one to disentangle the theoretical processes of central interest and to provide measures of associated latent constructs via the standard tools of model estimation. Note 1. A technical report with details on the normative model (Klauer, 2008) can be obtained from the author upon request.

References Anderson, J. R. (1990). The adaptive character of thought. Hillsdale, NJ: Erlbaum. Baron, J. (2012). The point of normative models in judgment and decision making. Frontiers in Psychology, 3, 577. Campbell, D. T. (1963). Social attitudes and other acquired behavioral dispositions. In S. Koch (Ed.), Psychology: A study of a science (Vol. 6, pp. 94–172). New York: McGrawHill. Conrey, F. R., Sherman, J. W., Gawronski, B., Hugenberg, K., & Groom, C. (2005). Separating multiple processes in implicit social cognition: The quad-model of implicit task performance. Journal of Personality and Social Psychology, 89, 469–487.

388

FORMAL THEORIES

Correll, J., Park, B., Judd, C. M., & Wittenbrink, B. (2002). The police officer’s dilemma: Using ethnicity to disambiguate potentially threatening individuals. Journal of Personality and Social Psychology, 83, 1314–1329. Fiedler, K. (2000). Beware of samples! A cognitive-ecological sampling approach to judgment biases. Psychological Review, 107, 659–676. Fiedler, K., & Freytag, P. (2004). Pseudo-contingencies. Journal of Personality and Social Psychology, 87, 453–467. Fiedler, K., Freytag, P., & Meiser, T. (2009). Pseudocontingencies: A long-overlooked cognitive illusion. Psychological Review, 116, 187–206. Fiedler, K., Freytag, P., & Unkelbach, C. (2007). Pseudocontingencies in a simulated classroom. Journal of Personality and Social Psychology, 92, 665–677. Gelman, A., Carlin, J. B., Stern, H. S., & Rubin, B. D. (2004). Bayesian data analyses (2nd ed.). Boca Raton, FL: Chapman & Hall/CRC. Greenwald, A. G., Draine, S. C., & Abrams, R. L. (1996). Three cognitive markers of unconscious semantic activation. Science, 273, 1699–1702. Hamilton, D. L., & Gifford, R. K. (1976). Illusory correlation in interpersonal perception: A cognitive basis of stereotypic judgments. Journal of Experimental Social Psychology, 12, 392–407. Hamilton, D. L., & Sherman, S. J. (1990). Illusory correlations: Implications for stereotype theory and research. In D. Bar-Tal, C. F. Graumann, A. W. Kruglanski, & W. Stroebe (Eds.), Stereotyping and prejudice: Changing conceptions (pp. 59–82). New York: Springer. Hammdon, J. L. (1973). Two sources of error in ecological correlations. American Sociological Review, 38, 764–777. Hintzman, D. L. (1991). Why are formal models useful in psychology? In W. E. Hockley & S. Lewandowsky (Eds.), Relating theory and data: Essays on human memory in honor of Bennet B. Murdock (pp. 39–56). Hillsdale, NJ: Erlbaum. Kaiser, F. G., Byrka, K., & Hartig, T. (2010). Reviving Campbell’s paradigm for attitude research. Personality and Social Psychology Review, 14, 351–367. Klauer, K. C. (2006). Hierarchical multinomial processing tree models: A latent-class approach. Psychometrika, 71, 7–31. Klauer, K. C. (2008). Pseudocontingencies: A normative model. Unpublished manuscript, University of Freiburg, Germany. Klauer, K. C. (2010). Hierarchical multinomial processing tree models: A latent-trait approach. Psychometrika, 75, 70–98. Klauer, K. C. (2014). Random-walk and diffusion models. In J. Sherman, B. Gawronski, & Y. Trope (Eds.), Dual process theories of the social mind (pp. 139–152). New York: Guilford Press. Klauer, K. C., & Kellen, D. (2011). The flexibility of models of recognition memory: An analysis by the minimum-description length principle. Journal of Mathematical Psychology, 55, 430–450. Klauer, K. C., & Meiser, T. (2000). A source-monitoring analysis of illusory correlations. Personality and Social Psychology Bulletin, 26, 1074–1093. Klauer, K. C., Stahl, C., & Voss, A. (2012). Multinomial models and diffusion models. In K. C. Klauer, A. Voss, & C. Stahl (Eds.), Cognitive methods in social psychology (abridged ed., pp. 331–354). New York: Guilford Press. Klauer, K. C., Voss, A., Schmitz, F., & Teige-Mocigemba, S. (2007). Process components of the Implicit Association Test: A diffusion-model analysis. Journal of Personality and Social Psychology, 93, 353–368. Klauer, K. C., & Wegener, I. (1998). Unraveling social categorization in the “Who said what?” paradigm. Journal of Personality and Social Psychology, 75, 1155–1178.



Mathematical Modeling 389

Kruglanski, A. W., & Stroebe, W. (2005). The influence of beliefs and goals on attitudes: Issues of structure, function, and dynamics. In D. Albarracín, B. T. Johnson, & M. P. Zanna (Eds.), The handbook of attitudes (pp. 323–368). Mahwah, NJ: Erlbaum. Marr, D. (1982). Vision. San Francisco: Freeman. Meiser, T., & Hewstone, M. (2004). Cognitive processes in stereotype formation: The role of correct contingency learning for biased group judgments. Journal of Personality and Social Psychology, 87, 599–614. Meissner, F., & Rothermund, K. (2013). Estimating the contributions of associations and recoding in the Implicit Association Test: The ReAL model for the IAT. Journal of Personality and Social Psychology, 104, 45–69. Payne, B. K., & Bishara, A. J. (2009). An integrative review of process dissociation and related models in social cognition. European Review of Social Psychology, 20, 272–314. Schaller, M. (1994). The role of statistical reasoning in the formation, preservation and prevention of group stereotypes. British Journal of Social Psychology, 33, 47–61. Simpson, E. H. (1951). The interpretation of interaction in contingency tables. Journal of the Royal Statistical Society, 13, 238–241. Smith, E. R., & Zaraté, M. A. (1992). Exemplar-based model of social judgment. Psychological Review, 99, 3–21. Tajfel, H. (1969). The cognitive aspects of prejudice. Journal of Social Issues, 25, 79–97. Taylor, S. E., Fiske, S. T., Etcoff, N. J., & Ruderman, A. J. (1978). Categorical and contextual bases of person memory and stereotyping. Journal of Personality and Social Psychology, 36, 778–793. Ulrich, R. (2009). Uncovering unobservable cognitive mechanisms: The contribution of mathematical models. In F. Rösler, C. Ranganath, B. Röder, & R. H. Kluwe (Eds.), Neuroimaging of human memory: Linking cognitive processes to neural systems (pp. 25–41). New York: Oxford University Press.

19 Agent-Based Modeling Eliot R. Smith Asaf Beasley

O

ften the most important outcomes for society and individuals— ranging from whom we marry and what jobs we get to the formation and influence of norms, to fluctuations in housing and financial markets, to escalation of intergroup conflicts—are not consequences of the explicit choices of lone individuals, but rather result from repeated interactions among multiple individuals over time. Even when every person in a situation has similar goals, the dynamics of their interactions can have unintended consequences. Consider, for example, “pluralistic ignorance,” a situation in which most people disagree with a norm but falsely assume that the majority of those around them support it. This leads to the paradoxical behavior of a group engaging in an activity that most members find problematic (e.g., fraternity members participating in drinking rituals; Prentice & Miller, 1993) because each person thinks he is one of the few who oppose the behavior and does not want to be seen as a dissenter. Fearing discovery, such persons may actually enforce the norm publicly, which only perpetuates the false belief that this norm is widely endorsed (Centola, Willer, & Macy, 2005). To help understand and explain such dynamic interactions, agent-based modeling (ABM, also termed multiagent modeling) is an ideal approach. As we will see, ABM allows the theorist to specify the rules of behavior assumed to be followed by individuals (agents), which may be based on empirical 390

Agent-Based Modeling 391

evidence or on theoretical reasoning. The model can then be run to display the outcomes that occur when multiple agents following those rules interact over time. This chapter introduces this approach to theory construction by providing an example, some background, and key definitions, followed by a description of several areas of social-psychological research interest in which ABMs have been or can be developed. We then discuss a few larger, metatheoretical issues as a way of situating this approach among others represented in the volume, and we close with a few concluding remarks.

Background and Definitions A classic example of an ABM is the economist Thomas Schelling’s model of segregation (Schelling, 1971). The model explored how a seemingly reasonable preference of homeowners and homebuyers to avoid being a racial minority in their neighborhood can lead to nearly complete segregation of neighborhoods. In this model, agents (representing households) of two races are represented by different color squares in a lattice. Initially, agents are randomly distributed and therefore are intermixed without any segregation. The agents abide by the following rule: If they are in the minority color among the 9 squares including themselves and the 8 surrounding squares (i.e., their “neighborhood”), they move to a nearby empty space on the grid where they are not in the minority. By iterating this process a number of times until no further movements occur, under a broad range of assumptions and thresholds for what constitutes a minority (for example 30% or 50%), the outcome usually is extreme segregation, with few if any mixed neighborhoods. What is striking about this model is that it generates that outcome even though no agent is assumed to have the goal of extreme segregation. If everyone has the understandable preference to not be a minority in their neighborhood and moves accordingly, a high level of segregation will ensue because those movements inevitably increase segregation in both the neighborhood the agent leaves and the one it moves to. Thus, the model illustrates the potentially surprising patterns that can emerge from simple behavioral rules as they operate over time. The model also shows that large-scale segregated patterns can occur where no central, controlling authority intentionally generates the pattern (e.g., through legal or informal restriction of particular races to distinct neighborhoods). This property of macro-level behavior, termed self-organization, is usually a feature of the most interesting ABMs. Despite its simplicity, Schelling’s segregation model illustrates many of the typical features of agents that are represented in ABMs (Railsback & Grimm, 2012; Smith & Conrey, 2007). First, agents are discrete, possessing discernible boundaries that separate them from their external environment and other agents. Rather than being homogeneous, each agent has its own unique properties (e.g., in Schelling’s model, its race and its location). Second, agents are autonomous, acting according to internally dictated rules

392

FORMAL THEORIES

rather than being controlled by some external executive or other force. Third, agents are usually goal directed in that their actions are attempts toward some individual purpose, such as finding a mate, obtaining food to survive, or getting as many points as possible in a game played with other agents. Fourth, agents have local, limited information (e.g., information only about their own spatial neighborhood) and lack complete knowledge of their environment and other agents with whom they interact. Although these four features are typical, agents also vary on some dimensions. In some models (like Schelling’s), the behavioral rules of agents may be fixed and unchanging. Alternatively, agents may be assumed to be adaptive, learning and changing through repeated interactions, or evolving over successive generations through a Darwinian selection process. For example, in a model of mate choice, an agent repeatedly rejected by potential mates may adapt by lowering its aspiration level and begin making offers to potential mates with lower levels of desirability. Most commonly, agents are programmed into a virtual (computer) environment, but they can also be embodied (robotic), such as the robotic players featured in the RoboCup soccer competition (www.robocup.org/robocup-soccer). Finally, agents can represent any discrete entity a researcher is interested in studying, such as neurons, single-celled or higher-level organisms, individuals, groups, corporations, or nations. In most social-psychological applications, of course, agents represent individual humans. Based on this definition of agent, a multiagent system is a system in which multiple interdependent agents interact in a (usually) virtual environment. The interdependence may be direct, as when the agents interact directly with each other—for example, eating each other in a predator– prey ecosystem model. Or interdependence may be indirect, when agents constitute an environment that influences other agents, as in Schelling’s segregation model, or when they compete for shared environmental resources such as food. Finally, an agent-based model (or multiagent model) is a multiagent system that is explicitly intended to represent the postulates of a scientific theory. We believe that such models are ideally suited to capture social-psychological phenomena that involve multiple interdependent actors, with dynamic interactions that occur repeatedly over time, and include reciprocal influences, where actors impact each other bidirectionally. As with any theory, considerations of parsimony, clarity, and simplicity are paramount. Agent-based modeling is related to a general movement that has touched a number of scientific fields, the study of complex systems (Weaver, 1948). These are systems of interacting entities that as a whole display behaviors, or accomplish tasks, that are not a direct consequence (e.g., a linear summation) of the behaviors or capacities of individual entities. For example, no single bee appears to have full knowledge of how to build a hive, but somehow thousands of interacting bees are able to accomplish this feat. Birds readily form dynamic, highly organized flocks with no leader orchestrating their

Agent-Based Modeling 393

actions. ABM is one tool to understand how specific behaviors of interconnected individuals, each with limited information and control of their environment, can result in interesting, often nonobvious patterns at a macro level. These nonintuitive or difficult to infer patterns are often described as “emergent.” Though commonly used, the concept of emergence is difficult to clearly define (Epstein, 1999). In one definition, “emergent” simply means surprising, unintuitive. But as Epstein and Axtell (1996, p. 35) note, this usage “begs the question, ‘surprising to whom?’.” Outcomes that are surprising for one researcher may seem readily predictable for another (perhaps one whose intuitions have been refined by extensive experimentation with ABMs). A second definition is that outcomes are “emergent” if they are definable only at a higher level than the individual entities (agents) in the model. For example, segregation (synchronization, consensus, etc.) are properties of a population of multiple agents and cannot be defined for a single agent. A final meaning, related to both of these, is that an outcome is “emergent” if it is not built in to the model’s behavior, for example, by being explicitly represented as a goal of individual agents, but that nevertheless results from repeated interactions among these agents. In this sense, segregation is an emergent outcome of Schelling’s original model, but it would not be emergent if the model’s agents sought to move to highly segregated neighborhoods, building in the outcome. Railsback and Grimm’s (2012) textbook captures several of these aspects in a three-part conceptual definition of an emergent outcome: “It is not simply the sum of the properties of the model’s individuals; it is a different type of result than individual-level properties; and it cannot easily be predicted from the properties of the individuals” (p. 101).

Illustrative Social-Psychological Models Multiagent models have been formulated in many core areas of socialpsychological relevance.

Social Influence Many models address situations in which individuals are affected by others with whom they interact, whether through conformity (moving toward the others’ opinions or behavior) or through contrast (moving away from others). What patterns in the population emerge when such influence takes place repeatedly over time (Mason, Conrey, & Smith, 2007)? A key question is whether all agents will collapse toward uniform attitudes or behavior or whether diversity will persist in the population (Nowak, Szamrej, & Latané, 1990). An interesting model of social influence (Centola et al., 2005) examines how unpopular norms (ones that most people personally disagree with)

394

FORMAL THEORIES

could spread through a population. Here agents are represented by squares in a lattice structure, their neighbors comprising the squares that surround them. When a small group of “true believer” agents enforcing an unpopular norm are clustered together, it is possible for the norm to spread (cascade) through the population. Other agents near the initial group of enforcers, based on their local view of the world, overestimate the global popularity of the norm and also enforce it (despite their private disagreement) in order to avoid being seen as dissenters. This enforcement in turn gives other agents in their neighborhood the false impression that the norm is supported, leading them to enforce it in the same way, eventually resulting in a cascade of enforcement based on pluralistic ignorance. However, this result only holds when agents are embedded in a neighborhood, connected only to agents adjacent to them in the lattice. If every agent in the population is connected to all others (so that all agents have information about the entire population), or if true believers are randomly distributed in the population rather than clustered, the norm fails to spread. This model highlights the importance of how agents are connected (network structure) and also shows how a counterintuitive outcome—the spread of a norm that nearly everyone disagrees with—can occur under the right circumstances.

Mate Choice Choosing a mate poses many issues. People have to estimate the desirability of potential mates and must decide when to make an offer to one who is available now versus searching further for a potentially better alternative, in the face of great uncertainty. What decision rules produce good outcomes in this task, especially considering the two-sided nature of the decision (each agent is being evaluated as well as evaluating others)? An early social-psychological model in this domain by Kalick and Hamilton (1986) showed how agents who are trying to maximize the attractiveness of their partners nevertheless end up forming couples whose levels of attractiveness are correlated; one need not assume that the agents are seeking partners with similar levels of attractiveness. This occurs because the more attractive agents in the population get paired up and removed from the “dating pool” early, leaving a population of “singles” whose average attractiveness grows lower and lower over time. Naturally, then, as couples are formed over time, both partners in the later-forming ones tend to be lower in attractiveness than partners in couples that formed early, generating the correlation. A more recent model by Simao and Todd (2002) incorporates additional elements of human mate choice. For instance, there is a flirtation period when agents learn their own level of attractiveness to the opposite sex, which enables them to develop realistic aspiration levels—a sense of the caliber of a mate they can hope to secure. In addition, agents are embedded in social networks, instead of randomly interacting as in Kalick and Hamilton, and

Agent-Based Modeling 395

new agents join and leave these networks over time (as in real life) instead of the total population remaining static. The key finding was that giving agents a “courting period,” when they tentatively form couples but for a time can switch to a more attractive mate that may become available, increases withincouple attractiveness correlations. This is a striking, nonobvious effect of courting periods, which we often think of as serving the purpose of allowing potential mates to get to know each other to make better-informed decisions about whether to finally commit, or of allowing agents to stay in the market longer in case a more desirable mate comes along.

Group Search for Problem Solutions It often happens that many individuals have partial information about solutions to a problem. How can they combine their information to speed the group’s convergence on a high-quality solution to the problem, allowing the group to perform better than any individual could based solely on his or her own information? Might overly rapid convergence on a moderately good solution foreclose further search that could uncover a better one (Hutchins, 1991)? Similarly, when many perceivers have limited knowledge about a target person, can they combine their information (through gossip) to arrive at a fuller picture of the individual, perhaps by sharing knowledge about important yet rare or difficult-to-observe behaviors such as deceit or aggression (Smith & Collins, 2009)? Several models have addressed these and similar questions (e.g., Kennedy, 2009). One notable agent-based approach to studying group information search and sharing comes from Rendell, Boyd, Cownden, Enquist, Eriksson, et al. (2010). Rendell et al. set up a tournament and invited researchers to submit strategies for a social learning challenge. In their environment, agents seek payoffs by playing a game akin to a slot machine with 100 different levers, where each lever produces different probabilistic payoffs. The payoffs for each lever randomly change after a certain number of iterations, so agents must continuously learn what the best strategy is at a given time. At each time point, agents can choose one of three behaviors: (1) OBSERVE which levers a random other agent utilizes and see its approximate payoff; (2) INNOVATE by learning what the payoff is for pulling a random lever for which they do not currently know the payoff; or (3) EXPLOIT by choosing a particular lever that they have learned through INNOVATION or OBSERVATION. Only the EXPLOIT behavior earns the agent an actual payoff. In this model, OBSERVATION is a type of social learning (learning what other agents are doing), and INNOVATION is a type of individual learning. In the tournament, the winning strategy and the runner up relied almost completely on social learning. This is somewhat surprising in the context of the model because there was no added cost for individual learning. Moreover, in many cases (over 50% of the time) agents who used OBSERVE failed to learn anything new, because they saw the payoff for a lever they already

396

FORMAL THEORIES

knew. Even so, the fact that agents tend to enact the most profitable strategies they know, thereby filtering out inferior strategies, means that OBSERVATION, social learning, is an extremely powerful information acquisition technique.

Cooperation Often positive outcomes can be obtained only when two or more individuals work together, but attempts at cooperation may leave an agent vulnerable to exploitation by noncooperative others. How can cooperation emerge (e.g., through learning or evolution) and be maintained? Axelrod and Hamilton (1981)’s pioneering work with the prisoner’s dilemma game inspired a great deal of modeling effort around these questions, which continues to the present (e.g., Nowak, 2006). An illustrative model (Kameda, Takezawa, & Hastie, 2003) examines the emergence of the norm of sharing large food resources such as hunted game within a group (beyond kin), a practice that anthropologists find to be common. The model assumes that at each time period one randomly chosen group member acquires a food resource. Other members may demand that it be shared. If the acquirer is unwilling to comply with such a demand, fights ensue, resulting in costs to the loser(s), and if the acquirer loses any fight, the resource is shared with all who made the demand. Results of simulated evolution show that under a wide range of parameter values the sharing strategy (demanding that others share their resources, willingly sharing one’s own) can emerge. In fact, sharing dominates other potential strategies such as selfishness (demanding sharing by others, refusing to share when one is the acquirer) or “bourgeois” respect for private property (neither demanding that others share nor willingly sharing one’s own resources). Kameda et al. argue that their work sheds light on the emergence and sustainability of prosocial or cooperative norms, in the absence of formal sanctioning structures to enforce compliance.

Social Networks Often connections or links between individuals are structured so that each can interact with specific linked others as in some models already discussed, rather than with just anyone. What are the characteristics of networks that arise through individual choices or dyadic interaction (e.g., Sutcliffe, Wang, & Dunbar, 2012)? What are the consequences of the structure of the network for other forms of social behavior such as influence or cooperation (e.g., Centola & Macy, 2007)? Many behaviors, such as adopting a new technology, participating in a social movement, or spreading an urban myth, may depend on influence from multiple sources. One source (such as a specific friend in one’s social network) exerting influence on multiple occasions will likely not suffice for

Agent-Based Modeling 397

people to enact costly or controversial behaviors. When collective behaviors require more than one source of influence, they are referred to as “complex contagions,” in contrast to “simple contagions,” such as the spread of information or disease, which require only a single source of influence or infection. Though previous work has found that “weak ties,” connections linking distant segments of a network, facilitate simple contagions, such farreaching ties actually hinder the spread of complex contagions. This was demonstrated under a variety conditions in a series of simulations (Centola & Macy, 2007). When ties in strongly interconnected sections of a network were randomly replaced by weak ties bridging to more distant sections of the network, the spread of complex contagion decreased. This is because there is a low probability that such long-distance links will result in two or more connections from the same source region to a new region. Without such a “wide bridge” consisting of two or more connections, the contagion cannot spread because individual agents will not be exposed to two or more sources of influence supporting the new behavior. These findings underscore the notion that seemingly similar phenomena (simple contagion and complex contagion) may differ greatly in terms of the network structures that facilitate them. Here modeling demonstrates such differences as well as their underlying cause.

Top-Down and Bottom-Up Uses of ABMs Multiagent models can be used in two distinct ways. First, a researcher may work in top-down fashion, beginning with a group-level phenomenon of interest, and formulating models to investigate what alternative agent behaviors or interaction rules might generate it. The goal of this approach is to broaden theoretical thinking about potential alternative mechanisms underlying a phenomenon, often to get away from the straightforward assumption that individual agents are seeking exactly the observed outcomes. Schelling’s segregation model (1971) is an example. Beginning with the observation that segregation is common, the model reveals that agents who simply desire not to be in the minority in their neighborhoods (but do not want extreme segregation) can end up creating segregation. Such modeling results can inspire empirical work that directly tests the hypothesized behavioral rules. As Epstein (1999, p. 48) noted, “Agent-based models may . . . furnish laboratory research with counterintuitive hypotheses regarding individual behavior. Some, apparently bizarre, system of individual agent rules may generate macrostructures that mimic the observed ones. Is it possible that those are, in fact, the operative micro-rules? It might be fruitful to design laboratory experiments to test hypotheses arising from the unexpected generative sufficiency of certain rules.” Second, models can also be used in bottom-up fashion, starting with a pro-

398

FORMAL THEORIES

posed set of agent behaviors or interaction rules, to observe what populationlevel patterns they produce. This constitutes a rigorous, reproducible thought experiment, aimed at obtaining new insights about how mechanisms operate and interact when embodied in multiple agents over time. The distinction between the top-down and bottom-up approaches has to do with the modeler’s specific interests and the data that are available; this distinction is not an intrinsic aspect of the model. For example, someone might have experimental data demonstrating that people prefer mates who are more attractive, regardless of the perceiver’s own level of attractiveness. They might then run an ABM to see what population outcomes, such as the level of correlation between the attractiveness levels of partners, emerge when agents interact and form couples based on this preference (representing a bottom-up approach). Others (e.g., Kalick & Hamilton, 1986) may start with the observation of a population-level pattern (a moderate to high correlation between attractiveness levels of dating or married partners) and construct a model testing whether this pattern emerges if agents are assumed to simply prefer the most attractive available mates (rather than to prefer mates with similar levels of attractiveness). This is a top-down application. In both instances the model may be identical, but the focus and direction of analysis are different. Whether top-down or bottom-up approaches are preferred, a high priority is given to a model’s simplicity and transparency (rather than on fitting all details of some data; Railsback & Grimm, 2012). The key point is that a model represents a theory that by definition is always simplified and abstracted. The goal is not for a model to represent all the complexity of reality, and efforts in that direction will inevitably damage the model’s ability to provide insight and understanding. Some models have the goal of prediction, in cases where the modeled system is well understood (Heath, Hill, & Ciarallo, 2009). More often, however, ABMs support understanding and explanation but not direct prediction. For example, a model cited by Epstein (2008) explored how housing prices are impacted by interest rates, the number of first-time buyers, and other factors. One major insight from this model is that the number of firsttime buyers in a market is extremely important (even more important than interest rates) to housing prices. In 2008, after the banking crisis, the housing market declined as banks increased the size of required down-payments, hence decreasing the number of first-time buyers who could enter the market. The model did not predict the housing crash (it was not intended to do that), but it did help explain why this decision by banks hurt the market. In general, models can help us explain observed outcomes, while in contrast predicting outcomes is difficult and depends on many factors on which we are not likely to have full information. Many powerful explanatory theories do not allow predictions, in the same way that plate tectonics explains earthquakes, but (at least currently) does not allow their prediction with specificity.

Agent-Based Modeling 399

Metatheoretical Issues To promote understanding of the specific place of ABM among other theoretical approaches represented in this volume, we offer additional characterizations and evaluations of the ABM approach.

Typical Patterns of Explanations Multiagent models usually explain a group- or population-level pattern, on the basis of assumptions about the lower levels (individual agent behaviors, agent–agent and agent–environment interactions). Often the goal is to demonstrate emergence in the sense that the higher-level pattern does not occur because it is the explicit goal of any agent, but results as an unintended byproduct. For example, Simao and Todd’s (2002) model reveals that a courting period in a mate choice model, where agents are situated in changing social networks, has an unexpected impact on the macro-level pattern of the correlation between attractiveness levels within couples.

Core Metaphors or Assumptions of the ABM Approach The idea of crossing levels is key, with individual agent behaviors assumed to aggregate to produce larger-scale patterns. “A crucial lesson of . . . agentbased models . . . is that even perfect knowledge of individual decision rules does not always allow us to predict macroscopic structure. We get macro-surprises despite complete micro-knowledge. Agent-based models allow us to study the micro-to-macro mapping” (Epstein, 1999, p. 48, emphasis in original).

Contrast with Variable-Based Modeling A different approach to modeling, which we could term variable-based modeling (VBM), is more common than ABM in social psychology. A VBM focuses on patterns of causal relationships among variables (e.g., a change in X causes a change in Y), rather than on interactions among agents. Smith and Conrey (2007; see also Cedarman, 2005) provide more extensive discussion of contrasts between these approaches, but here are a few key points. 1.  VBMs offer concise descriptions of systems, enabling easy prediction or extrapolation (e.g., values of Y expected for particular values of X), while ABMs must be run to generate predicted outcomes (and usually averages of many runs have to be taken, because aspects of agent behavior are assumed to vary randomly). 2.  VBMs almost always require simplifying assumptions, such as linear relationships, rational decision making by individuals in response to environmental contingencies, and homogeneous agents (the same patterns

400

FORMAL THEORIES

of relationships are expected to describe all agents). In contrast, ABMs are more flexible, with theorists able to specify nonlinear decision rules, agents that learn and change their behavior over time, or subpopulations of agents that use different rules. 3.  Structural equation models, the most common type of VBM in social psychology, generally require stringent causal-ordering assumptions (e.g., no multidirectional flows of causation), although other types of VBMs (such as coupled differential equations) do allow for multiple variables to dynamically influence each other. In contrast, agents in an ABM can affect others in unconstrained causal sequences; agent A’s action may influence B who in turn influences A at a later time. 4.  Perhaps most important, compared to VBMs, ABMs provide a different type of explanation and understanding of a system, what has been termed “generative” explanation (Cedarman, 2005; Epstein, 1999). An ABM that yields an outcome provides a sufficiency proof that the specified set of behavioral rules can generate that outcome, a mode of understanding that is compatible with the emphasis of today’s social-psychological theories on underlying processes that interact to generate observed outcomes. VBMs describe process–outcome links as statistical dependencies (e.g., correlations) among variables, which seem to provide less rich insights: “process theorists would regard them as insufficient and superficial substitutes for the deeper understanding yielded by a generative explanation” (Cederman, 2005, p. 868). ABMs aspire to support this deeper understanding by considering phenomena as generated by the interaction of multiple underlying (perhaps unobserved) processes, rather than by describing the phenomena with parameters estimating statistical relationships among variables.

Structural Features of ABMs and Implications for Model Validation The essential structural features of ABMs are as follows: (1) There are at least two levels (agents, populations, and perhaps additional levels in between, such as dyads or small groups of agents); (2) there are multiple agents; and (3) those agents are interdependent, whether because they directly interact or because they share a common environment (e.g., they use the same finite resources). Multiagent models can be confirmed or disconfirmed in two distinct ways. First, researchers compare the population-level patterns to those found in the real system to see whether the model generates similar outcomes. Second, researchers seek to validate their assumptions about individual agent behavior rules (e.g., in lab experiments). Validation of a model can and should take place at both levels (Edmonds, 2010; Moss & Edmonds, 2005). In practice, rigorous model validation is often difficult. A model with

Agent-Based Modeling 401

many parameters can be tuned to fit data, but without enhancing understanding. The addition of too many parameters can be problematic for this reason. Pitt and colleagues have proposed a framework for validating models that takes into account the extent to which they fit data, as well as penalizing models that contain too many parameters (Pitt, Myung, & Zhang, 2002). A model does not need to be completely valid or accurate to have value for the theorist. The simple act of programming a model, explicitly specifying the rules agents should follow and the environment in which they should exist, can suggest new forms of data to collect. For example, a researcher creating an ABM of social influence might grapple with the possibility that when one individual (A) succeeds or fails in influencing another (B), over time there might be effects on A. Might A become more confident in the opinion due to the success in influencing B? Might A become more susceptible to influence (through a kind of reciprocity) when B seeks to influence A, in turn, on a later occasion? Such questions can be answered, of course, by appropriately designed studies. But they will not even be asked when influence is considered purely as a one-shot, unidirectional effect of person A on person B (Mason et al., 2007). Another possibility is that a model may generate predictions that can be tested by new types of data. For example, a courting and mate choice model might display specific patterns such as a relation between a couple’s length of courtship and their probability of breakup, which could then motivate the collection of new data from experiments or surveys that could test the hypothesis.

Levels of Analysis and Their Interrelations As previously described, the individual and the population are distinct levels in an ABM (and one could also pick out intermediate levels such as groups or dyads of agents if agent interactions are rich and complex). The lower-level behaviors generate the outcomes that are observed at the population level; for example, prey and predator behaviors lead to cyclic patterns in the sizes of the two populations. But the causal flow between levels goes both ways: Population patterns also influence individual agents because the agents are interdependent. For example, if the predator population grows larger, the chance of any one prey individual encountering a predator becomes higher.

Level Represented by an Individual Agent In social-psychological applications of ABM, an agent will ordinarily be assumed to represent an individual person. However, there are other possibilities. Agents could be assumed to represent larger-scale entities such as firms (common in economic ABMs), couples, nations, or social groups of other sorts—anything that can sense and act on its environment and pursue its goals autonomously. Although the model would be complex, it would even be possible to represent a number of individuals who can choose to join

402

FORMAL THEORIES

or exit groups, and to put efforts toward group goals, as well as to represent the groups themselves as agents with their own goals, rules of intergroup interaction, and so on (see Bornstein, 2003). Going in the other direction, agent-based models can capture some of the assumptions of neural network or connectionist models (Smith, 1998), with agents representing lower-level entities (individual neurons or smallscale assemblies of cells) that are richly interconnected and able to send signals to each other. Selfridge (1959/1988) fancifully described an early neuralnetwork architecture intended to visually recognize letters as a multiagent model, called “Pandemonium.” This architecture contained “demons” of several types: feature demons who examine the input for specific visual features (e.g., a vertical bar) and yell if they see those features; letter demons who listen to feature demons and yell if they hear from the feature demons that make up their specific letter (e.g., vertical bar and closed loop for capital P); and a single decision demon who listens to all the letter demons and issues as output the letter demon who yells loudest. Pandemonium underlines the parallels between neural-network architectures and multiagent models in that both postulate autonomous agents using simple decision rules and operating on only local inputs (e.g., only receiving information from specific others).

Trade-offs between Explanatory Breadth (Generality) and Predictive Power (Specificity) This and other trade-offs are ubiquitous in modeling. Our preference is for models that (like the best theories of all types) maximize simplicity and transparency. Such models predict outcome patterns only generically and abstractly, and could not be expected to predict fine details of a specific realworld situation. Others, however, have used ABMs in a more specific, datadriven way—for example Dean et al.’s geographic model of a group of Anasazi Native Americans and their population changes and movements over time driven by climate fluctuations and intergroup conflicts (described in Epstein, 1999). This model sought “to develop, in collaboration with anthropologists, microspecifications—ethnographically plausible rules of agent behavior—that will generate the true history. The computational challenge, in other words, is to place artificial Anasazi where the true ones were in 800 AD and see if—under the postulated rules—the simulated evolution matches the true one” (Epstein, 1999, p. 44).

Avoiding Circular Explanations With multiagent models, circularity is usually not a problem given the clear separation of levels. In fact, many examples show how such models can help avoid otherwise tempting explanatory circularity. For example, it is desirable to avoid assuming that observed segregation implies that agents want seg-

Agent-Based Modeling 403

regation, and models help by showing that other assumptions about what agents want also lead to patterned segregation (Schelling, 1971). It is good to avoid assuming that an observed correlation across couples between the partners’ attractiveness level means that individuals actually prefer similar partners, and models help by showing that other assumptions can also generate that correlation (Kalick & Hamilton, 1986).

Falsifiability of ABMs Modeling is sometimes undertaken in a spirit of “game playing” that fails to meet the requirements of science (such as reproducibility and falsifiability). Edmonds (2010) labels as “floating models” those that are “justified vaguely with reference to some phenomena of interest, use many assumptions that are justified solely in terms of their surface plausibility to the modeler, that are fitted loosely to some known data for outcomes, but are not general enough to be considered as a sort of pseudo-mathematics. Such . . . [models] are often closer to an expression of the conceptions of the modeler rather than a model of anything observed—they are closest to an analogy—a computational analogy” (section 7.1). However, not all models are this loose and informal. Systematic exploration of the behavior of a model under different conditions can be as rigorous as any theory-driven thinking, and falsifiability can be high because of the two levels at which models can be validated. Falsifiability also suffers when a model is overly complex. This is another reason, besides transparency and understandability, that parsimony is a virtue in any type of theory or model. Naive observers’ usual responses to an ABM are typically along the lines of “but you left out X, Y, and Z!,” demanding complexification in the name of the (misguided) goal of increased realism. Theorists need to have a principled response to such demands, in order to avoid losing the necessary simplicity and transparency of a model. A model needs to incorporate those elements or processes that are essential to generate the phenomenon of interest, but no more. With overly complex models, one cannot understand the reasons for their behavior any more than one can by examining the modeled natural system itself.

Unique Challenges for ABMs The ABM approach does have some unique challenges. First is the need for some level of computational literacy, which is not widespread among social psychologists. On the positive side, NetLogo (Wilensky, 1999), which is widely used for ABM, is quite accessible, and one of us has taught severalsession ABM workshops many times, invariably succeeding in bringing typical graduate students to the point of programming their own simple models and experimenting with them. Another significant challenge is the need to rigorously document mod-

404

FORMAL THEORIES

els. Many published models are difficult or impossible to replicate because of inadequate documentation or incomprehensible code (Edmonds & Hales, 2003; Meadows & Cliff, 2012). In other cases, model results may reflect programming errors rather than the intended assumptions, or results may unexpectedly depend on highly specific details of an implementation (see especially Huberman & Glance, 1993). Fortunately, standards for documenting the assumptions of a model are emerging (see Railsback & Grimm, 2012). An important challenge is to take advantage of the potentially mutual reinforcing nature of empirical research and modeling. Ideally, we might envision iterative cycles in which a model (informed by data) generates a novel hypothesis, which can be studied empirically, these empirical results in turn leading to modifications of the model, and so on. However, this complete cycle rarely seems to happen. Most often, just a model informed by data is published. According to Edmonds (2010), often existing data are insufficient to justify specific assumptions such as the agent behavioral rules or model parameter values, or to validate the model’s macro-level behavior. In fact, it is unusual for an empirical researcher to collect data with the thought of how it could inform an ABM, or for a modeler to create models specifically with an eye toward generating new, testable empirical hypotheses. Fortunately, tools like Netlogo make it possible for experimentalists to start incorporating ABMs into their research. Synergy between empirical and modeling work will often involve collaborations in which empirical researchers learn enough about ABM to team up with modelers, to begin potentially fruitful cycles where empirical work and models reciprocally inform each other. The result should be going beyond simple, intuitive “floating models” to those that are explicitly and systematically tested and validated (Edmonds, 2010). As a positive sign, Heath et al.’s survey of modeling practices (2009) finds that the extent of validation is increasing strongly over time. The percentage of published models categorized as offering no validation at all decreased from 75% in 1999 to 12% in 2008.

Unique Strengths and Limitations ABMs are most helpful for crossing levels, describing and analyzing the results when multiple interdependent agents interact following specific assumed behavioral rules. These models, however, do have their limits. For example, many theories in psychology involve assumptions about multiple interacting processes within a single individual (such as theories of interacting cognitive and motivational processes), but the focus of such theories on the isolated individual means that ABMs are not relevant. However, in social psychology specifically, many of our theories do involve interpersonal interaction, influence or learning from other people, effects of group norms, differential treatment of ingroup and outgroup members, and the like—all areas in which explicit modeling of many interacting individuals can help clarify and refine theory.

Agent-Based Modeling 405

Conclusion Most work in social psychology has taken a reductionist approach. The field has excelled in finding specific effects in controlled laboratory studies, breaking them down into component processes, and isolating the conditions that impact them. This is valuable, necessary work, but its success means that more “synthetic” ways of understanding phenomena have been neglected. We know a great deal about how individuals are impacted by specific conditions in our experiments (for example, how people are influenced by knowledge about others’ attitudes), but this alone will not allow us to infer the patterns that emerge when those individuals interact in specific network configurations and repeatedly influence each over time (Mason et al., 2007). ABM is an indispensable tool for capturing these critical, complex social phenomena that cross levels. The time is ripe for social psychologists to increase their use of ABMs. While a few models exist in social psychology, researchers in other fields, like economics, political science, sociology, and even physics, have often taken the lead in building ABMs of human social behavior. Unfortunately, most such models do not use empirically validated rules for individual agent behavior, instead often adopting simplifying (but unrealistic) assumptions such as rational utility maximization (Mason et al., 2007). Social psychologists have the data, or know how to conduct studies to collect the data, that can specify plausible rules at the level of individual agents. On the other hand, we often have less data on more macro, population-level patterns and outcomes, where sociologists, economists, or political scientists may have more insight. ABM can serve as a focus for collaborations between researchers in social psychology and other fields, allowing them to join forces to create models that theoretically connect data at multiple levels of analysis (individual and population). This is something we rarely see in the social sciences. Regardless of whether such lofty interdisciplinary aims are met, the very act of constructing a model can not only suggest new data to collect, but also spark theoretical reflection on processes of dynamic interaction among interdependent individuals, which are often left out of focus in many theories and in lab-based research paradigms. This direction seems essential to developing a fuller understanding of human behavior in its social context (Mason et al., 2007; Smith & Semin, 2004) by conceptualizing social-psychological phenomena as emergent outcomes of dynamically interacting processes. References Axelrod, R., & Hamilton, W. D. (1981). The evolution of cooperation. Science, 211, 306–318. Bornstein, G. (2003). Intergroup conflict: Individual, group, and collective interests. Personality and Social Psychology Review, 7, 129–145. Cederman, L. (2005). Computational models of social forms: Advancing generative process theory. American Journal of Sociology, 110, 864–893.

406

FORMAL THEORIES

Centola, D., & Macy, M. (2007). Complex contagions and the weakness of long ties. American Journal of Sociology, 113, 702–734. Centola, D., Willer, R., & Macy, M. (2005). The emperor’s dilemma: A computational model of self-enforcing norms. American Journal of Sociology, 110, 1009–1040. Edmonds, B. (2010). Bootstrapping knowledge about social phenomena using simulation models. Journal of Artificial Societies and Social Simulation, 13, 1. Edmonds, B., & Hales, D. (2003) Replication, replication and replication: Some hard lessons from model alignment. Journal of Artificial Societies and Social Simulation, 6, 11. Epstein, J. (1999). Agent-based computational models and generative social science. Complexity, 4, 41–60. Epstein, J. (2008). Why model? Journal of Artificial Societies and Social Simulation, 11, 4. Epstein, J. M., & Axtell, R. (1996). Growing artificial societies: Social science from the bottom up. Cambridge, MA: MIT Press. Heath, B., Hill, R., & Ciarallo, F. (2009). A survey of agent-based modeling practices (January 1998 to July 2008). Journal of Artificial Societies and Social Simulation, 12, 4. Huberman, B., & Glance, N. (1993). Evolutionary games and computer simulations. Proceedings of the National Academy of Sciences, 90, 7715–7718. Hutchins, E. (1991). The social organization of distributed cognition. In L. B. Resnick, J. M. Levine, & S. D. Teasley (Eds.), Perspectives on socially shared cognition (pp. 283–307). Washington, DC: American Psychological Association. Kalick, S. M., & Hamilton, T. E. (1986). The matching hypothesis reexamined. Journal of Personality and Social Psychology, 51, 673–682. Kameda, T., Takezawa, M., & Hastie, R. (2003). The logic of social sharing: An evolutionary game analysis of adaptive norm development. Personality and Social Psychology Review, 7, 2–19. Kennedy, J. (2009). Social optimization in the presence of cognitive local optima: Effects of social network topology and interaction mode. Topics in Cognitive Science, 1, 498–522. Mason, W. A., Conrey, F. R., & Smith, E. (2007). Situating social influence processes: Dynamic, multidirectional flows of influence within social networks. Personality and Social Psychology Review, 11, 279–300. Meadows, M., & Cliff, D. (2012). Reexamining the relative agreement model of opinion dynamics. Journal of Artificial Societies and Social Simulation, 15, 4. Moss, S., & Edmonds, B. (2005). Sociology and simulation: Statistical and qualitative crossvalidation. American Journal of Sociology, 110, 37. Nowak, A., Szamrej, J., & Latané, B. (1990). From private attitude to public opinion: A dynamic theory of social impact. Psychological Review, 97, 362–376. Nowak, M. (2006). Five rules for the evolution of cooperation. Science, 314, 1560–1563. Pitt, M. A., Myung, I. J., & Zhang, S. (2002). Toward a method of selecting among computational models of cognition. Psychological Review, 109, 472–491. Prentice, D. A., & Miller, D. T. (1993). Pluralistic ignorance and alcohol use on campus: Some consequences of misperceiving the social norm. Journal of Personality and Social Psychology, 64, 243–256. Press, W. H., & Dyson, F. J. (2012). Iterated prisoner’s dilemma contains strategies that dominate any evolutionary opponent. Proceedings of the National Academy of Sciences, 109, 10409–10413. Railsback, S. F., & Grimm, V. (2012). Agent-based and individual-based modeling: A practical introduction. Princeton, NJ: Princeton University Press. Rendell, L., Boyd, R., Cownden, D., Enquist, M., Eriksson, K., Feldman, M. W., et al. (2010). Why copy others? Insights from the Social Learning Strategies Tournament. Science, 328, 208–213.

Agent-Based Modeling 407 Schelling, T. C. (1971). Dynamic models of segregation. Epstein, 1999; Epstein, 1999; Journal of Mathematical Sociology, 1(2), 143–186. Selfridge, O. G. (1988). Pandemonium: A paradigm for learning. In J. A. Anderson & E. Rosenfeld (Eds.), Neurocomputing: Foundations of research (pp. 117–122). Cambridge, MA: MIT Press. (Original work published 1959) Simao, J., & Todd, P. (2002). Modeling mate choice in monogamous mating systems with courtship. Adaptive Behavior, 10, 113–136. Smith, E. R. (1998). Mental representation and memory. In D. Gilbert, S. Fiske, & G. Lindzey (Eds.), Handbook of social psychology (4th ed., Vol. 1, pp. 391–445). New York: McGraw-Hill. Smith, E. R., & Collins, E. C. (2009). Contextualizing person perception: Distributed social cognition. Psychological Review, 116, 343–364. Smith, E. R., & Conrey, F. R. (2007). Agent-based modeling: A new approach for theorybuilding in social psychology. Personality and Social Psychology Review, 11, 87–104. Smith, E. R., & Semin, G. R. (2004). Socially situated cognition: Cognition in its social context. Advances in Experimental Social Psychology, 36, 53–117. Sutcliffe, A., Wang, D., & Dunbar, R. (2012). Social relationships and the emergence of social networks. Journal of Artificial Societies and Social Simulation, 15(4), 3. Weaver, W. (1948). Science and complexity. American Scientist, 36, 536–544. Wilensky, U. (1999). NetLogo. http://ccl.northwestern.edu/netlogo. Center for Connected Learning and Computer-Based Modeling, Northwestern University, Evanston, IL.



Author Index

Aaker, J. L., 118, 331 Aberle, D. F., 268 Abrams, D. B., 168 Abrams, R. L., 372 Adorno, T. W., 268 Agnew, C. R., 312 Aguirre, G. K., 187, 188 Ainsworth, M. S., 267 Ajzen, I., 43, 52, 75, 229, 230, 238, 247, 249, 251 Allport, G. W., 166, 233, 306 Alsabban, S., 218 Ambady, N., 364 Ames, D. R., 121 Anderson, J. R., 199, 372 Anderson, N. H., 75, 76, 356 Andersson, J., 292 Ariely, D., 320 Arnold, M. B., 88, 90 Aron, A. R., 140 Aronson, E., 9 Aronson, J. A., 113, 192 Arriaga, X. B., 305, 307, 312, 316, 320, 323 Artistico, D., 168 Asch, S. E., 42, 266, 279 Ataca, B., 333 Atran, S., 54 Aue, T., 98 Austin, J. L., 42 Averill, J. R., 87, 90, 96 Aviezer, H., 98, 295

Axelrod, R., 350, 396 Axtell, R., 393 Aydede, M., 284 Ayduk, O., 157, 169

B Babcock, L., 111 Bacon, F., 126 Baddeley, A. D., 135 Bain, P., 54 Balcetis, E., 109 Balota, D. A., 50 Bandettini, P. A., 188 Bandura, A., 109, 118, 167, 352 Banks, W. C., 36 Barash, D. P., 267 Barch, D. M., 140 Bargh, J. A., 50, 52, 67, 80, 109, 121, 122, 134, 147, 291, 293, 294 Barkow, J. H., 229, 231, 237 Barndollar, K., 121 Barnes-Holmes, D., 32, 33, 37, 66 Baron, J., 380 Baron, R. M., 70, 72, 94 Barrett, J. L., 111 Barrett, L. F., 89, 94, 96, 295 Barsalou, L. W., 50, 94, 287, 289, 290, 295, 365 Bartel, P., 196 Bartkowski, W., 347 Bartlett, F., 43, 284

409

410 Author Index Bastian, B., 54, 206 Baumeister, R. F., 52, 73, 111, 267 Baumgardner, M. H., 4 Bauserman, R., 19 Beasley, A., 6 Beasley, A, 390 Beauregard, K. S., 120 Bechtel, W., 26, 38 Beck, A. T., 186 Becker, A. P., 17 Becker, G., 111, 120 Beckers, T., 27, 30 Beer, J. S., 17, 67, 77, 78, 94, 183, 184, 186, 187, 188, 189, 190, 192, 195, 197, 198, 199 Beevers, C. G., 186 Bekkering, H., 289 Bellugi, U., 186 Bem, D. J., 165 Benet-Martínez, V., 331, 336 Bentin, S., 190, 295 Bentley, M., 86 Berntson, G. G., 92, 289 Berscheid, E., 308, 324 Bettman, J. R., 69 Bhanji, J. P., 195 Bishara, A. J., 385 Blair, E., 173 Blairy, S., 294 Blewitt, M. E., 220 Blickle, G., 267 Bliss-Moreau, E., 94 Blizinsky, K. D., 332 Blom, S., 289 Bloom, P., 17 Bodenhausen, G. V., 3, 13, 17, 65, 70, 71, 72, 74, 76, 78, 85, 112, 114, 133, 134, 136, 138, 142, 143, 145, 149, 192, 196, 232, 235, 237, 252, 270, 319, 321, 373, 387 Bodner, R., 114 Boettger, R., 273 Bohlin, G., 165 Bohner, G., 33 Boomsma, D. I., 211 Boone, A. L., 111 Bördgen, S., 98 Borghi, A. M., 289 Boring, E. G., 267, 269 Bornstein, G., 402 Boroditsky, L., 289 Borsboom, D., 165, 166, 174 Botvinick, M. M., 140 Bouchard, T. J., 210, 211 Boudreau, L. A., 94

Bourgois, P., 295 Bowdle, B. F., 331 Boyd, R., 395 Braver, T. S., 140 Brehm, J. W., 16, 311 Brendl, C. M., 294 Brennan, G., 117 Bresler, C., 100 Breus, M., 52 Brewer, M. B., 14, 52, 138, 336 Brigham, J. C., 291 Brighton, H., 366 Brinol, P., 137 Brooks, R. A., 289 Brown, B., 192 Brown, D. F., 118 Brown, G. D. A., 366 Brown, R., 88 Bruner, J., 43, 57, 296 Bruner, J. S., 328 Brunstein, J. C., 121 Bryson, S. E., 186 Buccino, G., 288 Bugental, D. B., 309 Burnstein, E., 352 Burrows, L., 80 Burton, S. S., 173 Buss, D. M., 225, 230, 231, 239, 308 Buxton, R. B., 187 Byrka, K., 380

C Cabeza, R., 199 Cacioppo, J. T., 14, 92, 98, 139, 141, 144, 190, 288, 289, 293 Caggiano, V., 288 Caldwell, T. L., 78, 157, 158, 170, 171, 272, 314 Camerer, C., 111 Campbell, C. G., 166 Campbell, D. T., 267, 380 Campbell, W. K., 120, 316 Campos, J. J., 89, 95 Cannistraci, C. J., 195 Cannon, W. B., 92 Capaldi, E. J., 7, 162 Caporael, L. R., 79, 292 Caprara, G. V., 167 Caramazza, A., 289, 290 Carlborg, O., 208 Carlin, J. B., 377 Carlsmith, K. M., 275 Carlson, N. R., 38

Author Index 411 Carlston, D. E., 67, 77 Carrera, P., 98 Carroll, J. M., 295 Carter, C. S., 140 Cartwright, N., 250 Carver, C. S., 111 Casasanto, D., 289 Cattell, R. B., 229, 232 Cavalli-Sforza, L., 332 Cederman, L., 399, 400 Centola, D., 390, 393, 396, 397 Cervone, D., 78, 157, 158, 159, 167, 168, 170, 272, 314 Cesario, J., 51, 67, 78, 80 Chaiken, S., 14, 77, 134, 138, 141, 144, 149, 192 Chambers, J. R., 112 Chambliss, W., 117 Chartrand, T. L., 122, 293 Chen, E. E., 331 Chen, M., 80, 293, 294 Chen, M. K., 17 Chen, S., 77, 149 Cheng, C. Y., 336 Cheng, P. W., 27, 29 Chiao, J. Y., 332 Chiesa, M., 38 Chiu, C.-Y., 46, 54, 331 Choi, I., 48, 334, 337 Christensen, K., 211, 215 Chun, W. Y., 13 Cialdini, R. B., 114 Ciarallo, F., 398 Clancey, W. J., 296 Clark, A., 289, 292 Clark, M. S., 309, 313, 321 Clary, E. G., 115 Claverie, B., 190 Cleeremans, A., 80 Cliff, D., 404 Cloninger, C. R., 165, 169 Clore, G. L., 90, 96, 103 Cohen, A. B., 330 Cohen, A. K., 268 Cohen, D., 330, 338 Cohen, G. L., 113, 116, 124 Cohen, J. D., 140, 192 Cole, E. R., 338, 339 Coles, M. G. H., 189 Collins, A., 96 Collins, E. C., 395 Collins, W., 308 Coltheart, M., 190, 200 Conrey, F. R., 349, 385, 391, 393, 399

Conway, L. G., 78 Coon, H. M., 330 Cooper, W. S., 352 Corr, P. J., 165 Correll, J., 372, 386 Cosmides, L., 92, 214, 229, 231, 237, 269 Costa, P. T., 163, 164, 165, 166 Costa, P. T., Jr., 166 Cottrell, C. A., 234 Coventry, W. L., 212 Cownden, D., 395 Craighero, L., 288 Cramer, A. O. J., 166 Crandall, C. S., 234 Craver, C. F., 38 Crawford, L. E., 289 Creighton, L. A., 14, 132, 134, 138 Crenshaw, K., 338 Crick, F. H., 280 Crosignani, P. G., 211 Cross, S. E., 333 Cuddy, A. J. C., 54, 80 Cummins, R., 152 Cunningham, W. A., 195

D Dahl, A., 89 Dale, A. M., 188 Daly, M., 236 Damasio, A., 90 D’Andrade, R., 330 Daniel, L. G., 172 Danker, J. F., 199 Danziger, K., 174 D’Argembeau, A., 198 Darley, J. M., 36, 192, 275 Dar-Nimrod, I., 54, 206 Darwin, C. R., 65, 66, 231, 235, 280 Davis, A. K., 268 Davis, K. E., 74 Daw, N. D., 197 Dayan, P., 197 De Bruin, E. N., 310 de Gelder, B., 295 de Geus, E. J., 211 de Groot, J. H. B., 288 De Houwer, J., 24, 27, 30, 32, 33, 37, 50, 66, 67, 68, 70, 71, 72, 93, 99, 120, 133, 134, 139, 145, 147, 150, 151, 152, 317, 372 De Leersnyder, J., 337 De Rivera, J., 95 de Vries, N. K., 261 Dean, K. K., 124

412 Author Index Debbané, M., 197 Deco, G., 197 DeCoster, J., 14, 147, 149 DeFries, J. C., 210 Delgado, A., 334 DeMarree, K. G., 67 DeMartino, B., 192 Dennett, D. C., 231 Denrell, J., 353, 354, 355 Desmond, J. E., 187 D’Esposito, M., 187, 188, 192 Deutsch, M., 347 Deutsch, R., 13, 14, 34, 69, 71, 78, 79, 120, 132, 134, 136, 138, 140, 144, 145, 147, 149, 310, 316 Devine, P. G., 14, 136, 285 Dewey, J., 284 Dhar, R., 69 Diamond, J., 234 Dijksterhuis, A., 69 Dijkstra, K., 289 Dilthey, W., 56 Dimberg, U., 294 DiRago, A. C., 218 Disner, S. G., 186 Dixon, R., 292 Dolan, R. J., 192, 197 Dougherty, M., 347 Dove, G., 289 Dovidio, J. F., 291 Downs, D. L., 122 Doyen, S., 80 Doyle, T. F., 186 Draine, S. C., 372 Duckworth, A. L., 111 Duffy, E., 86 Dugan, P. M., 363 Duhem, P., 7, 8 Dunbar, R., 396 Duncan, L. E., 218 Dunn, E. W., 15 Dunn, J. C., 141, 142 Dunning, D., 78, 108, 109, 112, 113, 116, 117, 119, 120, 121 Durante, K. M., 132 Dweck, C. S., 54, 109

E Eagly, A. H., 14, 78, 134, 138, 148 Eastwick, P. W., 320 Eder, A. B., 294 Edmonds, B., 400, 403, 404 Edwards, D., 86, 97

Edwards, K., 99 Eelen, P., 99 Egan, L. C., 17 Eiser, J., 352, 354, 355 Ekman, P., 87, 92, 95, 98, 294 Elliot, A. J., 109, 118 Ellis, B. J., 10, 230, 231, 232, 235, 237, 238 Ellsworth, P. C., 99, 100, 191 Elman, J. L., 347, 364 Elmehed, K., 294 Elson, S. B., 274 Eng, S. J., 235 Enquist, M., 395 Eom, K., 78, 269, 328 Epley, N., 322 Epstein, J. M., 393, 397, 398, 399, 400, 402 Epstein, S., 138, 149, 158, 167 Erb, H.-P., 13, 144 Erber, R., 292 Erdelyi, M. H., 15 Eriksson, K., 395 Etcoff, N. J., 385 Evans, J. S. B. T., 132, 134, 138, 140, 149

F Fabiani, M., 188 Fairchild, A. J., 139 Fanselow, M. S., 35 Farias, A. R., 284, 289, 290 Farrell, D., 317 Fazio, R. H., 14, 34, 71, 74, 137, 138, 316, 352, 354, 355 Feather, N. T., 43, 53 Fehr, E., 114 Feldman, M., 332 Ferguson, M. J., 109, 121 Fernández-Dols, J. M., 98 Ferrari, P. F., 288 Ferreira, M. B., 34, 291, 292 Fessler, D. M. T., 233, 235 Festinger, L., 16 Fetchenhauer, D., 117, 119 Feyerabend, P., 8 Fieberg, E., 54 Fiedler, K., 6, 79, 347, 356, 357, 358, 359, 360, 361, 363, 364, 374, 375, 376 Fincher, C. L., 234, 332 Fincher, K. M., 124, 266, 275, 277, 278, 279, 321 Finkel, E. J., 316, 320, 322 Finkenauer, C., 321 Fischer, A. H., 95, 287, 293, 295 Fischer, G. W., 359

Author Index 413 Fischer, K. W., 192 Fish, D., 88 Fishbach, A., 109, 121 Fishbein, M., 43, 52, 75, 229, 230, 238, 247, 249, 250, 251, 254, 262 Fiske, A. P., 268, 309, 328 Fiske, S. T., 14, 54, 58, 80, 291, 385 Flagan, T., 197, 198, 199 Flaxman, S. M., 234 Fleeson, W., 158, 313 Fodor, J. A., 285 Fogassi, L., 288 Fogel, A., 95 Fombonne, E., 186 Fong, G. T., 72, 139, 152 Forgas, J. P., 103, 173 Foroni, F., 289, 296 Förster, J., 15, 121, 293, 295 Fournier, M. A., 158 Fowler, J. H., 276 Fraley, R. C., 169 French, R., 361 Frenkel-Brunswik, E., 268 Freytag, P., 356, 358, 359, 361, 374, 375 Fridlund, A. J., 95, 98, 101, 102 Friedman, R. S., 15, 121 Friese, M., 132, 138, 142 Friesen, W. V., 87, 98, 294 Frijda, N. H., 87, 92, 98, 101 Friston, K. J., 187 Fritz, M. S., 139 Fujimura, T., 294 Funder, D. C., 73, 166, 314 Fung, H. H., 333, 334

G Gabriel, S., 336 Gaertner, L., 320 Gaertner, S. L., 291 Galton, F., 351 Garcia-Marques, L., 34, 291, 292 Gardner, H., 25, 26, 33 Gardner, W. L., 52, 124, 291, 331, 336 Garrido, M. V., 283, 284, 289, 290, 291, 292, 320 Gawronski, B., 3, 13, 14, 17, 33, 37, 65, 66, 70, 71, 72, 74, 76, 78, 85, 112, 114, 132, 133, 134, 136, 137, 138, 140, 142, 143, 145, 146, 149, 192, 196, 232, 235, 237, 252, 270, 319, 321, 373, 385, 387 Gazzaniga, M. S., 183, 186, 187 Gee, T. L., 157 Geertz, C., 42, 43

Gelfand, M., 43, 56, 58 Gelles, R. J., 236 Gelman, A., 377 Gelman, S., 206 Gendron, M., 295 Gercek-Swing, B., 333 Gergen, K. J., 57, 279 Gervais, M. C., 210 Gettys, C., 347 Gibson, G., 215 Gibson, J. J., 289 Giere, R. N., 76, 161, 162 Gifford, R. K., 360, 361, 374 Gigerenzer, G., 18, 140, 142, 145, 151, 226, 228, 239, 247, 260, 366 Gilbert, D. T., 14, 17, 29, 74, 134, 136, 138, 145, 149 Gilovich, T., 276 Glance, N., 404 Glenberg, A. M., 289 Glick, P., 54, 80 Glover, G. H., 187 Glynn, P., 188 Gobel, M. S., 339 Gobster, P. H., 334 Goethals, G. R., 359 Goffman, E., 278 Goldman, R., 141 Goldstein, D. G., 247, 260 Gollwitzer, P. M., 52, 111, 121 Gonzalez, R., 310, 331 Goode, M. R., 49 Goode, W., 236 Gorlin, M., 69 Gotlib, I. H., 99 Gottesman, I. I., 210 Gould, O., 292 Grabenhorst, F., 197 Graham, J., 186, 332 Gramm, K., 361 Gratton, G., 188 Gray, J., 169 Green, M. C., 274 Greenberg, J., 52, 122, 123, 268 Greene, J. D., 117, 192 Greening, S. G., 186 Greenwald, A. G., 4, 11, 34, 139, 372 Greve, W., 7, 52, 53, 147, 251, 252, 254, 255 Griffin, D., 276 Griffiths, P. E., 86 Grimm, V., 391, 393, 398, 404 Groom, C., 385 Gross, J. J., 87, 89, 191 Grossmann, I., 339

414 Author Index Grünbaum, A., 15 Grunedal, S., 294 Guglielmo, S., 54, 55 Guillem, F., 190

H Hagekull, B., 165 Haidt, J., 135, 138 Haigh, E. A., 186 Hales, D., 404 Halgren, E., 190 Hamaker, E. L., 157 Hamann, D., 99 Hamann, S., 90 Hamilton, D. L., 67, 77, 285, 290, 291, 292, 360, 361, 363, 374, 384 Hamilton, T. E., 347, 394, 398, 403 Hamilton, W. D., 396 Hamlin, A., 117 Hammdon, J. L., 374 Haney, C., 36 Hanson, N. R., 160, 164, 166 Hardistry, D. J., 55 Harman, G., 11 Harmon-Jones, E., 187, 188, 189, 195 Harré, R., 162, 166 Harris, V. A., 74 Hartig, T., 380 Hartmann, D., 152 Haselton, M. G., 239 Haslam, N. O., 41, 46, 54, 58, 206 Hassin, R., 109, 295 Hastie, R., 347, 348, 396 Haubner, T., 211 Hayes, A. F., 70, 72 Heath, B., 398 Heath, C., 112 Heerey, E. H., 186 Heider, F., 111, 230, 237, 238 Heine, S. J., 54, 122, 123, 206, 218, 330 Heinrich, J., 233 Heise, D. R., 66 Hemmeter, U., 361 Hempel, C. G., 8, 68, 140, 161 Henrich, J., 218, 276 Henson, R. N. A., 188, 197 Herrera, P., 295 Herrnstein, R. J., 352 Hertwig, R., 355, 364 Hess, P., 356 Hess, U., 287, 293, 294, 295, 296 Hewstone, M., 361, 374 Hicks, B., 218

Higgins, E. T., 3, 6, 51, 53, 67, 78, 118, 168, 224, 292 Hill, C. T., 75 Hill, R., 398 Hill, S. E., 132 Hill, W. G., 208 Hills, T. T., 355 Hintzman, D. L., 358, 373 Hirschfeld, L. A., 54 Hitch, G. J., 135 Hodges, B. H., 75 Hoeger, R., 132 Hoffmann, A., 219 Hofmann, C., 361 Hofmann, W., 132, 138, 142 Hogan, R., 267 Hollingshead, A. B., 292 Hollis, M., 246 Holmes, J. G., 306, 312, 314, 316, 318, 321, 322 Holtbend, T., 98 Holtgraves, T. M., 42 Holyoak K. H., 27 Homans, G. C., 309 Hommel, B., 294 Hong, Y.-Y., 46, 54, 331, 336 Horowitz, A. L., 188 Horstmann, G., 92 Hruschka, D. J., 233 Hu, L.-T., 359 Huberman, B., 404 Hubert, M., 351 Huettal, S. A., 187, 188 Hugenberg, K., 385 Hughes, B. L., 195, 199 Hull, C. L., 120 Humphreys, G. W., 197 Hutchins, E., 292, 395 Hutzler, F., 198 Huxley, T. H., 236

I Iacoboni, M., 288 Iacono, W. G., 218 Ickes, W., 310, 314 Ijzerman, H., 291 Imada, T., 331, 332 Insko, C. A., 317 Iriki, A., 288 Ishii, K., 332 Ito, K., 331 Iuzzini, J., 320 Ivry, R. B., 183

Author Index 415 J Jablonka, E., 207 Jacoby, J., 72 James, W., 87, 90, 100, 224, 225, 226, 238, 284, 289 Jencius, S., 168 Ji, L. J., 336 John, O. P., 192 Johnson, A. H., 111 Johnson, B., 148 Johnson, B. T., 291 Johnson, C., 360 Johnson, E. J., 55 Johnson, T., 276 Johnson, W., 19, 78, 205, 208, 210, 211, 213, 218, 219 Johnson-Laird, P. N., 6 Joireman, J. A., 310 Jonas, E., 268 Jones, E. E., 74, 75 Joormann, J., 99 Josephs, O., 187 Jost, J. T., 268 Judd, C. M., 143, 291, 372

K Kagan, J., 165, 167 Kahneman, D., 18, 69, 192, 276 Kaiser, F. G., 380 Kalick, S. M., 347, 394, 398, 403 Kameda, T., 396 Kanfer, R., 118 Kanwisher, N., 183, 190, 196 Kaplan, R., 352 Kappas, A., 88, 89 Karasawa, M., 48, 330 Kareev, Y., 364 Karnath, H. O., 184 Kaschak, M. P., 289 Kashima, E., 50 Kashima, Y., 41, 42, 43, 44, 46, 47, 48, 49, 50, 54, 56, 57, 58, 59, 66, 80, 311, 312, 322 Kashy, D. A., 125, 315 Kassel, J., 168 Katz, D., 125 Kay, A. C., 51, 268 Kellen, D., 386 Keller, M. C., 212, 218 Kelley, H. H., 26, 246, 250, 254, 256, 306, 307, 308, 309, 312, 315, 318, 320, 322, 323, 324 Keltner, D., 99, 186, 192, 330

Kemmelmeier, M., 330, 359 Kennedy, J., 395 Kenny, D. A., 70, 72, 310, 315 Kenrick, D. T., 225, 231, 308 Kerekes, A. R. Z., 50 Keren, G., 133, 135, 140, 142, 147, 150, 151 Ketelaar, T., 10, 20, 78, 126, 224, 230, 231, 232, 235, 237, 238, 260, 269, 308 Kim, H. S., 78, 269, 328, 330, 331, 337, 338, 339 Kirk, A. K., 118 Kirkpatrick, L. A., 233 Kirsh, D., 292 Kirsner, K., 141, 142 Kitayama, S., 47, 48, 328, 330, 332, 333 Klauer, K. C., 6, 71, 79, 361, 371, 381, 382, 385, 386, 387 Klein, O., 80, 218 Klein D. J., 92 Kluger A. N., 9 Knee, R. C., 315, 319 Knight, R. T., 186, 189, 192 Knobe, J., 54, 55 Knowles, M. L., 124, 291 Knutson, B. K., 333, 334 Kober, H., 94 Koestner, R., 121 Kohl, D., 54 Kolb, B., 184, 186, 187 Koole, S. L., 159, 167 Korenberg, J. R., 186 Koseki, H., 219 Krambeck, H. J., 114 Kraus, M. W., 330 Kreibig, S. D., 87 Krieglmeyer, R., 120 Krist, H., 54 Kristel, O. V., 274 Krueger, R. F., 211, 218 Kruger, J. M., 112 Kruglanski, A. W., 3, 6, 13, 34, 59, 140, 144, 151, 225, 347, 380 Krull, D. S., 145 Krumhuber, E., 88 Kuhl, J., 159, 167 Kuhn, T. S., 12, 13, 287, 293 Kumaran, D., 192 Kumashiro, M., 322 Kunda, Z., 10, 109, 276, 291, 347 Kutzner, F. L. W., 6, 79, 347, 358, 362, 363 Kwan, L., 331

416 Author Index L LaBar, K. S., 199 Labiouse, C., 361 Lacey, A. R., 112, 114 Laird, J. D., 100 Lakatos, I., 9, 10, 11, 12, 75, 76, 136, 143, 232, 235, 237, 238, 239, 253 Lakens, D., 289 Lamb, M. J., 207 Lambie, J. A., 87, 101 Lamiell, J. T., 163 Lamm, H., 359 Lamoreaux, M., 331 Larsen, G., 169 Latané, B., 36, 347, 348, 393 Latham, G. P., 109 Laurie, C. C., 208 Lavelle, L. A., 98 Lavender, T., 294 Lazarus, R. S., 85, 92, 93, 96, 100, 101, 168 Le Mens, G., 354 Le Rouzic, A., 208 Leach, C. W., 95 Leary, M. R., 122, 267 LeBel, E. P., 12, 71 Ledermann, T., 315 Lee, A. Y., 118, 331, 336 Lee, F., 336 Lee, S. W. S., 331, 335, 336 Lee, T., 188 Lee-Chai, A. Y., 121 Lefcourt, H. M., 170 Leippe, M. R., 4 Lemley, R. E., 273 Leonardelli, G. J., 291 Lepper, M. R., 117 Lerner, J. S., 273, 274 Lerner, M. J., 268, 276 Lerner, S. C., 268 Leu, J., 336 Leuenberger, A., 113 Leung, A. K. Y., 338 Levenson, R. W., 87, 334 Leventhal, H., 85, 89, 93, 100, 101 Levine, J., 292 Levinson, D. J., 268 Levy, M. J., 268 Lewin, K., 3, 43, 160, 224, 225, 226, 232, 236, 306, 307, 308, 314, 324 Lewis, M., 192 Lewis, M. D., 94, 95, 169 Liberman, A., 14, 134, 138 Liberman, N., 15, 121

Lieberman, M. D., 17, 136, 140, 149 Lilienfeld, S. O., 20 Lin, Y. H. W., 317 Lindberg, M. J., 43 Lindquist, K. A., 94, 99, 295 Lindsey, S., 74, 141 Linnaeus, C., 65 Linville, P. W., 359, 360 Lipkus, I., 318 Little, B. R., 157 Lloyd, G. E. R., 123 Locke, E. A., 109 Loersch, C., 50, 51, 67 Loewenstein, G., 111, 121 Lombardo, M. V., 195 Lou, H. C., 197 Loughnan, S., 54 Louie, J. Y., 331, 334 Lovibond, P. F., 145 Lucas, G. M., 124 Luce, M. F., 69 Luce, R. D., 307 Lumsden, C. J., 332 Luszczynska, A., 168 Lutz, C. A., 86

M Mace, W., 100 Machado, A., 6 Machery, E., 290 Mackie, D. M., 291 MacKinnon, D. P., 139 Macrae, C. N., 67, 73, 291 Macy, M., 390, 396, 397 Maher, B., 209 Mahon, B. Z., 289, 290 Main, A., 89 Malle, B. F., 54, 55 Malone, P. S., 29, 145 Mandler, G., 86 Mangun, G. R., 183 Mannetti, L., 13 Manstead, A. S. R., 78, 84, 88, 95, 101, 191, 267 Marcel, A. J., 87, 101 March, J. G., 353 Maringer, M., 296 Markman, A. B., 294 Markman, K. D., 43 Markowitsch, H. J., 196 Markus, H. R., 47, 168, 328, 330, 331, 333, 337 Marquardt, N., 132

Author Index 417 Marr, D., 24, 25, 26, 27, 28, 29, 38, 50, 67, 71, 73, 77, 78, 79, 93, 94, 133, 134, 372 Marsh, H. W., 157 Martin, L. L., 289, 296 Martin, R. A., 169, 170 Martz, J. M., 317 Maslow, A. H., 124 Mason, W. A., 393, 401, 405 Masuda, T., 331 Mather, M., 190 Matsumoto, D., 48, 98 Matsumoto, H., 333 Matz, D., 98 Mauss, I. B., 87 Mayer, N. D., 78, 157, 173, 272, 314 Mayo, R., 352 McAdams, D. P., 159 McCarter, L., 87 McCarthy, G., 188, 190 McCartney, K., 214 McClamrock, R., 26 McClearn, G. E., 210 McClelland, D. C., 121 McClelland, J. L., 162, 169, 361, 367 McCloskey, M., 54 McClure, J., 27, 29 McConnell, A. R., 74 McCourt, M. E., 93 McCoy, K., 334 McCrae, R. R., 163, 164, 165, 166 McDougall, W., 269 McGrath, J. E., 7 McGregor, H. A., 109, 118 McGue, M., 211, 215, 218 McGuffin, P., 210, 218 McGuire, W. J., 273 Mead, M., 284 Mead, N. L., 49 Meadows, M., 404 Medin, D. L., 54 Meehl, P., 229, 231, 232 Meier, B. P., 289 Meiser, T., 34, 361, 374, 381, 382 Meissner, F., 385 Mellenbergh, G. J., 165, 174 Mellor, A. K., 233, 235, 239 Mendoza-Denton, R., 157, 169, 313 Menon, T., 46 Mermillod, M., 296 Merton, R. K., 274, 332 Mesquita, B., 295, 330, 337 Metcalfe, J., 149 Miao, F., 330 Miles, L. K., 67, 73

Milgram, S., 36, 37 Milinski, M., 114 Miller, D. T., 276, 390 Miller, J. G., 336 Miller, M. B., 172 Mills, J., 309 Mischel, W., 149, 158, 164, 165, 167, 174, 313 Mitchell, C. J., 145, 146 Mitchell, D. V., 186 Mitchell, J. P., 190, 197, 291 Miyamoto, Y., 333 Mochon, D., 320 Molden, D. C., 124 Molenaar, P. C. M., 166 Moonen, C. T. W., 188 Moore, G. E., 206 Moors, A., 24, 32, 37, 50, 67, 90, 92, 93, 99, 132, 133, 134, 139, 146, 147, 151, 152, 317, 372 Moran, J. M., 190, 197 Morgan, M. S., 162 Morling, B., 331, 333 Morris, M. W., 46, 331, 336 Morrison, M., 162 Moscovici, S., 331, 366 Moskowitz, D. S., 158 Moss, S., 400 Mullen, B., 359, 360 Muller, D., 143 Mumford, J. A., 187 Murata, A., 330 Murgatroyd, C., 219 Murray, D. R., 332 Murray, H. A., 269 Murray, R. J., 197, 198 Murray, S. L., 312, 316, 318 Myers, D. G., 359 Myung, I. J., 401

N Na, J., 330, 334 Nasby, W., 173 Navarrete, C. D., 233, 235 Nayak, A., 188 Neal, D. T., 125, 126 Nelson, L. D., 113 Nesse, R. M., 239 Nesselroad, J. R., 229 Nettle, D., 239 Neuberg, S. L., 14, 231, 233, 234, 237, 291 Neumann, R., 99, 293 Newell, A., 232

418 Author Index Newell, B. R., 15, 17 Newman, P. L., 96 Newport, E. L., 364 Niaura, R., 168 Nichols, T. E., 187, 199 Nickel, S., 361 Niedenthal, P. M., 94, 296 Nisbett, R. E., 43, 48, 117, 273, 306, 311, 328, 331, 336, 337 Nobre, A. C., 190 Noftle, E. E., 313 Norasakkunkit, V., 333 Norenzayan, A., 48, 218, 337 Northoff, G., 197, 198 Nosek, B. A., 34 Novick, L. R., 29 Nowak, A., 347, 348, 349, 350, 366, 393 Nowak, M., 396 Nozick, R., 160, 164, 166 Nunley, E. P., 90 Nystrom, L. E., 192

O Oatley, K., 86, 95 Ochsner, K. N., 17, 136, 190, 191, 197, 198 O’Doherty, N. D., 197 Ogawa, S., 188 Ogden, E., 347 Oishi, S., 332 Onwuegbuzie, A. J., 172 Oppenheim, P., 140, 161 Oriña, M., 320 Orom, H., 168 Ortony, A., 54, 90, 96, 169 Osgood, C. E., 66 Osler, M., 215 Ostrom, T. M., 285 Osuch, E. A., 186 Otten, W., 310 Ouellette, J., 125 Ouwerkerk, J. W., 321 Oyserman, D., 330, 331, 335, 336

P Pacini, R., 149 Paldi, A., 219 Palma, T. A., 284, 289, 292 Pareto, V., 111, 120 Park, B., 291, 372 Park, J. H., 48, 233, 234, 239 Parkinson, B., 78, 84, 86, 92, 94, 95, 96, 97, 101, 191

Paul, S. T., 50 Paunonen, S. V., 165 Payne, B. K., 16, 50, 51, 67, 137, 385 Payne, J. W., 69 Pearl, J., 165 Pendry, L. F., 291 Peng, K., 48, 337 Penke, L., 19, 78, 205, 208, 210, 211, 219 Penrod, S., 347 Pepitone, A., 269, 279 Pereira-Pasarin, L. P., 292 Perry, R. J., 186 Pervin, L. A., 159 Peters, K. R., 12, 71 Petty, R. E., 14, 67, 137, 139, 141, 144, 276, 289, 293 Phaf, R. H., 293 Phelps, E. A., 195 Piaget, J., 289 Pichon, C., 80 Pickering, A. D., 165 Pickett, C. L., 291 Pierro, A., 13 Piff, P. K., 330 Pinder, C. C., 109 Pinker, S., 19, 231 Pinkus, R. T., 318 Pitt, M. A., 401 Plaks, J. E., 51, 67 Plaut, D. C., 146 Pleskac, T. J., 364 Plessner, H., 356 Plomin, R., 164, 210, 211, 212 Poldrack, R. A., 77, 140, 187, 188, 191, 199 Polich, J., 189 Popper, K. R., 4, 5, 6, 7, 8, 9, 10, 11, 14, 15, 20, 142, 161, 232, 252, 253 Posthuma, D., 211 Pratkanis, A. R., 4 Preacher, K. J., 70, 72 Prelec, D., 114 Prentice, D. A., 390 Prescott, T. J., 352 Priester, J. R., 289, 292, 293 Prinz, W., 52, 289 Proctor, R. W., 7, 162 Profet, M., 234 Proulx, T., 43, 122 Przybeck, T. R., 165 Puhlik-Doris, P., 169 Pulvermüller, F., 296 Purcell, S., 212, 213 Pyszczynski, T., 52

Author Index 419 Q Qin, P., 197, 198 Quine, W. V. O., 7, 8, 9, 10, 11, 15, 20, 53, 74 Quinn, J. M., 125

R Rabin, M., 114 Radloff, C. E., 359 Raiffa, H., 307 Railsback, S. F., 391, 393, 398, 404 Rajaran, S., 292 Ramaswamy, J., 332 Rashid, M., 331 Raymond, P., 292 Regier, T., 142, 145, 151 Reis, H. T., 305, 306, 308, 312, 313, 314, 321, 322 Reisenzein, R., 92, 98 Rempel, J. K., 318, 321 Rendell, L., 395 Rensick, L., 292 Rescorla, R. A., 278, 361 Rice, S., 251 Righart, R., 295 Righetti, F., 321 Rilling, J. K., 192 Rind, B., 19, 20 Risch, N., 218 Risen, J. L., 17 Ritchey, M., 199 Rizzolatti, G., 288 Robbins, P, 284 Robbins, T. W., 140 Roberts, B. W., 157 Robinson, M. D., 93, 289 Robinson, P. H., 275 Roefs, A., 132 Rogers, R. W., 109 Rogers S. J., 186 Rolls, E. T., 197 Rönnberg, J., 292 Rorden, C., 184 Roseman, I. J., 92 Rosenstock, I. M., 109 Rosenthal, R., 364 Ross, E. A., 306 Ross, L., 43, 306, 311 Rothermund, K., 294, 385 Rotshein, P., 197 Rotteveel, M., 293 Rougier, A., 190 Rowe, D. C., 164 Rowley, G. L., 172

Roy, M., 197 Rozin, P., 279, 280 Rubin, B. L., 211, 377 Rubin, L. H., 158, 170 Ruderman, A. J., 385 Rui, J., 197 Ruiz-Belda, M. A., 98 Rumelhart, D. E., 162, 169, 361, 367 Rusbult, C. E., 305, 307, 309, 310, 312, 313, 316, 317, 318, 321, 322 Russell, J. A., 86, 88, 89, 90, 96, 98, 295 Russer, S., 361 Rydell, R. J., 74 Ryle, G., 42

S Salmon, W. C., 158, 159, 160, 161 Salovey, P, 359 Samper, A., 69 Sánchez, F., 98 Sande, G. N., 359 Sanfey, A. G., 192 Sanford, R. N., 268 Santos, A. S., 290, 291 Santos, L. R., 17 Sarbin, T. R., 89 Sartre, J-P., 89, 101 Sasaki, J. Y., 331, 338 Sassenberg, K., 72 Sato, W., 294 Scabini, D., 186, 192 Scarantino, A., 86 Scarr, S., 214 Schachter, S., 92, 96 Schacter, D. L., 17, 136 Schacter, S., 224 Schaer, M., 197 Schaller, M., 78, 225, 231, 233, 234, 237, 239, 332, 374 Scheier, M. F., 111 Schelling, T. C., 391, 392, 393, 397, 403 Scherer, K. R., 85, 89, 91, 93, 99, 101, 191 Schlenker, B., 57 Schlösser, T., 119 Schmalhausen, I. I., 218 Schmidt, K. M., 114 Schmitz, F., 385 Scholz, U., 168 Schooler, J. W., 52 Schooler, L., 364 Schooler, T. Y., 74, 141 Schopler, J., 318, 320

420 Author Index Schul, Y., 133, 135, 140, 142, 147, 150, 151, 352 Schultheiss, O. C., 121 Schunk, D. H., 118 Schwarz, N., 18, 33, 34, 71, 103, 173, 290, 331 Schwarzer, R., 168 Sears, D. O., 218 Sedikides, C., 120 Segal, N. L., 211 Selfridge, O. G., 402 Sell, A., 214, 217 Semin, G. R., 283, 284, 286, 287, 288, 289, 291, 292, 295, 296, 320, 405 Semmann, D., 114 Seppala, E., 330 Sevincer, A. T., 48 Seymour, B., 192 Seymour, P., 197 Shadel, W. G., 168 Shah, J., 53 Shanks, D. R., 15, 17, 27 Sheeran, P., 254 Sheldon, S., 121 Shelley, M., 233 Sherif, C. W., 233 Sherif, M., 233 Sherman, D. A., 113 Sherman, D. K., 113, 116, 124, 339 Sherman, J. W., 70, 79, 132, 385 Sherman, P. V., 234 Sherman, S. J., 361, 384 Shimamura, A. P, 192 Shoda, Y., 158, 164, 167, 313 Shohamy, D., 197 Shook, N., 352 Shweder, R., 328 Siebert, M., 196 Siebler, F., 146 Siegel, P. B., 208 Silva, F. J., 6 Simao, J., 394, 399 Simmel, M., 111 Simmons, W. K., 290 Simon, L., 52 Simpson, E. A., 288 Simpson, E. H., 374 Simpson, J. A., 225, 231, 318, 321 Sinclair, L., 291 Singer, J. E., 225, 226, 232 Skinner, B. F., 152 Skitka, L. J., 277 Slatkin, M, 219 Slezak, P., 296

Sloman, S. A., 141, 142, 145, 149 Slovic, P, 18 Slovik, L. F., 318 Smedslund, G., 7 Smeesters, D., 51 Smirnov, O., 276 Smith, C. A., 168 Smith, E. R., 6, 14, 147, 149, 284, 285, 287, 291, 295, 296, 349, 358, 361, 363, 385, 390, 391, 393, 395, 399, 402, 405 Smith, R. E., 167 Smith, W. J., 317 Snibbe, A. C., 330, 331 Snyder, M., 115, 272, 290, 314 Solarz, A. K., 293 Solomon, R. C., 89 Solomon, S., 52, 347 Soltani, M., 189 Sommerville, R. B., 192 Song, A. W., 188 Sonnby-Borgström, M., 294 Sorenson, E. R., 98 South, S. C., 218 Spencer, D. D., 190 Spencer, S. J., 72, 139, 152 Spengler, D., 219 Spinath, F. M., 208, 210, 219 Sporns, O., 196 Stacy, A. W., 132 Staddon, J. E. R., 152 Stafford, T., 352 Stahl, C., 382 Stasser, G., 348, 351 Steele, C. M., 113 Stephens, N. M., 330 Stepper, S., 289, 296 Stern, H. S., 377 Storbeck, J., 93 Strack, F., 13, 14, 34, 78, 132, 134, 136, 138, 140, 144, 145, 147, 149, 289, 293, 296 Straus, M. A., 236 Strick, M., 69 Stroebe, W., 380 Studtmann, M., 92 Suci, G. J., 66 Sudman, S., 291 Sul, S., 334 Suls, J., 112 Sun, R., 347 Sunbay, Z., 333 Sutcliffe, A., 396 Sutton, F. X., 268 Suzuki, N., 294

Author Index 421 Švrakić, D. M., 165, 169 Swartz, T. S., 92 Szamrej, J., 347, 393

T Tajfel, H., 233, 374 Takemura, K., 332 Takezawa, M., 396 Tambor, E. S., 122 Tangney, J. P., 111, 192 Tannenbaum, P., 66 Tauber, B., 356 Taylor, S. E., 290, 339, 385 Tazelaar, M. A., 321 Teige-Mocigemba, S., 71, 385 Terdal, S. K., 122 Tesser, A., 116, 124, 359 Tetlock, P. E., 66, 124, 266, 267, 268, 273, 274, 275, 276, 277, 278, 279, 321 Thagard, P., 11, 20, 169, 347 Thaler, R., 111 Thibaut, J. W., 306, 307, 308, 309, 324 Thomas, E. A., 210 Thomas, S., 170 Thompson, E. P., 144, 151 Thompson, R. A., 89 Thorisdottir, H., 268 Thorndike, E. L., 352 Thorngate, W., 306 Thornhill, R., 234, 332 Thunberg, M., 294 Tice, D. M., 111 Tiedens, L. Z., 95 Tikochinsky, J., 9 Timmermans, B., 30 Todd, P., 394, 399 Todorov, A., 98 Tomkins, S. S., 100 Tooby, J., 92, 214, 229, 231, 237 Toulmin, S., 161 Tourangeau, R., 100 Towels-Schwen, T., 138 Townsend, S. S. M., 330 Trafimow, D., 6, 75, 127, 245, 251, 253, 254, 255, 256, 258, 261, 262, 310 Tramacere, A., 288 Traub, R. E., 172 Triandis, H. C., 43, 47 Troetschel, R., 121 Trolier, T. K., 290, 363 Tromovitch, P., 19 Trope, Y., 70, 74, 98, 132, 295 Trut, L. N., 210

Tsai, J. L., 330, 331, 333, 334 Turkheimer, E., 207, 210, 219 Turner, J. C., 233 Tversky, A., 18, 192 Twain, M., 213

U Uchida, Y., 331 Uher, R., 218 Uleman, J. S., 147 Ullian, J. S., 8, 11, 15, 20, 74 Ulrich, R., 381, 384 Unkelbach, C., 361, 374, 375 Uskul, A. K., 48, 333

V Valins, S., 92, 289 van der Maas, H. L. J., 166 van der Pligt, J., 261 Van der Schalk, J., 295 van Elk, M., 289 Van Essen, D. C., 199 van Heerden, J., 165, 174 Van Lange, P. A. M., 3, 305, 307, 309, 310, 313, 314, 316, 321 Van Overwalle, F., 30, 146, 360, 361 Van Rooy, D., 361, 362 van Schie, H. T., 289 Vandorpe, S., 30 Vanhoomissen, T., 360, 361 Varnum, M. E. W., 330 Verette, J., 318 Vickaryous, N. K., 219 Vidmar, N., 276 Viechtbauer, W., 157 Visscher, P. M., 211, 212 Vogel, T., 358 Vohs, K. D., 49, 52, 73, 111, 122 von Soden, W., 161 Voracek, M., 211 Voss, A., 382, 385 Vygotsky, L., 284 Vytal, K., 99

W Waddington, C. H., 205, 206, 218, 219 Wager, T. D., 94, 197, 199 Wagner, A., 361 Waldorp, L. J., 166 Wallach, L., 7, 146, 147, 309 Wallach, M. A., 7, 146, 147, 309 Walle, E. A., 89 Walther, E., 356, 361

422 Author Index Walton, K. E., 157 Wang, D., 396 Wang, H., 331 Watson, J. B., 152 Watson, J. D., 280 Watson, R. I., 267 Weaver, W., 392 Webbink, D., 211 Weber, E. U., 55 Wegener, D. T., 139, 276 Wegener, I., 385 Wegner, D. M., 292 Weinberger, J., 121 Weiner, B., 247, 254, 256 Weir, K., 169 Weitlauf, J., 167 Wells, G. L., 289 Werth, L., 132 Westen, D., 15 Wheeler, S. C., 51, 67 Whishaw, I. Q., 184, 186, 187 White, P. H., 276 Whitelaw, E., 219 Whitney, G. A., 318 Wickham, R. E., 315, 319 Wiers, R. W., 132 Wiest, C., 92 Wiggins, J. S., 163, 168 Wilensky, U., 403 Wilhelm, F. L., 87 Wilkening, F., 54 Willer, R., 390 Williams, L. E., 52, 291 Williamson, P. C., 186 Willingham, B., 98 Wilson, C. D., 290 Wilson, E. O., 332 Wilson, M., 169, 236, 290 Wilson, T. D., 15, 74, 141 Windschitl, P. D., 112 Wing, E. A., 199 Witt, M., 320

Wittenbrink, B., 291, 372 Wittgenstein, L., 88 Woellert, F., 356 Wollstonecraft, M., 239 Wood, W., 78, 125, 126 Woodward, J., 158, 159, 160, 161, 165 Woolcock, J., 49 Wright, J. C., 158 Wright, P. M., 167 Wu, J. J., 276 Wundt, W., 57 Wurf, E., 168 Wyer, R. S., Jr., 73

X Xu, J., 54

Y Yahalom, N., 352 Yamagishi, T., 75 Yamazaki, Y., 288 Yap, A. J., 336 Yarkoni, T., 199 Yaxley, R. H., 289 Yovetich, N. A., 316 Yzerbyt, V. Y., 143

Z Zajonc, R. B., 93, 100, 101 Zaki, J., 190, 197 Zanna, M. P., 72, 139, 152 Zarahn, E., 187 Zaraté, M. A., 291, 385 Zhang, S., 336, 401 Zhang, Z., 336 Zhong, C. B., 291 Zhou, X., 52 Zimbardo, P. G., 36 Zukier, H., 273 Zuroff, D. C., 158, 159 Zwaan, R. A., 289

Subject Index

An f following a page number indicates a figure; a t following a page number indicates a table; an n following a page number indicates a note. Abstract rules, 145–146 Action meaning-based explanation and, 43, 52–54 passions versus, 88–89 Active gene–environment correlations, 214. See also Genetic influences Actor–partner interdependence model (APIM), 315–316. See also Interdependence theory Acts of Meaning, 43 Adaptation interdependence theory and, 315–321 social functionalism and, 280 Adaptationist program, 231 Adaptive action, 289–291 Affect valuation theory, 333–335 Affective compatibility, 292–293 Affective primacy debate, 100–101 Affirming the consequent fallacy, 16–17 Affordance, 313–315 Agent-based modeling metatheoretical issues, 399–404 overview, 390–393, 403–405 simulation approaches and, 349–350 social-psychology theory and, 393–398 top-down and bottom-up uses of, 397–398

Aggregation, 358–364, 362f, 363f Aggregation models, 362 Algorithmic level of analysis. See also Levels of analysis behavioral functional level of analysis and the informational functional level of analysis and, 30–34, 32f emotion theories and, 93–94 overview, 25–28, 93 research and, 34–38 social cognition as a level of analysis and, 67–68 social-cognitive theories and, 72–73 Algorithmic models, 80n, 381–385, 383f Amygdala, 197 Analysis levels. See Levels of analysis Anger, 97–99 Anomalies, 235–236 Appealing to ignorance fallacy, 17–18 Appraisals cognitive neuroscience and, 191 emotion theories and, 85–86, 91–94, 97–99 personality psychology and, 167–169 Approach motivations, 118–120 Assessments, 168–172 Association, 195–196 Associative principles, 145–146

423

424

Subject Index

Assumptions agent-based modeling and, 399 cultural psychological theory and, 335–339 duality models and, 136–138, 137f, 148–151 evolutionary theories and, 228–229, 234–235 interdependence theory and, 307–308 Attitudes rational actor theories and, 248 simulation approaches and, 352–358, 355f, 357f social-cognitive theories and, 66 Attractiveness, 394–395 Automata theory, 152n Automaticity, 152n Autonomic nervous system (ANS) activity, 87–88 Autonomous agents, 391–392 Avoidance motivations, 118–120

B Balance theory, 230–231 Bayesian computation, 381–385, 383f Bayes’s theorem, 378 Behavioral affirmation, 322 Behavioral ecology view, 102 Behavioral effects, 69, 70–73 Behavioral functional level of analysis. See also Computational level of analysis algorithmic level of analysis and, 30–34, 32f overview, 28–30, 39n Behavioral priming, 68. See also Priming Behavioral processes, 329–332, 330f Behaviorism interdependence theory and, 321–322 meaning-based explanation and, 57 Belief-evaluation pairs, 261 Between-person analyses, 166 Bias, 380 Bias postulate, 276 Bi-conditional relations, 71 Big Five personality traits, 165–167. See also Personality psychology Biological processes cultural psychological theory and, 329–332, 330f, 332–333 simulation approaches and, 351–352 socially situated cognition and, 287–288

Blocking. See Discounting Blood oxygenation level dependent (BOLD) contrast, 188 Bodily influences, 295 Brain-imaging approaches. See Neuroimaging approaches Brunswikian induction algorithm for social cognition (BIAS) model, 356–358, 357f

C Categories demarcating emotion and, 90 dimensions versus, 89–90 emotion theories and, 96 Causal attribution discounting, 35–36 levels of analysis and, 27–28, 28, 29 Causal explanation. See also Explanation cultural psychological theory and, 335–337 emotion theories and, 99–100 meaning-based explanation and, 59 personality psychology and, 159–160, 164–165 priming and, 48–52 social-cognitive theories and, 72, 80n Causal judgment, 27–28, 31–32 Central nervous system, 93 Change, 337–338 Choices, 310–311 Circularity agent-based modeling and, 402–403 rational actor theories and, 252 Classical logic, 4–5 Clustering, 138–140 Cognition distribution of, 291–292 social-cognitive theories and, 78 socially situated cognition and, 289–291 Cognitive bias, 359 Cognitive neuroscience. See also Neuroscience approach; Socialcognitive theories consistency fallacy and, 200 explanation and, 190–192 interpretation and, 193f neuroscience methodologies used in, 184–190, 185t overview, 183–184, 200–201 social functionalism and, 279



Subject Index 425

Cognitive processes, 329–332, 330f Cognitive-affective processing systems (CAPS) model, 313–314 Collective constructionist theory, 333 Collectivism collective constructionist theory and, 333 cultural psychological theory and, 332–333, 336 meaning-based explanation and, 47–48 Common sense, 86 Comparison level, 317–319, 320 Complexity, 386–387 Computational level of analysis. See also Computational theories; Levels of analysis agent-based modeling and, 403 behavioral functional level of analysis and the informational functional level of analysis and, 28–30 emotion theories and, 93–94 overview, 25–28, 93 social cognition as a level of analysis and, 67–68 social-cognitive theories and, 72 Computational power, 145–146 Computational theories. See also Computational level of analysis algorithmic models, 381–385, 383f normative model of pseudocontingencies and, 374–380, 375t, 376t, 377t, 378f, 379f overview, 385–387 varieties of, 371–372 Computer simulation overview, 347–349, 364–366 regression and polarization and, 358–364, 362f, 363f types of, 349–350 Concept, emotion theories and, 95–97 Conceptual models, 162–163 Connectionist model, 354 Conservatism, 11–12 Consistency fallacy, 200 Construal principle, 43 Constructivism, 333 Contextual moderation, 72 Contingency learning, 374 Contingent-feedback condition, 353 Cooperation, 396 Core affect, 90

Core metaphors, 399. See also Assumptions Correlational sense, 260–262 Corroboration, 231–232 Criterion S, 141–143 Critical involvement, 195–196 Cross-situational consistency, 167–169 Cultural comparisons, 46–48 Cultural frames, 336 Cultural priming methods, 335–337 Cultural psychological theory dominant assumptions in, 335–339 middle-range theories of the mutual constitution framework, 332–335 mutual constitution of culture, 329–332, 330f overview, 328–329, 339–340

D Deduction, 4–7, 228–230, 231–232 Deductive reasoning, 226–227, 230–231 Deductive-nomological model, 159–160 Definition versus hypothesis, 146–147 rational actor theories and, 251–255, 263 Degenerative problem-shift, 10 Dehumanization, 278–279 Delta P model, 27–28, 30–31 Delta-based feedback learning models, 361–362 Demarcation problem, 4 Denying the antecedent fallacy, 18–19 Depriving the motive, 112–113 Description, 249–251 Differences, 191–192 Dimensions, 89–90 Direction, 110 Discounting, 35–36 Discrete agents, 391–392. See also Agentbased modeling Disjunctive fallacy, 19 Dispositional inference, 74–75 Dissonance, 278 Distal motivations, 118 Distributed cognition, 291–292 Distributed connectionist models, 50 DNA, 207–208, 280 Duality, 143–147 Duality models challenges created by, 143–151 overview, 132–143, 137f, 151–153 Dual-process theories, 13, 14

426

Subject Index

Dual-system theories, 136 Duhem–Quine thesis, 8, 15 Dynamic constructivist approach, 335–337

E Ecological correlations, 374 EEGs, 184, 185t, 187–190. See also Neuroimaging approaches Effective situation, 310 Effort after meaning, 43 Elaboration-likelihood model, 139 Embodied knowledge, 283 Embodiment, 288–289, 294–296 Emergent social cognition, 283 Emotion. See also Emotion theories characterizing, 85–88 cognitive neuroscience and, 191 demarcating, 90 Emotion regulation, 89, 191 Emotion theories. See also Emotion cognitive neuroscience and, 191 levels of analysis and, 91–97 logical and inferential issues, 97–102 overview, 84–85, 102–103 social-cognitive theories and, 78 states versus processes, 88–91 Emotional expressions, 293–295 Emotional processes, 329–332, 330f Empirical data, 97–99 Empirical phenomena, 140–143. See also Phenomena Environmental influences cultural psychological theory and, 332–333 genetic influences and, 208, 211–219 interdependence theory and, 322 socially situated cognition and, 290–291 Environmental models, 350 Epigenetic phenomena, 219–220 Epistemological status, 142 Equifinality, 113 Error management theory, 239n Error-reducing delta algorithm, 362f Essentialism, 206 Evaluative learning, 356–358, 357f Event-related potentials (ERPs), 184, 185t, 187, 188–190. See also Neuroimaging approaches Evocative gene–environment correlations, 214. See also Genetic influences Evolutionary biology, 231

Evolutionary theories motivational theories and, 126 overview, 224–230, 227f, 236–238 simulation approaches and, 351–352 social functionalism and, 275–276 social psychology and, 233–238 social-cognitive theories and, 78–79 testing, 231–232 what makes a theory evolutionary, 230–231 Existentialism, 43 Experimental dissociations, 141–143 Experimental levels, 148 Experimental semiotics meaning-based explanation and, 56–58 overview, 41–42, 58–59 Experimental sense, 260–262 Explanandum. See also Explanation duality models and, 140–143 emotion theories and, 84–85, 95–97 personality psychology and, 158–159, 171–172 social-cognitive theories and, 68–73 Explanans. See also Explanation duality models and, 140–143 emotion theories and, 96–97 social-cognitive theories and, 68–73, 73 Explanation. See also Causal explanation; Generalizability agent-based modeling and, 398, 399, 402 cognitive neuroscience and, 190–192 cultural psychological theory and, 329 duality models and, 138–143, 143–151, 152n evolutionary theories and, 225, 226–230, 227f, 230, 239n genetic influences and, 219–221 interdependence theory and, 319–321 levels of analysis and, 31–32 meaning-based explanation and, 44–46, 45f motivation and, 109–111 motivational theories and, 114–115 personality psychology and, 158–167 precision and, 16 rational actor theories and, 249–251 simulation approaches and, 350, 365–366 social cognition as a level of analysis and, 68 social functionalism and, 267



Subject Index 427

social-cognitive theories and, 68–73 theory and, 3 Explicit motivations, 121–122 Expressive motivation, 116–117 Extensional meanings, 47–48 Extensional relationships, 44 Extrinsic motivation, 117–118. See also Motivation

F Facial behavior emotion theories and, 98–99, 101–102 social functionalism and, 279 socially situated cognition and, 295 Fair-but-biased-yet-correctible model (FBC), 276 Fairness postulate, 276 Fallacies logical fallacies, 16–20, 97–102 mathematical models and, 380 motivational theories and, 114–115 Fallibilism, 5–6 Falsifiability agent-based modeling and, 403 emotion theories and, 100–102 evolutionary theories and, 231–232 overview, 5–6 pragmatics of, 7–9 rational actor theories and, 251–255, 263 theory evaluation and, 4 Feedforward models, 350 Five-factor theory, 163–167 Floating models, 403 Focal lesions, 185t, 186, 196 Frequency estimates, 375–380, 375t, 376t, 377t, 378f, 379f Functional approach. See also Behavioral functional level of analysis; Informational functional level of analysis duality models and, 150 motivational theories and, 124–125 overview, 36–38 Functional connectivity, 198–199 Functional levels, 91–93, 94–95 Functional magnetic resonance imaging (fMRI). See also Neuroimaging approaches cognitive neuroscience and, 184, 185t, 187–188, 197–198

consistency fallacy and, 200 interpretation and, 193–194, 193f Functionalism, 321–322 Functionalist hard cores, 269–275, 271f, 272f, 273f

G Gene–culture coevolution theory, 332– 333 Gene–environment correlation (rGE), 213–219, 220–221 Gene–environment interactions (G x E), 208, 212–213, 220–221 Generalizability. See also Explanation agent-based modeling and, 402 emotion theories and, 85–86 evolutionary theories and, 227–228 simulation approaches and, 354 theory evaluation and, 14 Generalized self-efficacy, 167–169 Genetic influences explanation and, 219–221 gene–environment correlation (rGE) and, 213–219 influences of on social behaviors, 210–213 overview, 205–209 social functionalism and, 280 social psychology and, 219–221 Genomewide association studies (GWAS), 208, 209 Gestalt tradition, 321–322 Given situation, 310 Goal-directed agents, 391–392. See also Agent-based modeling Goals, 246

H Habit, 125–126 Heritability, 207–209. See also Genetic influences Heuristics, 273 Heuristic-systematic model of persuasion, 134–135 History of psychology, 56–58 Holistic theory evaluation, 7–9, 294–295 Homeostatic motivations, 120–121 Homogeneity, 358–364, 362f, 363f Humor, 169–172 Humor Styles Questionnaire (HSQ), 169–170

428

Subject Index

Hyperpunitiveness, 277 Hypopunitiveness, 277 Hypothesis, 146–147 Hypothetico-deductive method, 142–143

I Identification, 111 Illusory correlations mathematical models and, 374 multinomial processing-tree model of, 381–385, 383f simulation approaches and, 360–364, 362f, 363f Implementational level of analysis. See also Levels of analysis emotion theories and, 93–94 overview, 25–28, 93 social cognition as a level of analysis and, 67–68 Implicit motivations, 121–122 Independence meaning-based explanation and, 47–48, 52 social-cognitive theories and, 70–71 Individual differences cultural psychological theory and, 337–338 interdependence theory and, 313–315 Individual levels, 94–95 Individualism cultural psychological theory and, 332–333, 336 meaning-based explanation and, 47– 48 Induction, 4–7, 227–228, 231–232 Inductive–hypothetic–deductive method, 229–231 Inference, 196–200 Inferential error emotion theories and, 97–102 logical fallacies and, 16–20 Infinity, 252 Influence, 396–397 Information processing, 80n Informational functional level of analysis, 28–30, 30–34, 32f. See also Computational level of analysis Informational influence, 149 Inheritability, 207–209. See also Genetic influences Input–output relations, 70–71

Inputs, 133–135. See also Mental representations Institutional constraints, 275–276 Instrumental learning, 352–358, 355f, 357f Instrumental motivation, 116–117 Integrative social-functionalist framework, 268, 269–275, 271f, 272f, 273f. See also Social functionalism Integrative theory, 76–77 Intellectual history, 56–58 Intention, 52–55 Intentional relationships, 44 Interactions, 148–151 Interdependence, 47–48, 52 Interdependence theory advantages and disadvantages of, 321–324 agent-based modeling and, 393–398 explanation and predictive value of, 319–321 metatheoretical framework of, 307–315 overview, 305–308, 324 structural features of, 315–321 Interindividual differences, 166 Interpretation cognitive neuroscience and, 193–200, 193f genetic influences and, 206–207 meaning-based explanation and, 44–46, 45f Intersectionality, 338–339 Intracranial recording, 187. See also Neuroimaging approaches Intrapsychic formulation, 278–279 Intrapsychic level, 94–95 Intrapsychic retributivist drives, 275–276 Intrinsic motivation, 117–118. See also Motivation Invariance assumption, 294–295 Invariant regularities, 292–295

J Judgment tasks, 98

K Knowledge-and-appraisal personality architecture (KAPA) model, 168–169

L Learning, 352–358, 355f, 357f Learning theory, 278



Subject Index 429

Levels of analysis agent-based modeling and, 400–402 cultural psychological theory and, 330–332, 330f emotion theories and, 91–97 extending beyond Marr’s ideas regarding, 28–38, 32f interdependence theory and, 321–324 Marr’s three levels of analysis, 25–28 overview, 24–25, 38 social cognition as, 67–68 socially situated cognition and, 285–287 Limited information, 392. See also Agentbased modeling Logic, 145–146 Logic of falsification, 4–7. See also Falsifiability Logical fallacies, 16–20, 97–102. See also Fallacies Logical incoherence, 6

M Magnetic resonance imaging (MRI), 188. See also Neuroimaging approaches Magnetoencephalogram (MEG), 187. See also Neuroimaging approaches Manipulability cognitive neuroscience and, 192 personality psychology and, 160–161, 165–166 Marr’s three levels of analysis, 25–28. See also Levels of analysis Mate choice, 394–395 Mathematical modeling algorithmic models, 381–385, 383f normative model of pseudocontingencies and, 374–380, 375t, 376t, 377t, 378f, 379f overview, 385–387 varieties of, 371–372 verbal theories and, 372–373 Mattering, 260–262, 264 Maximizing motivations, 120–121 Mean predicted frequencies, 379t Meaning-based explanation examples of, 46–55 history of psychology and, 56–58 overview, 41–46, 45f, 58–59 Meaningful social action, 42–43 Mechanistic explanation, 80n Medial orbitofrontal cortex, 199

Mediation duality models and, 138–140 social-cognitive theories and, 72 Mental constructs, 69, 70–73 Mental mediation, 72 Mental representations duality models and, 133–135 social-cognitive theories and, 79–80 Metacognition, 310–311 Metaphors, 399 Metatheoretical assumptions, 307–315, 321–322 Metatheoretical criteria agent-based modeling and, 399–404 evolutionary theories and, 226–227, 227f, 228–229, 235 overview, 239n theory evaluation and, 11–16 Metatheory, 228–230, 230, 235, 237–238, 239n Michelangelo Phenomenon, 322 Middle-range theories, 333–335 Mimicry, 293–295 MINERVA model, 358–359, 361–362, 363 Mirror neuron system, 288 Modeling, 249–251 Moderation duality models and, 138–140 social-cognitive theories and, 72 Modus tollens, 8–9, 19–20 Moralistic fallacy, 19–20 Motivation. See also Motivational theories distinctions among motives, 123–124 interdependence theory and, 310–311 methods to implicate, 112–114 overview, 108–109 pursuit of a master motive, 122–124 rationality and, 246 related concepts, 124–127 relations among motives, 123 varieties of motives and, 116–122 what motivation explains, 109–111 Motivational bias, 359 Motivational processes, 329–332, 330f Motivational theories. See also Motivation errors in affirming the consequent, 112 identification and, 111 methods to implicate motivation, 112–114 nominal fallacy and, 114–115 overview, 108–109, 127

430

Subject Index

Motivational theories (continued) pursuit of a master motive, 122–124 related concepts, 124–127 scope and, 115–116 varieties of motives and, 116–122 Multifinality, 115 Multinomial processing-tree model, 381–385, 383f Multivariate representational similarity analyses (MRSA), 199–200 Mutual access hypothesis, 236 Mutual constitution framework cultural psychological theory and, 329–332, 330f middle-range theories of, 332–335

N Narrative coherence, 159 Naturalistic fallacy, 20n Necessity, 113–114 Negative heuristic, 10 Nepotism, 235–236 Neural pathways, 93 Neuroimaging approaches. See also Functional magnetic resonance imaging (fMRI); Neuroscience approach cognitive neuroscience and, 184–190, 185t emotion theories and, 93–94, 98–99 Neuroscience approach. See also Cognitive neuroscience emotion theories and, 93–94 overview, 37–38 social-cognitive theories and, 77 socially situated cognition and, 288 Nominal fallacy, 18, 114–115 Nomothetic statement, 228, 239n Normative model of pseudocontingencies, 374–380, 375t, 376t, 377t, 378f, 379f, 387 Nuclear magnetic resonance imaging (NMRI or MRI), 188. See also Neuroimaging approaches

O Object of inquiry, 287–288 Objectification, 278–279 Objective properties, 311–313, 322 Ontology, 163 Operating conditions, 134, 147–148

Operating principles, 134 Orbitofrontal cortex, 199 Outputs, 133–135. See also Mental representations; Responses

P Paradigm shifts, 13 Parallel tests, 254–255 Parsimony mathematical models and, 386–387 moralistic fallacy and, 19–20 theory evaluation and, 12–14 Partial dualities, 144–145 Passions, 88–89 Passive gene–environment correlations, 214. See also Genetic influences Pathogen prevalence hypothesis, 332–333 Patient population approach, 184–187, 185t Perception emotion theories and, 96 interdependence theory and, 310–311 meaningful social action and, 42–43 Persistence, 110 Person factors, 313 Personality coherence, 158–159 Personality psychology cognitive neuroscience and, 184–187, 185t evaluation of, 163–167 examples of, 167–174 humor and, 169–172 overview, 157–158, 174 scientific explanation and, 158–167 Personality traits, 165–167. See also Personality psychology Personality-in-context, 169–172 Phenomena duality models and, 140–143 emotion theories and, 96, 96–97 evolutionary theories and, 239n Phi coefficient, 378–379, 378f Phrenology, 196–197 Polarization, 358–364, 362f, 363f Popperian method of falsification, 232 Positive heuristic, 10 Positivist approach, 4 Posterior distribution, 377–378, 379f Precision emotion theories and, 85–86 theory evaluation and, 15–16



Subject Index 431

Prediction agent-based modeling and, 398 duality models and, 136–143, 137f, 152n–153n interdependence theory and, 319–321 levels of analysis and, 31–32 meaning-based explanation and, 47–48, 52 personality psychology and, 161 precision and, 16 simulation approaches and, 350, 365–366 social-cognitive theories and, 80n theory and, 3 Predictive posterior distribution, 378–379 Predictive power, 402. See also Prediction Prefrontal cortex, 197–198 Priming cultural psychological theory and, 335–337 meaning-based explanation and, 48–52 motivational theories and, 121–122 social cognition as a level of analysis and, 68 social-cognitive theories and, 71, 80n Primitiveness, 252 Prior distribution, 377–379, 378f Problem solving, 395–396 Processes, 88–91 Progressive problem-shift, 10 Propositional model of associative learning, 145–146 Proximal causes, 99–100 Proximal motivation, 118 Pseudocontingencies, 374–380, 375t, 376t, 377t, 378f, 379f, 386–387 Psychological investigations, 93–94 Psychological process cognitive neuroscience and, 191–192 duality models and, 133–136 Psychological reactance, 311–312 Psychological variables, 192

R Rational actor theories definitions and falsification and, 251–255 mattering and, 260–262 modeling, description, and explanation, 249–251 overview, 245–249, 263–264

rationality and, 262–263 sufficiency and, 257–260 unit coherence, 255–257 Rationality, 245–249, 262–263 Reality, 95–97 Reason, 56–58 Recurrent feedback models, 350 Reflective-impulsive model, 144–145 Refutability social-cognitive theories and, 74–77 theory evaluation and, 14–15 Regression, 72, 358–364, 362f, 363f Reinforcement, 252 Relationships, 394–395, 396–397. See also Interdependence theory Reliability, 172–174 Representations, mental duality models and, 133–135 social-cognitive theories and, 79–80 Response syndromes, 87–88 Responses, 133–135 Restructuring the field, 314 Reverse inference, 196–200 Reward, 352–358, 355f, 357f Risk regulation model, 312

S Sample size, 363–364 Satisfying the motive, 112–113 Schema, 169 Schematicity, 173–174 Scientific research programs, 9–11 Scientific revolutions, 13 Scope motivational theories and, 115–116 social-cognitive theories and, 74–77 Segregation model, 391–392 Self versus other, 358–364, 362f, 363f Self-appraisals, 168–169. See also Appraisals Self-consciousness, 173–174 Self-control, 110–111 Self-correction postulate, 276 Self-efficacy appraisal, 167–169 Self-esteem, 116 Self-evaluation, 198 Self-interest, 309–311 Self-knowledge, 359 Self-organization, 391 Self-perception, 310–311 Self-regulation, 110–111, 274

432

Subject Index

Self-schemas, 169 Self-worth, 116 Semiotic model of psychological explanations meaning-based explanation and, 44–46, 45f priming and, 51–52 Sets of theoretical assumptions, 8 Sexual behavior, 394–395 Sharing, 396 Similarities, 191–192 Simulation models. See also Computer simulation overview, 347, 364–366 regression and polarization and, 358–364, 362f, 363f types of, 349–350 Single-process theories, 13 Situated behavior meanings, 292–295 Situated cognition movement, 284. See also Socially situated cognition Situational influences interdependence theory and, 311–313, 315–316 levels of analysis and, 286 overview, 283 socially situated cognition and, 289–291, 294–295 Small samples, 363–364 Social behavior, 205–207, 210–213 Social cognition. See also Social-cognitive theories explanation and, 68–70 as a level of analysis, 67–68 Social functionalism core requirements of, 269–275, 271f, 272f, 273f deficiency in, 275–279 forms of, 267–269, 268t overview, 266–267, 279–280 Social impact theory, 348 Social influence, 393–394 Social interaction, 169–172 Social learning, 352–358, 355f, 357f Social networks, 396–397 Social neuroscience, 77 Social processes, 184–187, 185t Social psychology evolutionary theories and, 233–238 genetic influences and, 219–221 Social value orientations, 314

Social-cognitive theories. See also Socially situated cognition cognitive neuroscience and, 184–187, 185t explanation and, 68–70 interdependence theory and, 321–322 overview, 65–66, 79–80 relation to other theoretical approaches, 77–79 scope and refutability of, 74–77 social cognition as a level of analysis, 67–68 testing, 70–73 Socially situated cognition. See also Socialcognitive theories from invariant to situated behavior meanings, 292–295 levels of analysis and, 285–287 overview, 283–285, 295–296 pillars of, 287–292 Social-psychological theory. See Interdependence theory Socioecological factors, 332–333 Solutions, 395–396 Specificity, 402. See also Prediction Standard model of psychological explanations meaning-based explanation and, 44–46, 45f priming and, 51–52 States emotion theories and, 88–91 versus processes, 88–91 Statistical fit, 384–385, 386 Statistical models duality models and, 139–140 gene–environment correlation (rGE) and, 215–216 levels of analysis and, 27–28 Stereotyping duality models and, 133–134, 148 mathematical models and, 374 Strong situations, 314 Structural equation models, 400 Structural levels agent-based modeling and, 400–401 emotion theories and, 91–93, 94–95 interdependence theory and, 317–319 Structuralism, 321–322 Subjective experience, 86–87 Subsymbolic model, 354–355, 355f



Subject Index 433

Sufficiency motivational theories and, 113–114 rational actor theories and, 257–260, 264 Survival advantages, 351–352 Symbolic model, 354–355, 355f Symbolic representations interdependence theory and, 320–321 meaning-based explanation and, 50 Symbols, 145–146 Systematic processing, 134–135

T Technologies, 93 Teleology, 126–127 Temporal considerations, 318 Termination, 110–111 Testability, 386–387 Theory agent-based modeling and, 398 evaluation of, 4–7 levels of analysis and, 31–32 metatheoretical criteria for the evaluation of, 11–16 overview, 3–4, 224–230, 227f simulation approaches and, 359–364, 362f, 363f Theory of planned behavior (TPB) mattering and, 261 rational actor theories and, 247 Theory of reasoned action (TRA) definitions and falsification and, 251–255 modeling, description, and explanation, 249–251 overview, 263–264 rational actor theories and, 247 sufficiency and, 257–260 Theory-constructing role, 226–227 Thought experiments, 261–262 Tilt, 108–109 Transactive memory systems, 291–292

Transcranial magnetic stimulation (TMS), 187. See also Neuroimaging approaches Transformation of motivation, 310 Transformational activity, 316 Transparency, 350 Triad of functionalist frameworks, 268– 269, 268t. See also Social functionalism True score theory, 254–255 Trust, 119 Twin studies gene–environment correlation (rGE) and, 215–216 genetic influences and, 211–213

U Unconscious thought effect, 69 Unimodel, 144–145 Unit coherence, 255–257

V Validation agent-based modeling and, 400–401 multinomial processing-tree model of illusory correlations and, 384–385 Valuation rules, 314 Variable-based modeling agent-based modeling and, 399 simulation approaches and, 349–350 Variation, 337–338 Verbal theories, 372–373 Verification, 4 Virtus dormativa explanations, 239n Voluntary settlement hypothesis, 332–333

W Weak situations, 314

X Xenophobia, 233–235