Configurational Theory and Methods in Organizational Research 9781781907795, 9781781907788

Citation preview

CONFIGURATIONAL THEORY AND METHODS IN ORGANIZATIONAL RESEARCH

RESEARCH IN THE SOCIOLOGY OF ORGANIZATIONS Series Editor: Michael Lounsbury Recent Volumes: Volume 17:

Organizational Politics

Volume 18:

Social Capital of Organizations

Volume 19:

Social Structure and Organizations Revisited

Volume 20:

The Governance of Relations in Markets and Organizations

Volume 21:

Postmodernism and Management: Pros, Cons and the Alternative

Volume 22:

Legitimacy Processes in Organizations

Volume 23:

Transformation in Cultural Industries

Volume 24:

Professional Service Firms

Volume 25:

The Sociology of Entrepreneurship

Volume 26:

Studying Difference between Organizations: Comparative Approaches to Organizational Research

Volume 27:

Institutions and Ideology

Volume 28:

Stanford’s Organization Theory Renaissance, 1970–2000

Volume 29:

Technology and Organization: Essays in Honour of Joan Woodward

Volume 30A: Markets on Trial: The Economic Sociology of the US Financial Crisis: Part A Volume 30B: Markets on Trial: The Economic Sociology of the US Financial Crisis: Part B Volume 31:

Categories in Markets: Origins and Evolution

Volume 32:

Philosophy and Organization Theory

Volume 33:

Communities and Organizations

Volume 34:

Rethinking Power in Organizations, Institutions, and Markets

Volume 35:

Reinventing Hierarchy and Bureaucracy – From the Bureau to Network Organisations

Volume 36:

The Garbage Can Model of Organizational Choice – Looking Forward at Forty

Volume 37:

Managing ‘Human Resources’ by Exploiting and Exploring People’s Potentials

RESEARCH IN THE SOCIOLOGY OF ORGANIZATIONS VOLUME 38

CONFIGURATIONAL THEORY AND METHODS IN ORGANIZATIONAL RESEARCH EDITED BY

PEER C. FISS University of Southern California, USA

BART CAMBRE´ Antwerp Management School, Belgium

AXEL MARX University of Leuven, Belgium

United Kingdom – North America – Japan India – Malaysia – China

Emerald Group Publishing Limited Howard House, Wagon Lane, Bingley BD16 1WA, UK First edition 2013 Copyright r 2013 Emerald Group Publishing Limited Reprints and permission service Contact: [email protected] No part of this book may be reproduced, stored in a retrieval system, transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise without either the prior written permission of the publisher or a licence permitting restricted copying issued in the UK by The Copyright Licensing Agency and in the USA by The Copyright Clearance Center. Any opinions expressed in the chapters are those of the authors. Whilst Emerald makes every effort to ensure the quality and accuracy of its content, Emerald makes no representation implied or otherwise, as to the chapters’ suitability and application and disclaims any warranties, express or implied, to their use. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN: 978-1-78190-778-8 ISSN: 0733-558X (Series)

ISOQAR certified Management System, awarded to Emerald for adherence to Environmental standard ISO 14001:2004. Certificate Number 1985 ISO 14001

CONTENTS LIST OF CONTRIBUTORS

ix

ADVISORY BOARD

xiii

FOREWORD: THE DISTINCTIVENESS OF CONFIGURATIONAL RESEARCH Charles C. Ragin

xv

CHAPTER 1 CONFIGURATIONAL THEORY AND METHODS IN ORGANIZATIONAL RESEARCH: INTRODUCTION Peer C. Fiss, Axel Marx and Bart Cambre´

1

CHAPTER 2 CRISP-SET QUALITATIVE COMPARATIVE ANALYSIS IN ORGANIZATIONAL STUDIES Axel Marx, Bart Cambre´ and Benoıˆt Rihoux

23

CHAPTER 3 THE TWO QCAs: FROM A SMALL-N TO A LARGE-N SET THEORETIC APPROACH Thomas Greckhamer, Vilmos F. Misangyi and Peer C. Fiss

49

CHAPTER 4 CONFIGURATIONAL ANALYSIS AND ORGANIZATION DESIGN: TOWARDS A THEORY OF STRUCTURAL HETEROGENEITY Anna Grandori and Santi Furnari

77

CHAPTER 5 THE ANALYSIS OF TEMPORALLY ORDERED CONFIGURATIONS: CHALLENGES AND SOLUTIONS Tony Hak, Ferdinand Jaspers and Jan Dul

107

v

vi

CONTENTS

CHAPTER 6 UNDERSTANDING COMPLEMENTARITIES AS ORGANIZATIONAL CONFIGURATIONS: USING SET THEORETICAL METHODS Gregory Jackson and Na Ni

129

CHAPTER 7 CORPORATE GOVERNANCE AND CONFIGURATION RESEARCH: THE CASE OF FOREIGN IPOs LISTING IN LONDON R. Greg Bell, Ruth V. Aguilera and Igor Filatotchev

159

CHAPTER 8 CORPORATE SOCIAL RESPONSIBILITY: A MULTILEVEL EXPLANATION OF WHY MANAGERS DO GOOD Donal Crilly

181

CHAPTER 9 THE VALUE OF CONFIGURATIONAL APPROACHES FOR STUDYING DIGITAL BUSINESS STRATEGY YoungKi Park and Omar A. El Sawy

205

CHAPTER 10 THE CONFIGURATIONAL APPROACH IN ORGANIZATIONAL NETWORK RESEARCH Jo¨rg Raab, Robin H. Lemaire and Keith G. Provan

225

CHAPTER 11 COUNTRY-SPECIFICITY AND INDUSTRY PERFORMANCE: A CONFIGURATIONAL ANALYSIS OF THE EUROPEAN GENERIC MEDICINES INDUSTRY Kalle Pajunen and Ville Airo

255

CHAPTER 12 APPLYING FUZZY SET METHODOLOGY TO EVALUATE SUBSTITUTES FOR LEADERSHIP J. Lee Whittington, Victoria McKee, Vicki L. Goodwin and R. Greg Bell

279

Contents

vii

CHAPTER 13 WE TRY HARDER: SOME REFLECTIONS ON CONFIGURATIONAL THEORY AND METHODS David J. Ketchen, Jr.

303

CHAPTER 14 CONCLUSION: THE PATH FORWARD Bart Cambre´, Peer C. Fiss and Axel Marx

311

LIST OF CONTRIBUTORS Ruth V. Aguilera

Department of Business Administration, University of Illinois at Urbana-Champaign, Champaign, IL, USA

Ville Airo

Aalto University, Espoo, Finland

R. Greg Bell

College of Business, University of Dallas, Irving, TX, USA

Bart Cambre´

Antwerp Management School, Antwerp, Belgium

Donal Crilly

Strategy and Entrepreneurship, London Business School, London, UK

Jan Dul

Rotterdam School of Management, Erasmus University, Rotterdam, the Netherlands

Omar A. El Sawy

Marshall School of Business, University of Southern California, Los Angeles, CA, USA

Igor Filatotchev

Cass Business School, City University London, London, UK; Institute of International Business, Vienna University of Economics and Business, Vienna, Austria

Peer C. Fiss

Marshall School of Business, University of Southern California, Los Angeles, CA, USA

Santi Furnari

Cass Business School, City University London, London, UK

Vicki L. Goodwin

College of Business, University of North Texas, Denton, TX, USA

Anna Grandori

Bocconi University, Milan, Italy

Thomas Greckhamer

Ourso College of Business, Louisiana State University, Baton Rouge, LA, USA ix

x

LIST OF CONTRIBUTORS

Tony Hak

Rotterdam School of Management, Erasmus University, Rotterdam, the Netherlands

Gregory Jackson

School of Business & Economics, Freie Universita¨t Berlin, Berlin, Germany

Ferdinand Jaspers

Rotterdam School of Management, Erasmus University, Rotterdam, the Netherlands

David J. Ketchen, Jr. College of Business, Auburn University, Auburn, AL, USA Robin H. Lemaire

School of Public & International Affairs, Virginia Tech University, Blacksburg, VA, USA

Axel Marx

Leuven Centre for Global Governance Studies, University of Leuven, Leuven, Belgium

Victoria McKee

College of Business, University of North Texas, Denton, TX, USA

Vilmos F. Misangyi

Smeal College of Business, Pennsylvania State University, University Park, PA, USA

Na Ni

Department of Management and Marketing, Hong Kong Polytechnic University, Hong Kong

Kalle Pajunen

Turku School of Economics, University of Turku, Turku, Finland

YoungKi Park

College of Business Administration, University of Akron, Akron, OH, USA

Keith G. Provan

Eller College of Management, University of Arizona, Tucson, AZ, USA

Jo¨rg Raab

Department of Organization Studies, Tilburg University, Tilburg, The Netherlands

Charles C. Ragin

Department of Sociology, University of California, Irvine, CA, USA; Centre for Welfare State Research, University of Southern Denmark, Odense, Denmark

xi

List of Contributors

Benoıˆt Rihoux

Centre de Science Politique et de Politique Compare´e, Universite´ catholique de Louvain, Louvain-La-Neuve, Belgium

J. Lee Whittington

College of Business, University of Dallas, Irving, TX, USA

ADVISORY BOARD SERIES EDITOR Michael Lounsbury Associate Dean of Research Thornton A. Graham Chair University of Alberta School of Business and National Institute for Nanotechnology, Alberta, Canada

ADVISOR BOARD MEMBERS Howard E. Aldrich University of North Carolina, USA

Frank R. Dobbin Harvard University, USA

Stephen R. Barley Stanford University, USA

Royston Greenwood University of Alberta, Canada

Nicole Biggart University of California at Davis, USA Elisabeth S. Clemens University of Chicago, USA

Mauro Guillen The Wharton School, University of Pennsylvania, USA

Jeannette Colyvas Northwestern University

Paul M. Hirsch Northwestern University, USA

Barbara Czarniawska Go¨teborg University, Sweden

Brayden King Northwestern University

Gerald F. Davis University of Michigan, USA

Renate Meyer Vienna University of Economics and Business Administration, Austria

Marie-Laure Djelic ESSEC Business School, France

xiii

xiv

ADVISORY BOARD

Mark Mizruchi University of Michigan, USA

Marc Schneiberg Reed College

Walter W. Powell Stanford University, USA

W. Richard Scott Stanford University, USA

Hayagreeva Rao Stanford University, USA

Haridimos Tsoukas ALBA, Greece

FOREWORD: THE DISTINCTIVENESS OF CONFIGURATIONAL RESEARCH Conventional quantitative research in the social sciences today is largely based on an understanding of analysis that is antithetical to configurational thinking. To analyze a phenomenon is to break it into its constituent parts and then to examine how the parts fit together, a two-step process. A common way of accomplishing the first step – breaking things into parts – is to conceptualize variables that can be used to characterize differences across cases.1 In conventional quantitative research the second step – examining how the parts fit together – is accomplished primarily through various forms of cross-case analysis using correlational techniques (e.g., multiple regression). Thus, in conventional quantitative research, assessments of cross-case correlational patterns provide the primary basis for statements about how the parts of cases are connected to each other. Quantitatively oriented researchers studying organizations have produced an abundance of such studies, relating specific aspects of organizations to other aspects based on correlations observed across a set of comparable organizations. Case studies of organizations, by contrast, often reveal that multiple aspects of cases are strongly linked together in coherent bundles. In research I conducted with Bruce Kogut and John Paul MacDuffie, we found that production strategies in automobile assembly plants are often bundled in specific, prototypical ways. Complementarities associated with specific bundles of practices often yield greater productivity (Kogut, MacDuffie, & Ragin, 2004). In essence, we focused on how organizational aspects were linked together within cases, which is the hallmark of configurational research. This way of viewing how the parts of cases fit together offers a strong contrast with conventional, correlation-based approach. In configurational research, combinations of case aspects are not disaggregated into separate independent variables and then treated as raw material for correlational analysis. Rather, they are viewed in terms of the nexus they constitute. The conventional approach to analysis presents a second obstacle to configurational thinking in its focus on the ‘‘net effects’’ of competing xv

xvi

FOREWORD

‘‘independent variables’’ on a ‘‘dependent variable.’’ The usual idea is that different variables influence the dependent variable to different degrees and that a central goal of analysis is to isolate, to the extent possible, the relative impact of each independent variable. A famous example of the net effects way of conducting social science is Herrnstein and Murray’s The Bell Curve. They ask: which is more consequential for an individual’s success, socioeconomic status of the respondent’s family of origin or the respondent’s ‘‘intelligence’’ as measured by test scores? The idea that variables compete against each other in a contest to explain variation in an outcome is antithetical to configurational thinking. From a configurational viewpoint, the ‘‘effect’’ of a causally relevant condition depends on context – on the other casually relevant conditions found in the case in question. Thus, configurational research follows the lead of qualitative researchers in their commitment to the idea that the effects of causes are context dependent. Given the right configuration of conditions, the effect of cause X on an outcome may be decisive; in other configurations, its effect may be trivial; in still other contexts, the cause may have a counterproductive impact. Scholars who study cases as configurations have a related interest in the ‘‘limited diversity’’ of the cases they study. Consider the researcher who is interested in five causal conditions. The vector space defined by these five causal conditions has 32 sectors (25). Researchers who use naturally occurring (i.e., nonexperimental) data almost always find that many sectors are void or virtually void of cases. These empty sectors attest to the limited diversity of a given set of cases; a fully diverse set of cases would occupy all 32 sectors. From the perspective of conventional quantitative research it is easy to explain why some sectors have no cases or very few cases. The five variables in the example may be correlated in various ways, which has the natural consequence of distributing cases unevenly in the five-dimensional vector space. In other words, the limited diversity of naturally occurring social phenomena is seen as a straightforward artifact of correlational links among variables. Thus, even very large-N data sets may exhibit limited diversity. From a configurational viewpoint, by contrast, limited diversity is rich in its implications. It is not a mere artifact of correlated variables. Why are cases limited in their diversity? Correlation certainly plays a role, but so do set-theoretic connections. Consider a simple example, the link between education and occupational prestige, which is shown in Fig. 1 using data from the General Social Survey. There are virtually no cases in the upperleft triangle of the scatterplot, where low education/high occupational prestige respondents would reside. This void is not a product of the positive

xvii

Foreword

Fig. 1.

Limited Diversity in a Two-Dimensional Vector Space.

correlation between education and occupational prestige. The lower triangle with data could be flipped and the correlation would remain the same. The void exists for a very good substantive reason – high prestige occupations almost always have educational qualifications. Thus, there is a set theoretic relation between education and occupational prestige such that the set of respondents with high occupational prestige constitute a more or less consistent subset of the respondents with high education.2 This set-theoretic connection explains the limited diversity of cases in the two-dimensional vector space shown in Fig. 1. More generally, whenever there are sectors without cases (i.e., constraints on diversity), it is important to ask why these constraints exist. There are often very good reasons for these voids. Consider a more elaborate example of limited diversity. Studies of variation in the generosity of welfare states across the advanced industrial societies focus on a variety of conditions, but especially those related to unions and the strength of left parties. Table 1 presents configurational data on three such conditions: strength of left parties (as indicated by years of rule by left parties), unionization (as indicated by the extent of union

xviii

Table 1.

FOREWORD

Configurations of Left Party and Labor Relations Across the Advanced Industrial Countries.

Strong Left Parties 1 0 0 0 0 1 1 1

Strong Unions

Corporatist Relations

Number of Countries

1 0 0 1 1 0 0 1

1 0 1 0 1 0 1 0

6 5 3 1 1 0 0 0

membership), and degree of corporatism (based on an index measuring the degree of formal coordination among employers, unions, and the government).3 With three conditions defining the analytic space, there are eight sectors. Five of these eight sectors are populated with cases; three are completely void of cases. Algebraically, the three empty sectors can be expressed as follows: empty ¼ left  ð unions þ  corporatismÞ where ‘‘B’’ indicates negation, ‘‘+’’ indicates logical or, and ‘‘’’ indicates logical and. The empty cells combine the existence of strong left parties and the absence of either strong unions or corporatism, showing that strong left parties are linked to the co-presence of both strong unions and corporatism (as in the first data row of the table). It is important to ask why these voids exist. However, answering this question would require a lengthy excursion into the history of labor politics in the advanced industrial societies and is beyond the scope of this forward. The point is that viewing the evidence in a configurational manner provides insights that are simply impossible to glean from a matrix of bivariate correlations, the analytic foundation for most conventional quantitative techniques. Perhaps more striking from a configurational viewpoint are the two high frequency rows of Table 1 – the first two data rows. These two rows (1) together capture almost 70% of the cases, (2) are both internally consistent from the viewpoint of substantive knowledge, and (3) are at opposite ends of the three-dimensional vector space. It is not uncommon for a small number of high frequency configurations to capture most of the cases. In Redesigning Social Inquiry (Ragin, 2008), I show that this pattern

xix

Foreword

holds even for a large-N, individual-level data set. What is more striking is the fact that the two high-frequency configurations are internally consistent and opposite. In essence, the evidence in Table 1 suggests that there are two main types of advanced industrial societies when viewed through the lens of left parties and labor relations. The evidence suggests further that there are institutional, political, and inertial forces that support diametrically divergent paths, one combining strong left parties, strong unions and corporatist labor relations and the other rejecting these three institutions. In this light, the other three configurations listed in Table 1 (totaling five cases) constitute imperfect instances of the two central types. The first empirical example of configurational analysis presented here uses individual-level data (Fig. 1); the second uses country-level data (Table 1). The study of organizations lies in between these two micro-to-macro extremes. Configurational methods are especially well-suited for research on organizations because the study of organizations is very much focused on the question of how the parts of a case fit together. A common observation is that organizational practices often come in bundles that defy easy disassembly. Practitioners often find that it is very difficult to change or reform one part or aspect of an organization without also changing many others. Sometimes minor changes can cascade through an organization, leading to substantial changes in the organization as a whole. It is also true that because the parts of an organization are so ‘‘sticky,’’ that meaningful change is often thwarted. In short, because organizational features are configurationally linked, organizational change is often configurational as well. When parts are strongly interconnected, as they are in organizations, they are best examined in a configurational and unitary manner. In a given population of comparable organizations, there is usually only a very a limited number of ways of putting organizations together, and these limitations on organizational diversity are meaningful and interpretable. It is becoming increasingly clear that correlational techniques offer a very limited and primitive way to study the wholes formed by strongly interconnected parts. This much-needed volume provides unequivocal evidence of the promise of configurational methods in organizational research.

NOTES 1. It is worth noting that while the idea of an extractable variable is central to social science, many scholars in the humanities consider the idea of the variable to be an obstacle to understanding, interpretation, and representation.

xx

FOREWORD

2. Goertz, Hak, and Dul (2013) explain why it is often more important to examine the ‘‘white spaces’’ in scatterplots than focus simply on the distribution of data points. 3. The data in Table 1 are based on the fuzzy set membership scores of countries in the three sets that define the dimensions of the eight-sector cube.

Charles C. Ragin

REFERENCES Goertz, G., Hak, T., & Dul, J. (2013). Ceilings and floors: Where are there no observations? Sociological Methods and Research, 42(1), 3–40. Kogut, B., MacDuffie, J. P., & Ragin, C. (2004). Prototypes and strategy: Assigning causal credit using fuzzy sets. European Management Review, 1(2), 114–131. Ragin, C. (2008). Redesigning social inquiry: Fuzzy sets and beyond. Chicago, IL: University of Chicago Press.

CHAPTER 1 CONFIGURATIONAL THEORY AND METHODS IN ORGANIZATIONAL RESEARCH: INTRODUCTION Peer C. Fiss, Axel Marx and Bart Cambre´ ABSTRACT The notion of configuration – that the whole is best understood from a systemic perspective and should be viewed as a constellation of interconnected elements – is arguably one of the central ideas of organization studies. Yet, this idea also remains one of the field’s least understood aspects. In this volume and its introduction, we outline a new perspective for understanding configuration. Our starting point is the emergence of set theoretic configurational methods, and especially Qualitative Comparative Analysis (QCA), which provides novel ways for analyzing configurations. Our volume goes beyond introducing a new method to the fields of management and organization, as these methods furthermore offer an opportunity to rethink our understanding of the field and to develop different ways of theorizing the rich complexity of relationships that characterize organizational life. In this introduction, we introduce some of the key themes that differentiate the approach taken here from previous work on organizational configurations and provide

Configurational Theory and Methods in Organizational Research Research in the Sociology of Organizations, Volume 38, 1–22 Copyright r 2013 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0733-558X/doi:10.1108/S0733-558X(2013)0000038005

1

2

PEER C. FISS ET AL.

evidence for the emerging renaissance of the configurational approach in organizational theory and research. Keywords: Configurational theory; configurational methods; set theory; Qualitative Comparative Analysis (QCA); management; organization studies

INTRODUCTION The notion of configuration – that the whole is best understood from a systemic perspective and should be viewed as a constellation of interconnected elements – is arguably one of the central ideas of organization studies, stemming back to the writings of founding fathers such as Max Weber (1922[1978]). That this notion has at the same time remained one of the field’s least understood aspects is one of the greater paradoxes of organization studies. While the emergence of systems thinking (Katz & Kahn, 1978; Lawrence & Lorsch, 1967) in organization and management theory presented one of its most defining developments, configurational theory and analysis itself – while showing considerable advances (e.g., Meyer, Tsui, & Hinings, 1993; Miller, 1996) – has yet to live up to its promise. Indeed, after the notion of configuration became a central feature of organization theory during the 1970s and 1980s (e.g., Child, 1972; Miles & Snow, 1978; Mintzberg, 1983), and while some key contributions emerged in the 1990s and early 2000s (e.g., Child, 2002; Doty, Glick, & Huber, 1993; Ketchen, Thomas, & Snow, 1993), the development of the configurational approach appears to have stalled. This failure points to a dual challenge to configurational theory: the need to develop theory that can account for the complexity of configurations, a complexity that grows exponentially as more elements are added to the system, along with a methodology that can account for the complexity of such interconnected elements that bring about outcomes jointly and synergistically rather than individually and in a linear fashion. While these challenges appear daunting, we believe the current volume provides further evidence that we are currently witnessing an emerging renaissance of the configurational approach in organization studies. The background to this development is a need to account for the growing complexity of organizational life, coupled with an increasing number of new theories to account for this complexity (e.g., Suddaby, Hardy, & Huy, 2011).

Configurational Theory and Methods in Organizational Research: Introduction

3

While a number of more recent works have aimed to address this need to better account for the configurational nature of organizational phenomena (e.g., Desarbo, Di Benedetto, Song, & Sinha, 2005; Marlin, Ketchen, & Lamont, 2007; Siggelkow, 2001, 2002), the specific approach that unites the contributions of this volume presents perhaps a more fundamental shift in that it aims to reorient both theoretical conceptualization and methodological approach towards configurations based on ideas that have their roots in a set theoretic understanding of the world (Ragin, 1987; Zadeh, 1972). Our starting point is the emergence of set theoretic configurational methods, and especially Qualitative Comparative Analysis (QCA) in the social sciences. QCA has its origins in a rich tradition of comparative casebased sociology and has been systematized, further developed and transformed into a coherent approach by Charles C. Ragin (1987, 2000, 2008). In recent years, several new innovations have been introduced further broadening the scope and performance of the method (Rihoux & Marx, 2013; Schneider & Wagemann, 2012). QCA was originally developed as a middle way between the case-oriented (or ‘‘qualitative’’), and the variable-oriented (or ‘‘quantitative’’) approaches; a ‘‘synthetic strategy’’ that would ‘‘integrate the best features of the caseoriented approach with the best features of the variable-oriented approach’’ (Ragin, 1987, p. 84). It provides the researcher with a novel set of tools for disentangling complex causal relationships. While these methods emerged in political science and sociology, where their comparative nature made them attractive to researchers aiming to understand the configurational nature of a limited set of cases (Marx et al., 2013; Rihoux & Marx, 2013), their ability to handle causal complexity also makes them particularly attractive to organization and management scholars who believe that ‘‘organizations are best understood as clusters of interconnected structures and practices, rather than as modular or loosely coupled entities whose components can be understood in isolation’’ (Fiss, 2007, p. 1180). These methods are distinct in the sense that they combine set theory and Boolean algebra to offer researchers a set of new methodological tools to analyze how configurations of explanatory conditions result in observable changes or discontinuities in an outcome.1 To be sure, neither the use of set theory nor the focus on configurations are methodologically new and several methodological tools exist (see Table 1). What is distinct is the combination of set theory and configurational approaches as a method. The different contributions in this volume apply these methods and show in depth their potential to analyze configurations and contribute to theory development in a diverse field of applications in management and organization science.

4

PEER C. FISS ET AL.

Table 1. Set Theoretical Configurational Analysis and Adjacent Methods. Set Theory Set theory

Non set theory

Configurational

Focus of the Volume

Cluster analysis

Non-configurational

Fuzzy-set regressions

Conventional correlational analysis

Moreover, the purpose of our volume goes beyond merely introducing a new method to management science. More to the point, we believe the contributions also provide an opportunity to rethink our understanding of the field of organization studies and to perhaps offer a different way to theorize the rich complexity of relationships that characterize organizational life. We further develop this idea in this introduction and further elaborate the distinctive features of this configurational methodological and theoretical approach. Our introduction ends with an overview of the different contributions in the volume.

THE INTERPLAY OF THEORY AND METHODS The idea that theory and methods are closely interconnected and tend to evolve in tandem is certainly not novel. As Sørensen, Van Maanen, and Mitchell (2007, p. 1146) have noted, ‘‘method can generate and shape theory, just as theory can generate and shape method.’’ As the process of creating representations of social life involves a dialogue of ideas (theory) and evidence (data) (Ragin, 1994), we should not be surprised to see both talk and back-talk in this exchange; how we get to know the world is as consequential as the ideas we start out with. In the natural sciences, the role of instrumentation and calibration has long been understood to be a central feature in the evolution of theoretical paradigms (Kuhn, 1970). It was the construction and operation of the Large Hadron Collider that eventually allowed physicists to detect the Higgs boson, providing for support for the Standard Model of particle physics; the failure to discover the Higgs boson would have required consideration for other theoretical accounts underlying the Higgs mechanism. Yet, the analogy to the natural sciences goes deeper than the shift in overall theoretical accounts. At a very basic level, researchers in the natural sciences routinely calibrate their measuring

Configurational Theory and Methods in Organizational Research: Introduction

5

devices ‘‘so they match or conform to dependably known standards’’ (Ragin, 2008, p. 72). As Ragin points out, calibration is also known to social science researchers in the form of indices such as the Human Development Indicator, which allows international comparisons based on a country’s quality of life based on life expectancy, education, and income. Yet, the use of finely calibrated measures tends to be the exception rather than the rule in the social sciences. In management, too often we tend to use performance measures such as ROE or ROA in a sample-dependent way, paying relatively little attention to the actual meaning of where on the overall scale a firm would need to fall in order to qualify as ‘‘high performing’’ relative to an external standard as opposed to the sample in question. Taking this interplay between theory and methods as a starting point, our goal in the current volume is to start a conversation about the ways in which a configurational approach may reshape both the ways in which we theorize organizations and how we empirically and theoretically engage with our data. At the same time, we are keenly aware that much work still remains and that the process of establishing a novel theoretical perspective will be a lengthy one. As Kuhn noted, ‘‘a new scientific truth does not triumph by convincing its opponents and making them see the light, but rather because its opponents eventually die, and a new generation grows up that is familiar with it’’ (1970, p. 150). Our own goals here are much more modest in that we do not aim to supplant current theories as much as complement them with a different perspective.

THE CONFIGURATIONAL CHALLENGE Before delving more deeply into the approach suggested here, it is helpful to consider the challenges associated with taking a configurational perspective. Most centrally, this challenge involves dealing with increased levels of complexity that have to be accounted for both theoretically and methodologically. The causes of such complexity are outlined by Ackoff (1981, pp. 15–16), who notes that a complex system satisfies three conditions: ‘‘(1) The behavior of each element has an effect on the behavior of the whole. (2) The behavior of the elements and their effects on the whole are interdependent. The way each element behaves and the way it affects the whole depends on how at least one other element behaves. (3) However subgroups of the elements are formed, each has an effect on the behavior of the whole and none has an independent effect on it.’’ Configurational theory and methodology thus have to account for complex interdependencies that

6

PEER C. FISS ET AL.

run counter to the ‘‘the more we observe variable X, the more we should observe outcome Y’’ statements that tend to dominate current theorizing in organization and management studies and are based on a correlational understanding that may in fact bear at times little resemblance to the true causal structure of the relationships in question; particularly when that relationship is complex and causality is conjunctural and reflective of what Mackie (1974) called INUS conditions that are insufficient but nonredundant parts of a condition which is itself unnecessary but sufficient for the occurrence of the effect. Instead, a configurational approach is more likely to be interested in what Ragin has called ‘‘chemical causation’’ (Ragin, 1987) and ‘‘causal recipes,’’ (Ragin, 2000) that is, constellations of causal factors that jointly bring about an outcome. The challenge of the configurational approach is further complicated by the fact that much configurational theorizing tends to be informed by a logic of consistency – that is, by the idea that all elements of a configuration are equally important and present necessary conditions for either its existence or effectiveness. This logic flows from the holistic nature of configurations which holds that the configuration should be viewed as a whole, not as a collection of elements. Yet, this holistic view presents a problematic assumption that is likely to lead both researchers and manages astray (Fiss, 2011). Most empirically observed configurations are likely to contain not only indispensable parts but also inconsistencies and trivial elements. Yet, identifying what really matters for the configuration to be effective and what is perhaps expendable is a nontrivial problem given that our understanding of the causal processes involved is almost always incomplete. Building on prior theories regarding the challenges of understanding causal relationships in typologies and taxonomies, we may think of this issue in configurational theory as the ‘‘Blue Butterfly Problem,’’ where ‘‘the creation of a class [of] blue butterflies is irrelevant for the understanding of the anatomical structure of Lepidoptera’’ (Leach, 1961; quoted in Pinder & Moore, 1979, p. 109). The challenge is a dual one; what matters to the configuration may not always be evident, and what would appear to matter may in fact be quite irrelevant. If complexity is both the strength and the challenge of the configurational approach, it is the sources of this complexity that present the greatest opportunities for theoretical advancement. Yet, theorizing these sources is a task that remains largely incomplete. Prior work, however, offers some insight as to where such theoretical advancement may both begin and connect to other theoretical accounts. For instance, in one of the key pieces outlining the configurational approach, Meyer et al. define organizational

Configurational Theory and Methods in Organizational Research: Introduction

7

configurations as ‘‘any multidimensional configuration of conceptually distinct characteristics that commonly occur together’’ (1993, p. 1175). This definition conceptualizes configurations in terms of the co-occurrence of distinct characteristics, thus using commonality as a reference point (Fiss, 2009). Yet, considering the realm of possibilities, why is it that we tend to observe only a relatively limited set of configurations that have empirical instances? As Miller (1981, 1986) notes, there are at least three reasons for this state of affairs. First, competitive pressures from the environment tend to weed out unsustainable models, thus pointing to the role of external selection pressures emphasized both by economic theories of competition and sociological theories of population dynamics. Note, however, that this perspective does not require the presence of any internal consistency or fit. This is where the second reason comes in: organizations tend to be drawn to certain configurations that are internally harmonious and mutually reinforcing, demonstrating alignment among elements of structure, strategy, process, and environment (Miller, 1990). This suggests an internal selection mechanism, usually based on the experience of what works, or at least the impression that such arrangements appear to have worked for another organization. Such arguments open the door for both organizational learning and institutional arguments on top of efficiency-based ones. In this regard, Fombrun (1989) has pointed to the symbolic-cultural aspect of organizational configurations; the processes that bring about a limited set of configurations are likely to operate at multiple levels, including competitive and evolutionary forces alongside sociopolitical and cultural ones. Yet, the challenge of exploring the interaction between these theoretically distinct forces remains so far largely unmet. Third, organizational change frequently tends to be noncontinuous and episodic, suggesting that hybrid forms are less likely to be explored (Miller & Friesen, 1984). Such arguments would also find support from a perspective of localized search around performance peaks on a rugged landscape (Levinthal, 1997). Finally, there is of course also a mathematical reason for why not all theoretically possible configurations actually have empirical instances, as the number of possible configurations increases exponentially with the number of attributes considered. The theoretical challenges of the configurational approach have been complemented by equally vexing methodological ones. Configurational theory has arguably been held back by a mismatch between theory and methods; ‘‘while theoretical discussions of configurational theory thus stress nonlinearity, synergistic effects, and equifinality, empirical research has so far largely drawn on econometric methods that by their very nature tend to

8

PEER C. FISS ET AL.

imply linearity, additive effects, and unifinality’’ (Fiss, 2007, p. 1181). Commonly used methods such as cluster analysis, deviation scores, and interaction effects all have severe deficiencies in their ability to disentangle the complex causal processes inherent in organizational configurations, thus providing only limited insights into what lies at the core of the configurational approach (e.g., Fiss, 2007). Given the strong interplay between theory and methods we outlined above, it is evident that advances in the configurational realm will have to be as much methodological as theoretical. We now aim to sketch the outline of such a path forward for the configurational perspective.

THE CONFIGURATIONAL PERSPECTIVE: A SKETCH The perspective we suggest here and that pervades the contributions in this volume is at this time perhaps best understood not so much as a novel theory of organizations but as a meta-theoretical perspective of organizational phenomena; it does not necessarily challenge the power of mechanisms suggested by prior organizational theories such as resource dependence, contingency theory, institutional theory, population ecology, or transaction cost economics, but instead it suggests that detecting and understanding the proper operation of these mechanisms may require novel and different ways of detecting and examining them. Yet, the perspective we outline here is more than merely a novel methodology – it challenges not only empirical research strategies, but relies on a number of different theoretical concepts to bring about a novel way of thinking about organization studies. We do not aim to provide an exhaustive discussion of these ideas in this introduction, but merely aim to sketch some of the themes that are the hallmarks of this perspective (for a more extensive introduction see Fiss, 2007, 2011; Schneider & Wagemann, 2012). Perhaps the most fundamental aspect of the configurational approach we aim to outline is its reliance on sets and set–subset relationships rather than variables and correlations. The shift may appear to be a subtle one, but it is in fact a significant one. As Ragin (2000, 2008) notes, sets are not variables, although they may be based on data that is usually employed to create variables and although they may have the same fine-grained texture to them as variables. However, sets are superior to most variables in that they are not only precise but are also calibrated based on either prior theoretical or substantive knowledge about the concept they are meant to represent. This reliance on external theoretical criteria to calibrate sets makes them

Configurational Theory and Methods in Organizational Research: Introduction

9

particularly powerful tools in the hands of researchers aiming to gain a substantive understanding of the nature of organizational phenomena. For instance, instead of using variables that are mean-centered and thus usually sample-dependent, the calibration of fuzzy sets based on external criteria forces the researcher to be explicit in determining what it means to be at any given level of a scale or other dimension of relevance, for example, performance. This greater theoretical precision forces the researcher to be explicit about how their measures are constructed. Hence, a set theoretic approach starts from the idea that attributes of cases are not best described in terms of variables but in terms of setrelations. Variables aim to capture a dimension of variation across cases and distributes cases on this variation. A set assess whether, or that what degree, a case is a member of a set and then analyses the intersection between sets. For example, a country can be a member of the set of countries with orthodox budgetary policies. Sets are theoretical constructs. The criteria for set membership are defined by the researchers and are often calibrated against an external standard. Membership in sets need not be black or white, absent or present, but can vary by the degree to which they satisfy membership criteria. In QCA, one often makes the distinction between crisp sets, which are dichotomous in nature (in or out) or fuzzy-sets, which range from 0 to 1, which allow for more fine-grained assessment of set membership. Both types of sets are applied in the different contributions in this volume. Fuzzy-sets can take different ranges across sets in analysis. For some sets one can easily work with dichotomous crisp sets. For example, firms are either certified or not, or are publicly traded or not. For other sets, such as financial performance, more fine-grained information and varying degrees of membership can be used. The assignment of set membership scores follows from the definition and operationalization of the set in question and the calibration to an external standard. Fuzzy sets can take many gradations from dichotomous to continuous and are characterized by the fact that their floor value and ceiling value has substantial meaning. In this way fuzzy sets are both quantitative as well as qualitative. Full membership to a set and full nonmembership to a set are qualitative states and assessments. In order to illustrate the difference consider, for example, the measure of yearly budget deficits and the set of countries with orthodox budget policies within the European Union in the context of new adopted rules (six pact following stability pact rules) following the financial crisis. Country A can have a deficit of 2.0%, country B of 2.7%, country C of 2.9%, country D of 3.1%, and country E of 3.2%. Although the measured variation between countries A and B is larger than between Country C and

10

PEER C. FISS ET AL.

countries D and E this variation might be far less theoretically relevant since the new rules stipulate that a maximum of 3% deficit is acceptable and concurs to budget orthodoxy while budget deficits in excess of 3% are problematic and generate a whole host of measures including sanctions. In terms of a standard variable approach, we have variation between 2% and 3.2%. In set theoretic terms, we have countries with set membership or nonmembership to the set of countries with orthodox budget policies. Nonmembership to this set can be used to explain a range of qualitative relevant outcomes such as social protests. The assessment of set membership is calibrated against an external parameter, namely the rules laid down in the stability pact. Several contributions in this volume use fuzzy sets and show how they are operationalized. While the idea of calibration is perhaps still a relatively familiar one to organizational researchers, the shift from correlations to set–subset relationships is a more demanding one to make. Correlational analysis is a standard tool of the social scientist, and in the hands of the skilled researcher they are powerful means especially of isolating the effect of individual causes. Yet, it is this very aspect that has led to the dominance of what Ragin (2008) has termed ‘‘net effects thinking,’’ that is, an analytical meta-theory in which ‘‘each independent variable is assumed to be capable of influencing the level or probability of the outcome regardless of the values or levels of other variables (i.e., regardless of the varied contexts defined by these variables)’’ (Ragin, 2008, pp. 177–178, emphasis in the original). In contrast, the configurational approach, and particular that based on QCA, places contextual effects at its very center; the effect of an individual causal condition (measured as set membership) may crucially depend on context, that is, the presence or absence of one or several other causal conditions. The configurational approach thus shifts from generally assuming additivity (and allowing for deviation from this model mainly by means of interaction terms) to a view that generally assumes interaction between elements, be it in the form of positive or negative complementarities. We should be quick to add, however, that the configurational approach does not require synergistic relationships and still allows for additive ones. However, it is fair to say that interdependence is assumed to be the norm, rather than the exception. Furthermore, it is the set–subset relationship, rather than the correlation, that more easily allows for the creation of fully interactive models of causes, as set–subset relationships are direct relationships rather than correlational tendencies and are not constrained by issues such as multicollinearity. The set–subset relationship thus allows researchers to more easily assess the particular configuration of contingencies derived from

Configurational Theory and Methods in Organizational Research: Introduction

11

the context in which every organization is located (Child, 1977; Moores & Yuen, 2001). The shift to set–subset relationship comes with another important benefit: it greatly facilitates the analysis of necessity and sufficiency in causal relationships. Both necessity and sufficiency are fundamental aspects of causation (e.g., Goertz & Levy, 2007), yet correlational analysis is not well geared towards analyzing relationships in terms of whether causes are necessary, sufficient, both, or neither. The standard pattern of ‘‘the more we observe variable X, the more we should observe outcome Y’’ that is typical of much of current theorizing and dovetails with correlational analysis in fact suggests a fairly simple pattern where a given variable is assumed to be simultaneously necessary and sufficient. Yet, this is a strong assumption to impose upon both theory and evidence. For instance, take the field of corporate governance that is examined by Bell, Aguilera & Filatotchev (2013, pp. 159–180). It would seem plausible that good governance is a necessary condition for the presence of sustained high firm performance; without it, such continued performance would likely be threatened by a host of issues. Yet, the presence of good governance by itself is not guarantee that a firm will be able to keep achieving such sustained performance; there are many well-governed firms that nevertheless fail to achieve such returns. In other words, it would appear that while good governance is a necessary condition, it is not sufficient for bringing about sustained performance. From a correlational perspective, such a pattern is problematic as it is not additive but in fact resembles a multiplicative model where the outcome would approach zero even if only one of several predictor variables approaches zero. If good governance was indeed a necessary but not sufficient condition, it would help account for the failure of corporate governance researchers to find support for a consistent relationship between governance practices such as CEO duality and performance (e.g., Dalton, Daily, Ellstrand, & Johnson, 1998). From a set theoretic point of view, however, such a pattern would be perfectly consistent with a necessary but not sufficient condition, suggesting that such a view would offer a powerful tool for analyzing relationships that are more complex than simultaneous necessity and sufficiency (Fiss, 2011). If the presence of INUS conditions – or conjunctural causation, in Ragin’s (2000) terms – is the rule rather than the exception, then the need for a shift in both theoretical accounts and empirical approaches would seem evident. Indeed, organizations would appear to be a prime field where we might witness situations of causal complexity, where causes may combine in a number of ways to bring about outcomes of interest, leading to situations

12

PEER C. FISS ET AL.

where individual causes may be neither necessary nor sufficient, although of course any other combination of the presence and or absence of necessity and sufficiency may be possible. If so, a methodology such as QCA that was developed to deal with such situations along with novel ways of theorizing these relationships may offer a promising way forward and perhaps a way to resolve a number of long-standing puzzles in organization theory including board composition, the effect of strategic planning, market share, strategic groups, or generic strategies (e.g., Nicolai & Kieser, 2002), to name but a few. There is yet another important difference associated with the shift from a correlational understanding to a configurational approach based on set theoretic understanding of the world. As we have noted, correlations are symmetric – statements such as ‘‘the more we observe variable X, the more we should observe outcome Y’’ imply that the reverse is also true; the less we observe variable X, the less we should observe the outcome, and indeed from a purely correlational view one might reverse predictor variable and outcome without affecting the correlation itself. While correlation thus is symmetric, much of our theorizing about organizations should in fact involve asymmetric causation where the set of factors that bring about an outcome may be different from the set of factors associated with the absence of the outcome; for instance, ‘‘the configurations leading to very high performance are frequently different from those leading to merely high or average performance’’ (Fiss, 2011, p. 411). The contrast here becomes perhaps even more evident when considering not so much continuous as binary outcomes. The preferred correlational tools for such situations – logit or probit regression and their derivatives – simultaneously model the presence and the absence of the outcome, making it impossible for such approaches to model the presence and the absence of the outcome separately. Consider, for instance, the large literature on the adoption of organizational practices, which has made extensive use of binary outcomes to model adoption. Yet, it would seem that the factors that predict when firms adopt a practice may be quite different from those that lead firms to abstain from adoption and not merely their inverse. For instance, the absence of sufficient slack resources or incompatibility in terms of technological, political and cultural fit (Ansari, Fiss, & Zajac, 2010) may all account for the failure to adopt, yet having resources and sufficient levels of fit may be insufficient in explaining adoption; not every firm that might adopt actually will. As we have outlined previously, one of the key issue of configurational theory relates to dealing with the issue that the number of theoretically

Configurational Theory and Methods in Organizational Research: Introduction

13

possible configurations may become overwhelming even with a relatively limited number of relevant characteristics. This means that in a truth table listing all possible configurations we will frequently or indeed usually observe a number of cells that do not contain empirical observations, which presents special problems for the analysis of such tables. In addition, the number of empirical observations may be quite small relative to the number of causal conditions, further complicating the use of standard multivariate analyses. In QCA, this is known as a situation of limited diversity (see also the contribution by Charles Ragin, 2013, pp. xv–xx). While limited diversity presents a challenge for many conventional forms of analysis, this is not the case for a set theoretic approach based on QCA. As Ragin (2008) demonstrates, the researcher can use counterfactual analysis to overcome the challenges of limited diversity, allowing the drawing of inferences based on both ‘‘easy’’ and ‘‘difficult’’ counterfactuals. For organizational researchers, the notion of limited diversity is important for at least two reasons. First, limited diversity allows researchers to detect patterns of both presence and absence within the multidimensional property space; patterns that ‘‘may offer insights by making explicit the otherwise implicit and widely shared assumptions about what design elements should or should not go together’’ (Fiss, 2007, p. 1189). Second, knowledge about the presence and absence of certain elements may for instance also be used to extrapolate to nonexistent configurations, thus offering a strategy for extending configurational thinking from the existing empirical universe to the world of the possible. Finally, the configurational approach carries particular relevance for organization studies because it focuses our attention on the concept of equifinality, that is, the notion that ‘‘a system can reach the same final state from different initial conditions and by a variety of different paths’’ (Katz & Kahn, 1978, p. 30). In other words different causal paths may lead to the same outcome. This implies that a set theoretic configurational approach develops a conception of causality that allows for complexity (Ragin, 1987, 2008). In early contributions Charles Ragin referred to multiple conjunctural causation which means, first of all, that it is most often a combination of explanatory sets that eventually produces an outcome. Secondly, several different combinations of sets may produce the same outcome. Thirdly depending on the interaction with other sets, a given set may very well have a different impact on the outcome. This notion of equifinal configurations presents both a challenge and an opportunity for organization theory and therefore has been the focus of a number of recent works (e.g., Doty et al., 1993; Fiss, 2007;

14

PEER C. FISS ET AL.

Gresov & Drazin, 1997; Payne, 2006; Marlin et al., 2007). Equifinality in organizations may for instance arise when different structural design alternatives are available to deal with environmental contingencies, resulting in the same functional effect (Gresov & Drazin, 1997). The notion of equifinality hence accounts for the persistence of a variety of design choices that can lead to the desired outcome, making it a key yet undertheorized element of organization theory (e.g., Ashmos & Huber, 1987; Short, Payne, & Ketchen, 2008). The set theoretic configurational approach we have outlined here again differs from standard methods and theorizing in that equifinality is assumed to be the norm rather than the exception. From a methodological point of view, both crisp and fuzzy set QCA are specifically geared towards helping the researcher identify not only equifinal configurations but also provide measures of their empirical importance in terms of the coverage measure, which in essence describes how empirically important each of the equifinal configurations is. Further, it allows for the analysis of both first and second order equifinality, where first-order equifinality refers to equifinal types that exhibit different core characteristics (e.g., type A vs. type B), while second-order equifinality refers to neutral permutations within a given firstorder equifinal type (e.g., type A1 vs. A2 y An) (Fiss, 2011, p. 398). As such, the set theoretic approach provides us with tools for a more fine-grained and complex analysis of equifinality that goes beyond merely identifying its existence and towards allowing the researcher to identify its specific nature and significance. Indeed, set theoretic approaches allow researchers to determine the degree of explanatory parsimony or complexity they want to achieve.

OUTLINE OF THE VOLUME The purpose of our introduction has been twofold: to locate our contribution within the broader field of organization studies and to introduce some of the key themes that differentiate the approach taken here from previous work on organizational configurations. Yet, the role of an introduction is to set the stage for the main contribution, which is offered by the following chapters. Chapter 2 by Marx, Cambre´, and Rihoux (2013, pp. 23–47) on ‘‘Crisp-Set Qualitative Comparative Analysis in Organizational Studies’’ starts with a stylized presentation of two dominant research strategies, case-based

Configurational Theory and Methods in Organizational Research: Introduction

15

research and variable-based research, and how crisp-set QCA relates to them. Subsequently, the authors further introduce crisp-set QCA as a step-wise approach and discuss its different applications in organization studies. The chapter then turns to a presentation of some distinctive strengths of the approach which include the reduction of complexity by pooling cases together through the use of truth tables, achieving parsimony through minimization, analyzing causal complexity and using different sorts of data. Finally, Marx and co-authors discuss the main criticisms that have been raised with regard to crisp-set QCA and propose some solutions. Chapter 3, ‘‘The Two QCAs: From a Small-N to a Large-N Set Theoretic Approach’’ by Greckhamer, Misangyi, and Fiss (2013, pp. 49–75), aims to provide guidance to prospective researchers interested in opening up QCA’s potential for widespread use in organization studies involving large-N settings, both as an alternative and as a complement to conventional regression analyses. For this purpose, they compare small-N and large-N QCA with respect to theoretical assumptions and objectives, processes and decisions involved in building the causal model, selecting the sample as well as analyzing the data and interpreting the analytical results. Chapter 4 on ‘‘Configurational Analysis and Organization Design: Towards a Theory of Structural Heterogeneity’’ by Grandori and Furnari (2013, pp. 77–105) reconstructs the roots, evolution and some prospects of configurational analysis in organization theory and organizational economics. First the chapter reveals the presence of elements of configurational analysis on many organization theory and organizational economics approaches. Secondly, the authors identify ‘‘structural heterogeneity’’ as an organizational property that can be distinctively studied by the configurational analysis. They then further elaborate and substantiate this notion using an empirical analysis of a multisector sample of firms. Chapter 5 by Hak, Jaspers, and Dul (2013, pp. 107–127) on the ‘‘The Analysis of Temporally Ordered Configurations: Challenges and Solutions’’ focuses on a specific application of configurational methods in the context of analyzing processes, that is, a complex of activities that unfolds over time. In this context the order in which conditions appear in a configuration is of key-importance. In order to capture this, the authors develop the idea of temporally ordered configurations which can be defined as those configurations in which conditions occur in a specific temporal order. The chapter illustrates the aims, characteristics, and limitations of approaches that have been proposed as tools for the analysis of temporal order with an example. After discussing several approaches that deal with temporal order the

16

PEER C. FISS ET AL.

authors introduce an alternative approach, Temporal Necessary Condition Analysis (TNCA). Chapter 6 by Jackson and Ni (2013, pp. 129–158), entitled ‘‘Understanding Complementarities as Organizational Configurations: Using Set Theoretical Methods,’’ reviews the emerging literature on complementarities to identify a series of conceptual challenges related to understanding complementarities as organizational configurations, and examines the methodological challenges in studying how such elements combine to produce joint effects on performance. The chapter argues that new set theoretic methods using QCA may present a very useful methodological alternative to studying complementarities. The authors illustrate this potential by re-analyzing past work by Aoki, Jackson, and Miyajima (2007) on relationships between ownership structure, board structure, and employment practices of listed firms in Japan to show evidence of complementarities associated with hybrid configurations that combine market and relational forms of organization. Chapter 7 by Bell, Aguilera, and Filatotchev (2013, pp. 159–180) on ‘‘Corporate Governance and Configuration Research: The Case of Foreign IPOs Listing in London’’ applies configurational methods to show how firm-level governance practices interact with informational asymmetries associated with a firm’s industry. By examining foreign Initial Public Offerings that have chosen to list on London stock exchanges, the authors demonstrate that an assessment of the firm-level corporate governance configurations is incomplete without taking into account the firm’s industry affiliation. Their use of fsQCA underscores the possibilities configurational approaches have in advancing theories of corporate governance. In Chapter 8 on ‘‘Corporate Social Responsibility: A Multilevel Explanation of Why Managers Do Good,’’ Crilly (2013, pp. 181–204) investigates the multilevel essentials of managerial behavior. Managers frequently confront dilemmas where maximizing shareholder value is incompatible with enhancing social welfare. Most explanations of responses to these dilemmas center on a single level of analysis and a single discipline. The novel approach that he suggests is to simultaneously study individual characteristics of managers and the social context in which they and their organization exist. Using a fuzzy set QCA using data on 335 managers of overseas subsidiaries of three multinational firms headquartered in the Netherlands, Crilly identifies the combined influence of effects at multiple level of analysis in explaining how managers respond to pressures for social responsibility. Park and El Sawy (2013, pp. 205–224) deal with the topic of digital business strategy in Chapter 9, on ‘‘The Value of Configurational Approaches for

Configurational Theory and Methods in Organizational Research: Introduction

17

Studying Digital Business Strategy.’’ They show how configurational approaches can help us to better understand the complex phenomenon of digital ecodynamics. They argue that configurational approaches are especially helpful as in inquiring system for exploring the holistic nature of digital ecodynamics. Support for their arguments comes from an empirical study that explores how IT systems, organizational dynamic capability, and environmental turbulence simultaneously and systematically combine to create competitive firm performance. They apply fuzzy set QCA to a sample of 106 Korean firms, showing how configurational approaches can create new practical insights in digital ecodynamics by offering multiple strategic options to organizations. Chapter 10 by Raab, Lemaire, and Provan (2013, pp. 225–253) explores ‘‘The Configurational Approach in Organizational Network Research.’’ It explains how a configurational approach and set theoretic methods can contribute to a deeper and more nuanced understanding of organizational networks and network relations. This is especially true for the study of ‘‘whole networks’’ of organizations where data collection difficulties often limit the sample size. The authors present two empirical examples of current research on whole networks, demonstrating how QCA can be used to analyze organizational networks. They then discuss the methodological and theoretical implications of the configurational approach for future organizational network research. Chapter 11 by Pajunen and Airo (2013, pp. 255–278) on ‘‘CountrySpecificity and Industry Performance: A Configurational Analysis of the European Generic Medicines Industry,’’ links the configurational approach with the topic of institutional complementarities. The identification of country specific advantages for business activities is one of the most crucial issues of strategic management and international business literatures. The authors address this issue by examining location specific conditions for a successful generic medicines industry within 24 European countries. The findings of their fuzzy-set QCA show that there are no necessary conditions for the high performance or absence of the high performance industry. By revealing the causal complexity related to the issue, however, they show that the high performance or lack of it results from a configuration of conditions. Specifically, Pajunen and Airo identify two sufficient causal configurations to both outcomes. These findings provide clear implications for generic medicines industry firms who are planning location choices of their operations. In addition, this study provides a methodological advancement to explain and understand which country elements matter more, for what outcomes, and under what conditions.

18

PEER C. FISS ET AL.

Whittington, McKee, Goodwin, and Bell (2013, pp. 279–302) aim to blaze a novel path for leadership research in Chapter 12 by ‘‘Applying Fuzzy Set Methodology to Evaluate Substitutes for Leadership.’’ They start from the finding that transformational leadership has been found to positively influence employee attitudes and behaviors. However, research also has shown that a variety of task and motivational factors lead to similar outcomes. Yet, little research has explored the potential interaction of transformational leadership with these other factors. The authors utilize fuzzy set QCA to explore the ways these factors may interact to produce positive employee outcomes. Specifically, they find that high levels of employee commitment and performance can be achieved in the absence of a transformational leader through various ‘‘bundles’’ of enriched jobs, challenging goals, and high quality leader–follower relationships. Our volume closes with two final chapters. First, in a response piece entitled ‘‘We Try Harder: Some Reflections on Configurational Theory and Methods,’’ David Ketchen (2013, pp. 303–309) provides a view from the vantage point of an eminent organizational scholar who has long worked in the field of configurational research. His thoughtful reflections provide insight in both the goals of the individual chapters and the gist of the overall volume. Finally, in our conclusion, we offer some further thoughts on the way forward for configurational theory and methods in organization studies. This volume – a transatlantic collaboration in terms of both the editors and the authors – would not have come about without a dedicated set of scholars that generously offered their time and energy in creating what we believe is a terrific set of contributions. It is their insights that make or break a work such as this one. We would also like to thank the editor of Research in the Sociology of Organizations, Michael Lounsbury, for his encouragement and the gracious invitation to begin the conversations that eventually led to this volume. Lastly, we would be amiss if we did not acknowledge that all of us are indebted to Charles Ragin, whose path breaking work is an inspiration and in many ways remains the starting point of this book.

NOTE 1. A note on the use of terms. Throughout this chapter and the other chapters in this volume we use the term conditions for the explanans and outcome for explanandum. This corresponds to a degree to the notions of independent and dependent variables as used in more conventional methodological approaches.

Configurational Theory and Methods in Organizational Research: Introduction

19

However, as we argue in this chapter, and has been argued by Charles Ragin (1987, 2008) extensively, there are also significant ontological differences since a settheoretic configurational approach departs from the notion of independently operating variables.

REFERENCES Ackoff, R. L. (1981). Creating the corporate future: Plan or be planned for. New York, NY: Wiley. Ansari, S. M., Fiss, P., & Zajac, E. (2010). Made to fit: How practices vary as they diffuse. Academy of Management Review, 35, 67–92. Aoki, M., Jackson, G., & Miyajima, H. (2007). Corporate governance in Japan: Institutional change and organizational diversity. Oxford: Oxford University Press. Ashmos, D. P., & Huber, G. P. (1987). The systems paradigm in organization theory: Correcting the record and suggesting the future. Academy of Management Review, 12, 607–621. Bell, G. R., Aguilera, R. V., & Filatotchev, I. (2013). Corporate governance and configuration research: The case of foreign IPOs listing in London. In P. C. Fiss, B. Cambre´ & A. Marx (Eds.), Configurational theory and methods in organizational research. Research in the Sociology of Organizations. Bingley, UK: Emerald Group Publishing. Child, J. (1972). Organization structure and strategies of control: A replication of the Aston study. Administrative Science Quarterly, 17, 163–177. Child, J. (1977). Organizations: A guide to problems and practice. New York, NY: Harper & Row. Child, J. (2002). A configurational analysis of international joint ventures: Drawing upon experience in China. Organization Studies, 23, 781–815. Crilly, D. (2013). Corporate social responsibility: A multilevel explanation of why managers do good. In P. C. Fiss, B. Cambre´ & A. Marx (Eds.), Configurational theory and methods in organizational research. Research in the Sociology of Organizations. Bingley, UK: Emerald Group Publishing. Dalton, D. R., Daily, C. M., Ellstrand, A. E., & Johnson, J. L. (1998). Meta-analytic reviews of board composition, leadership structure, and financial performance. Strategic Management Journal, 19, 269–290. Desarbo, W. S., Di Benedetto, C. A., Song, M., & Sinha, I. (2005). Revisiting the miles and snow strategic framework: Uncovering interrelationships between strategic types, capabilities, environmental uncertainty, and firm performance. Strategic Management Journal, 26, 47–74. Doty, D. H., Glick, W. H., & Huber, G. P. (1993). Fit, equifinality, and organizational effectiveness: A test of two configurational theories. Academy of Management Journal, 36, 1196–1250. Fiss, P. C. (2007). A set-theoretic approach to organizational configurations. Academy of Management Review, 32, 1180–1198. Fiss, P. C. (2009). Case studies and the configurational analysis of organizational phenomena. In C. Ragin & D. Byrne (Eds.), Handbook of case study methods (pp. 424–440). Thousand Oaks, CA: Sage.

20

PEER C. FISS ET AL.

Fiss, P. C. (2011). Building better causal theories: A fuzzy set approach to typologies in organization research. Academy of Management Journal, 54, 393–420. Fombrun, C. J. (1989). Convergent dynamics in the production of organizational configurations. Journal of Management Studies, 26, 439–458. Goertz, G., & Levy, J. S. (2007). Causal explanation, necessary conditions, and case studies. In G. Goertz & J. S. Levy (Eds.), Explaining war and peace: Case studies and necessary condition counterfactuals (pp. 9–45). New York, NY: Routledge. Grandori, A., & Furnari, S. (2013). Configurational analysis and organization design: Towards a theory of structural heterogeneity. In P. C. Fiss, B. Cambre´ & A. Marx (Eds.), Configurational theory and methods in organizational research. Research in the Sociology of Organizations. Bingley, UK: Emerald Group Publishing. Greckhamer, T., Misangyi, V., & Fiss, P. C. (2013). The two QCAs: From a small-N to a large-N set theoretic approach. In P. C. Fiss, B. Cambre´ & A. Marx (Eds.), Configurational theory and methods in organizational research. Research in the Sociology of Organizations. Bingley, UK: Emerald Group Publishing. Gresov, C., & Drazin, R. (1997). Equifinality: Functional equivalence in organization design. Academy of Management Review, 22, 403–428. Hak, T., Jaspers, F., & Dul, J. (2013). The analysis of temporally ordered configurations: Challenges and solutions. In P. C. Fiss, B. Cambre´ & A. Marx (Eds.), Configurational theory and methods in organizational research. Research in the Sociology of Organizations. Bingley, UK: Emerald Group Publishing. Jackson, G., & Ni, N. (2013). Understanding complementarities as organizational configurations: Using set theoretical methods. In P. C. Fiss, B. Cambre´ & A. Marx (Eds.), Configurational theory and methods in organizational research. Research in the Sociology of Organizations. Bingley, UK: Emerald Group Publishing. Katz, D., & Kahn, R. L. (1978). The social psychology of organizations (2nd ed.). New York, NY: Wiley. Ketchen, D. J., Jr., (2013). We try harder: Some reflections on configurational theory and methods. In P. C. Fiss, B. Cambre´ & A. Marx (Eds.), Configurational theory and methods in organizational research. Research in the Sociology of Organizations. Bingley, UK: Emerald Group Publishing. Ketchen, D. J., Thomas, J. B., & Snow, C. C. (1993). Organizational configurations and performance: A comparison of theoretical approaches. Academy of Management Journal, 36, 1278–1313. Kuhn, T. S. (1970). The structure of scientific revolutions. Chicago, IL: Chicago University. Lawrence, P. R., & Lorsch, J. W. (1967). Organization and environment: Managing differentiation and integration. Boston, MA: Harvard University Press. Leach, E. R. (1961). Rethinking anthropology. New York, NY: Humanities Press. Levinthal, D. A. (1997). Adaptation on rugged landscapes. Management Science, 43, 934–950. Mackie, J. L. (1974). The cement of the universe: A study of causation. Oxford: Clarendon Press. Marlin, D., Ketchen, D. J., & Lamont, B. (2007). Equifinality and the strategic groupperformance relationship. Journal of Managerial, 19, 208–232. Marx, A., Cambre´, B., & Rihoux, B. (2013). Crisp-set qualitative comparative analysis in organizational studies. In P. C. Fiss, B. Cambre´ & A. Marx (Eds.), Configurational theory and methods in organizational research. Research in the Sociology of Organizations. Bingley, UK: Emerald Group Publishing.

Configurational Theory and Methods in Organizational Research: Introduction

21

Marx, A., Rihoux, B., & Ragin C. (2013). The origins, development and application of qualitative comparative analysis (QCA): The first 25 years. European Political Science Review. doi: http://dx.doi.org/10.1017/S1755773912000318 [This article was published online on February 22, 2013]. Meyer, A. D., Tsui, A. S., & Hinings, C. R. (1993). Configurational approaches to organizational analysis. Academy of Management Journal, 36, 1175–1195. Miles, R. E., & Snow, C. C. (1978). Organizational strategy, structure, and process [A. D. Meyer, collaborator; H. J. Coleman Jr., contributor]. New York, NY: McGraw Hill. Miller, D. (1981). Toward a new contingency approach: The search for organizational gestalts. Journal of Management Studies, 18, 1–27. Miller, D. (1986). Configurations of strategy and structure: A synthesis. Strategic Management Journal, 7, 233–249. Miller, D. (1990). Organizational configurations: Cohesion, change, and prediction. Human Relations, 43, 771–789. Miller, D. (1996). Configurations revisited. Strategic Management Journal, 17, 505–512. Miller, D., & Friesen, P. H. (1984). Organizations: A quantum view. Englewood Cliffs, NJ: Prentice-Hall. Mintzberg, H. (1983). Structures in fives: Designing effective organizations. Englewood Cliffs, NJ: Prentice-Hall. Moores, K., & Yuen, S. (2001). Management accounting systems and organizational configurations: A life-cycle perspective. Accounting, Organizations, and Society, 26, 351–389. Nicolai, A., & Kieser, A. (2002). Trotz eklatanter Erfolglosigkeit: Die Erfolgsfaktorenforschung weiter auf Erfolgskurs. Die Betriebswirtschaft (DBW), 62, 579–596. Park, Y., & El Sawy, O. A. (2013). The value of configurational approaches for studying digital business strategy. In P. C. Fiss, B. Cambre´ & A. Marx (Eds.), Configurational theory and methods in organizational research. Research in the Sociology of Organizations. Bingley, UK: Emerald Group Publishing. Pajunen, K., & Airo, V. (2013). Country-specificity and industry performance: A configurational analysis of the European generic medicines industry. In P. C. Fiss, B. Cambre´ & A. Marx (Eds.), Configurational theory and methods in organizational research. Research in the Sociology of Organizations. Bingley, UK: Emerald Group Publishing. Payne, G. T. (2006). Examining configurations and firm performance in a suboptimal equifinality context. Organization Science, 17, 756–770. Pinder, C. C., & Moore, L. F. (1979). The resurrection of taxonomy to aid the development of middle range theories of organizational behavior. Administrative Science Quarterly, 24, 99–118. Raab, J., Lemaire, R. H., & Provan, K. G. (2013). The configurational approach in organizational network research. In P. C. Fiss, B. Cambre´ & A. Marx (Eds.), Configurational theory and methods in organizational research. Research in the Sociology of Organizations. Bingley, UK: Emerald Group Publishing. Ragin, C. C. (1987). The comparative method: Moving beyond qualitative and quantitative strategies. Berkeley, CA: University of California Press. Ragin, C. C. (1994). Constructing social research: The unity and diversity of method. Thousand Oaks, CA: Pine Forge Press. Ragin, C. C. (2000). Fuzzy set social science. Chicago, IL: University of Chicago Press. Ragin, C. C. (2008). Redesigning social inquiry: Fuzzy sets and beyond. Chicago, IL: University of Chicago Press.

22

PEER C. FISS ET AL.

Ragin, C. C. (2013). Foreword: The distinctiveness of configurational research. In P. C. Fiss, B. Cambre´ & A. Marx (Eds.), Configurational theory and methods in organizational research. Research in the Sociology of Organizations. Bingley, UK: Emerald Group Publishing. Rihoux, B., & Marx, A. (2013). Qualitative comparative analysis at 25: State of play and agenda. Political Research Quarterly 66, 167–170. Schneider, C. Q., & Wagemann, C. (2012). Set-theoretic methods for the social sciences: A guide to qualitative comparative analysis. Cambridge: Cambridge University Press. Short, J. C., Payne, G. T., & Ketchen, D. J. (2008). Research on organizational configurations: Past accomplishments and future challenges. Journal of Management, 34, 1053–1079. Siggelkow, N. (2001). Change in the presence of fit: The rise, the fall, and the renaissance of Liz Claiborne. Academy of Management Journal, 49, 838–857. Siggelkow, N. (2002). Evolution towards fit. Administrative Science Quarterly, 47, 25–159. Sørensen, J. B., Van Maanen, J., & Mitchell, T. R. (2007). The interplay of theory and methods. Academy of Management Review, 32, 1145–1154. Suddaby, R., Hardy, C., & Huy, Q. (2011). Where are the new theories of organizations? Academy of Management Review, 36, 236–246. Weber, M. (1922[1978]). Economy and society: An outline of interpretive sociology. Berkeley, CA: University of California Press. Zadeh, L. (1972). A fuzzy-set-theoretical interpretation of linguistic hedges. Journal of Cybernetics, 2, 4–34.

CHAPTER 2 CRISP-SET QUALITATIVE COMPARATIVE ANALYSIS IN ORGANIZATIONAL STUDIES Axel Marx, Bart Cambre´ and Benoıˆ t Rihoux ABSTRACT Qualitative Comparative Analysis (QCA), initiated by Charles C. Ragin, is a research strategy with distinctive added value for organization studies. QCA constitutes in essence two configurational approaches, each grounded in set theory. One approach uses crisp-sets (dichotomous variables) to analyze cases. The other approach uses fuzzy-sets. While the use of fuzzy-sets has been increasing over the last few years, the crisp-set (csQCA) approach is still used in a majority of empirical applications. This chapter discusses in-depth the application of csQCA in organization studies. This chapter starts with a stylized presentation of two dominant research strategies, case-based research and variablebased research, and how csQCA relates to them. Subsequently, csQCA is further introduced and the different applications in organization studies are discussed. This section ends with a brief step-wise ‘‘how to’’ presentation. The chapter then turns to a presentation of the main distinctive strengths of the approach. In the final part, the chapter

Configurational Theory and Methods in Organizational Research Research in the Sociology of Organizations, Volume 38, 23–47 Copyright r 2013 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0733-558X/doi:10.1108/S0733-558X(2013)0000038006

23

24

AXEL MARX ET AL.

discusses extensively the main criticisms which have been raised with regard to (cs)QCA and draws out some of the main implications of this discussion. Keywords: Qualitative Comparative Analysis (QCA); comparative methods; case methods; organization studies; management

INTRODUCTION Qualitative Comparative Analysis (QCA), initiated by Charles C. Ragin, constitutes a research strategy with distinctive added value for organization studies (Fiss, 2007; Marx, Rihoux, & Ragin, 2013; Ragin, 1994, 2000, 2008a; Rihoux & Ragin, 2009). Ragin developed QCA, a comparative case-oriented research technique based on Boolean algebra and set theory, for the analysis of a medium number of cases. The broader aim was also to develop a new research approach which combined some strengths of qualitative (or ‘‘caseoriented’’) and quantitative (or ‘‘variable-oriented’’) research methods (Ragin, 1987). In an era of increased attention to case studies in many social sciences (Box-Steffensmeier, Brady, & Collier, 2008, 2004; George & Bennett, 2005; Gerring, 2007; Poteete, Janssen, & Ostrom, 2010) QCA holds the potential to provide a unique set of tools to systematically examine similarities and differences of a set of comparable cases and identify structural conditions that lead to an outcome. QCA is specifically suited to analyze aggregate units of analysis such as organizations. Consequently, this research approach corresponds to recent pleas in organization and management journals to perform more types of configurational analysis instead of variance-based analysis (Fiss, 2011; Kogut, 2000). QCA constitutes in essence two configurational approaches each grounded in set theory. One approach uses crisp-sets (dichotomous variables) to analyze cases. The other approach uses fuzzy-sets (introduced in the next chapter). While the use of fuzzy-sets has been increasing over the last few years, the first approach to QCA, outlined in Ragin’s seminal book The Comparative Method (1987) was based on the use of crisp sets, and crisp set QCA (csQCA) is still used in a majority of empirical applications (Rihoux, Alamos, Bol, Marx, & Rezsohazy, 2013), as is shown in Fig. 1. Although one can see a clear and recent increase in the number of fsQCA articles in management and organization studies, articles with a crisp set approach still remain significant and even dominant.

25

Crisp-Set QCA in Organizational Studies 60

50

40

30

20

10

0 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 19 19 19 19 19 20 20 20 20 20 20 20 20 20 20 20 20

mvQCA

fsQCA

csQCA

Fig. 1. Empirical QCA-applications Published in Management and Organization Studies Journals. Source: Authors’ calculations on the basis of the COMPASSS website.

This chapter starts with a stylized presentation of two dominant research strategies, case-based research and variable-based research, and how csQCA relates to them. Subsequently, csQCA is further introduced and the different applications in organization studies are discussed. This section ends with a brief step-wise ‘‘how to’’ presentation. The chapter then turns to a presentation of the main distinctive strengths of the approach. In the fourth part, the chapter discusses extensively the main criticisms which have been raised with regard to (cs)QCA.

DIFFERENT RESEARCH STRATEGIES Scholars studying organizations often use either a case-oriented (qualitative) approach or a variable-oriented (quantitative) approach for the study of

26

AXEL MARX ET AL.

organizations. Case-oriented strategies are distinctive in that they are centrally concerned with making sense of how all relevant aspects of a case combine in order to generate a certain outcome (Ragin & Becker, 1992). In these strategies, cases are selected because they are substantively and/or theoretically significant (Eisenhardt, 1989; Gerring, 2007). Case research has been extensively used in organizations studies (Van Maanen, 1998). Caseoriented researchers in organization studies typically want to answer ‘‘how’’questions (Yin, 1994). Therefore the observation or case is crucial, as the observed relationships actually occur in each case under investigation and one can only understand the observed relationships in the case-specific context (Ragin, 1987, 1994, 2000; Ragin & Becker, 1992). Variable-oriented research strategies, by contrast, are generally concerned with the problem of assessing the relationship between a limited number of aspects of cases (or variables) across a large number of cases or observations, usually pursuing the goal of inferring general patterns between aspects (or variables) of cases (observations) that hold for a population (Ragin, 2000). Variable-oriented researchers typically want to assess the effect from one variable on the other. For example, they want to know what the effect is of firm size or age on survival of firms (Carroll & Hannan, 1999), the effect of network position on careers (Burt, 1992) or the effect of distribution of ownership shares on the performance of joint ventures (Child, 2002). In this research strategy, the individual case as such is not important. The conclusions or observed relationships are based on probabilities and not on fixed effects within each case. A variable-oriented research strategy does not aim to explain an outcome for every case. These two research strategies have been tremendously successful in their own right, but also display shortcomings, some of which QCA aims to overcome. Single case research often results in a thick description of cases without the possibility of testing theories. This is the problem of indeterminate research designs; one cannot make more causal inferences than one has observations. It is impossible to decide which causal hypotheses are true in a single case-design, because ‘‘each observation can help us make one inference at most’’ (King et al., 1994, p. 119). As a result, a research design based on less cases/observations than explanatory variables can only be used to develop a causal hypothesis. However, it does not enable one to test this hypothesis. In other words, single case studies are important to develop models, but not to test models or explore diversity within research populations. A second problem related to single case-studies or studies of a few cases concerns selection bias. In order to make generalizations from

Crisp-Set QCA in Organizational Studies

27

cases one has to know how representative they are. This is often a difficulty in case research: admittedly many case researchers are not concerned with regard to the external validity of their research, especially not with regard to some form of ‘‘universal’’ generalization. There is no general theory of case selection in the context of case study research and many different strategies are used (see Miles & Huberman, 1994, pp. 27–34; Gerring, 2007). This makes it difficult to relate the research results to a wider population from which the case was selected. The latter is especially relevant in organization studies. Several researchers focus on a few cases to learn more about a wider set of cases and in order to develop more middle-range types of theories. The testing of models (applicability of a theory to a wider population) and the careful, mostly random, selection of cases is often performed in variableoriented research. However, this research strategy also displays some limitations. Variable-oriented research usually does not allow researchers to reach an in-depth understanding of the cases under investigation. As a result, variable-oriented research is only applicable to a certain subset of research questions relating to co-variation of variables. Secondly, variableoriented research sometimes generates inconsistent results. For example, different studies on the relationship between corporate social performance and financial performance produce different results. Some studies argue that corporate social responsibility has no effect on financial performance. Other studies argue that it has a negative effect. Still other studies argue that it has a positive effect. This is often explained by the fact that the mechanisms at play between variables are ill-understood and can differ in relation to the configuration in which they occur. In other words, depending on the interaction with other factors a certain variable can have a different effect. This has recently resulted in the use of configurational approaches in management research. In essence, ‘‘a configurational approach suggests that organizations are best understood as clusters of interconnected structures and practices, rather than as modular or loosely coupled entities whose components can be understood in isolation’’ (Fiss, 2007, p. 1180). John Child (2002), for example, uses a configurational approach to analyze the relationship between distribution of ownership shares control and performance in international joint ventures, explicitly arguing that a variance (variable-oriented) approach overlooks inter-case nuances and important interaction effects between variables. QCA was developed to overcome some of the problems of a case-oriented and variable-oriented approach.

28

AXEL MARX ET AL.

INTRODUCING CRISP-SET QUALITATIVE COMPARATIVE ANALYSIS (csQCA) The aim of csQCA was to develop a research strategy which develops a middle road between the case-oriented (qualitative) and the variable-oriented (quantitative) approaches (Ragin, 1987, p. 12ff, 1991, 1997). The goal of this systematic comparative case strategy is to ‘‘integrate the best features of the case-oriented approach with the best features of the variable-oriented approach’’ (Ragin, 1987, p. 84). This approach consists of four central features (Ragin, 1987, 1994, 2000; Rihoux, 2008; Rihoux & Ragin, 2009). First, it is a case approach which implies that each individual case is considered as a complex entity (a whole – a configuration of conditions/ variables) which needs to be comprehended and which should not be forgotten in the course of the analysis. Different parts of each case are understood in relation to one another and in terms of the total picture that they form together as a case. Cases in this context are regarded as configurations of conditions/variables. Second, it is comparative in the sense that it explores and finds similarities and differences across comparable cases by comparing configurations of conditions. Third, it allows for equifinality or multiple conjunctural causation (Ragin, 1987, 2000; Rihoux, 2003; Maggetti, 2007). This implies that: (1) most often, it is a combination of conditions that produces a phenomenon, the outcome; (2) several different combinations of conditions may produce the same outcome; and (3) depending on the context, on the ‘‘conjuncture,’’ a given condition may very well have a different impact on the outcome, that is, in some configurations the presence of a condition might generate an outcome, in other configurations the absence of the same condition might generate an outcome. This implies that different causal paths – each path being relevant in a distinct way – may lead to the same outcome. (Berg-Schlosser et al., 2009, pp. 8–10; De Meur & Rihoux, 2002, pp. 28–30). Fourth, it is systematic in the sense that it uses a formal logic to compare cases, explore causal diversity, and reduce the wealth of case information. The analytical procedure which facilitates this diversity-analysis is Boolean logic. Boolean analysis allows one to identify causal regularities that are parsimonious, that is, which combine the fewest possible conditions within a set of conditions that are considered in an analysis. Although one can observe a diversity in ways csQCA is applied, in general performing a csQCA analysis requires nine distinct steps as is illustrated in Box 1 (see also Rihoux & De Meur, 2009; Rihoux & Lobe, 2009).

Crisp-Set QCA in Organizational Studies

Box 1. Step-Wise Approach to csQCA. 1. Decide what outcome needs to be investigated. 2. Define the research population and select the cases for analysis with sufficient variation on the outcome. Several case-selection strategies are available (see Gerring, 2007). However, for a csQCA analysis researchers are advised to use a Most Similar Different Outcome (MSDO)-design (Przeworski & Teune, 1970). 3. List the most significant conditions, other than the scoping conditions used to define the research population, which might contribute to an explanation of the outcome. Several condition selection strategies can be used (Amenta & Poulsen, 1994; Berg-Schlosser & De Meur, 2009). 4. Define each condition and outcome as a binary condition. In a csQCA analysis both the presence and absence of a condition or outcome are meaningful. As a result, each explanatory factor is discussed and operationalized as a crisp-set condition which will be used in a csQCA analysis (binary code 1 or 0). This implies that for each case an explanatory condition is coded 1 if the condition is present for that case and 0 if the condition is absent in that case. Provide transparent measurement procedures for coding condition as either being absent or present. For some conditions this is straightforward. For other conditions, with more variation, there might be more discussion. In this case other csQCA-related techniques such as fuzzy sets and mvQCA are available (Cronqvist & Berg-Schlosser, 2009; Rihoux & Ragin, 2009). 5. Code each condition for each case and bring this information together in a data matrix. 6. Analyze the data matrix by specifying an explanatory model and resolving contradictions through the development of a truth table. Rihoux and De Meur (2009, pp. 48–56) discuss extensively several strategies to resolve contradictions. 7. Analyze the model with no contradictions and generate the most parsimonious explanation on the basis of the minimization procedure which is available in csQCA. 8. Analyze the presence of necessary conditions (or configurations of conditions). csQCA software allows researchers to analyze the presence of necessary (configurations) of conditions. A necessary condition is a condition which is present in all (or almost all)

29

30

AXEL MARX ET AL.

cases where the outcome is present (Ragin, 2000, p. 203; see also Eliason & Stryker, 2009). The analysis of necessary conditions has received increased attention in the social sciences (Goertz & Starr, 2003; Goertz, 2006a, 2006b; Dul & Hak, 2008). 9. Interpret the resulting explanatory models, both models which explain the presence of an outcome and as the models that explain the absence of an outcome. This often requires a return to the cases in order to find out how the configuration of the explanatory conditions lead to the outcome, that is, unpack the dynamics of cases (Rihoux & Lobe, 2009). The last step allows researchers to identify mechanisms which link explanatory conditions to an outcome.

STRENGTHS OF csQCA AS A RESEARCH STRATEGY Throughout different applications one can observe at least four distinctive strengths of applying a comparative case analysis with csQCA. These include the reduction of complexity by pooling cases together through the use of truth tables, achieving parsimony through minimization, analyzing causal complexity, and using different sorts of data. Each is elaborated in turn.

Reducing Complexity through the Truth Table A systematic comparison of cases allows researchers to reduce complexity and find general patterns in a limited number of cases. In this way, csQCA allows one to move away from full complexity (a description of each particular case with its own idiosyncrancies) to a more parsimonious explanation. The reduction of complexity occurs in two distinct steps: through the use of truth table and through the use of minimization procedures. First of all, csQCA allows researchers to pool cases together in identical ‘‘configurations.’’ Technically this is done via the creation of truth tables which list all theoretical possible combinations (2K where K= number of conditions) of configurations. In case of five conditions and one outcome, a truth table consists of 32 rows (i.e., 25). Each case is placed in one row. A row can contain several cases or none. Table 1 shows an example, based on Bakker et al. (2011). These authors studied the knowledge transfer from temporary project teams to the

31

Crisp-Set QCA in Organizational Studies

Table 1. Truth Table with Five Conditions and Outcome. Config.

R

C

T

A

M

Z

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0

1 1 1 1 0 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1 0 0 0 0

1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0

1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0

A B,C  D  E,F,G,H   I   J,K   L  M N O   P,Q   R,S       

Source: Bakker et al. (2011).

parent organizations. Based on theory, five conditions were studied: the relational (R), cognitive (C) and temporal (T) embeddedness of the project team, the absorptive capacity of the parent organization (A) and the motivation of the project team to exchange knowledge (M). The outcome (Z) was successful knowledge transfer from project team to parent organization. Five conditions indicate 25 possible configurations. In column Z (outcome), the cases (here described by letters) can be listed.

32

AXEL MARX ET AL.

This truth table (Table 1) allows us to synthesize the evidence substantially: out of 32 possible configurations (i.e., 25), only 12 actually contain observed cases. Hence, knowledge transfer does not happen in a ‘‘pure random’’ way, but is more likely to happen due to some configurations of causal conditions.

Achieving Parsimony through Minimization Besides reducing complexity through truth tables, csQCA also allows researchers to reduce complexity further and achieve maximum parsimony via minimizing configurations. The difference between reducing complexity and achieving full parsimony is not only one of degree but is fundamentally linked to the research goals. Researchers might be interested in how cases pool together without being interested in further reducing complexity via minimization. The minimization procedure is based on the following procedure: if two Boolean expressions differ in only one causal condition, yet produce the same outcome, then the causal condition that distinguishes the two expressions can be considered irrelevant and can be removed to create a simpler, combined expression. This is done by systematically comparing all cases. In theory, with multiple case studies, a researcher needs to perform N(N1)/2 paired comparisons if one intends to systematically compare all cases. In the example provided this would imply 136 paired comparisons. If one would analyze 50 cases one should in total perform 1,225 paired comparisons. It is impossible to keep track of this without software tools and an analytic technique. csQCA allows researchers to systematically compare all cases and eliminate irrelevant causes via the minimization procedure. Hence, csQCA reduces complexity in a second round via reducing the number of configurations and reducing the number of causal conditions in a configuration via the elimination of irrelevant causes. For instance, in the study by Bakker et al. (2011), it was found that R d Cd T d A d M ! Z (case 8) and RdCd  TdAdM ! Z (case 9).1 So whether temporal embeddedness was present (T) or not (BT), the transfer of knowledge was successful (Z) due to the combination of relational and cognitive embeddedness, the absorptive capacity of the parent organization and the motivation of the project team. Therefore, T is irrelevant for these two cases and can be excluded from this equation, resulting in a new equation: RdCdAdM ! Z.

Crisp-Set QCA in Organizational Studies

33

Analyzing Causal Complexity csQCA allows for an elaborate causal analysis of cases. This analysis contains three distinct aspects. First, the approach clearly allows the identification of multiple causal paths to an outcome (equifinality) and frames the relevance of the presence or absence of causes in its configuration with other causes. In an often cited study by Roscigno and Hodson (2004), the authors studied the organizational and social foundations of worker resistance. They found that strike occurrence could be caused by several configurations. One configuration was the combination of workplaces with significant conflict on the shop floor (C), union presence (U), and a bureaucratized work structure (B) (2004, p. 26). Written as a Boolean statement: C d U dB ! Z. However, a strike can also occur due to another configuration,2 this is when there is conflict on the shop floor (C), union presence (U), individualized abuse (A) and a poor organization (O), or written as Boolean statement: C d U d A d  O ! Z. So both configurations can cause a strike, indicating that there are multiple paths to the same outcome. Secondly, the analysis allow researchers to identify nontrivial necessary and sufficient (configurations of) conditions. In the example above by Roscigno and Hodson (2004), two sufficient combinations of conditions were detected (C  U  B; and C  U  A  B O). Each configuration in itself is sufficient to produce the outcome. However, these configurations are not necessary, as other configurations do exist that lead to the same outcome. Within these two configurations, conflict (C) and union presence (U) are necessary attributes, present in both configurations. They are necessary conditions for the outcome to occur. However, they are not sufficient conditions, because in itself they do not produce the outcome, they need to be combined with other conditions (B; A and BO respectively). Thirdly, the approach allows the identification of mechanisms. In addition to identifying equifinality, a systematic analysis also makes it possible to gain an insight into the specific causal mechanisms that link these explanatory factors to certain outcomes if one returns to the cases. More specifically, it makes it possible to map out processes (and causal sequences) within cases. Since processes are important to identify causal mechanisms, various authors (Pierson, 2003) advocate sequential analysis of causal processes as an additional research stage in case research, thereby supplementing data set observations with causal process observations (Blatter & Blume, 2008; Brady & Collier, 2004; George & Bennett, 2005; Mahoney & Rueschemeyer, 2003). Data set observations such as correlations, associations, necessary

34

AXEL MARX ET AL.

conditions, and configurations of conditions say little or nothing about the causal processes or mechanisms linking conditions with one another (Rihoux & Lobe, 2009). Process observations focus specifically on identifying the causal mechanisms that generate a particular outcome (Hedstro¨m & Swedberg, 1998; Hedstro¨m & Ylikoski, 2010; Stokke, 2007). It should be noted that the opportunity for process analysis does not automatically follow from a QCA application. The results of a QCA analysis facilitate greatly the analysis of mechanisms since it identifies different conditions under which an outcome is generated (Marx, 2008). In the study by Roscigno and Hodson (2004), the authors refer to records of organizational ethnographies of 82 cases in England and the United States. They use them to deepen their understanding and to illustrate the causal mechanisms that link the various conditions. For example, they point to the lack of management competences (2004, p. 28) to understand the condition of conflict. More specifically, they describe based on ethnographic stories that an inappropriate use of (only) formal managerial power caused an ongoing conflict on the shop floor, eventually resulting in a collective worker response and a strike.

Using the Distinctive Strength of Case Studies to Use Different Types of Data A fourth strength of small-N or intermediate-N approach lies in the collection and treatment of various sorts of data. On the one hand, qualitative data or data collection techniques may form part of a csQCA research design, but multiple-case research is not limited to such data collection techniques. Data gathered from other sources, such as the available statistics or data from structured questionnaires are also used. csQCA is thus not limited to participatory observation or conducting in-depth interviews, even though these are very useful data collection techniques. Systematically gathering data using various methods and systematically analyzing data using various techniques are an intrinsic part of csQCA research design. csQCA researchers typically use a variety of data collection techniques and data sources. For instance, in a study on the organizational antecedents of repetitive strain injuries in jobs, Marx and Van Hootegem (2007) used three different data sources in order to increase the accuracy of the measurement and to make proper judgments on assigning conditions to cases. First of all, data collected through a survey with 331 people was used. Secondly, clinical tests on all workers were used to measure RSI. Finally, video registration was used to assess specific job characteristics, since the employees were

Crisp-Set QCA in Organizational Studies

35

video-taped during all the phases of their normal jobs. Together, these three data sources allowed the researchers to gain in-depth understanding of the phenomenon studies.

CRITICISMS ON csQCA The use of csQCA has been the focus of much fruitful intellectual debate and criticism during the last decade (Lieberson, 1991, 1994, 2004; Goldthrope, 2000; Mahoney, 2000; Savolainen, 1994; Seawright, 2004, 2005; Sewell, 1996). We concentrate on three aspects since they are related to key issues in applications of csQCA. Several other criticisms have been put forward, including issues of measurement and calibration, the use of simplifying assumptions and others (see De Meur, Rihoux, & Yamasaki, 2009). Since other multi-value and fuzzy-set approaches exist, researchers can easily switch to these alternatives if measurement issues are at stake. However, as indicated above, many researchers are still using crisp-sets. Moreover, it should be noted that for many concepts in the social sciences there is no real hierarchy between using crisp-sets or dichotomous variables and ordinal or continuous variables (Schneider & Wagemann, 2012). Depending on the research question one might be preferable over another. We focus on two issues which are specifically related to the use of the crisp-set approach of QCA: the assumption that no omitted variables are excluded from an analysis (this is related to model specification), and the linked issue of the sensitivity to individual cases (this is related to population construction and case selection).

The Assumption of Contradictions A first critique focuses on the assumption QCA makes with regard to the conditions which are included in an explanatory model. Seawright (2004, 2005) objected that csQCA makes very strong assumptions with regard to the omitted-variable bias, that is, the fact that the applied explanatory model overlooks an important explanatory condition (King et al., 1994, pp. 168ff.). Seawright (2005) argued that csQCA makes the very restrictive assumption that no relevant conditions are excluded from an explanatory model. The origin of the assumption formulated by Seawright lies in the process by which csQCA generates an explanatory model. The key decision

36

AXEL MARX ET AL.

criterium for selecting an explanatory model which is further analyzed in csQCA is the presence of contradictions (Ragin, 1987, pp. 113–118; Rihoux & De Meur, 2009, pp. 48–50). Contradictions occur in csQCA when an identical configuration of conditions accounts for both the presence and absence of an outcome. In csQCA terms, a contradiction occurs when: AdB ! Z A d B ! Z Contradictions occur in the transformation of a data-matrix in a truth table. The occurrence of contradictions is signaled in the output (truth table) of a csQCA analysis. The output presents the number of configurations which lead to the presence of the outcome (1) the number of configurations of conditions which lead to the absence of the outcome (0) and the number of configurations which lead both to the presence and absence of the outcome (i.e., contradictory configurations – C-configurations). As Ragin (1987, p. 118) notes the ‘‘lesson here is that an existing data set should not be considered an irrevocable starting point. In qualitative comparative work, the representation of the empirical world in terms of a truth table is a crucially part of the investigation’’. This transformation reveals contradictions which should be resolved, primarily by identifying omitted causal variables (Ragin, 1987, p. 113; see also Rihoux & De Meur, 2009, pp. 48–49 for complementary strategies). Hence, the development of an explanatory model in csQCA goes hand in hand with resolving contradictions. This back and forth process of including and excluding theoretically and empirically relevant conditions in a model until a model has been identified with no or only a few contradictions is the key mechanism to develop an explanatory model for analytic purposes. The importance of the issue of contradictions in model construction via csQCA is repeatedly stressed by csQCA-users. Ragin (2005, p. 34; see also Ragin & Kogut, 2006) argued that a csQCA-analysis forces ‘‘The resolution of contradictions [y] deepens knowledge and understanding of cases and also may expand and elaborate theory.’’ In their textbook on QCA, Rihoux and De Meur (2009, pp. 48–56) discuss extensively several strategies to resolve contradictions in csQCA. For several scholars (Ragin, 2006, 2008a; Schneider & Wagemann, 2012) the focus on assessing the explanatory value of a model has shifted from contradictions to the introduction of new measures in QCA which aim to capture the explanatory power of a model, namely consistency. Consistency ‘‘assesses the degree to which the cases sharing a given condition or

Crisp-Set QCA in Organizational Studies

37

combination of conditions [y] agree in displaying the outcome in question’’ (Ragin, 2006, p. 292). In other words, consistency in crisp-set relations is the proportion of cases with a given cause or combination of causes that also display the outcome (Ragin, 2006, 2008b, p. 77). Charles Ragin (2006, p. 293) advises researchers to craft models which generate high consistency measures. Low consistency measures flag problems with the explanatory model such as omitted variables or measurement error. However, contradictions and consistency are interrelated (Marx & Dusa, 2011). Consistency in csQCA is measured on the level of a row of a truth table and, as indicated, is the proportion of cases with a given cause or combination of causes that also display the outcome. If, for example, 17 out of the 20 cases displaying a cause or causal combination also display the outcome, then consistency is 0.85 (Ragin, 2006, p. 293). This indicates that three cases do not display the outcome, that is, the outcome is absent and hence generates contradictions. In previous versions of csQCA, the result would be a C-configuration, indicating a contradiction which has to be resolved. With the introduction of consistency, the stringency of resolving all contradictions has made place for a measure which allows for some more error (actually including an error function in the analysis). The focus on resolving contradictions and hence developing models with a high consistency score are strongly related to the issue of the assumption on omitted variables as it is discussed by Seawright (2005 – see quote above). However, the issue in QCA is not whether QCA assumes no missing variables; an assumption that indeed is very hard to make (King et al., 1994, p. 182) and has never been suggested by QCA users or developers. The key issue is rather that contradictions or low consistency scores are naturally occurring phenomena in QCA when explanatory models are not correctly specified due to the omission of crucial explanatory variables, measurement error, or high heterogeneity of the research population. This assumption, although not explicitly formulated in these terms, was challenged by Lieberson (2004) and subsequently tested by Marx (2010) and Marx and Dusa (2011). Lieberson (2004) argued that QCA is not able to make a distinction between real and random data and always generates an explanatory model which is used in subsequent analytic steps of QCA. Lieberson (2004) in his argument overlooked a crucial step in the research process of QCA, namely resolving contradictions. Marx (2010) reformulated this objection in reference to the issue of contradictions. It was argued that csQCA is based on the assumption of naturally occurring contradictions if the explanatory model was flawed. It was hypothesized that contradictions should occur always when csQCA is

38

AXEL MARX ET AL.

applied to the analysis of random data. This hypothesis was tested on the basis of a simulation in which random datasets were analyzed by csQCA. The chapter found that contradictions are not naturally occurring phenomena. In some instances csQCA generated no contradictions on the basis of random data. The latter was a function of the design of the explanatory model in terms of number of cases and conditions included in the analysis. In some situations, when there are too many conditions included in a model or when the proportion of conditions on cases is high, csQCA is not able to distinguish real from random data. This is a result of the use of Boolean algebra which under these circumstances creates a situation for which no analytical reduction is possible and one is confronted with the fact that each case is unique (Aarebrot & Bakka, 1997; Marx, 2010). A further refinement of the testing of the hypothesis followed in Marx and Dusa (2011) which made used of new computing techniques in R to generate precise estimates on the relationship between cases and conditions. On the basis of an analysis of over 5 million random datasets that generated benchmarks tables which inform researchers on when it is safe to assume that contradictions or low levels of consistency will occur on random data. These benchmark tables can be used to assess the reliability of an explanatory model which is used in csQCA. Marx and Dusa (2011) developed a benchmark table, for each combination of conditions (up to 11) and cases (up to 300), which enables researchers to assess the chances of accepting a model for further analysis on the basis of random data. If this chance is too high, a researcher should not proceed with analyzing the selected model in csQCA since he/she will not be able to distinguish an analysis on the basis of random data versus real data. Hence, the benchmarks assess whether or not a model can be accepted for further analysis, and guide researchers in model specification. As in more quantitative approaches, csQCA also requires researchers to increase the number of cases if one wants to take on board additional condition variables. The threshold for accepting a model is set at 10%, that is, there is a 10% chance of accepting a model which could also have been generated on random data. More stringent benchmarks are 5% and 1%, respectively a 5% and 1% chance of accepting a model which could also be generated on random data. Table 2 summarizes the benchmark tables and shows how many cases are at least needed to perform a csQCA-analysis for a given number of conditions. For example, if one has 4 conditions and wants to pass the 10% benchmark test one needs at least 12 cases. One will need 17 cases if a researcher wants to pass the 1% threshold. If a researcher has

39

Crisp-Set QCA in Organizational Studies

Table 2.

Benchmark Table for Model Specification Assessment.

Conditions

COr2 COr3 COr4 COr5 COr6 COr7 COr8 COr9 COr10 COr11

Threshold 10%

5%

1%

CAZ6 CAZ9 CAZ12 CAZ17 CAZ24 CAZ33 CAZ49 CAZ69 CAZ97 CAZ139

CAZ8 CAZ11 CAZ15 CAZ20 CAZ29 CAZ39 CAZ55 CAZ78 CAZ112 CAZ154

CAZ11 CAZ14 CAZ17 CAZ25 CAZ34 CAZ47 CAZ66 CAZ92 CAZ129 CAZ181

Notes: CA, cases; CO, conditions.

11 cases or less with 4 explanatory conditions he/she should not proceed with the analysis. Sensitivity to Individual Cases A second critique which was formulated is that csQCA is too sensitive to individual cases, since the inclusion or exclusion of a single case can modify the results of an analysis (Goldthorpe, 1997). This argument needs to be qualified in several respects. First, if one assumes that the number of conditions stays equal, the inclusion of new cases is not problematic. Two situations can occur. First, a case is added to a row in a truth table which already contains other empirical cases. This will result in the fact that this casual path explains more cases and has a higher coverage. Second, the inclusion of new cases can also result in the discovery of new observed empirical cases which can fill the property space, that is, diminish the number of rows in a truth table without observations. The latter will require the researcher to make less simplifying assumptions when analyzing the model. In other words, one discovers an additional causal path to an outcome. It may not be very significant in terms of number of observed cases, but it is a new causal path which might be theoretically relevant. The existence of multiple causality is, as argued above, one of its strengths (see Section 4.3). In both instances, researchers can examine which causal paths (i.e., combinations of conditions) are more ‘‘traveled’’ than others, and

40

AXEL MARX ET AL.

hence could be considered as more important (see, e.g., Rihoux, 2001, 2003). In the recent developments of csQCA a new measure, coverage, is introduced to precisely assess the importance of the number of cases for each causal path. Coverage ‘‘assesses the degree to which a cause or causal combination ‘accounts for’ instances of an outcome. When there are several paths to the same outcome, the coverage of any given causal combination may be small. Thus, coverage gauges empirical relevance or importance’’ (Ragin, 2006, p. 292). As a result, the inclusion of a new case is not problematic ceteris paribus. If one does not assume that the number of conditions remain equal but also increases with the inclusion of new cases than the inclusion can be problematic due to a violation of model specification requirements outlined above. The exclusion of cases is potentially more problematic since the exclusion of a case can result in the disappearance of contradictions or the occurrence of higher consistency scores. If the exclusion of cases is not conducted transparently and is not supported by theoretical or methodological arguments it might influence the results and hence is troublesome. In order to address this issue it is advisable that researchers applying QCA provide sufficient information on the research population and the selection of cases. In organization studies one is often confronted with a limited number of cases. A limited research population is a result of the unit homogeneity requirement (King et al., 1994). The importance of analyzing, for example, specific sectors or industries is stressed by Griffin and Mahon (1997). They argue that many variables vary significantly between sectors and that this context effect should be taken into account in the research design. If one conducts multi-industry studies the assumption is that many of the variables under investigation, once controlled for sector, work in the same way across industries. This assumption is often false and should be investigated more in-depth. Hence, as Carroll and Hannan (1999, p. 91) argue, the complexity of mechanisms driving organization–environment relationships requires one to analyze specific (sectoral) populations. Similar arguments have been made for size and national economic context. Size has been linked to many different kinds of effects. In relation to, for example, institutional impacts on organizations it has been argued that larger organizations tend to be more resource rich, differentiated, more visible to external publics, including governance bodies and NGOs, etc. (Edelman, 1992; Greening & Gray, 1994; Scott, 2001). Many authors have also argued that there are distinct differences between nation-states in relation to how economies are organized and corporations interact with each other (see, e.g., Amsden, 2003).

Crisp-Set QCA in Organizational Studies

41

Taking these scoping conditions (Ragin, 2000, pp. 61–63) into account, researchers are often confronted with a limited research population (SmallN or intermediate-N). For csQCA purposes it is then often advisable to include the whole population, or a theoretical bounded subset, in the research design and collect data for all cases. The exclusion of cases from this population should then be supported by clear decision rules. It should be noted that a strict delimitation of the research population in organization studies might raise another methodological issue which is related to the assumption of case-independence. Cases in csQCA are compared under the assumption that they do not influence each other, an assumption shared with many variable-oriented techniques of analysis. In a comparative analysis of organizations (cases) this assumption might be problematic since some organizational theories stress the importance of the interdependence of organizations. Especially network theories (Podolny & Page, 1998; Powell & Smith-Doerr, 1994), institutional theories (Dimaggio & Powell, 1983; Powell & DiMaggio, 1991; Scott, 2001), and resource dependence theory (Pfeffer & Salancik, 2003) stress the importance of interrelatedness. In order to understand the adaptation of innovation, the expansion of capabilities, competitive advantage, the reduction of uncertainty, and the acquisition of information by organizations one needs to understand their interrelatedness. In other words, an outcome in one case might be explained simply because another case, earlier in time, generated the outcome. In this sense, an effect akin to endogeneity might occur where one observes across cases a loop of causality between the outcome and explanatory conditions. There are several possible strategies to address this difficult issue in csQCA. One of them is to add a specific ‘‘diffusion’’ condition (linked to some antecedent cases, typically) in the model. Another strategy, downstream of the computer-run part of csQCA proper, is to bring in these cross-case effects in the interpretation of the minimal formula. Yet another, more radical strategy, is to reconsider the ‘‘casing’’ procedure, for instance by modifying the level of analysis and by empirically defining cases not as full organizations but rather as organizational subunits.

CONCLUSION It is now 25 years since Charles Ragin published The Comparative Method in 1987 and introduced a new methodology – Crisp-Set Qualitative Comparative Analysis – to the social sciences. Early users quickly discovered the potential of the method. However, several critics pointed out to some

42

AXEL MARX ET AL.

weaknesses such as the crude, dichotomous, measurement of concepts. As a result, several new tools such as fuzzy-set and multi-value QCA were developed and introduced. Notwithstanding these new developments many researchers kept on using csQCA for its own distinctive advantage including its ability to generate parsimonious explanation on the basis of several case studies taking into account the possibility of multiple conjunctural causation. However, as indicated above, the use of csQCA is not disappearing and several recent contributions, including ones in this volume, highlight its usefulness. In this chapter we introduced csQCA, discussed its strengths, assessed some of the criticisms which are especially relevant for csQCA and provided some suggestions for ways forward. The chapter stressed that csQCA allows one to study configurations, that is, a specific combination of attributes that is common to a number of cases. Secondly, it was stressed that csQCA allows for the possibility that there may be several combinations of conditions that generate the same general outcome, that is, equifinality. Thirdly its potential to reduce complexity and generate parsimonious explanations, through the elimination of irrelevant causes, was highlighted. In terms of criticisms, two interrelated issues were discussed and analyzed. The first concerned a key assumption on which csQCA is build, namely that contradictions ‘‘naturally’’ occur and that no omitted variables are excluded from an analysis. Solving contradictions and striving for high consistency scores are important in csQCA since contradictions force researchers to pay great attention to the development and refinement of explanatory models since models should be able to explain every empirical observation and contradictions should be resolved. The second criticism concerned the sensitivity to individual cases. Both are interrelated since they both touch on the selection of conditions and cases, and their mutual dependence, which go into a csQCA analysis and have important implications for the case selection framework. The chapter assessed the assumption of contradictions on which csQCA and showed that this assumption is not always valid. The latter is dependent on the number of conditions one takes into account. This limiting condition was further elaborated and a benchmark table which enables the assessment of the validity of the assumption of contradictions was presented. Secondly, the chapter assessed the sensitivity to individual cases and the effect of the inclusion and exclusion of a case. In both cases, the assessment of the criticisms points to the importance of population construction and case selection as a way to keep the research population homogenous and the number of explanatory conditions under control. Notwithstanding these

Crisp-Set QCA in Organizational Studies

43

criticisms the chapter clearly identified the potential added value of csQCA for configurational analysis in organization studies. Crisp-set studies will probably remain predominantly in the near future (see Fig. 1) because they allow researchers to study complex phenomena such as configurations of interconnected structures.

ACKNOWLEDGMENT The authors thank Peer Fiss and one anonymous reviewer for valuable feedback on a first draft of the chapter. Special thanks to Priscilla Alamos for research assistance on Fig. 1.

NOTES 1. In a hypothetical logical statement A  C + B  BD - Z, ‘‘+’’ denotes the logical operator or, ‘‘’’ denotes the logical operator and, while ‘‘B’’ denotes the logical not. ‘‘-’’ denotes the logical implication operator. So what this actually states is that a combination of 4 practices used to measure market and C or a combination of B and not-D leads to Z (see also Fiss, 2007; Rihoux & De Meur, 2009, pp. 34–35). 2. Roscigno and Hodson (2004) describe more configurations causing strike. We here selected two configurations to explain the topic of equifinality.

REFERENCES Aarebrot, F. H., & Bakka, P. H. (1997). Die vergleichende methode in der Politikwissenschaft. In D. Berg-Schlosser & F. Mu¨ller-Rommel (Eds.), Vergleichende politikwissenschaft (3rd ed, pp. 49–66). Opladen: Leske & Budrich. Amenta, E., & Poulsen, J. (1994). Where to begin: A survey of five approaches to selecting independent variables for qualitative comparative analysis. Sociological Methods and Research, 23(1), 22–53. Amsden, A. (2003). The rise of ‘‘the rest’’: Challenges to the west from late-industrializing economies. Oxford: Oxford University Press. Bakker, R. M., Cambre´, B., Korlaar, L., & Raab, J. (2011). Managing the project learning paradox: A set-theoretic approach toward project knowledge transfer. International Journal of Project Management, 29(5), 494–503. Berg-Schlosser, D., & De Meur, G. (2009). Comparative research design: Case and variable selection. In B. Rihoux & C. Ragin (Eds.), Configurational comparative methods.

44

AXEL MARX ET AL.

Qualitative comparative analysis (QCA) and related techniques (pp. 19–32). Thousand Oaks, CA: Sage. Berg-Schlosser, D., De Meur, G., Rihoux, B., & Ragin, C. C. (2009). Qualitative comparative analysis (QCA) as an approach. In B. Rihoux & C. Ragin (Eds.), Configurational comparative methods. Qualitative comparative analysis (QCA) and related techniques (pp. 1–18). Thousand Oaks, CA: Sage. Blatter, J., & Blume, T. (2008). In search of co-variance, causal mechanisms or congruence? Towards a plural understanding of case studies. Swiss Political Science Review, 14(2), 315–356. Box-Steffensmeier, J., Brady, H., & Collier, D. (2008). The Oxford handbook of political methodology. Oxford: Oxford University Press. Brady, H., & Collier, D. (2004). Rethinking social inquiry: Diverse tools, shared standards. Berkeley, CA: Rowman & Littlefield. Burt, R. (1992). Structural holes: The social structure of competition. Cambridge: Harvard University Press. Carroll, G. R., & Hannan, M. T. (1999). The demography of corporations and industries. Princeton: Princeton University Press. Child, J. (2002). A configurational analysis of international joint ventures drawing upon experience in China. Organization Studies, 23(5), 781–815. Cronqvist, L., & Berg-Schlosser, D. (2009). Multi-value QCA (mvQCA). In B. Rihoux & C. Ragin (Eds.), Configurational comparative methods: Qualitative comparative analysis (QCA) and related techniques (pp. 69–86). Thousand Oaks, CA: Sage. Curchod, C. (2002). La me´thode comparative en sciences de gestion: Vers une approche qualiquantitative de la re´alite´ manage´riale. COMPASSS Working Paper. De Meur, G., & Rihoux, B. (2002). L’analyse quali-quantitative compare´e (AQQC-QCA): Approche, techniques et applications en sciences humaines. Louvain-la-Neuve: AcademiaBruylant. De Meur, G., Rihoux, B., & Yamasaki, S. (2009). Addressing the critiques of csQCA. In B. Rihoux & C. Ragin (Eds.), Configurational comparative methods. Qualitative comparative analysis (QCA) and related techniques (pp. 139–160). Thousand Oaks, CA: Sage. Dimaggio, P., & Powell, W. (1983). The iron cage revisited-institutional isomorphism and collective rationality in organizational fields. American Sociological Review, 48(2), 147–160. Dul, J., & Hak, T. (2008). Case study methodology in business research. Amsterdam: Elsevier. Edelman, L. (1992). Legal ambiguity and symbolic structures: Organizational mediation of civil rights law. American Journal of Sociology, 97, 1531–1576. Eisenhardt, K. (1989). Building theories from case-study research. Academy of Management Review, 14(4), 532–550. Eliason, S., & Stryker, R. (2009). Goodness-of-fit tests and descriptive measures in fuzzy-set analysis. Sociological Methods and Research, 38, 102–146. Fiss, P. (2007). A set-theoretic approach to organizational configurations. Academy of Management Review, 32, 1180–1198. Fiss, P. (2011). Building better causal theories: A fuzzy set approach to typologies in organization research. Academy of Management Journal, 54, 393–420. George, A., & Bennett, A. (2005). Case research and Theory development. Cambridge, MA: MIT Press.

Crisp-Set QCA in Organizational Studies

45

Gerring, J. (2007). Case study research: Principles and practice. Cambridge: Cambridge University Press. Goertz, G. (2006a). Assessing the trivialness, relevance, and relative importance of necessary or sufficient conditions in social science. Studies in Comparative International Development, 41(2), 88–109. Goertz, G. (2006b). Social science concepts: A user’s guide. Princeton, NJ: Princeton University Press. Goertz, G., & Starr, H. (2003). Necessary conditions: Theory, methodology, and applications. New York, NY: Rowman & Littlefield. Goldthorpe, J. H. (1997). Current issues in comparative macrosociology: A debate on methodological issues. Comparative Social Research, 16, 1–26. Goldthrope, J. H. (2000). On sociology: Numbers, narratives, and the integration of research and theory. Oxford: Oxford University Press. Greening, D., & Gray, B. (1994). Testing a model of organizational response to social and political issues. Academy of Management Journal, 37, 467–498. Griffin, J., & Mahon, J. (1997). The corporate social performance and corporate financial performance debate. Business & Society, 36(1), 5–32. Hedstro¨m, P., & Swedberg, R. (Eds.). (1998). Social mechanisms: An analytical approach to social theory. Cambridge: Cambridge University Press. Hedstro¨m, P., & Ylikoski, P. (2010). Causal mechanisms in the social sciences. Annual Review of Sociology, 36, 49–67. King, G., Keohane, R., & Verba, S. (1994). Designing social enquiry: Scientific inference in qualitative research. Princeton, NJ: Princeton University Press. Kogut, B. (2000). The transatlantic exchange of ideas and practices: National institutions and diffusion. Working Paper No. WP 00-13, The Reginald H. Jones Center, The Wharton School Working Paper. University of Pennsylvania. Lieberson, S. (1991). Small N’s and big conclusions: An examination of the reasoning in comparative studies based on a small number of cases. Social Forces, 70, 307–320. Lieberson, S. (1994). More on the uneasy case for using Mill-type methods in small N comparative studies. Social Forces, 72, 1225–1237. Lieberson, S. (2004). Comments on the use and utility of QCA. Qualitative Methods: Newsletter of the American Political Science Association Organized Section on Qualitative Methods, 2, 13–14. Maggetti, M. (2007). De facto independence after delegation: A fuzzy-set analysis. Regulation and Governance, 1, 271–294. Mahoney, J. (2000). Strategies of causal inference in small-N analysis. Sociological Methods and Research, 28, 387–424. Mahoney, J., & Rueschemeyer, D. (Eds.). (2003). Comparative historical analysis in the social sciences. Cambridge: Cambridge University Press. Marx, A. (2008). Limits to non-state market regulation: A qualitative comparative analysis of the international sport footwear industry and the fair labor association. Regulation & Governance, 2, 253–273. Marx, A. (2010). Crisp set qualitative comparative analysis (csQCA) and model specification. International Journal of Multiple Research Approaches, 4(2), 138–158. Marx, A., & Dusa, A. (2011). Crisp-set qualitative comparative analysis (csQCA), contradictions and consistency: Benchmarks for model specification. Methodological Innovations, 6(2), 103–148.

46

AXEL MARX ET AL.

Marx, A., Rihoux, B., & Ragin, C. (2013). The origins, development and application of qualitative comparative analysis (QCA): The first 25 years. European Political Science Review. doi:http://dx.doi.org/10.1017/S1755773912000318 Marx, A., & Van Hootegem, G. (2007). Comparative configurational case analysis of ergonomic injuries. Journal of Business Research, 60, 522–530. Miles, M., & Huberman, A. (1994). Qualitative data analysis (2nd ed.). Thousand Oaks, CA: Sage Publications. Pfeffer, J., & Salancik, G. R. (2003). The external control of organizations: A resource dependence perspective. Stanford, CA: Stanford Business Books. Pierson, P. (2003). Big, slow-moving, andyinvisible. In J. Mahoney & D. Rueschemeyer (Eds.), Comparative historical analysis in the social sciences. Cambridge: Cambridge University Press. Podolny, J., & Page, K. (1998). Network forms of organization. Annual Review of Sociology, 24, 57–76. Poteete, A., Janssen, M., & Ostrom, E. (2010). Working together: Collective action, the commons and multiple methods in practice. Princeton, NJ: Princeton University Press. Powell, W., & DiMaggio, P. (Eds.). (1991). The new institutionalism in organizational analysis. Chicago, IL: Chicago University Press. Powell, W., & Smith-Doerr, L. (1994). Networks and economic life. In N. Smelser & R. Swedberg (Eds.), Handbook of economic sociology (pp. 368–402). Princeton, NJ: Princeton University Press. Przeworski, A., & Teune, H. (1970). The logic of comparative social inquiry. New York, NY: Wiley. Ragin, C. C. (1987). The comparative method. Moving beyond qualitative and quantitative strategies. Berkeley, CA: University of California Press. Ragin, C. C. (1991). Introduction: The problem of balancing discourse on cases and variables in comparative social science. In C. C. Ragin (Ed.), Issues and alternatives in comparative social research (pp. 1–8). Leiden: E.J. Brill. Ragin, C. C. (1994). Constructing social research: The unity and diversity of method. Thousand Oaks, CA: Pine Forge Press. Ragin, C. C. (1997). Turning the tables: How case-oriented methods challenge variable-oriented methods. Comparative Social Research, 16, 27–42. Ragin, C. C. (2000). Fuzzy-set social science. Chicago, IL: Chicago University Press. Ragin, C. C. (2005). Core versus tangential assumptions in comparative research. Studies in Comparative International Development, 40(1), 33–38. Ragin, C. C. (2006). Set relations in social research: Evaluating their consistency and coverage. Political Analysis, 14(3), 291–310. Ragin, C. C. (2008a). Redesigning social inquiry. Fuzzy sets and beyond. Chicago, IL: Chicago University Press. Ragin, C. C. (2008b). fsQCA manual. Retrieved from http://www.u.arizona.edu/Bcragin/ fsQCA/software.shtml. Accessed on April 13. Ragin, C. C., & Becker, H. (Eds.). (1992). What is a case? Exploring the foundations of social inquiry. Cambridge: Cambridge University Press. Ragin, C. C., & Kogut, B. (2006). Exploring complexity when diversity is limited: Institutional complementarity in theories of rule of law and national systems revisited. European Management Review, 3(1), 44–59.

Crisp-Set QCA in Organizational Studies

47

Rihoux, B. (2001). Les partis politiques: Organisations en changement. Le test des e´cologistes. Paris: L’Harmattan. Rihoux, B. (2003). Bridging the gap between the qualitative and quantitative worlds? A retrospective and prospective view on qualitative comparative analysis. Field Methods, 15(4), 351–365. Rihoux, B. (2008). Case-oriented configurational research: Qualitative comparative analysis (CSQCA), Fuzzy sets and related techniques. In J. Box-Steffensmeier, H. Brady & D. Collier (Eds.), The Oxford handbook of political methodology (pp. 722–736). Oxford: Oxford University Press. Rihoux, B., Alamos, P., Bol, D., Marx, A. & Rezsohazy, I. (2013). From niche to mainstream method? A comprehensive mapping of QCA applications in journal articles from 1984 to 2011, Political Research Quarterly, 66(1), 175–184. Rihoux, B., & De Meur, G. (2009). Crisp-set qualitative comparative analysis (csQCA). In B. Rihoux & C. Ragin (Eds.), Configurational comparative methods. Qualitative comparative analysis (QCA) and related techniques (pp. 33–68). Thousand Oaks, IL: Sage. Rihoux, B., & Lobe, B. (2009). The case for qualitative comparative analysis (QCA): Adding leverage for thick cross-case comparison. In D. Byrne & C. Ragin (Eds.), The Sage handbook of case-based methods (pp. 222–243). London: Sage. Rihoux, B., & Ragin, C. (Eds.). (2009). Configurational comparative methods: Qualitative comparative analysis (QCA) and related techniques. Thousand Oaks: Sage. Roscigno, V. T., & Hodson, R. (2004). The organizational and social foundations of worker resistance. American Sociological Review, 69(1), 14–39. Savolainen, J. (1994). The rationality of drawing big conclusions based on small samples: In defense of Mill’s methods. Social Forces, 72, 1217–1224. Schneider, C. Q., & Wagemann, C. (2006). Reducing complexity in qualitative comparative analysis (QCA): Remote and proximate factors and the consolidation of democracy. European Journal of Political Research, 45, 751–786. Scott, W. R. (2001). Institutions and organizations. Thousand Oaks, CA: Sage Publications. Seawright, J. (2004). Qualitative comparative analysis vis-a-vis regression. Qualitative Methods: Newsletter of the American Political Science Association Organized Section on Qualitative Methods, 2, 14–17. Seawright, J. (2005). Qualitative comparative analysis vis-a`-vis regression. Studies in Comparative International Development, 40(5), 3–26. Sewell, W. (1996). Three temporalities: Toward an eventful sociology. In T. McDonald (Ed.), The historic turn in the human sciences (pp. 245–280). Ann Arbor, MI: The University of Michigan Press. Stokke, O. (2007). Qualitative comparative analysis, shaming and international regime effectiveness. Journal of Business Research, 60(5), 501–511. Van Maanen, J. (Ed.). (1998). Qualitative studies of organizations. Thousand Oaks, CA: Sage. Yin, R. K. (1994). Case study research: Design and methods (2nd ed.). Thousand Oaks, CA: Sage.

CHAPTER 3 THE TWO QCAs: FROM A SMALL-N TO A LARGE-N SET THEORETIC APPROACH Thomas Greckhamer, Vilmos F. Misangyi and Peer C. Fiss ABSTRACT Although QCA was originally developed specifically for small-N settings, recent studies have shown its potential for large-N organization studies. In this chapter, we provide guidance to prospective researchers with the goal of opening up QCA’s potential for widespread use in organization studies involving large-N settings, both as an alternative and as a complement to conventional regression analyses. We compare small-N and large-N QCA with respect to theoretical assumptions and objectives, processes and decisions involved in building the causal model, selecting the sample, as well as analyzing the data and interpreting the results. Finally, we discuss the prospects for large-N configurational analysis in organization studies and related fields going forward. Keywords: QCA; fuzzy sets; set theoretic methods; case-oriented research; comparative analysis; configurational methods

Configurational Theory and Methods in Organizational Research Research in the Sociology of Organizations, Volume 38, 49–75 Copyright r 2013 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0733-558X/doi:10.1108/S0733-558X(2013)0000038007

49

50

THOMAS GRECKHAMER ET AL.

INTRODUCTION Configurational thinking has a long tradition in organization studies. Yet, the promise of configurational research has remained largely unfulfilled, not least because of a lack of adequate methodological tools to match the theoretical assumptions of the configurational approach (Fiss, 2007). Recently, however, a methodological framework has emerged that is particularly well suited for viewing organizations as configurations and examining the interdependence of the causal effects underlying organizational outcomes: Qualitative Comparative Analysis (QCA) (Ragin, 1987, 2000, 2008). As a research strategy, The Comparative Method generally and QCA1 specifically were originally developed to extend the systematic, in-depth, qualitative approach exemplified by the comparative case study design to research settings entailing more than a few cases. Although most prior studies using QCA have involved relatively small-N settings (10–50 cases), recent studies (e.g., Fiss, 2011; Greckhamer, Misangyi, Elms, & Lacey, 2008) have shown the promise of QCA as a useful tool for analyzing large-N situations (i.e., more than 50 cases). While remaining configurational in its theoretical and methodological approach, the application of QCA to large-N research situations inevitably involves a departure from some of the underlying ideas of the original smallN QCA approach. The primary purpose of this chapter is to provide a theoretical framework and practical guidance for the use of QCA in large-N applications and thereby to open up QCA for wider usage in organizational studies. To accomplish this objective, in this chapter we use the small-N approach to QCA as a springboard for discussing the respective considerations, strengths, and trade-offs involved in extending applications of QCA to large-N settings. Specifically, we compare small-N and large-N QCA with respect to their theoretical assumptions and objectives, processes and decisions involved in building the causal model, selecting the sample, as well as analyzing the data and interpreting the analyses. We particularly aim to outline QCA’s potential for widespread use involving large-N settings, both as an alternative and as a complement to conventional regression-oriented statistical approaches. In addition to facilitating the use of large-N applications, our comparison of the small-N and large-N QCA approaches provides guidance for researchers choosing between these two approaches in a manner that capitalizes on their advantages while avoiding their pitfalls. Furthermore, although we pay particular attention to applying QCA to the study of

The Two QCAs

51

organizations, our general arguments readily apply to QCA independent of its field of application.

THE COMMON PROPERTIES AND ASSUMPTIONS OF SMALL-N AND LARGE-N QCA APPROACHES Starting from the general observation that most empirical social science research involves a comparison among cases, Ragin (1987) developed the QCA approach to account for two fundamental insights that are frequently neglected by empirical cross-case analyses: (1) that cases are best viewed as configurations of attributes or causal conditions (hereafter we use these terms interchangeably) and (2) that causality tends to be complex and conjunctive as outcomes typically occur as a result of several different combinations of causal conditions. QCA affords researchers with the formal analytical tools and procedures to capture the diversity of causal combinations that constitute cases, to both map this diversity of cases and to systematically analyze combinations of causal conditions that are linked to an outcome of interest under study. In short, the two fundamental assumptions – that cases are configurations of causal attributes and thus there is a need to study the diversity of cases and its attendant causal complexity – apply to all research settings, regardless of whether they involve a small or large sample size. We thus begin by briefly highlighting these commonalities between small-N and large-N QCA that enable them to account for the configurational nature of causal complexity: their set theoretic perspective, using Boolean algebra to map and systematically analyze the diversity of cases and causal relations, and their multiple conjunctive conception of causality informed by the set theoretic perspective. An essential property of QCA, for both its small-N and large-N approaches, is that it is set theoretic in nature; it conceptualizes the connection between causal conditions and outcomes in terms of set membership and subset relations (Fiss, 2007; Ragin, 2000, 2008). This means that both the outcomes of interest and the conditions expected to be causally linked to these outcomes2 are viewed as sets and that each case is assessed for its membership in each of these sets, making the process of determining set memberships (i.e., calibration) the key to capturing the meaningful diversity of cases. In crisp set QCA (csQCA), set memberships are evaluated in a dichotomous (‘‘crisp’’) manner, which captures differences in kind. Based on

52

THOMAS GRECKHAMER ET AL.

theoretical or empirical knowledge, cases are thus classified as either ‘‘fully in’’ (1) or ‘‘fully out’’ (0) of the sets. For example, a specific organization may belong or not belong to the set of large organizations. The fuzzy set approach (fsQCA), on the other hand, allows the researcher to capture both differences in kind as well as degree; in addition to the two qualitative states of full membership and full nonmembership, a case may have partial membership in a set and thereby be assigned scores in the range from 0 to 1. To continue with the example of large organizations, rather than simply being classified as fully in or out of the set of large organizations, a specific organization may be assessed as having partial membership in the set (e.g., it may be ‘‘more in than out’’ of the set of large organizations). Thus, all approaches to QCA – csQCA and fsQCA, small-N and large-N QCA – involve the calibration of set memberships and the specification of these critical qualitative anchors (Ragin, 2000, 2008). Set memberships form the basis of the truth table approach to typology utilized in QCA (Ragin, 2000, 2008), which captures the (limited) diversity of cases. The truth table is a chart with 2k rows (k=number of included sets) which displays all logically possible combinations of the included theoretical attributes under study. Thus, it is the key tool of set theoretic analysis as it enables the researcher to map both the empirically occurring configurations of attributes as well as those logically possible configurations that do not occur. As Ragin (1987, 2000) points out, truth tables usually contain hypothetical combinations that lack empirical instances, which underscores the limited diversity of many social phenomena – the attributes of cases tend to occur in coherent patterns, including in organizations (e.g., Meyer, Tsui, & Hinings, 1993). As noted above, QCA conceptualizes causal relations among social phenomena as set relations. This perspective allows for the analysis of causal complexity through the construction and examination of arguments regarding the necessity and/or sufficiency of causal conditions – combinations of the theoretically relevant causal attributes under study – for a particular outcome. Examination of whether any combinations may be necessary and/or sufficient for the occurrence of an outcome involves examining the subset relations: when set memberships in the outcome are a subset of the attribute set memberships (i.e., all occurrences of the outcome exhibit the same causal attribute(s)), a causal condition can be argued to be necessary for an outcome to occur. On the other hand, when the causal condition is a subset of the outcome (i.e., all cases with the particular attribute(s) will display the outcome), a causal condition can be argued to be sufficient for the occurrence of an outcome. This mapping of set

53

The Two QCAs

memberships and analysis of their subset relations is enabled through the combinatorial logic of Boolean algebra,3 and can be applied to both small-and large-N research contexts. See Greckhamer et al. (2008) for a comprehensive demonstration of these properties to a large-N organizational setting. In sum, the basic premise of QCA is that aspects of cases should be examined as combinations of set memberships and that a single difference between cases may constitute a difference in kind. This approach – both its small-N and large-N approaches – thus permits researchers to capture and explore the diversity of organizations through configurations. Furthermore, this also means that both small-N and large-N QCA are premised on the notion of a multiple conjunctural understanding of causality; causality is conjunctural in that causes are seen as operating in combination rather than independently, and multiple (i.e., equifinal) because more than one combination may produce the same outcome (Becker, 1992; Ragin, 2000). This implies that outcomes of interest rarely have a single cause, causes rarely operate in isolation, and specific causes may have opposite (i.e., asymmetrical) effects depending upon context.

CONTRASTING SMALL-N AND LARGE-N QCA To strengthen the theoretical and practical basis of large-N applications of QCA and provide guidance to interested researchers, we utilize the small-N approach as a point of departure. Our goals are to clarify the differences between small-N and large-N QCA approaches with respect to their theoretical assumptions and objectives, the processes, and decisions involved in building the causal model, selecting the sample, as well as analyzing the data and interpreting the analytical results. In addition to elucidating the large-N approach, our hope is that this comparison may serve as a guide for future researchers in deciding which of the two approaches to implement. An overview and summary of the main points of comparison is contained in Table 1.

Objectives: Reasoning and Primary Goals As discussed in the previous section, both small-N and large-N applications of QCA lend themselves to empirical analyses of the configurational nature of causal relationships. However, the potential objectives – and thus

Interpretation of findings

Frequency threshold

Resolving contradicting observations Coverage

Results of necessity and sufficiency are interpreted by returning back to cases; case knowledge is used to build theory

Minimum typically higher (3+); tradeoff between potential for deductive analysis and inclusion of rare configurations Results of sufficiency and necessity are interpreted primarily as patterns across many cases without returning back to cases; statistical inferences are possible

Various strategies; large-N may benefit some while distance from cases may limit others Relatively lower – large coverage desirable but not necessary

Consistency Z.80 (i.e., ‘‘Almost always sufficient’’) is convention.

Consistency =1 (i.e., ‘‘Always sufficient’’) is plausible (though minimum threshold consistency of .80 can be used). Various strategies; intimacy with cases may benefit some while small-N may limit others Typically high – all cases accounted for after iterations of building the model based on indepth case knowledge Minimum typically one or two cases

Typically 4–8

Number of causal conditions Analyses processes Consistency

50+ Relatively distant, based on knowledge of conceptual relationships Theoretical or random sampling; random sample may be inappropriate for large-N research primarily interested in diversity Typically 6–12

12–50 Relatively close, based on knowledge of each case

Theoretical sampling based on theoretical relevance or significance of the case

Inductive or deductive Theory building and testing

Large-N QCA

Mostly inductive Theory building

Sample/case selection

Objectives Reasoning Primary Goal Sample and causal model Number of cases Relationship to cases

Small- and Large-N QCA Approaches.

Small-N QCA

Table 1.

54 THOMAS GRECKHAMER ET AL.

The Two QCAs

55

analytical reasoning – will tend to differ across small-N and large-N approaches. To begin with, although QCA is capable of incorporating probabilistic criteria to account for randomness and error (see Ragin, 2000, pp. 109–115), it has typically not been viewed as a hypotheticodeductive technique (Ragin, 2006, 2008). In short, the QCA approach has been described as a tool that contributes ‘‘to theory building by providing a rigorous way to combine verbal statements with logical relationships’’ (Fiss, 2007, p. 1181). As a result, small-N studies utilizing QCA have tended to aim at theory building, and have primarily employed inductive reasoning. Nevertheless, small-N QCA applications could be used to test theories deductively by constructing (non-probabilistic) evidence to either support or refute theories, as is true for case study research designs more generally (e.g. Bitektine, 2008; Ridder, Hoon, & McCandless, 2009; Yin, 1994). Large-N inquiries utilizing QCA in organization studies have also been designed with the primary goal of theory elaboration. For example, Greckhamer et al. (2008) explored how configurations of industry, corporate and business-unit factors affect business performance on a sample of 2,841 business-units spanning 2,451 corporations and 184 industries.4 While this study demonstrates the utility of large-N QCA for inductive, theory building inquiry, no methodological reasons hold back large-N QCA approaches from being used deductively (e.g., see Fiss, 2011). Hypothetico-deductive large-N QCA applications are not only possible but in our view present one of the most promising areas to extend the set theoretic approach. In this regard, as opposed to theorizing and trying to isolate the independent effects of single causes, QCA’s configurational nature both enables and guides the researcher to theorize about combinations of causal attributes that are necessary or sufficient for the occurrence of an outcome and enables the testing of hypotheses of these causal relations of necessity and sufficiency. For this purpose, as mentioned above, QCA enables the use of probabilistic criteria. Hence, a researcher may specify a set of hypotheses (e.g., ‘‘configuration X will be sufficient for outcome Y’’), set parameters surrounding the acceptable threshold for the consistency (i.e., Z.80; Ragin, 2000, 2008) of the hypothesized set relation(s) required to constitute evidence of support of the hypotheses, as well as probabilistic criteria (i.e., po.05) used to assess whether this consistency, that is, the proportion of cases displaying the hypothesized configuration (e.g., ‘‘X’’) and manifesting the outcome (e.g., ‘‘Y’’), is significantly greater than the designated threshold (for a detailed explanation, see Ragin, 2000, pp. 115–119).

56

THOMAS GRECKHAMER ET AL.

The challenges for conducting hypothetico-deductive large-N QCA studies are primarily theoretical, as opposed to methodological, in nature. The reason for this is that due to the dominant position of (net effectsoriented) general linear regression approaches in organizational research, unsurprisingly extant theorizing has primarily focused upon identifying the (strengths and direction of) relationships of single causes with outcomes of interests (see Abbott, 1988). The challenge, then, lies in the fact that QCA’s set theoretic approach requires researchers to shift away from ‘‘net effects’’ thinking and instead focus their attention on how causes combine to bring about outcomes. Thus, as opposed to theorizing about and trying to isolate the independent effects of single causes (i.e., their ‘‘net effects’’), QCA’s configurational approach both enables and guides researchers toward theorizing about causality in terms of the necessity and/or sufficiency of combinations of attributes (Fiss, 2007; Ragin, 2000, 2008). For example, Fiss’s (2011) study of 139 high-tech UK firms takes a deductive orientation to theorize and empirically examine the core and periphery aspects of organizational configurations (i.e., Miles and Snow’s typology of strategic organizations). This study illustrates that the long tradition of configurational theorizing by organizational and strategy scholars (e.g., Fiss, 2007; Meyer et al., 1993; Miller, 1986, 1996; Miller & Friesen, 1978, 1984; Mintzberg, 1979) can serve as rich foundation for future deductive large-N QCA research. Furthermore, considering the difficulty in interpreting multiway interactions (e.g., 3-way, 4-way, etc.) in regression-based analyses (e.g., Aiken & West, 1991), another avenue forward for large-N QCA studies could be to build on more micro-oriented theories that have theorized such interactions (e.g., Oldham & Cummings, 1996; Skarlicki & Folger, 1997; Wood, Michela, & Giordano, 2000). For instance, QCA is well suited to examine the hypothesis that creative performance is highest when individuals with creative personalities work in complex jobs and have supportive and un-controlling supervisors (i.e., a four-way interaction; Oldham & Cummings, 1996). In summary, both small-N and large-N QCA studies can be used for either theory testing or theory building. Large-N QCA studies, however, are relatively better positioned for theory testing as they can conform to the widely held expectations in organization studies that hypothetico-deductive theory testing be tightly coupled with statistical testing of probabilistic criteria. Indeed, we envision that deductive theory testing could become the typical mode of inquiry for large-N QCA studies and suggest that such research can build upon extant theorizing on organizational configurations, typologies, and multi-way interactions to chart its course.

57

The Two QCAs

Building the Sample and Causal Model The Number of Cases Not surprisingly, the number of cases under study is a point of difference across small-N and large-N QCA studies. Nevertheless, a few points are worth highlighting here. First, small-N studies typically involve between 12 and 50 cases. QCA has occasionally been applied to analyze samples of 12 or even fewer cases (see Marx, 2010). However, in such cases a systematic comparison of the necessity or sufficiency of attribute combinations depends on substantial in-depth cross-case and within-case analyses. QCA may be utilized to formalize cross-case comparisons; however if not combined with these in-depth cross-case and within case analyses, QCA analyses with 10 or fewer cases do not provide sufficient evidence to construct robust causal models (Marx, 2010). Systematic comparison of causal connections across more than 10–12 cases becomes quite unwieldy without a tool for systematic comparison such as QCA; a deep, rich investigation which is the signature of qualitative and case oriented research is still possible when examining 12–50 cases via QCA. Thus, QCA offers researchers a tool that supports both a deep qualitative analysis of cases and a systematic comparison. Large-N QCA studies typically will involve more than 50 and, as for example Greckhamer et al. (2008) demonstrate, QCA may handle even thousands of cases. Indeed, theoretically the sample size would be limited only by hardware and software limitations rather than methodological impediments. The Researcher’s Relationship to Cases The relationship that the researcher has to the cases under study differs somewhat between small-N and large-N QCA studies. As just described, researchers involved in small-N QCA settings will typically have a relatively deep knowledge of each case. Inevitably, such a close relationship is decreasingly feasible as the analysis involves 50, a few hundred, or even thousands of cases. In this way, large-N QCA somewhat resembles regression analysis approaches commonly used to study large samples. Yet, this is where the resemblance to statistical analysis ends. Unlike in correlational analysis, in which ‘‘measures vary around an inductively derived, sample-specific mean’’ (Ragin, 2008, p. 8), the set memberships of each theoretical attribute in the QCA approach must be calibrated. As described above, this means that the researcher must establish, a priori to testing, qualitative anchors to capture differences in kind (i.e., full membership and full nonmembership) as well as differences in degree (partial membership in the continuum between 0 and 1) based upon both

58

THOMAS GRECKHAMER ET AL.

theoretical and substantive knowledge (Ragin, 2000, 2008).5 Thus, while the researcher’s relationship to the cases in large-N QCA research is somewhat less intimate than in small-N research, the designation of anchors still requires greater familiarity with the data – both theoretically and empirically – than is commonly expected in standard correlational analyses. Accordingly, QCA pushes researchers to fully understand how to calibrate the particular attributes under study, even though prior theory is unlikely to always be a reliable guide here, as we discuss further below. Overall, the researchers’ relationship to the studied cases differs across small-N and large-N QCA studies, from relatively close to relatively distant. Nonetheless, the QCA method to large-N studies both requires and affords researchers a closer and more intimate relationship with the data than is required in large-N correlational studies. Case Selection In general, QCA uses a purposive sampling method; because QCA examines commonalities across the same outcome in cases (i.e., subset relations), researchers begin with the outcome of interest they wish to study in order to identify the population of cases of theoretical interest. For example, if one is interested in understanding the causes of success of mergers of manufacturing firms one would accordingly identify manufacturing firms that experienced mergers. In small-N studies, the sample is typically selected by purposefully selecting cases where either (1) all of the cases fall into the identified domain (i.e. the entire population) and thus within the established theoretical boundaries (e.g. all manufacturing firms that experienced mergers), or (2) a few relatively representative cases are selected from the larger population of cases for purposeful in-depth study (e.g., select a number of representative cases of mergers of manufacturing firms). Furthermore, this purposeful sampling may be an iterative procedure that is guided by the original research question and the relevant theory, justifying the inclusion of each case on theoretical grounds (e.g., Rihoux & Ragin, 2009). Purposive sampling should also be used in large-N QCA studies. The large-N QCA researcher may again choose to study the whole population of theoretically relevant firms (i.e., the set of cases relevant to a question) or select some sub-sample. For example, a researcher interested in understanding the causality of performance in a certain industry with many competitors may study all the competitors in the industry. Alternatively, the researcher might construct a stratified sample of the theoretically relevant population (i.e., all firms competing in the industry) of cases representative

The Two QCAs

59

of the population’s diversity of cases. Note that drawing a random sample of a population may not always be the best choice for large-N QCA researchers for two reasons. First, the logic of generalizing from random samples to populations in regression analyses presupposes a substantial degree of homogeneity of cases in the population (Ragin, 2000), and is consequently basing generalizations on properties of central tendency, variability, and the shape of sampling distributions (e.g., Schwab, 1999). Thus, when using a random sample in large-N QCA studies, researchers may only generalize beyond the sample if it is reasonable to believe that the sample is a representative one. Second, random sampling is not appropriate for large-N researchers primarily interested in exploring the diversity of cases. This is due to the fact that a random sample may not represent the full diversity of cases – some configurations that occur only very rarely in a larger population (say, configurations represented by only one or two organizations in a population of 1,000 or more cases) may not be represented in the sample, thus requiring oversampling. For example, a random sample may not represent relatively uniquely differentiated organizations that represent very successful but rare configurations, which may not be desirable for organizational scholars. The Number of Causal Conditions An important consideration in QCA is the number of causal conditions included in the causal model under study. In small-N settings, researchers need to pay careful attention to the number of causal conditions in relation to the number of cases when designing a QCA study (Lieberson, 2001, 2004; Marx, 2010; Marx & Dusa, 2011; Rihoux & Ragin, 2009). This is because as the number of conditions and thereby the number of logically possible configurations of conditions increases, each case will tend to become its own unique configuration. Such situations make it difficult for QCA to find any commonality across cases in explaining the outcome as well as to rule out ill-specified or even nonsensical theoretical models. Simulations of csQCA models by Marx and colleagues (Marx, 2010; Marx & Dusa, 2011) demonstrate that the possibilities of finding an explanatory model in random data increases as the proportion of conditions to cases exceeds certain thresholds. Their findings suggest that a consequence of exceeding this threshold is the violation of QCA’s core assumption that ill-specified or atheoretical models will have low consistency, thereby violating the assumption that high consistency implies validity of the set theoretic relation and thus the analyzed model (Marx & Dusa, 2011). This implies that in cases where these thresholds of proportions of conditions to cases are exceeded, the use of

60

THOMAS GRECKHAMER ET AL.

QCA should be accompanied by in-depth cross-case and within cases analyses; even in such situations QCA provides a systematic mapping of causal conjunctions as well as introducing simplifying assumptions in a manner that are closely connected to the cases (e.g., Stokke, 2007). Overall, researchers need to be mindful of Marx and colleague’s (Marx, 2010, Marx & Dusa, 2011) tentative guidelines when conducting and specifying csQCA models in small-N QCA studies.6 Large-N applications, on the other hand, do not as readily face this same problem. Nevertheless, a limit to the number of conditions still exists within large-N settings, mainly due to reasons of complexity. This is because each additional condition (k) doubles the number of logically possible configurations (i.e., logically possible configurations ¼ 2k). For example, increasing the number of conditions from 10 to 12 quadruples the number of logically possible configurations from 1,024 to 4,096 (210 ¼ 1,024; 212 ¼ 4,096), and adding one more condition doubles this number yet again (213 ¼ 8,192). One implication of this exponentially increasing complexity is that as the number of conditions examined increases, so too does the difficulty of interpreting the findings, because of an increase of both the number of configurations that may be sufficient (and/or necessary) for the occurrence of an outcome and of the complexity of the configurations themselves. Indeed, given this complexity, the results of an analysis involving more than 8–10 conditions are likely to be intractable (Ragin, 2008). We foresee another implication that will directly impact researchers conducting large-N QCA studies: following the prevailing logic in research applying general linear regression approaches (e.g., Davis, 2010; Edwards, 2008; Williams, Vandenberg, & Edwards, 2009), it is likely that reviewers (and peers) will request the inclusion of more ‘‘control variables’’ in their QCA models. This extant expectation for the use of control variables constitutes a potential (and perhaps formidable) barrier for the acceptance of large-N QCA research in highly regarded journals. While this will require researchers to make compelling conceptual arguments for their specification of included conditions – which as argued above constitutes a vast opportunity given the relative dearth of configurational theorizing – this will also involve methodological arguments that go beyond the already wellarticulated ‘‘case-oriented’’ versus ‘‘variable-oriented’’ arguments (e.g., Fiss, 2007; Ragin, 2000, 2008). Researchers conducting large-N QCA studies here can draw on criticism recently leveled against practices surrounding the use of control variables in regression analyses. Specifically, researchers have recently critiqued the ‘‘rampant and relatively unthinking use of control variables’’ (Davis, 2010, p. 701). This critique draws on the observation that control variables may frequently

61

The Two QCAs

affect the interpretation of effects primarily ascribed to substantive variables of interest (Edwards, 2008, pp. 481–482). Additionally, as measurement errors in control variables increase, so too does the interpretation of the focal variables under study (Williams et al., 2009). Thus, rather than simply accounting for alternative explanations (and using up degrees of freedom) as conventionally believed, this recent literature suggests that control variables bias the results regarding the substantive variables under study. Furthermore, although statistical control is very different in nature from experimental control (Ragin, 1987), Davis (2010) argues that another reason for the overuse of control variables is that researchers (and reviewers) fail to recognize that their studies are quasi-experiments and thus rather than taking the appropriate steps to remediate the resulting design weaknesses (e.g., Cook & Campbell, 1979), they add control variables to their models, which often is an inexpensive and convenient remedy considering the increasingly abundant data in many areas of organizational research. In sum, decisions regarding the number of causal conditions to include in QCA studies are vital. Small-N researchers including too many conditions may inadvertently render their results meaningless. Large-N QCA researchers have the option of including a greater number of conditions (from 6 to up to 12 conditions) but are likely to face additional hurdles in explaining why they did not include more conditions in their modeling. To navigate these hurdles, large-N applications of QCA may (at least initially) need to clearly articulate why their specifications of perhaps 7–8 (or fewer) conditions is not only adequate but appropriate. Methodologically, this will require a shift away from conventional ‘‘net effects’’ notions to configurational thinking in the evaluation of large-N QCA research. At the same time, large-N QCA researchers who are prepared to provide theoretical arguments for the conditions included (and excluded) from their specifications and are able to enhance a study’s validity by ruling out alternative explanations via remedies other than control variables (e.g., Cook & Campbell, 1979) or through alternative tests (e.g. Fiss, Sharapov, & Cronqvist, 2013) will be better positioned to overcome expectations and requests to include more control variables in their QCA specifications, in addition to developing more soundly specified causal models.

Analyses Processes Consistency Consistency ‘‘indicates how closely a perfect subset relation is approximated’’ (Ragin, 2008, p. 44). The basic notion of consistency is perhaps most

62

THOMAS GRECKHAMER ET AL.

easily conveyed in connection with csQCA as it describes the proportion of cases belonging to any particular configuration (for in-depth discussions of consistency see Ragin 2000, 2008). For example, assume that for a certain configuration of causal attributes 26 of the 30 firms sharing the configuration also display the outcome (and thus 4 do not), the consistency of this configuration would equal 0.867 (i.e., about 87% of cases in the configuration share the outcome). While it is desirable to have consistency as close to 1 as possible, (near) perfect consistency is more likely to be obtained in small-N studies (Ragin, 2006). Regardless of the sample size, Ragin (2008) has suggested a minimum consistency of .80 for inferring a subset relationship, and extant organizational research has been at or above this recommendation (e.g., Crilly, 2011; Fiss, 2011; Greckhamer, 2011). Applying this consistency threshold, the example configuration can be said to be consistently linked to the outcome, and given appropriate theoretical justification can be said to be a sufficient causal combination for the occurrence of the outcome. There are two main considerations that somewhat differ across small-N and large-N studies with respect to consistency. To begin with, while the use of probabilistic criteria and benchmarks is limited in small-N studies, they are a viable option in large-N settings. Even a finding of perfect consistency in small-N settings may not support an argument that a causal combination is sufficient at a statistically significant level, because depending upon how small the sample is, the evidence may not be adequate to rule out that the finding has occurred by chance (Ragin, 2000). Large-N settings afford the possibility to determine whether relationships of necessity or sufficiency occur at a statistically significant level. Second, and as discussed in more detail below, consistency may be shaped by possibilities to resolve contradictory configurations. The presence of contradictory configurations by definition lowers consistency scores. Hence, strategies for the resolution of contradictory configurations become vital to potentially enhance the consistency of QCA results, and these strategies differ between small-N and large-N QCA studies. In any case, researchers should observe the recommended minimum consistency thresholds (i.e., Z.80; Ragin, 2008) to confidently draw inferences from their findings. Resolving Contradictory Configurations Contradictory configurations occur when cases in the same configuration show different outcomes. For QCA, contradictions weaken set theoretic consistency, making it more difficult to draw inferences about causal relationships. Ragin (2008) and Rihoux and Ragin (2009) offer several

The Two QCAs

63

strategies for resolving these logical contradictions, which entail a mix of theoretically and empirically driven recommendations. Those authors’ recommendations were tailored to small-N studies, and below we highlight how the various strategies they proposed may be extended to large-N studies. The first potential strategy is to review case selection rules. Here, researchers would question whether all of the selected cases are actually part of the relevant population. For instance, a researcher interested in studying the causes of performance in large firms could reassess whether all firms in the sample are indeed ‘‘large.’’ Indeed, if contradictions are attributed to ‘‘borderline’’ cases (i.e., cases near the specified threshold of what constitutes large firms), then dropping these cases may well be warranted. As is true for all modes of inquiry, and particularly for QCA, the research process is very much an iterative one: theory and empirics should come together to refine the research design and thus strengthen the inquiry (Ragin, 2000, 2008). In the current running example, the theoretical threshold for what constitutes a ‘‘large’’ firm is ambiguous at best, and thus this type of iteration very much would help to inform theory (as well as potentially resolve occurring contradictions). A second strategy to deal with contradictory configurations involves the addition, removal, or replacement of one or more of the theoretically important causal conditions in the model. While this strategy is perhaps more accessible in large-N as compared to small-N studies – due to a reduced danger of surpassing the threshold of maximum proportion of causal conditions to cases – it is vital that large-N researchers rely on extant theory to identify attributes that have the potential to differentiate contradictory cases and thereby resolve contradictory configurations. While exploratory data analysis can be a useful tool and be part of the iterative nature of developing the causal model, data mining and fishing expeditions are no more valid in QCA than they are in regression-oriented analyses. A third recommendation to resolve contradictory configurations involves the re-examination of the ways in which sets – including the outcome set – are operationalized and calibrated. Inappropriate calibration of sets (i.e., the qualitative thresholds of full membership, full nonmembership, and degrees of membership are not well specified) will place cases that should be differentiated into the same configuration (i.e., differences in kind are not captured). This strategy appears to be a potentially very fruitful strategy for large-N samples. Because large-N researchers lack intimate knowledge of each individual case, and moreover because extant theory will often prove to be ambiguous in guiding the specification of the anchors used in calibration

64

THOMAS GRECKHAMER ET AL.

(e.g., what constitutes a ‘‘large’’ firm), researchers might have to rely on empirical knowledge such as central tendencies in the data when initially calibrating sets (e.g., mean, median, quartiles). However, as discussed above, contradictory configurations may provide an opportunity to advance theory; an examination of the cases falling into contradictory configurations, combined with extant theory, may improve specifications of the anchors in calibration (i.e., rather than simply central tendencies) and thus help to refine theory in this regard. Relatedly, large-N QCA researchers should be mindful of the quality of the data underlying the sets and how any potential problems with the quality of the data (e.g., random or systematic errors in archival sources) may be contributing to the occurrence of contradictory configurations. These recommendations also apply to the outcome, and contradictions may be resolved by reevaluating whether the outcome has been defined and calibrated properly. A fourth strategy involves developing deeper knowledge of each of the cases involved in a study so as to identify aspects of the cases that would help to resolve the occurring contradictions. While at first consideration this strategy does not appear to be practicable for large-N studies, it may nevertheless be possible considering that only the cases falling into the contradictory configuration(s) need to be so explored. For example, if the number of cases displaying the contradiction is limited (e.g. 15 or 20 cases), then it might be possible for the researcher to gain more detailed knowledge on these particular cases that helps to both resolve the contradictions and to develop a more in-depth knowledge of the cases under study. Additionally, even if the number of cases is quite large, an alternative possibility could be to take an in-depth qualitative look at a randomly selected sample of contradictory cases (e.g., Eisenhardt, 1989; Yin, 1994). A final strategy is to rely on a frequency criterion. For example, if only one in twenty cases is contradictory (e.g., 19 cases have the outcome of high performance and one has the outcome of not-high performance), one could make the judgment that this does not constitute a theoretically significant contradiction, but may more reasonably be assumed to involve factors such as coding error or randomness. As Rihoux and Ragin (2009) point out, this is the most controversial strategy as it is purely probabilistic in nature and does not lend itself to combining theoretical and substantive knowledge. Approaching contradictory configurations in this manner nevertheless constitutes a viable strategy assuming that the implications of applying this probabilistic logic are discussed as potential limitations. However, it also leaves the task of more deeply investigating the contradictory cases to subsequent studies. Overall, as is the case with this and all the other

The Two QCAs

65

strategies described, transparency is of the utmost importance when performing QCA research (Ragin, 2008), thus small-N and large-N researchers alike should report both the contradictory configurations and any steps taken to resolve it. Coverage Coverage is a measure of the proportion of cases that display the causal configuration; note that determining adequate consistency is a precondition for calculating a configuration’s coverage, because without it one cannot infer that a set relation between a configuration and an outcome exists in the first place (Ragin, 2008). Again, the concept of coverage is most easily conveyed in connection to csQCA, as this proportion is calculated by dividing the number of cases showing a specific configuration by the total number of instances of the outcome. Continuing with the example above, this configuration’s raw coverage would be calculated based upon the 26 out of 30 cases displaying the outcome; assuming that in this study there were 104 cases displaying the outcome, the raw coverage score for this particular configuration would be 0.25.7 In short, because QCA allows for equifinality, coverage provides an assessment of the relative empirical importance of each configuration that was found to be consistently linked to the outcome. It therefore is an important indicator for both small-N and large-N QCA studies. Small-N and large-N studies differ with respect to coverage in at least one important way. The first issue concerns the combined coverage of all configurations consistently linked to the outcome, or the solution coverage. Given the primary objective of small-N QCA studies to build or elaborate theory as well as the more intimate relationship the researcher has to cases in these settings, it is desirable and usually possible to attain near perfect solution coverage (i.e., after iterations of building the causal model, including application of the strategies to resolve contradictions as discussed above). Put differently, a causal model that accounts for all occurrences of an outcome can usually be developed if the number of cases under study is relatively small and allows for the building of in-depth knowledge to continuously revise and refine the model based upon qualitative exploration of the cases. To the extent that large-N QCA studies are more deductive in their focus, researchers likely have to settle for lower levels of solution coverage. Imperfect solution coverage indicates incomplete causal evidence that leaves some paths to the outcome unaccounted for. However, considering solution coverage roughly as analogous to an overall R2 in a regression analysis, it is worthwhile to remember that the explained variation is

66

THOMAS GRECKHAMER ET AL.

frequently quite low in regression-based organizational research – for the overall model specifications and particularly so for the substantive variables under study. Frequency Threshold An additional consideration pertains to decision-making surrounding the appropriate specification of the frequency threshold for configurations’ inclusion in causal analyses. That is, QCA maps all logically possible configurations as well as the full range of diversity of empirically existing configurations, and thus the researcher has to specify the minimum number of cases that must be observed for each configuration in order for it to be considered relevant for purposes of causal analysis of necessity and sufficiency. The appropriate minimum level of cases will depend upon the objectives of the study, which as discussed above, tend to differ between small-N and large-N QCA studies. In the context of small-N studies, it is common to specify a minimum frequency of one or two cases, given the small number of cases and the exploratory nature of such research as well as the researcher’s intimate knowledge with the cases. The minimum frequency for large-N studies, however, would potentially be much higher, particularly if the objective is a hypothetico-deductive model. In such studies, it may not be desirable to include configurations that occur among only very few cases in the analyses, because the presence of these low-frequency configurations might represent random forces or measurement errors (Ragin & Fiss, 2008). On the other hand, setting the minimum frequency too high may result in a reduction of the number of cases included in the analysis. In order to establish the proper level of this threshold, the researcher will typically have to use empirics as a guide. For example, Ragin and Fiss (2008) selected a frequency threshold that captured more than 80% of the cases for the analyses. If strictly adhering to an a priori set threshold (say for instance, a minimum frequency of 14 cases per configuration) results in a low inclusion of the overall cases in the analysis (60% for instance), then this implies that there is a relatively large degree of diversity present (implications could range from a small number of relatively well-represented configurations being excluded to a relatively large number of configurations containing relatively few cases being excluded) and thus excluding this diversity would be undesirable and should be avoided by selecting a proper frequency threshold that takes this tradeoff into account. It is also important to note that while researchers may exclude configurations not represented by some minimum number of cases from causal analyses, any such rare configurations nonetheless may highlight cases of

The Two QCAs

67

interest to explore in more depth to fully understand the diversity of cases under study. For example, some of these rarely occurring configurations may represent vital niches or populations of relatively newly created organizations to be explored further in a subsequent study (i.e., a follow-up small-N or case study analysis). Alternatively, the inclusion of rare configurations may be warranted theoretically: if, for example, an inquiry is about firm performance, and to the extent that competitive advantage is necessarily held by very few firms in an industry (e.g., Peteraf, 1993), excluding such rare configurations may be detrimental to the study’s research design. In such projects, researchers may be able to reduce the likelihood that rare configurations are the result of measurement errors; for example, in their study of firm performance, Greckhamer et al. (2008) aggregated performance and causal attributes over three time periods to increase the reliability of cases’ set membership. In short, unless the researcher has theoretical reasons for doing so, we suggest that in setting their minimum frequency thresholds, large-N researchers strike a balance between the inclusion of at least 80% (see also Ragin & Fiss, 2008) of the overall cases and a relatively high number of cases per configuration. Moreover, researchers conducting large-N studies may consider experimenting with both relatively high and relatively low frequency thresholds, which would yield more coarse-grained analyses focusing on the dominant configurations linked to the outcome of interest and more fine-grained analyses influenced by relatively rare configurations, respectively. Interpretation of Findings In the interpretation of findings, differences between small-N and large-N QCA come back full circle to the goals of the research: whereas small-N QCA studies are typically aimed at theory building and tend to follow an inductive logic, large-N QCA studies may be utilized to build or test theories, thus following an inductive or deductive logic. Because existing literature provides guidance with respect to interpreting small-N QCA findings (e.g., Ragin, 2000, 2008) as well as inductive large-N QCA findings (e.g., Greckhamer et al., 2008; Ragin & Fiss, 2008), here we focus on highlighting a few issues involved in the interpretation of deductive large-N QCA inquiries. First, considering the potential challenges involved in deducing hypotheses regarding multiple conjunctive causality, we emphasize that large-N studies following a hypothetico-deductive logic must clearly specify hypotheses about predicted relationships of causal necessity and/or sufficiency. For example, hypotheses such as ‘‘equifinal causal configurations will lead to outcome Y’’ or ‘‘configurations of firm practices will lead

68

THOMAS GRECKHAMER ET AL.

to high firm performance’’ are not readily testable and falsifiable; they simply reiterate configurational assumptions rendering them to be tautology. Building on the latter example (and the hindsight provided by the findings of Greckhamer et al., 2008), hypotheses such as ‘‘a high level of corporate slack resources is a sufficient condition by itself for achieving high performance among manufacturing firms’’ or ‘‘a combination of abundant corporate slack resources, large firm size, and industry stability leads to high performance among service firms’’ provide enough specificity to be testable (i.e., falsified). Second, as discussed above, large-N QCA applications including those of a hypothetico-deductive nature share the basic properties and assumptions of QCA and set theoretic methods; these assumptions shape (and limit) the extent to which the findings of large-N QCA studies lend themselves to empirical generalizations. In short, the extent to which researchers can generalize findings of relations supporting claims of necessity and sufficiency beyond their sample will depend upon the initial construction of the study sample and the incorporation of any simplifying assumptions. With respect to the former, researchers need to be mindful that in large-N QCA applications the emphasis remains on complexity, even in hypotheticodeductive studies. That is, in the inevitable tradeoff between complexity and generalizability of findings – ‘‘an appreciation of complexity sacrifices generality; an emphasis on generality encourages a neglect of complexity’’ (Ragin, 1987, p. 54) – QCA’s focus on complex causal combinations and the integrity of cases trades off generalizability for contextual realism and complexity. Thus, for example, a finding of support for the hypothesis that a combination of abundant corporate slack resources, large firm size and industry stability is sufficient for high organizational performance among a representative sample of S&P 500 service firms has limited generalizability in that it has no implications for cases beyond this population of organizations (i.e., S&P 1500 service firms; smaller service firms, etc.) nor does it have implications for S&P 500 service firms not displaying all elements of this configuration of attributes. In sum, to the extent that generalizability is desirable and given these properties of QCA, the study sample should be constructed at the outset with due consideration for representativeness – as discussed above, a representative random sample or a stratified sample that captures the diversity of cases – or alternatively include the population of cases. Simplifying assumptions about configurations not found in the dataset (i.e., easy and difficult counterfactuals taken into account in deriving the solutions; see Ragin, 2008) also affect interpretation. While a discussion of

The Two QCAs

69

the interpretation of the alternative solutions available in QCA (i.e., complex, parsimonious, intermediate) is beyond the purview of the current chapter, suffice it to say that any conclusions drawn from hypotheses tests and consequently any generalizations drawn from empirical results hinge on the plausibility of any included simplifying assumptions and the extent to which researchers can defend their inclusion on theoretical grounds (Ragin, 2000, 2008). Thus, not only should interpretations of results be made with these simplifying assumptions in mind, but also should researchers transparently explicate those simplifying assumptions that may affect their interpretations in their discussion of the results. Third, interpretations of large-N QCA studies, not unlike those of their small-N counterparts, are shaped by the fact that set theoretic relationships allow for asymmetric causal relationships, that is, the causal conditions leading to an outcome’s presence may be quite different from simply being the opposite of the causes leading to the outcome’s absence (Fiss, 2011; Ragin, 2008). Moreover, frequently researchers studying an outcome (e.g., firms with high performance) may not be interested in what leads to the absence of the outcome (e.g., firm with not high performance), but rather in an outcome that is best captured by means of a separate set (e.g., membership in the set of firms with low performance – in which membership would be calibrated according to theoretical and substantive knowledge as to what constitutes low performance). Therefore, any findings for hypothesized configurations of firm attributes leading to high firm performance, for example, cannot be generalized as also having implications for configurations leading to low performance. Instead, researchers would need to calibrate the set of firms with low performance. Indeed, as previous research has shown, the causal combinations that lead to not high or low performance may be quite different (and sometimes asymmetrical) to those leading to high firm performance (see Fiss, 2011; Greckhamer et al., 2008). In addition to these considerations for the interpretation of deductive large-N studies, interpretations of large-N QCA studies more generally will be affected by the difference in relationship that the researcher has with the cases themselves. As discussed above, the researcher’s relationship to the cases in small-N applications is much more intimate and thus interpretation of the results of causal analysis of necessity and sufficiency can greatly benefit from linking the observed cross-case patterns with in-depth knowledge of individual cases (Rihoux & Ragin, 2009). Conversely, in large-N studies a return to the cases may not be (immediately) possible or feasible, due to the greater distance of the researcher from the cases and the lack of case-specific knowledge needed to return to the cases. Thus, results

70

THOMAS GRECKHAMER ET AL.

of causal analysis of sufficiency and necessity are interpreted primarily as patterns across many cases and as we outlined above, researchers should take care not to interpret their findings beyond the boundaries inherent to their study’s research design.

THE PROSPECTS FOR LARGE-N CONFIGURATIONAL ANALYSIS QCA can lay claim to being one of the few genuine methodological innovations to have occurred in the social sciences over the last few decades (Gerring, 2001). While QCA’s initial development and proliferation was driven in small-N situations, this chapter suggests that there is both a need for a large-N QCA approach and potential to enhance QCA’s applicability to large-N situations. Our preceding discussion of contrasting small-N and large-N QCA shows that they share vital basic foci on configurations and complex causality and that they both employ a set theoretic perspective and a Boolean algebraic approach. At the same time, a large-N QCA approach differs from its small-N counterpart with respect to goals, assumptions, and research processes. In particular, in addition to its potential to support theory building shared with small-N approaches, large-N QCA can be utilized for hypothesis testing and deductive reasoning and by its very nature maintains a distance between the researcher and the cases. In these respects large-N QCA applications are analogous to conventional general linear approaches that currently dominate large-N organization studies; despite these analogies, however, the QCA approach differs from the general linear approach in vital fundamental assumptions constituting its configurational nature. Because these foundational differences are in detail discussed by Ragin (2000, 2008), our focus in this last section of the chapter is to highlight the value QCA contributes to large-N organization studies. In short, the large-N QCA approach can complement existing general linear approaches to the study of organizations in at least two fruitful ways: as a standalone configurational alternative to standard regression analyses or as a complementary component in mixed-methods approaches integrated with standard regression analyses. The first issue relates to the ability to generate novel theories and insights that are fundamentally configurational, making QCA a vibrant alternative with substantive application potential for large-N organizational research. In this regard, QCA provides an alternative understanding of causality for

The Two QCAs

71

organizational researchers by making the leap from net-effects thinking to configurational thinking, also emphasizing the diversity of organizations (e.g., Fiss, 2007, 2011; Greckhamer & Mossholder, 2011; Greckhamer et al., 2008; Kogut, Macduffie, & Ragin, 2004; Ragin, 2000, 2008). This is very important to the study of organizations as a mismatch between theories and methods currently pervades much of organization studies: while theoretical discussions about organizations frequently stress nonlinear relationships and equifinality, empirical research has typically drawn on general linear model methodologies that by their very nature tend to imply singular causation and linear relationships (Fiss, 2007). While these methodologies are powerful tools for empirical research in their own right, QCA offers an approach that allows researchers to (re)discover important phenomena and research questions that do not comply with a general linear understanding of reality they construe (see Abbott, 1988). Because they start from the assumption that theory building and testing as well as formulating predictions and generalizations regarding causal processes need to take into account the diversity of cases (here organizations) (Ragin, 2000), large-N QCA applications have the potential to make unique contributions to organizational research. The second way forward for large-N QCA relates to utilizing the method as a direct complement to conventional regression analyses and a suitable component of mixed-methods approaches. Such mixed-method studies could be utilized in a host of fruitful ways. For one, they could be used to answer (a) particular research question(s) by examining the same data from these alternative perspectives in a manner that employs the strength of each. For example, studies that hypothesize independent main effects (best tested by linear regression) as well as complex interaction effects (best examined via QCA) may be best served in this way, particularly when the hypothesized interaction effects involve multiple attributes (and thus may go untested in general linear approaches). Another potential use is that of triangulation: these alternative methods may serve as robustness checks for each other. For instance, one might utilize QCA to identify a particular configuration leading to the outcome in question by and then use solution membership as a predictor in a more standard regression analysis, allowing further for the addition of control variables that might make a QCA analysis too unwieldy. While considerable work remains to be done to explore the intersection and the potential complementarities between QCA and standard regression analysis, current efforts to explore these complementarities are under way (Fiss et al., 2013); we believe doing so presents a promising way forward.

72

THOMAS GRECKHAMER ET AL.

Additionally, large-N QCA can also be complemented by either small-N QCA or qualitative exploration of cases. That is, the results of large-N QCA studies could be used to identify cases of certain configurations as vital for understanding causal relations and to guide selection of cases for more indepth study of the causal mechanisms underlying patterns of relationships. For example, researchers studying the determinants of organizational performance could choose (a sample of) cases representing a configuration that is linked to the outcome of interest with high consistency, and conduct in-depth qualitative case studies to explore why combinations of attributes representing the configuration may lead to the outcome of interest. In closing, QCA holds great promise for both small-N and large-N social research in general and organization studies in particular, and both of these variants of QCA should become standard tools in the organizational researcher’s toolbox. The purpose of this chapter was to establish that due to its alternative perspective and complementary properties as compared to conventional general linear approaches, large-N QCA holds significant potential for organization studies. To help future researchers harness this potential, we provided guidance for large-N QCA applications by discussing the ways in which it departs from small-N QCA applications. In order for large-N QCA analysis to flourish, best practices and conventions still need to be developed. To do so, we hope that organizational scholars using QCA continue the dialogue on large-N applications of QCA in organization studies we aimed to begin with this chapter.

NOTES 1. In this chapter, we use Qualitative Comparative Analysis (QCA) to encompass both crisp-set (csQCA) and fuzzy-set (fsQCA) QCA and only use the more specific terms when warranted by the discussion. 2. Although ‘‘causal conditions’’ and ‘‘causality’’ is the terminology commonly used in the QCA approach, we fully recognize that QCA has the same limitations as other methodologies (i.e., regression-oriented approaches) when it comes to making causal inferences. See Greckhamer et al., (2008, footnote 1) and Ragin (2008, pp. 13–20) for more on this issue. 3. The primary means of designating and examining set relations are the two basic Boolean operators – logical and and logical or. The operator and represents the intersection of sets, and is used when conditions A and B combined may lead to an outcome. The operator or represents the union of sets, and is used when either one condition or another may lead to the same outcome. For more detailed explanations, see for example, Fiss (2007), Greckhamer et al. (2008) and Ragin (1987, 2000, 2008).

73

The Two QCAs

4. Examples of large-N QCA studies in other disciplines taking an inductive approach include Amoroso and Ragin (1999), Miethe and Drass (1999), Ragin and Bradshaw (1991), and Ragin and Fiss (2008). 5. As discussed above, csQCA requires the researcher to specify full membership, while fsQCA requires thresholds for full, non, and partial memberships; partial membership can be calibrated through setting of a cross-over point in continuous fuzzy sets or the setting of multiple anchors to calibrate multi-value fuzzy sets, for example four-value (e.g., Crilly, 2011), five-value or seven-value fuzzy sets (Ragin, 2000, 2008). 6. While their guidelines are tentative, Marx and Dusa (2011) suggest that csQCA models that exceed the proportion of conditions to cases recommended based on their simulation study should not be analyzed because the probability of generating results with random data increases beyond a 10% chance. The implications of their work for fuzzy set analysis have not yet been determined. 7. In addition to calculating a configuration’s raw coverage as demonstrated here, researchers can calculate each configuration’s unique coverage (i.e., the amount of coverage that does not overlap with other configurations) as well as the overall solution coverage (i.e., the proportion of cases showing the outcome falling into any of the configurations consistently linked to the outcome) (see Ragin, 2006, 2008). .

REFERENCES Abbott, A. (1988). The system of profession: An essay on the division of expert labor. Chicago, IL: University of Chicago Press. Aiken, L., & West, S. (1991). Multiple regression: Testing and interpreting interactions. Newbury Park, CA: Sage. Amoroso, L., & Ragin, C. (1999). Two approaches to understanding control of voluntary and involuntary job shifts among Germans and Foreigners from 1991 to 1996. Vierteljahreshefte zur Wirtschaftsforschung, 68, 222–229. Becker, H. (1992). Cases, causes, conjunctures, stories, and imagery. In C. Ragin & H. Becker (Eds.), What is a case? Exploring the foundations of social inquiry, (pp. 205–216). Cambridge, UK: Cambridge University Press. Bitektine, A. (2008). Prospective case study design: Qualitative method for deductive theory testing. Organizational Research Methods, 11, 160–180. Cook, T., & Campbell, D. (1979). Quasi-experimentation: Design and analysis issues for field settings. Chicago, IL: Rand McNally. Crilly, D. (2011). Predicting stakeholder orientation in the multinational enterprise: A mid-range theory. Journal of International Business Studies, 42, 694–717. Davis, G. (2010). Do theories of organizations progress? Organizational Research Methods, 13, 690–709. Edwards, J. (2008). To prosper, organizational psychology shouldyovercome methodological barriers to progress. Journal of Organizational Behavior, 29, 469–491. Eisenhardt, K. (1989). Building theories from case study research. Academy of Management Review, 14, 532–550. Fiss, P. (2007). Towards a set-theoretic approach for studying organizational configurations. Academy of Management Review, 32, 1180–1198.

74

THOMAS GRECKHAMER ET AL.

Fiss, P. (2011). Building better causal theories: A fuzzy set approach to typologies in organization research. Academy of Management Journal, 54, 393–420. Fiss, P., Sharapov, D., & Cronqvist, L. (2013). Opposites attract? Opportunities and challenges for integrating large-N QCA and econometric analysis. Political Research Quarterly, 66, 191–197. Gerring, J. (2001). Social science methodology: A criterial framework. Cambridge, UK: Cambridge University Press. Greckhamer, T. (2011). Cross-cultural differences in compensation level and inequality across occupations: A set-theoretic analysis. Organization Studies, 32, 85–115. Greckhamer, T., Misangyi, V. F., Elms, H., & Lacey, R. (2008). Using qualitative comparative analysis in strategic management research: An examination of combinations of industry, corporate, and business-unit effects. Organizational Research Methods, 11, 695–726. Greckhamer, T., & Mossholder, K. (2011). Qualitative comparative analysis and strategic management research: Current state and future prospects. In D. Ketchen & D. Bergh (Eds.), Building methodological bridges. Research methodology in strategy and management (pp. 259–288). Bingley, UK: Emerald. Kogut, B., Macduffie, J., & Ragin, C. (2004). Prototypes and strategy: Assigning causal credit using fuzzy sets. European Management Review, 1, 114–131. Lieberson, S. (2001). Review of fuzzy-set social science by Charles Ragin. Contemporary Sociology, 30, 331–334. Lieberson, S. (2004). Comments on the use and utility of QCA. Qualitative Methods. Newsletter of the American Political Science Association. Organized section on qualitative methods, 2, 13–14. Marx, A. (2010). Crisp-set qualitative comparative analysis (csQCA) and model specification: Benchmarks for future csQCA applications. International Journal of Multiple Research Approaches, 4, 138–158. Marx, A., & Dusa, A. (2011). Crisp-set qualitative comparative analysis (csQCA), contradictions and consistency benchmarks for model specification. Methodological Innovations Online, 6, 103–148. Meyer, A., Tsui, A., & Hinings, C. (1993). Configurational approaches to organizational analysis. Academy of Management Journal, 36, 1175–1195. Miethe, T., & Drass, K. (1999). Exploring the social context of instrumental and expressive homicides: An application of qualitative comparative analysis. Journal of Quantitative Criminology, 15, 1–21. Miller, D. (1986). Configurations of strategy and structure: Towards a synthesis. Strategic Management Journal, 7, 233–249. Miller, D. (1996). Configurations revisited. Strategic Management Journal, 17, 505–512. Miller, D., & Friesen, P. (1978). Archetypes of strategy formulation. Management Science, 24, 921–933. Miller, D., & Friesen, P. (1984). Organizations: A quantum view. Englewood Cliffs, NJ: Prentice Hall. Mintzberg, H. (1979). The structuring of organizations: A synthesis of the research. Englewood Cliffs, NJ: Prentice Hall. Oldham, G., & Cummings, A. (1996). Employee creativity: Personal and contextual factors at work. Academy of Management Journal, 39, 607–634. Peteraf, M. (1993). The cornerstones of competitive advantage: A resource-based view. Strategic Management Journal, 14, 179–191.

The Two QCAs

75

Ragin, C. (1987). The comparative method: Moving beyond qualitative and quantitative strategies. Berkeley, CA: University of California Press. Ragin, C. (2000). Fuzzy-set social science. Chicago, IL: University of Chicago Press. Ragin, C. (2006). Set relations in social research: Evaluating their consistency and coverage. Political Analysis, 14, 291–310. Ragin, C. (2008). Redesigning social inquiry: Fuzzy sets and beyond. Chicago, IL: University of Chicago Press. Ragin, C., & Bradshaw, Y. (1991). Statistical analysis of employment discrimination: A review and critique. Research in Social Stratification and Mobility, 10, 199–228. Ragin, C., & Fiss, P. (2008). Net effects versus configurations: An empirical demonstration. In C. Ragin (Ed.), Redesigning social inquiry: Fuzzy sets and beyond (pp. 190–212). Chicago, IL: University of Chicago Press. Ridder, H., Hoon, C., & McCandless, A. (2009). The theoretical contribution of case study research to the field of strategy and management. In D. Bergh & D. Ketchen (Eds.), Research methodology in strategy and management (pp. 137–175). Bingley, UK: Emerald. Rihoux, B., & Ragin, C. (2009). Configurational comparative methods: Qualitative comparative analysis (QCA) and related techniques. Los Angeles, CA: Sage. Schwab, D. (1999). Research methods for organizational studies. Mahway, NJ: Lawrence Erlbaum Associates. Skarlicki, D., & Folger, R. (1997). Retaliation in the workplace: The roles of distributive, procedural, and interactional justice. Journal of Applied Psychology, 82, 434–443. Stokke, O. (2007). Qualitative comparative analysis, shaming, and international regime effectiveness. Journal of Business Research, 60, 501–511. Williams, L., Vandenberg, R., & Edwards, J. (2009). Structural equation modeling in management research: A guide for improved analysis. The Academy of Management Annals, 3, 543–604. Wood, J., Michela, J., & Giordano, C. (2000). Downward comparison in everyday life: Reconciling self-enhancement models with the mood-cognition priming model. Journal of Personality and Social Psychology, 79, 563–579. Yin, R. (1994). Case study research: Design and methods (2nd ed.). Thousand Oaks, CA: Sage.

CHAPTER 4 CONFIGURATIONAL ANALYSIS AND ORGANIZATION DESIGN: TOWARDS A THEORY OF STRUCTURAL HETEROGENEITY Anna Grandori and Santi Furnari ABSTRACT This chapter reconstructs the roots of configurational analysis in organization theory and organizational economics, focusing on the elements of configurational thinking that are particularly relevant to organizational design; and outlining some future prospects for a configurational theory of organization design. We detect the presence of configurational ideas in many organization theories and organizational economics approaches. We argue that this, seldom acknowledged, continuity extends and enriches the implications of configurational analysis for organization design. In addition, we define and identify ‘structural heterogeneity’ as an organizational property that can be distinctively studied by configurational analysis, distinguishing between internal heterogeneity – diversity of organizational attributes within one configuration – and external heterogeneity – diversity of organizational configurations under the same environmental conditions. Some of the

Configurational Theory and Methods in Organizational Research Research in the Sociology of Organizations, Volume 38, 77–105 Copyright r 2013 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0733-558X/doi:10.1108/S0733-558X(2013)0000038008

77

78

ANNA GRANDORI AND SANTI FURNARI

insights that can be gained through a configurational analysis of structural heterogeneity are illustrated through a fs/QCA study of a multi-industry sample of firms. Keywords: Organizational configurations; contingency theory; organizational economics; heterogeneity; QCA

INTRODUCTION This chapter focuses on the conceptual and methodological contributions of a configurational approach to organization theory (OT) and design (OD). Although much work in configurational analysis (CA) has been descriptive, the way in which organizational forms are described in CA is conducive to contributing significantly to a (much called for) renewal of OD. To distinguish and develop these implications of CA, the chapter revisits the configurational elements already present in classic approaches to OD, especially in structural contingency theory (SCT) and in more recent and economic approaches such as transaction cost economics (TCE) and complementarity-based approaches. This excursus leads to disclosing greater continuity and knowledge accumulation with respect to earlier OT than formerly acknowledged. Although researchers in CA have frequently stressed the differences and ‘rivalry’ of CA as an approach with respect to other approaches (according to a very common but not particularly fruitful custom in OT), the continuity disclosed is not a ‘diminutio’ but the opposite: distinctive strengths, added value and possible further developments of CA emerge more clearly. Configurationism is clarified as a type of ‘analysis’ rather than substantive approach or theory that can, however, greatly contribute to renewing OD theories, also in terms of content. The chapter will demonstrate how the features of this analytic approach, including theoretical elements such as the definition of units of analysis as well as the methods for analysing interactions, affect the content of theoretical developments. Indeed, one of these possible developments is proposed in the second part of the chapter. CA could lead to constructing a significant ‘missing piece’ in OT. The missing piece is a theory of structural heterogeneity – intended as an explanation of the existence and effectiveness of different configurations under the same conditions – that is a theory of the ‘equifinality of forms’. The model of structural heterogeneity developed in the chapter includes this form of heterogeneity as well as another form

Configurational Analysis and Organization Design

79

that has thus far been conceptually dismissed or neglected, namely, the heterogeneity of organizational features within a configuration: what are the ‘conceptual differences’ among the variety of ‘conceptually distinct’ organizational elements clustered within configurations? What are the possible or impossible, necessary or optional, combinations of organizational elements in relation to performance outcomes? Should the ‘syndromes of attributes’ defining a configuration be ‘consistent’, ‘coherent’, ‘similar in kind’ as conventional wisdom would have it, or are there greater degrees of freedom? The chapter is organized as follows. The first section reviews the presence and features of CA elements in OT and OE approaches to OD. The second section constructs a typology of approaches to CA relevant to OD: a ‘map of configurationism’ along the two dimensions of internal and external ‘structural heterogeneity’. The third section presents an empirical analysis featuring the construction of new Structural Heterogeneity Indexes, of hypotheses on their variations, and propositions on the qualitative compositions of high-performing combinations of organizational elements and contingencies.

CONFIGURATIONISM IN ORGANIZATION DESIGN A widely shared view of ‘organization forms’ is that they are collections of attributes (Polos, Hannan, & Carroll, 2002). This view perhaps first emerged with Weber’s notion of bureaucracy. In fact, as these last authors noted, ‘If any approach to defining organizational forms can be regarded as the standard, it is one that regards forms as particular clusters of features. The example par excellence is Weber’s specification of rational-legal bureaucracy in terms of the nature of authority (y), procedures (y), and the employment relation of the official (y)’ (ibid. 2002, p. 87, emphasis in original). This feature-based conceptualization and operationalization of organization forms remained a central feature in almost all perspectives on organization forms and design, and is the root of a ‘configurational’ notion of forms as combinations of attributed, that has been more common than usually acknowledged. SCT, to start with, considered Weberian attributes and rendered them contingent. Methodologically, however, those attributes were considered to contribute to some ‘dimensions’ in an additive way. For example, in Aston operationalization (e.g. Pugh, Hickson, Hinigs, & Turner, 1969), the

80

ANNA GRANDORI AND SANTI FURNARI

dimension of ‘formalization’ was constructed by asking whether a series of elements such as organization charts, job descriptions, written operation manuals and procedures were present or absent, and summing up the 1s (indicating present). Likewise, in Lawrence and Lorsch (1967) the dimension of the ‘structuring of activities’ was constructed by summing up scales measuring the span and frequency of control, the detail of procedures and job descriptions, the number of hierarchical levels, while the dimension of integration was constructed by summing up the presence of practices ranging from procedures and programs to hierarchical coordination, to team coordination, to dedicated integration units. What precisely is then the difference between these approaches and a ‘configurational approach’ as an analysis of the ‘multidimensional constellation of conceptually distinct characteristics that commonly occur together’ (Meyer, Tsui, & Hinings, 1993)? A core and much emphasized difference is both substantive and methodological: the unit of analysis shifts from ‘dimensions’ (e.g. ‘degrees of’ formalization, centralization, standardization, differentiation, integration) to ‘qualitatively different’, ‘conceptually distinct’ attributes. Mintzberg (1979, 1983), for example gave an explicitly configurational version of SCT. He considered the main coordination mechanisms identified by SCT studies as core elements of organizing that are found in different combinations in different ‘forms’. This type of CA was denoted by a ‘taxonomic’ and empirical approach. Any ‘commonly occurring cluster’ is a configuration. Examples of archetypes defined in this way – de facto clustering of some array of organizational practices – include, in addition to Mintzberg’s five forms, Miles and Snow’s (1978) strategic and organizational types (prospectors, defenders, adaptors) (Doty, Glick, & Huber, 1993); Pugh et al.’s (1969) empirical taxonomy of bureaucracies; Child’s analysis of international JV configurations (2002) and many others (e.g. Miller & Friesen, 1984). However, there is also an ‘ideal-type’ or ‘typological’ configurationism, where Weber’s notion of the bureaucratic ideal-type can be considered the forerunner. In typological configurationism there is more pronounced theoretical effort in defining both why the constitutive elements are ‘conceptually distinct’ and the principle according to which they are expected to cluster (Doty & Glick, 1994). In contingency versions, this perspective appeared in mature contingency views (Drazin & Van de Ven, 1985). In this case, an ideal association of traits that is theoretically expected to work is defined, and real types of combinations are presumed to be more effective the closer they are to the ideal type. Drazin and Van de Ven (1985) called it a

Configurational Analysis and Organization Design

81

‘systemic fit’ notion, matured after earlier selection-based views (e.g. technology ‘determines’ structure, i.e. unfit forms are selected out) and bi-variate interaction views of fit (e.g. co-variance between technology and structure increases performance). In ‘systemic fit’ and ‘typological’ configurationism, the study of ‘internal fit’ among elements emerged in addition to the classic contingency theory concerning ‘external fit’. However, as Drazin and Van de Ven pointed out at the time, a methodological ‘gap’ prevented studying the interactions between ‘internal’ and ‘external’ fit (wherefore they used a stage-wise procedure). In addition, ‘coherence’ remains the underlying hypothesis on how the attributes should cluster, as per the traditional SCT view: ‘bureaucratic’/‘systematized’ organizational mechanisms are supposed to cluster together in one ideal-type configuration; alternative configurations are informed by alternative logics such as ‘organic’/‘developmental’, and are internally homogeneous. More recent configurational studies have highlighted at least two other important properties of configurations: the possibility that relations among the constitutive elements of configurations are ‘non-linear’ and that two or more configurations may be ‘equifinal’ in generating performance in certain given circumstances (Meyer et al., 1993; Short, Payne, & Ketchen, 2008; Gresov & Drazin, 1997). The notion of equifinality actually has a long history in organizational thought. Originally defined by the open system biologist Ludwig von Bertalanffy, and widely utilized in some organizational approaches such as socio-technical studies (Trist, Higgin, Murray, & Pollock, 1963), equifinality has been analysed empirically in a configurational perspective. For example, Gresov (1989) identified multiple, equifinal ODs of work units under specific combinations of conflicting contingencies, such as when units face low task uncertainty and high dependence; Galunic and Eisenhardt (1994) reported that different forms of compensation systems are equally effective in specialized retail stores, in contrast with agency theory predictions. Equifinality introduces more variety in possible combinations, while earlier taxonomic configurationism emphasized that organizational configurations are ‘surprisingly’ few in number (Meyer et al., 1993; Miller & Friesen, 1984). ‘Non-linearity’ in relations among organizational traits, or between these and contextual dimensions, are also non-distinctive of CA per se. Examples are any hypothesis and findings of U-shaped relations. However, nonlinearity assumes a stronger meaning in CA. For example, it means, ‘variables found to be positively related in one configuration may be unrelated or even inversely related in another’ (Meyer et al., 1993, p. 1178). Alternatively, nonlinearity may derive from positive or negative complementarities among

82

ANNA GRANDORI AND SANTI FURNARI

elements and multiple interaction effects that go beyond bi-variate interaction effects traditionally analysed in organization studies (Delery & Doty, 1996; Miller & Mintzberg, 1983). In sum, CA have resurrected and extended important concepts in OT, such as equifinality, multi-finality and non-linearity, which had somehow become forgotten as a central concern in organizational analysis and design, perhaps also due to the lack of powerful methods to study these phenomena. CA revitalized theorizing about these phenomena also due to new combinatorial, qualitative comparative analysis (QCA) methods. In conclusion, in our reconstruction, greater continuity than generally recognized is revealed between organizational configurational studies and earlier organizational studies. Rather than being a sort of ‘weakness’ this can be seen as a strength: CA contributed new insights to organizational analyses accruing previous knowledge in a cumulative way, thus extending previous theory rather than proposing the nth new theory. Together with those steps forward, modern CA in OT has been marked by some limitations. Theoretically, the ‘conceptual distinctions’ among elements are have been progressively lost, for increasingly adopting an empirical stance of analysing any ‘organizational practice’ that can be observed in practice. The central task of ‘contingency’ theory – that is the substantive specification, in theoretical terms, of which types of configurations are effective under what circumstances – has been rather abandoned. Finally, the laws behind the effective clustering of attributes have not actually been worked out. The principle of ‘coherence’ among organizational attributes has typically been invoked, implicitly or explicitly, but this principle has always been rather opaque. Empirically, CA ‘manifestos’ typically start out with the idea that effective configurations are ‘few’, and this is seen as a puzzling fact to be explained. However, this ‘fact’ actually depends heavily on the type of categorization employed. Empirical research has often shown that possible configurations may in fact be ‘many’, almost a continuum, if the analysis is fine-grained enough (e.g. many attributes are considered). For example, Child (2002) classified international JV organizational profiles into three groups on the single attribute of ownership (majority, parity, minority position of the international vs. domestic partner) but within each group, there are almost as many configurations as cases. Other studies (Letre´my & Cottrell, 2003), using connectionist methods based on distance among vectors of attributes, detected tens of configurations rather than ‘a few’: for example labour contract provisions (such as open ended/fixed term,

Configurational Analysis and Organization Design

83

presence of shifts, flexible schedule, part-time, etc.), ending up with 10 configurations (groups of contracts characterized by similar patterns and frequencies close to 100% in the presence of different provisions). Methodologically, the simultaneous analysis of internal and external fit remained a task for future research, arguably due to a time lag in the maturation of proper statistical methods. The properties of ‘equifinality’ and ‘non-linearity’ were more proclaimed rather than actually studied, arguably for much the same reason (Fiss, 2007). In addition, the mere ‘qualitative’ operationalization of organizational elements or practices, and their measurement as ‘present’ or ‘absent’, is not conducive to disclosing whether relations among them are linear or non-linear; a simultaneous consideration of ‘quantitative’ variations in their intensity or level of application would be needed. The strand of research in organizational economics based on ‘complementarity’ is seldom considered in reviews of configurational studies, but it is remarkably configurational in approach, and it actually shares many of the advantages and limitations of OT configurationism. It intended to address the problem of ‘internal fit’ and to clarify what’s behind it. With the notion of complementarity, OE has contributed to CA by defining internal fit and coherence more precisely. Milgrom and Roberts’ influential paper on complementarity and fit launched an entire stream of configurational studies based on the complementarity hypothesis, thereafter also influencing OT and HR studies (Delery et al., 1996; Ichniowski, Kathryn, & Prennushi, 1997; Laursen & Manke, 2002; Whittington, Pettigrew, Peck, Fenton, & Conyon, 1999). Milgrom and Roberts applied Edgeworth’s classic notion of complementarity among goods and services to strategic and organizational attributes. They stated that ‘attributes are complementary if doing (more of) any one of them increases the returns to doing (more of) the others’ (Milgrom & Roberts, 1995). Hence, complementarity has been defined as ‘supermodularity’ in the performance function f ðDx þ DyÞ4f ðDxÞ þ f ðDyÞ where x and y are any two complementary elements (e.g. goods and services, or organizational and strategic practices). In empirical research, ‘attributes’ have been operationalized as ‘practices’ and have included attributes that in OT would be classified in part as organizational – for example ‘pay for performance’, Taylorization of work, horizontal communication – in part as ‘contextual’ – for example longlinked vs. intensive technologies, mass market vs. niche strategies. This

84

ANNA GRANDORI AND SANTI FURNARI

approach may be seen more as a strength than a weakness from a configurational perspective. In fact, the distinction between ‘independent’ and ‘contextual’ variables vs. ‘organizational’ variables should loose relevance in a configurational approach: not only fit among organizational mechanism but also fit between these and strategic, technological or institutional practices can be studied, explained and clarified in terms of complementarity. Beyond these achievements, it should be noticed that some of the problems and limitations of CA in OT have remained or became even more prominent. First and foremost, the origins of ‘complementarity’ are no clearer than the origins of ‘coherence’ or ‘synergy’. This statement can be supported by examining the original illustrations of the complementarity framework given by Milgrom and Roberts – a comparison between ‘mass production’ and ‘flexible production’ practices, and an analysis of the Lincoln Electrics case. The ‘mass production’ array of attributes is claimed to be based on ‘the transfer line, interchangeable parts, and economies of scale’, and to include practices such as specialized machinery, long production runs, specialized skill jobs, central coordination and hierarchical planning, high inventories, vertical integration. The ‘flexible production’ array was characterized, by contrast, by a logic of ‘flexibility, speed, economies of scope and core competencies’, and identified by practices as flexible machines, short production runs, highly skilled cross-trained workers, worker initiative, horizontal communication and cross-functional teams, low inventories, reliance on outside suppliers. Hence, in this first example, complementarity and fit seem to stem from the ‘similarity in logic’ among practices. Another example is given next, the organization of Lincoln Electrics. The case was deemed famous for having revived ‘Taylorist’ practices as piece rate compensation based on time and motion studies, but able to offset and correct all the (in)famous problems of these systems by extensive employee ownership, a permanent employment policy with no layoffs even during severe crises, wide reliance on make rather than buy, the use of cross functional teams at a time when they were extremely rare in American manufacturing, flexible work rules and extensive firm-specific training. The authors argue that these distinctive features are complementary, but they do not notice that they are so for a different reason to ‘similarity in logic’: actually, some traits are drawn from a hard-nosed ‘Fordist’ capitalistic firm model and some traits from a flexible collective enterprise model; and it is thanks to their ‘difference in logic’ that they are able to balance, actually to counter-balance, each other.

Configurational Analysis and Organization Design

85

In subsequent research and in the main statements of the perspective, ‘consistency’ among organizational attributes or practices – in the sense of ‘congruence’, ‘similarity’ and ‘alignment’, hence the homogeneity of attributes – has been emphasized. In OE, Williamson (2004) addressed the issue openly and clearly, referring to the conceptual (Simonian) notion of similarity and difference ‘in kind’ among organizational attributes as an explanation of why we find them clustered in ‘coherent syndromes’. This assumption is also prevalent in OT, as we have argued (see also Doty et al., 1993). However, the notion of complementarity as homogeneity faces many counterexamples and counterarguments, as the Lincoln case, suggesting that both similarity and difference in kind as well as homogeneity and heterogeneity in traits can actually be sources of complementarity (Grandori & Furnari, 2009). In fact, there are many important and widely studied organization forms whose main advantage is precisely ‘incoherence’ or ‘diversity’ among the constitutive elements. Examples include all notions of external and internal ‘hybrids’ as forms mixing attributes drawn from different homogeneous ‘syndromes’ (such as ‘markets’, ‘hierarchies’ and collectives’) to improve the response to multiple/contrasting design requisites (e.g. Cohendet, Creplet, Diani, Dupoue¨t, & Schenk, 2004; Grandori, 1997; Hennart, 2013; Lindkvist, 2004; Zenger & Hesterly, 1997)1 and notions as organizational ‘ambidexterity’ (O’Reilly & Tushman, 2004) and ‘bimodality’ (Baharami, 1992): forms with simultaneously ‘opposite’ traits – centralization and decentralization, high regulation and high autonomy, individualism and collectivism – to foster performance in dynamic competitive conditions. Hence, the principle of complementarity as homogeneity is too simple, at best a particular case, since the association of organizational practices may add value precisely due to their heterogeneity rather than homogeneity. On the ‘theory loss’ issue, an empiricist approach to the definition of both practices and their possible combinations prevailed in OE to a greater extent than in OT, and the ‘list’ of practices became even more pronouncedly a ‘laundry list’. This is likely to ‘leave resources on the table’, that is, to concentrate on configurations that are all ‘sub-optimal’: what about outperforming outliers? Or even untried combinations? Where do the lists of practices come from? Are all the practices considered actually relevant? Responses to these questions would impart a much greater design power to CA. On methodological issues, CA in OE has indeed applied sophisticated methods and tools to test truly interactive effects among organizational traits. However, the math available has thus far not allowed including more than a few attributes (Athley & Stern, 1998), typically operationalized in

86

ANNA GRANDORI AND SANTI FURNARI

binary terms (presence/absence of a practice). These methodological limits have prevented studying complementarities on wide clusters of attributes (if not through standard correlation-based methods). The rest of this chapter is an effort to provide some advances on all the open issues identified in the conclusions of the reviews of CA both in OT and in OE.

A MAP OF ORGANIZATIONAL CONFIGURATIONISM Building on the above analysis of previous studies, we identify two important dimensions that can be used to characterize different approaches to CA as well as the different organizational configurations themselves. There are two types of structural heterogeneity (SH): the heterogeneity of organizational elements within a configuration (‘internal structural heterogeneity’) (ISH), and the heterogeneity of configurations effective under the same ‘contingencies’ (‘external structural heterogeneity’) (ESH). These two dimensions are used here to construct a ‘map of organizational configurationism’,2 which in turn will be useful in identifying the gaps/challenges for future research (Table 1). Cell A describes ‘the best way to organize in each given circumstance’ approach. In addition, that ‘best way’ is defined by a ‘coherent syndrome’ in which all elements are ‘of the same kind’. This approach was typical of early SCT, the most salient template being the ‘mechanistic’ vs. ‘organic’ systems partition, with the effectiveness of each system contingent on the uncertainty of the system’s task environments. Forms are therefore attributed the property of ‘structural unifinality’: each of these does well in one context and for one purpose. Cell B is more novel in that it points to the advantages of internal structural variety. Two different types of advantages have been highlighted in different models. In a simpler and early version, the heterogeneity of organizational attributes within the same organizational entity simply stems from its ‘differentiation’ into parts in turn adapted to their different task environments (as in Lawrence and Lorsch’s and Thompson’s SCT works). A more configurational notion of blending and mixing traits was that of ‘ambidexterity’ (O’Reilly & Tushman, 2004). This reformulation enriched classic contingency arguments especially in considering the contingency of forms on two qualitatively different types of organizational results that could be interesting to achieve simultaneously: efficiency and innovation,

External Heterogeneity (of configurations under the same conditions)

Yes

No

Table 1.

D – More than one ‘internally heterogeneous’ configuration can be effective under each configuration of conditions (e.g. organizational chemistry; fuzzy-set configurationism)

(e.g. ‘systemic fit’ approaches; complementary-based CA)

(e.g. ambidexterity, ‘bimodality’)

(e.g. classic SCT e.g. organic vs. mechanistic systems; classic TCE e.g. bureaucracies vs. clans) C – More than one ‘internally homogeneous’ configuration can be effective under each configuration of conditions

B – One ‘internally heterogeneous’ configuration is effective under a specified configuration of conditions

Yes

A – One ‘internally homogeneous’ configuration is effective under each configuration of conditions

No

Internal Heterogeneity (of attributes in the same configuration)

A Map of Organizational Configurationism.

Configurational Analysis and Organization Design 87

88

ANNA GRANDORI AND SANTI FURNARI

exploitation and exploration. While ‘ambidexterity’ has been analysed mainly in the context of large firms, and assumed to be realized by means of specializing and dedicating different parts of a structure to these different purposes, others have noted that in radically innovative contexts, and new entrepreneurial firms, the entire structure tends to exhibit ‘opposite’ characteristics at the same time: it is both centralized and decentralized, both formal and informal, both individualistic and communitarian – in a word it is ‘bi-modal’ (Bahrami, 1992). In one way or another, these forms have the property of ‘structural multifinality’, that is, they are able to achieve multiple purposes. Cell C defines a locus in which configurations are thought/found to be ‘coherent’ (internally homogeneous) clusters of attributes, but there may be more than one effective combination under the same ‘external’ circumstances, that is, there can be equifinality among forms. Roberts (2004) indicated examples of the ‘puzzle’ of different arrangements appearing to be equally effective under the same circumstances: for example the ‘disaggregated’, ‘let one thousand flowers bloom’ approach of BP vs. the ‘planned micro-economy’ approach of other successful firms in the petroleum industry, or the contrasting approaches adopted in organizing for innovation: a communitarian approach as, say, at Nokia, vs. the ‘highly powerful incentives’ found in other innovative firms. Cell C and Cell B is where most of CA has lived thus far; at least if we consider, as our map does, only those contributions that have specified/ modelled the type of heterogeneity envisaged, and are then useful from a design perspective. A limitation of many CA studies has actually been the lack of this kind of modelling. From an empiricist stance, a set of empirically observable practices were analysed in terms of clustering regularities and (eventually) relation to contingencies and performance. In this way, it could just happen by chance that some heterogeneous elements are found to be combined in a configuration, or that more than one configuration is effective in multiple circumstances, but these types of regularities – if detected at all – remain under-conceptualized, unexplained and are therefore scarcely applicable in design. Cell D is the most complex and novel, and the most fully configurational, in the sense that it takes into account both internal and external interaction effects among heterogeneous attributes of organization and context. Both the external and internal heterogeneity of forms is admitted. In this cell, both the possible equifinality of different configurations and the possible multifinality of each single configuration are admitted and enquired.

Configurational Analysis and Organization Design

89

On the basis of this wider picture, we note that the map can be interpreted as implying increasing degrees of freedom in design in moving from Cell A to Cell D. In other words, in all Cells the organizational configurations are contingent to the configuration of conditions, but there is a ‘degree’ in contingency – namely, different types of conditions constrain organizational solutions to a different extent. This notion is novel. In Cells A and B there is a one-to-one correspondence between structures and context. Greater degrees of freedom emerge when we move to Cell C, and even more so when we shift to Cell D. On the basis of this observation, new research questions and conjectures can thus be formulated: are these different degrees of contingency actually present in reality, rather than only being differences among approaches? That is to say, can internal and external heterogeneity vary under specifiable conditions? Which are they? In the next section, some propositions on the expected variations of external and internal heterogeneity of high-performing organizational configurations are developed and some exploratory empirical evidence is provided.

AN ANALYSIS OF INTERNAL AND EXTERNAL HETEROGENEITY The hypotheses on the nature and predictors of internal and external heterogeneity advanced here are grounded in the configurational approach to OD that we have developed in a series of previous studies (Grandori & Furnari, 2008, 2009). The explorative empirical evidence reported here comes from a data base on the adoption of a set of organizational practices that was also used in our previous research.3 The practices considered in the survey were identified and classified with the use of a theory-based typology of organizational elements (Grandori & Furnari, 2008): (1) market-like elements (M), such as individual-based or team-based pay for performance systems; (2) bureaucratic elements (B), such as formal rules and procedures for human resource evaluation and monitoring; (3) communitarian elements (C), such as teamwork and knowledge sharing systems.4 The intensity of use of each type of element was then measured on a 0–4 scale based on the number of organizational practices adopted for each type.5 We refer to this number as the ‘dose’ at which each type of element is infused in the organizational system Organizational configurations for high efficiency

90

ANNA GRANDORI AND SANTI FURNARI

and innovation have been identified through fs/QCA (see Appendix and Figs. A.1–A.4 for details). For the present analysis, new measures and data have been considered, and new hypotheses on how heterogeneity should vary with some important contingences have been developed. First, two indexes of structural heterogeneity have been constructed. An Internal Structural Heterogeneity Index (ISHI) is constructed as the number of types of elements (1, 2 or 3) represented in a given configuration over the total possible types of elements (which in our case is 3), multiplied by the sum of the doses at which the type of elements are used. For example, a configuration featuring two doses of only one type of element, say market, will have a ISHI ¼ (1/3)2 ¼ 0.66. Only doses equal or greater than two are considered here, as an internal heterogeneity at level 1 is a necessary condition for any type of high performance in this sample (see Grandori & Furnari, 2008 for empirical evidence on this finding). Hence, this index ranges from 0.66 to 12. The index takes the maximum value of 12 when all three types of elements are presented in a configuration, each used with the maximum intensity, which in our sample is four doses. An External Structural Heterogeneity Index (ESHI) is easier to devise. A simple ESHI is provided by the number of different effective organizational configurations under a given configuration of contingencies. Second, the data base has been integrated with additional data on ‘contingencies’, such as the size of the organizational system (SME vs. large firms) and the uncertainty of the task-environment (high-tech/low-tech sectors) (see Appendix for measures). This extension allows exploring two sets of research questions in a ‘contingent configurationism’ approach, bringing together the study of internal and external fit. Those questions and hypotheses are illustrated next, together with pertinent evidence. Contingent Heterogeneity A first type of questions defines a problem of ‘contingent heterogeneity’: When structural heterogeneity among the relevant organizational elements is higher or lower? Does the ‘degree of freedom’ in designing configurations vary across circumstances? Which circumstances? Building on previous theoretical development (Grandori, 2001), we advance a hypothesis that has never been explored before: H1. Higher degrees of complexity (system size, task uncertainty and innovation outcomes) are associated with lower equifinality (i.e. lower

91

Configurational Analysis and Organization Design

external heterogeneity), and with higher multi-functionality (i.e. higher internal heterogeneity) of organizational configurations. The rationale is that in ‘simple’ situations essentially anything can work (albeit at a different cost): rules, authority, communitarian coordination, incentives. It is when information complexity enters the picture that some begin to ‘fail’: if systems are large and computational complexity grows, centralized organization fails; if task-environments are highly variable the application of rule-like governance is impaired; if problems are unstructured and innovation is crucial, price and exit governance is put under strain. Hence, the set of feasible configurations should narrow down as size increases, task become more uncertain and innovation is the relevant outcome to be reached. H1 is innovative with respect to classic OD, according to which only forms enriched in bureaucratic elements should be effective in less uncertain/ simpler conditions, and only communitarian and ‘organic’ governance should be effective under uncertain/complex conditions. Figs. 1–4 plot the average values of the internal heterogeneity index in different combinations of conditions (types of outcomes to be achieved, size and sector contingencies). As expected, ISHI generally increases in combination with more ‘complex’ contingencies, represented by shifts from low-tech to high-tech, and from smaller to larger size, both in the achievement of efficiency and

9.00 8.00 7.00 6.00 5.00 4.00 3.00 2.00 1.00 0.00 Low-Tech

High-Tech

Internal Heterogeneity of Efficient Configurations

Fig. 1.

Internal Heterogeneity of Efficient Configurations in Low-Tech and HighTech Sectors.

92

ANNA GRANDORI AND SANTI FURNARI 9.00 8.00 7.00 6.00 5.00 4.00 3.00 2.00 1.00 0.00 Small-medium

Large

Internal Heterogeneity of Efficient Configurations

Fig. 2.

Internal Heterogeneity of Efficient Configurations in Small-Medium and Large Organizations.

9.00 8.00 7.00 6.00 5.00 4.00 3.00 2.00 1.00 0.00 Low-Tech

High-Tech

Internal Heterogeneity of Innovative Configurations

Fig. 3.

Internal Heterogeneity of Innovative Configurations in Low-Tech and High-Tech Sectors.

innovation, with some unexpected differences in the absolute levels of the Indexes and some signal that firm size does not behave as an indicator of complexity (Fig. 2). As to the unexpected levels of ISHI, comparing Figs. 1 and 3, we note that ISHI is higher for efficiency than for innovation in high-tech. This suggests an interesting possible explanation/refinement of H1 as concerns

93

Configurational Analysis and Organization Design 9.00 8.00 7.00 6.00 5.00 4.00 3.00 2.00 1.00 0.00 Small-medium

Large

Internal Heterogeneity of Innovative Configurations

Fig. 4.

Internal Heterogeneity of Innovative Configurations in Small-Medium and Large Organizations.

internal SH. It is possible that the heterogeneity of contingencies matters. The data behave as if a lower ISHI is needed if the contingencies faced are themselves more homogeneous, although more uncertain/difficult: that is, generating innovation in uncertain/high-tech sectors poses more homogenous demands on structure than achieving efficiency in high-tech/ uncertain conditions. Second, the downwards sloping line in Fig. 2 indicates that the ISHI for efficiency is lower in large sized firms rather than small sized. It seems that large firms competing on efficiency have to specialize their structure in one or a maximum of two directions. The second part of this analysis, on the qualitative composition of configurations, will specify which these directions are. The shape of the relationship in Fig. 2 can be interpreted with a cost argument. If competition is on costs and efficiency, the cost of investing in varied organizational practices as size grows may be weighted more, thereby pushing internal heterogeneity down. Efficient SMEs can afford and seem to require higher ISHI than efficient large firms. In competing for innovation, although ISHI is somewhat lower for SME, its value is nevertheless high. In this case, the explanation may be that it may be particularly difficult for smaller firms to keep up with innovation, substantive investments may be required in information and monitoring systems and a higher structural articulation than in less innovative SME may be required (some converging qualitative evidence that small firms that are particularly innovative have a particularly articulated structure has been found in research on

94

ANNA GRANDORI AND SANTI FURNARI

entrepreneurial firms). All in all, in any case, it seems that smaller size ‘complicates’ rather than ‘simplifies’ things. The analysis of External Heterogeneity also confirms H1, with some qualification on the role of firm size. Figs. A.1–A.4 show the results of a QCA of the configurations for high efficiency and high innovation, incorporating respectively uncertainty and size as two relevant contingencies.6 They indicate that the number of highperforming configurations is much higher for efficiency (18 configurations) than for innovation (10 configurations), as expected. If we add size and high technology as ‘contingencies’, the number of high-performing configurations, which we take as an indicator of ESHI, necessarily decreases, but not by the same amount in all conditions. Considering size as a relevant contingency, ESHI is higher for larger organizations (seven configurations for efficiency, three for innovation) than for smaller firms (three configurations for efficiency, one for innovation, see Figs. A.1 and A.3). In other words, larger firms enjoy greater degrees of freedom in organizing than smaller firms. This contrasts with the idea that size is a source of ‘complexity’ or otherwise of difficulty in organizing and further supports the finding and the interpretation emerging on ISHI: smaller size complicates business life and organization. When considering sectors, the number of high-performing configurations reduces from five in low-tech sectors to three in high-tech sectors. This is consistent with our hypothesis that higher task complexity narrows the portfolio of possible effective configurations. However, unexpectedly, this is valid only for efficiency outcomes. For innovation outcomes under uncertainty, instead, heterogeneity reappears, with the number of configurations expanding from two to four as we move from innovative low-tech firms to innovative high-tech firms (see Fig. A.4). Again it seems that the presence of ‘heterogeneous contingencies’ represents a particular difficulty by itself, reducing the degrees of freedom by posing constraints that are ‘different in kind’. Hence, we find lower ESHI and higher ISHI more frequently where contingencies are heterogeneous, rather than when they all point towards the pole that is generally presumed to be more complex, namely, the larger size/high-tech/innovation pursuit combination. Therefore, the ‘contingent heterogeneity hypothesis’ is supported in its general terms: there is variance in the degree of heterogeneity of forms across conditions. The more specific hypothesis that states that external heterogeneity is lower in organizing for innovation than for efficiency is also supported. However, adding further contingencies suggests that further refinements are possible. Putting together these findings on ESHI with those

95

Configurational Analysis and Organization Design

on ISHI, we note an interesting symmetry that would have been difficult to predict without QCA: maximal ISHI is detected for efficiency in high-tech, and minimal ESHI is observed in the opposite conditions: innovation in low-tech (and small size). These patterns suggest that a qualitative, configurational view of contingencies or situational variables, and not only of organizational variables, is useful. What makes a situation ‘difficult’ is not so much a high value of a ‘situational’ variable per se. Their combination is what could generate difficulties; and difficult combinations may not correspond to ‘high’ (or ‘low’) values of the variables per se. Difficulties seem to stem from lower levels of complementary among the conditions themselves. This would respond to the problem of ‘contrasting contingencies’ – early noted but never really addressed in SCT. It is in these combinations that the degrees of freedom especially decrease: since there are multiple non-redundant constraints (constraints that do not put the same demands on structures) the number of equifinal combinations is reduced – in some cases to one, as in achieving innovation in lower size conditions. Contingent Complementarity A second type of questions defines a problem of ‘contingent complementarities’: Which strings of ‘internally’ complementary organizational traits are complementary to which strings of ‘external’ conditions? Methodologically, a possible way to achieve this simultaneous specification of the ‘external’ and ‘internal’ fit is to include ‘contextual’ variables such as technology and sector uncertainty, size/complexity of activities, etc, in the ‘strings’ of elements whose complementarity is going to be assessed. Substantively, building on organization theory and previous theoretical development

Table 2.

H2.1–H2.4 on Contingent Complementarities.

Conditions

Outcomes Efficiency

Innovation

Low technology and size

H2.1 Equifinal enrichments in either B, or C or M

H2.2 Equifinal enrichments in either C or M

High technology and size

H2.3 Multifunctional enrichments in both B and M

H2.4 Multifunctional enrichments in all elements: B, and C, and M

96

ANNA GRANDORI AND SANTI FURNARI

(Grandori & Furnari, 2009), a second group of hypotheses are advanced and summarized in Table 2. H2.1. Configurations enriched in any element – either C or M or B – are associated with high efficiency in smaller and less uncertain activity systems. According to the theory behind H1, no type of element – M, B or C – faces conditions of ‘failure’ as long as system size is kept small, and uncertainty is not high. If all elements are viable, it is sensible to let the data reveal which type of element prevails, if any, in the specific sample studied. H2.2. Configurations enriched in either M or C elements are associated with high innovation in smaller and less uncertain activity systems. It has been observed that there are at least two types of organizational logics that could sustain the innovativeness of economic behaviours (Roberts, 2004): one is the knowledge sharing and goal sharing ‘clan-like’ or ‘communitarian’ logic (Ouchi, 1979, 1980), the other is the ‘highly powered incentives’ logic (Zenger & Hesterly, 1997). Here, we add the observation that those configurations may be mutually exclusive only under conditions of small size and low task complexity, due to a principle of coordination cost saving. H2.3. Configurations enriched in both M and B are associated with high efficiency in larger and more uncertain activity systems. The combination between high efficiency and high-tech/high uncertainty is seldom studied or even conceived; it is presumed that firms in these conditions should necessarily strive for innovation. However, the combination interestingly seems to represent situations of ‘routinized’ innovation, characterized by highly specialized tasks with known patterns of effective connections to systematically generating new products (pharmaceutical firms can be brought as examples); especially in large firms. H2.4. Configurations enriched in all B, M and C simultaneously are associated with high innovation in larger and more uncertain activity systems. The combination between high innovation outcomes and complex conditions – high-tech/high size – is the most difficult to manage. According to the theory behind our hypotheses most mechanisms – B, M and C alike – face problems, if not failures, if employed in a standalone fashion. Hence,

Configurational Analysis and Organization Design

97

we hypothesize that configurations should converge to few, even a single high-performing configuration, with maximal internal heterogeneity. H2.1–H2.4 summarize the broad trend that can be theoretically anticipated. This predictive effort leads to seeing a clearer and general expected pattern: as ‘contingencies’, broadly intended to include both the types of tasks to be mastered and the types of outcomes to be reached, become more challenging, the links between organizational elements pass from an ‘or’ link to an ‘and’ link among elements. Including also a specification of the identity of elements in configurations, the formulation of those hypotheses is going to be useful in interpreting results, and to detect and explain ‘unexpected’ results. In fact, in a sense, in CA one should always expect to find some unexpected results, since the combinations are so many that developing predictions for all of them seems either too costly for the marginal returns, or even logically impossible. Hence, it seems fair to admit that, while we made a point on the usefulness of theoretical prediction in CA, that type of analysis has an unavoidable empiricist aspect due to the number of possible combinations. Figs. A.1–A.4 contain information also on the components of configurations and can therefore be used for discussing the ‘contingent complementarities’ hypothesis, as follows. In organizing for efficiency, infusions of further elements (beyond the core) are necessary, but their quality (M, C or B) does not matter so much. This squares well with H2.1 – there are more ways of achieving efficiency than of achieving innovation. The results further suggest that this external heterogeneity is especially high for large sized firms striving for efficiency that should however limit their internal structural heterogeneity. In organizing for innovation, the identity of mechanisms matters more. In addition to fit to contingencies, high-performing infusions of elements seem to depend on complementarity with the elements that are already diffused in the initial/average conditions: in this sample (but this is likely to be common), B elements abound in large firms, whereby infusions of M and/ or C are called for. This situation is likely to be common, and in fact those infusions are typically the recommended cure to make large bureaucratic firms more ‘flexible’ (e.g. Zenger & Hesterly, 1997). By the same logic, a much less noted and conceptualized recommendation emerges: in lower size, infusions of M and/or B are beneficial. In fact, in smaller sized firms communitarian and informal practices abound on average, hence investments in C have low marginal returns there; rather, innovation is better served by investing in M and B practices, also taking into account that smaller size in combination with innovation is a ‘heterogeneous contingency set’, hence a difficult set.

98

ANNA GRANDORI AND SANTI FURNARI

As to hypothesis 2.4, stating that the complexity of conditions and outcomes should drive configurations towards a high ISHI structure, with intense enrichments in all elements M and B and C, some refinements emerge. Fully-fledged multimodal structures are rare. This confirms a result obtained in previous analyses (Grandori & Furnari, 2008): there is a ceiling to ISH, arguably due to decreasing marginal returns to investments in coordination in general and in the variety of coordination mechanisms in particular. However, the general hypothesis that more stringent/difficult conditions should drive ISHI up – hence produce a more stringent ‘and’ (rather than ‘or’) link among elements – squares well with the observations. Still, those more stringent/difficult conditions seem to be chiefly represented by situations of ‘heterogeneous contingencies’ rather than contingencies of any one type that we are used to consider more complex (e.g. large organizational size, high task complexity, innovation).

CONCLUSIONS The chapter reviewed and revealed the roots and constant presence of elements of CA in almost all the main approaches to organization analysis and design in OT and OE. This view of CA is a contribution in itself, as it has seldom been noted since the efforts of its proponents have been devoted more to stressing differences and discontinuities with other approaches. However, CA is more an analytic approach than a substantive approach or theory, alternative to other approaches and theories. The consequence of this interpretation is not a reduction but an expansion of its heuristic power, both in terms of its application domain and in terms of its potential to renew organization theory in its merits. The chapter also offers a ‘map’ of configurational studies relevant for OD. The map is a typology of approaches within CA based on two dimensions that emerged as key from the literature review. In fact, while all CA is based on the identification of ‘conceptually distinct elements’ and how they can be combined, there have been different ways of modelling the ‘laws of clustering’: some of the studies hypothesize that only ‘coherent’ and ‘similar’ elements can cluster, some envisage complementarities among elements that ‘differ in kind’, some hypothesize an one-to-one correspondence between one configuration of contextual contingencies and one effective organizational configuration, some envisage multiple effective configurations in the same conditions. A distinctive methodological contribution of this chapter has

Configurational Analysis and Organization Design

99

been to measure ‘structural heterogeneity’ and to advance and empirically explore some propositions on how it varies across contexts characterized by different levels of task uncertainty and complexity/size of the organized system. A new empirical application of QCA has been presented to demonstrate how this type of analysis can lead to substantive contributions, such as the positive relation between the heterogeneity of contingencies and the internal heterogeneity of structure, the higher equifinality of different configurations in large firms competing on efficiency (with respect to other conditions) and the substantive specification of which organizational elements are complementary under what conditions. Other methodological contributions of the type of CA proposed here reside in analytical options that can overcome some of the main limitations of previous CA studies identified in the course of the literature review. They include an analysis oriented to detecting effective strings of elements, rather than just ‘traits commonly occurring together’; and a simultaneous analysis of external and internal fit, reconnecting contingency theory and complementarity theory. In addition, the notion of ‘fit’ is made more precise by distinguishing ‘necessary and/or sufficient causes’ for observing performance outcomes; and by starting to specify what types of elements are expected to be complementary and when. Performance and situational variables have also been analysed as configurations of elements that ‘differ in kind’ (e.g. efficiency vs. innovation, high tech vs. mature industries) as much as organizational elements. These advances should contribute to taking a significant step forward in a CA towards a more fine-grained, chemistry-like, OD, based on the combination and infusion of ingredients with specified effects and purposes.

NOTES 1. These notions of ‘hybrids’ are configurational as they envisage the combination of mechanisms belonging to different structural alternatives, while Williamson’s notion (1991) of hybrids is not configurational since these forms are defined as ‘intermediate’ between markets and hierarchies, characterized by intermediate values of organizational traits typical of the extremes of the continuum. 2. We are very grateful to Peer Fiss for his constructive comments on a former version of this matrix. 3. The sample includes 75 firms drawn from the largest 600 independent organizations in Italy. Details on the measures of organizational practices and organizational performance (efficiency and innovation) are fully reported in Grandori and Furnari (2008).

100

ANNA GRANDORI AND SANTI FURNARI

4. Our original classification included a fourth class of ‘democratic elements’ (Grandori & Furnari, 2008). However, we do not consider this category here for reasons of analytical tractability. 5. More specifically, for each type of element, we identified four practices, measuring each of them with 4 sub-practices. For example, one of the four practices used to measure market-like elements (M) was ‘pay for performance’, a practice which was in turn measured with the presence/absence of four sub-practices (e.g. individual, team-based, firm-based, stock options type of pay for performance). We then consider a practice to be present in an organization (with the value of ‘1’) only if that organization adopted one sub-practice above the average number of subpractices adopted for the corresponding practice. This average level has been found to have an important property: the presence of all types of elements (M, B and C) at least at that level was found to be a necessary condition for high performance of any sort (either efficiency or innovation) in our previous QCA analysis (Grandori & Furnari, 2008). 6. The relatively low unique coverage scores of the organizational configurations detected may be due to the small sample size and to the multi-industry nature of the sample. Therefore, the empirical evidence reported in these tables can be interpreted as illustrative and exploratory.

REFERENCES Athley, S., Stern, S. (1998). An empirical framework for testing theories about complementarity in organization design. NBER Working Paper 6600, NBER, Cambridge, MA. Bahrami, H. (1992). The emerging flexible organization: Perspectives from Silicon Valley. California Management Review, 34(2), 33–52. Child, J. (2002). A configurational analysis of international joint ventures. Organization Studies, 23, 781–815. Cohendet, P., Creplet, F., Diani, M., Dupoue¨t, O., & Schenk, E. (2004). Matching communities and hierarchies within firms. Journal of Management and Governance, 8(1), 27–48. Delery, J. E., & Doty, D. H. (1996). Modes of theorizing in strategic human resource management: Tests of universalistic, contingency, and configurational performance predictions. Academy of Management Journal, 39, 802–835. Doty, D. H., & Glick, W. H. (1994). Typologies as a unique form of theory building: Toward improved understanding and modelling. Academy of Management Review, 19, 230–251. Doty, D. H., Glick, W. H., & Huber, G. P. (1993). Fit, equifinality, and organizational effectiveness: A test of two configurational theories. Academy of Management Journal, 36, 1196–1250. Drazin, R., & Van de Ven, A. H. (1985). Alternative forms of fit in contingency theory. Administrative Science Quarterly, 30, 514–539. Fiss, P. C. (2007). A set-theoretic approach to organizational configurations. Academy of Management Review, 32(4), 1180–1198. Fiss, P. C. (2011). Building better causal theories: A fuzzy set approach to typologies in organization research. Academy of Management Journal, 54, 393–420.

Configurational Analysis and Organization Design

101

Galunic, D. C., & Eisenhardt, K. M. (1994). Renewing the strategy-structure-performance paradigm. In L. L. Cummings & B. M. Staw (Eds.), Research in organizational behavior (Vol. 18, pp. 215–255). Greenwich, CT: JAI Press. Grandori, A. (1997). Governance structures, coordination mechanisms and cognitive models. Journal of Management and Governance, 1(1), 29–47. Grandori, A. (2001). Organization and economic behavior. London: Routledge. Grandori, A., & Furnari, S. (2008). A chemistry of organization: Combinatory analysis and design. Organization Studies, 29(2), 315–341. Grandori, A., & Furnari, S. (2009). Types of complementarity, combinative organization forms and structural heterogeneity: Beyond discrete structural alternatives. In M. Morroni (Ed.), Corporate governance, organization and the firm: Co-operation and outsourcing in a globalised market. London: Edward Elgar Publishers. Gresov, C. (1989). Exploring fit and misfit with multiple contingencies. Administrative Science Quarterly, 34, 431–453. Gresov, C., & Drazin, R. (1997). Equifinality: Functional equivalence in organization design. Academy of Management Review, 22(2), 403–428. Hennart, J. F. (2013). Internal and external hybrids and the nature of joint ventures. In A. Grandori (Ed.), Handbook of economic organization. Integrating economic and organization theory. Cheltenham: Edward Elgar. Ichniowski, C., Kathryn, S., & Prennushi, G. (1997). The effects of human resource management practices on productivity: A study of steel finishing lines. American Economic Review, 87(3), 291–313. Laursen, K., & Manke, V. (2001). Knowledge strategies, firm types and complementarity in human resource practices. Journal of Management and Governance, 5(1), 1–22. Lawrence, P. R., & Lorsch, J. W. (1967). Organization and environment: Managing differentiation and integration. Boston, MA: Harvard Business School Press. Letre´my, P., & Cottrell, M. (2003). Working times in atypical forms of employment: The special case of part-time work. In C. Lesage & M. Cottrell (Eds.), Advances in computational management science: Connectionist approaches in economics and management sciences (pp. 111–129). Dordrecth: Kluwer Academy Press. Lindkvist, L. (2004). Governing project based firms: Promoting market-like processes within hierarchies. Journal of Management and Governance, 8(1), 3–25. Meyer, A. D., Tsui, A. S., & Hinings, C. R. (1993). Configurational approaches to organizational analysis. Academy of Management Journal, 36(6), 1175–1195. Miles, R., & Snow, C. (1978). Organizational strategy: Structures and processes. New York, NY: McGraw Hill. Milgrom, P., & Roberts, J. (1995). Complementarities and fit: Strategy, structure and organizational change in manufacturing. Journal of Accounting and Economics, 19, 179–208. Miller, D., & Friesen, P. H. (1984). Organizations: A quantum view. Englewood Cliffs, NJ: Prentice Hall. Miller, D., & Mintzberg, H. (1983). The case for configurations. In G. Morgan (Ed.), Beyond method: Strategies for social research. Beverly Hills, CA: Sage. Mintzberg, H. (1979). The structuring of organizations. Englewood Cliffs, NJ: Prentice-Hall. Mintzberg, H. (1983). Structure in fives: Designing effective organizations. Englewood Cliffs, NJ: Prentice Hall. O’Reilly, C. A., & Tushman, M. (2004). The ambidextrous organization. Harvard Business Review, 82, 74–82.

102

ANNA GRANDORI AND SANTI FURNARI

Ouchi, W. G. (1979). A conceptual framework for design of organizational control mechanism. Management Science, 25, 833–848. Ouchi, W. G. (1980). Markets, bureaucracies and clans. Administrative Science Quarterly, 25, 129–141. Polos, L., Hannan, M. T., & Carroll, G. R. (2002). Foundations of a theory of social forms. Industrial and Corporate Change, 11(1), 85–116. Pugh, D. S., Hickson, D. J., Hinigs, C. R., & Turner, C. (1969). An empirical taxonomy of structures of work organizations. Administrative Science Quarterly, 14, 115–126. Ragin, C. C. (2008). Re-designing social inquiry. Chicago, IL: University of Chicago Press. Roberts, J. (2004). The modern firm. Organizational design for performance and growth. Oxford: Oxford University Press. Short, J. C., Payne, G. T., & Ketchen, D. J. (2008). Research on organizational configurations: Past accomplishments and future challenges. Journal of Management, 34, 1053–1079. Trist, E. L., Higgin, G. W., Murray, H., & Pollock, A. B. (1963). Organizational choice. London: Tavistock. Whittington, R., Pettigrew, A., Peck, S., Fenton, E., & Conyon, M. (1999). Change and complementarities in the new competitive landscape: An European panel study, 1992–1996. Organization Science, 10(5), 583–600. Williamson, O. E. (1991, June). Comparative economic organizations: The analysis of discrete structural alternatives. Administrative Science Quarterly, 36, 269–296. Williamson, O. E. (2004). In M. Augier & J. G. March (Eds.), Herbert Simon and organization theory: Lessons for the theory of the firm in models of a man (pp. 279–297). Cambridge, MA: MIT Press. Zenger, T. R., & Hesterly, W. S. (1997). The disaggregation of corporations: Selective intervention, high powered incentives and molecular units. Organization Science, 8(3), 209–222.

103

Configurational Analysis and Organization Design

APPENDIX To measure firm size (as a proxy of organizational complexity), we used the European Union enterprise size-classes. Specifically, firms with 250 or more employees were coded as large (1), while firms with fewer than 250 employees were coded as small (0). To measure environmental uncertainty, we classified the sectors in which the firms in our sample operated in the two groups, as illustrated in Table A.1: more research-intensive and technologyintensive sectors; and more traditional and mature sectors, with relatively known technologies. We used the truth table algorithm in the fs/QCA (2.5) software as described by Ragin (2008). A minimum threshold frequency of one case per configuration and a minimum consistency value of 0.66 were used to generate the truth table. The results reported in Figs. A.1–A.4 refer only to the ‘intermediate solution’ – that is those solutions that only include simplifying assumptions based on ‘easy’ counterfactuals because in this analysis we are not interested in distinguishing between core and peripheral solutions as in previous QCA studies (e.g. Fiss, 2011).

Table A.1.

‘High-Tech’ and ‘Low-Tech’ Sectors.

Industries Grouped as ‘Low-Tech’

Industries Grouped as ‘High-Tech’

    

    

Retail banking Construction Grocery distribution Steel Food and beverages

High-tech automotive Chemistry Energy and utilities Pharmaceuticals and bio-tech Software

104

Fig. A.1.

Fig. A.2.

ANNA GRANDORI AND SANTI FURNARI

Configurations for High Efficiency in SM and Large Organizations.

Configurations for High Efficiency in Low-Tech and High-Tech Sectors.

Configurational Analysis and Organization Design

Fig. A.3.

105

Configurations for High Innovation in Small-Medium and Large Organizations.

Fig. A.4. Configurations for High Innovation in Low-Tech and High-Tech Sectors.

CHAPTER 5 THE ANALYSIS OF TEMPORALLY ORDERED CONFIGURATIONS: CHALLENGES AND SOLUTIONS Tony Hak, Ferdinand Jaspers and Jan Dul ABSTRACT In organizational research the object of study is often a process, that is, a complex of events and activities that unfolds over time. In this chapter we focus on temporally ordered configurations, which can be defined as those configurations in which conditions occur in a specific temporal order. We illustrate the aims, characteristics, and limitations of several approaches that have been proposed as tools for the analysis of temporal order. Our illustration involves an analysis of the ‘‘gestation activities’’ of nascent entrepreneurs, that is, persons involved in the creation of a new firm. We aim to identify temporal sequences of gestation activities that generate or allow a successful outcome of the gestation process, while an occurrence of the same activities in another temporal order will not generate or allow that outcome. First we discuss Event Structure Analysis and Optimal Matching and conclude that these approaches cannot provide the kind of analysis that we are aiming at in this chapter. Then we discuss Temporal Qualitative Comparative Analysis, for which our analysis points to technical limitations that

Configurational Theory and Methods in Organizational Research Research in the Sociology of Organizations, Volume 38, 107–127 Copyright r 2013 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0733-558X/doi:10.1108/S0733-558X(2013)0000038009

107

108

TONY HAK ET AL.

constrain its application. We then present and discuss an alternative approach, Temporal Necessary Condition Analysis. Keywords: Temporally ordered configurations; necessary condition; sufficient condition; Temporal Qualitative Comparative Analysis (TQCA); Temporal Necessary Condition Analysis (TNCA)

INTRODUCTION In organizational research there is an increasing interest in the study of configurations, that is, of ‘‘multidimensional constellations of conceptually distinct characteristics that occur together’’ (Meyer, Tsui, & Hinings, 1993, p. 1175). Frequently, the object of study is a process, that is, a complex of activities that unfolds over time (e.g., an innovation project, reorganization, an implementation process). The characteristics that form the configuration are ‘‘conditions’’ (e.g., conditions A, B, and C) that are present (A, B, C) or absent (a, b, c). The notation ABC, thus, represents the observation that the three conditions A, B, and C are present in a process that is studied. Temporally ordered configurations can be defined as those configurations in which conditions occur in a specific temporal order (e.g., C-A-B, meaning that, in one case, C appears first, A next, and finally B). In this chapter we use the term ‘‘(temporal) sequence’’ for such a temporally ordered configuration. Note that the term ‘‘(temporal) order’’ is used here empirically as a synonym of the word ‘‘(temporal or chronological) pattern’’ and is not meant normatively (as opposed to ‘‘disorder’’). Specific temporal sequences might generate or allow outcomes that are not generated or allowed by the same configuration of conditions if they appear in another temporal order (e.g., A-B-C or B-A-C). The terms ‘‘generating’’ and ‘‘allowing’’ (an outcome), which are used here in order to avoid the term ‘‘cause,’’ are discussed later in the section on ‘‘QCA, Sufficiency and Necessity.’’ In many fields of social research the temporal order of events (i.e., the fact that these events occur in a specific temporal sequence) is, implicitly or explicitly, considered important for relevant outcomes. Many theories are inherently temporal in the sense that the arrows in chains of variables in conceptual models are interpreted as entailing a temporal lag or duration. Usually, the model itself only represents a-temporal associations between values of the variables in a chain of concepts, but the text that explains the

The Analysis of Temporally Ordered Configurations

109

theory represented by the model often entails episodes in which a high value of a variable is processually induced by a preceding high (or low) value of another variable. As has been noted in a large body of literature since the 1980s, the temporal nature of theories that are empirically assessed is not taken into account by the traditional methods of statistical ‘‘variance analysis’’ (Langley, 1999; Mohr, 1982; Markus & Robey, 1988; Pettigrew, 1999). These temporal features are not tested with such methods and, hence, the empirical status of temporal statements is that of mere commentary. We illustrate the aims, characteristics, and limitations of approaches that have been proposed as tools for the analysis of temporal order with an example. This example is an empirical investigation of ‘‘gestation activities’’ of nascent entrepreneurs, that is, persons involved in the creation of a new firm. The aim of the analysis is to identify temporal sequences of gestation activities (e.g., C-A-B) that generate or allow a successful outcome of the gestation process, while an occurrence of the same activities in another temporal order (e.g., A-B-C or B-A-C) will not generate or allow that outcome. The various analytic approaches will be evaluated in terms of their ability to achieve this aim. A distinction can be made between different types of temporal sequence (Abbott, 1995). Our data set represents only one such type, the nonrecurrent sequence of events, that is, a temporal sequence of which the (analytical) length cannot be longer than the total number of observed events and in which these events can occur only once. First we discuss Event Structure Analysis (ESA; Heise, 1989) and Optimal Matching (OM; Abbott, 1990, 1995) and conclude that these approaches cannot provide the kind of analysis that we are aiming at in this chapter. Then we discuss Temporal Qualitative Comparative Analysis (TQCA; Caren & Panofsky, 2005; Ragin & Strand, 2008), an approach that is developed specifically for the analytic problem discussed here. As yet, TQCA has not been applied in empirical studies of temporal sequences because of technical limitations. We then present an alternative approach, Temporal Necessary Condition Analysis (TNCA; based on Dul, Hak, Goertz, & Voss, 2010). We now present the data set that we have chosen for this illustration and discuss its characteristics.

DATA: GESTATION ACTIVITIES The data set for our analysis is taken from the second Panel Study of Entrepreneurial Dynamics (PSED II; Reynolds & Curtin, 2008). The data

110

TONY HAK ET AL.

obtained includes data on the nature of persons who are actively involved as nascent entrepreneurs, on the activities that they undertake during the start-up process, and on the characteristics of start-up efforts that become new firms. Our data set consists of the data regarding all ambitious hightech start-ups (N=15) in the PSED II data set. The data set is presented in Table 1. Each row corresponds to a nascent entrepreneur. The first column is the identification number. The next five columns refer to five different gestation activities:1 B D F E H

Start of research into the Business opportunity (including writing a business plan) Start of product or service Development First availability of Financial support for the gestation process First purchase of Equipment Hiring of a first employee

The five gestational activities in this analysis represent events (i.e., start of gestation activities) rather than states (i.e., doing gestation activities for a period of time). The numbers in the cells of Table 1 represent the month number in a series from 1 (i.e., August 2003) to 59 (i.e., July 2008). The last two columns show the outcome. Five gestation processes in this data set have actually resulted in a started firm (Start-up). Six nascent entrepreneurs have quitted the gestation process without having started a firm (Quit). For Table 1.

Data Set for Analysis.

ID

B

D

F

E

H

Start-Up

Quit

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

19 19 18 22 15 30 25 29 22 22 19 22 – 27 29

25 – 18 23 17 32 25 – – 20 – 23 29 27 29

– 25 18 23 18 24 25 20 – 22 – 20 29 27 –

– 25 32 22 17 31 26 20 – 25 – 22 – – –

– – – – – – – – – 25 – 22 – – –

– – – – 20 – 29 – – 25 – 25 – 40 –

59 33 – – – – – 32 36 – 41 – 46 – –

The Analysis of Temporally Ordered Configurations

111

the four remaining nascent entrepreneurs gestation was still ongoing at the time of the fifth wave of data collection. It is not known whether the nascent entrepreneurs that have not yet completed or quitted the start-up process will eventually succeed in starting a new firm and in continuing it (the data are right censored). We have chosen this data set initially because we are interested in the temporal order of successful gestation. Later, in the process of analysis, we discovered that this data set has two characteristics that make it ‘‘difficult’’ for analysis and hence is particularly useful for an evaluation of different analytic approaches. One of the difficulties is the large number of missing events. Our data show that a firm can start after only a limited number of gestational activities (three out of five in this data set). An approach to the analysis of the temporal order of the events in this data set must be able to deal with this characteristic of the data set. The other characteristic that presents a challenge to the analysis is the quite frequent co-occurrence (i.e., in the same month) of events. Obviously this is not an indication that gestational events occur at exactly the same time, but rather of the fact that relevant temporal order, if any, in this data set means ‘‘the temporal order of events that are more distant from each other than four weeks.’’ One could say that these data are imprecise because they do not specify the week or the exact calendar date of each event. Probably it is more accurate to state that the aim of the designers of PSED II has not been to develop (or to allow that users of the data develop) a process theory in terms of days or weeks but (only) in terms of longer periods of time. The fact that the temporal order of events within a (calendar) month is unknown in this data set implies that we aim at developing a process theory in which the temporal order within a time span of a month is not taken into account. This has an important practical implication for the analysis, namely that we must allow for the fact that co-occurrence of events in one month, such as D and E in case 5, can be consistent with both a theory or hypothesis that states that the one must precede the other (‘‘D must precede E’’) and with a theory that states that the reverse temporal order should occur (‘‘E must precede D’’). Table 2 presents the data from Table 1 in the form of temporal sequences. Obviously, these sequences only include those events that actually occurred. The occurrence of two or more events in the same month is indicated by slashes between events. Arrows represent the flow of time, that is, one or more months separate the occurrence of the respective events. Outcome is coded as 1 ( ¼ Start-up), 0 ( ¼ Quit), and – ( ¼ Ongoing). The aim of the analysis is to discover temporal sequences of these five gestational activities that are causally relevant for an outcome. In this

112

TONY HAK ET AL.

Table 2.

Temporal Sequences.

ID

Sequence

Outcome

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

B-D B-F/E B/D/F-E B/E-D/F B-D/E-F F-B-E-D B/D/F-E F/E-B B D-B/F-E/H B F-B/E/H-D D/F B/D/F B/D

0 0 – – 1 – 1 0 0 1 0 1 0 1 –

analysis, the desired outcome is defined as starting a firm, whereas quitting the gestation process without starting a firm is considered an undesired outcome or failure.2

EVENT STRUCTURE ANALYSIS (ESA) AND OPTIMAL MATCHING (OM) Two types of approaches have been developed for the empirical analysis of temporal order (Krook, 2006, pp. 9–10). One type of techniques focuses on the temporal order of pairs of events, and builds a model of how an organizational path can be constructed from such pairs. The best-known example of this type is ESA (Heise, 1989). ESA builds a pictorial model of pathways which have empirically shown to exist (with an accompanying text). The model looks like a flowchart with parallel routes and iterative loops. An example of such a flowchart representing the pathways of a case of entrepreneurial decision making can be found in Morse (1998, p. 112). If ESA is applied to our data set, the entrance to the model will be the decision to start a firm, and there will be two exits: a successful start-up and quit. The model that is built in ESA is a useful starting point for an analysis of temporal sequences. First, it allows the analyst to identify a limited set of pathways (if iterative loops are ignored) that lead to an exit, that is, of those

The Analysis of Temporally Ordered Configurations

113

pathways that ‘‘generate’’ that exit. Second, it allows the analyst to identify those temporal sequences and stations that must be passed in order to reach an exit point, that is, of temporal sequences that are necessary for an outcome to be generated. Third, these pathways are intrinsically temporal, which is exactly the type of pathway that we want to analyze in this chapter. However, although an ESA model allows this type of analysis, it is not itself an analytic instrument by which temporal sequences which generate or allow outcomes are identified. ESA, thus, is a form of ‘‘within-case’’ analysis that precedes ‘‘cross-case’’ analysis as aimed at in this chapter. A second type of approaches aims at such a ‘‘cross-case’’ analysis by mapping and comparing the structure of whole temporal sequences. Optimal matching (OM) is the best-known example of this type of approach. The principle of the optimal matching technique is the insight that, if we have a limited number of events (or states), every temporal sequence of these events or states can be derived from another one by applying a limited set of procedures: insertion, deletion, and substitution. This allows the analyst to calculate the ‘‘distance’’ between two temporal sequences. The simplest way of calculating the distance between two temporal sequences is to count the number of operations required for producing the one from the other. More complex approaches assign different weights to different operations. A substitution might, for instance, get a weight of 1.5 or 2 relative to an insertion or deletion. If all temporal sequences in a data set are compared with each other, the resulting distances can be represented in a so-called ‘‘distance matrix.’’ One may then submit the resulting distance matrix to any standard classification technique (e.g., cluster analysis, multidimensional scaling) to derive families of temporal sequences. A ‘‘most typical’’ temporal sequence may then be found by finding the temporal sequence that minimizes some (possibly weighted) function of the distances to all other temporal sequences (Abbott, 1990). A cluster might be represented by a ‘‘typical sequence,’’ that is, a sequence that has the smallest average distance to all other sequences in the cluster. Finally, the relation between cluster membership and specific outcomes might be statistically assessed. If such a relation is shown to exist, this result can be non-arbitrarily interpreted in a ‘‘narrative’’ way, that is, in the form of a story in which events occur in a temporal order that makes sense. The narration will closely follow the temporal order of events as represented in this typical sequence. The result of the optimal matching approach, thus, is a set of ‘‘typical sequences’’ (which do not need to exist empirically) that differ in the likelihood by which they ‘‘generate’’ the desired outcome. If that likelihood

114

TONY HAK ET AL.

equals or approaches 1.0, then we can consider that sequence a (temporally ordered) configuration that generates the outcome. However, observed likelihoods are much lower in practice, for the simple reason that optimal matching is not designed to cluster sequences on the basis of their outcome. Differences in likelihood can be expressed as odds ratios. For instance, in a recent application of OM to data on gestational activities (very similar to our data set), Gordon (2011, p. 11) concludes: ‘‘The marginal effect for sequence similarity is an 11.7% increase in the odds of becoming operational (b ¼ 0.111, z ¼ 3.873, p ¼ 0.000) and 5.8% increase in the odds of remaining ‘‘still trying’’ (b ¼ 0.060, z ¼ 1.951, p ¼ 0.051) as compared to termination.’’ Note that the problem that OM cannot identify sequences that ‘‘generate’’ an outcome cannot be overcome by applying optimal matching in a set of only those cases that have the desired outcome, because this approach ignores the possibility that there are other cases with similar sequences that result in a failure. An important limitation of this approach is that the applied permutation statistics have problems with ties as well as non-occurring events (Abbott, 1990, p. 383). As many data sets (like our example) contain ties or nonoccurring events the method normally cannot be used for assessing which sequences might generate or allow outcomes. This raises the question whether an approach could be developed in which sequences are clustered from the outset in such a way that their association with the outcome is part of the clustering technique. QCA is an obvious candidate technique for achieving this, because the core element of QCA – the truth table – is in essence a method of clustering configurations based on their association with a specific outcome.

TEMPORAL QUALITATIVE COMPARATIVE ANALYSIS (TQCA) We assume that QCA is familiar to the reader. Although QCA was developed initially for the a-temporal analysis of configurations, recently some proposals have been made for how it could be used for the analysis of temporal sequences. De Meur, Rihoux, and Yamasaki (2009) list five ‘‘solutions’’ that have been proposed to deal with temporal order in QCA. One of them is to combine QCA with other techniques such as Event Structure Analysis, discussed above. Another one is ‘‘returning to cases in a more qualitative manner,’’ which boils down to narratively adding temporal

115

The Analysis of Temporally Ordered Configurations

information to the non-temporal QCA result. Each of the three remaining solutions, though different in detail, integrates the temporal dimension into the definition of the conditions that are analyzed. The only one of these three solutions presented and discussed in the literature as a full-fledged method is TQCA proposed by Caren and Panofsky (2005) and, partially in response to them, by Ragin and Strand (2008). The procedure that turns QCA into TQCA is the substitution of conditions (such as B, D, F, E, and H in Table 1) by a set of other conditions that specify temporal relations between them. An example of such a new condition is ‘‘business research before development’’ (notated as B-D in the example below) with the codes 1 (when business research occurs before development) and 0 (when development occurs before business research). Note that ‘‘B-D’’ is just a label (or ‘‘variable name’’) and that code 0 indicates the temporal sequence D-B. In our data set this would imply the creation of 10 temporal conditions, one for each possible pair of conditions (B-D, B-F, etc.; see the example below). For each case, each of these 10 temporal conditions is coded as either 1 or 0. The usual QCA procedures can then be applied. The advantage of this procedure is that it allows, in principle, to use QCA software for the analysis. However, there are two problems with the application of TQCA that complicate the analysis of temporally ordered conditions, the problem that co-occurrences (‘‘ties’’) cannot be coded and the problem that a code cannot be assigned to a pair of which a condition is missing. Both problems have been discussed in the (small) literature on TQCA and they have, as yet, not been solved. We can illustrate both problems with case 14 in our data set. In this case conditions B, D, and F are tied, and conditions E and H are missing: ID

B

D

F

E

H

14

27

27

27

–

–

Ties are represented by question marks, and pairs with missing events by the symbol ‘‘—’’ in the following representation of the codes required for the application of TQCA: ID B-D B-F B-E B-H D-F D-E D-H F-E F-H E-H 14

?

?

—

—

?

—

—

—

—

—

116

TONY HAK ET AL.

Accidentally, the problem of missing events can partly be solved in our data set which is, as mentioned above, right censored, if we assume that events are not missing because of measurement error – they have occurred but have not been recorded – but only because they have not happened yet. Under this assumption, E and H in case 14 will necessarily be preceded by B, D, and F. This allows us to code these pairs accordingly: ID B-D B-F B-E B-H D-F D-E D-H F-E F-H E-H 14

?

?

1

1

?

1

1

1

1

—

However, E-H cannot be coded, and we still have a number of question marks indicating the unresolved issue of co-occurrence. The fact that TQCA cannot be applied to this data set or to other data sets with ties and missing events is regrettable because the basic ideas of QCA seem to be sound and applicable, in principle, to our data set. Therefore, it is useful to analyze in more depth why exactly QCA cannot handle ties and missing events.

QCA, SUFFICIENCY AND NECESSITY In QCA, the researcher seeks to identify the different configurations that are causally relevant for an outcome (Ragin & Strand, 2008, p. 431). However, this causal relevance can be of two very different types. It can mean (1) that the configuration generates the outcome (i.e., it is sufficient for the outcome) or (2) that the configuration allows the outcome to occur (i.e., it is necessary for the outcome). Establishing the one or the other type of causal relevance requires analytically distinct tasks (Ragin & Schneider, 2011). QCA aims at providing both of these two distinct types of analysis. In what respect is the analysis of sufficient configurations different from the analysis of necessary configurations? In their discussion of this difference, Ragin and Schneider (2011) discuss an example of a condition X1 which in a given data set occurs in cases that are successful as well as in cases that have failed. If it is the aim of the analysis to identify a configuration that is sufficient for success, then the researcher compares cases with and without the outcome [success] and tries to identify what was overlooked. The researcher concludes that X1 must be combined with X2 for the outcome to occur because the cases that combine these two conditions

The Analysis of Temporally Ordered Configurations

117

consistently exhibit the outcome, while X1 cases that lack X2 fail to exhibit the outcome. Thus, this foray into theory building results in a recipe for the outcome that is more elaborate and less inclusive than the initial recipe. [y] Observe that in this investigation, the objective is to establish that the causal condition or recipe is a subset of the outcome. [y] The resulting causal argument is made more restrictive [y] moving to a more combinatorial and nuanced conceptualization of causation. [y] Elaborating a causal argument in a combinatorial manner [means] that fewer instances of the outcome [success] are explained. [y] One of perhaps several recipes for the outcome has been clarified and refined. (Ragin & Schneider, 2011, pp. 159–160; emphasis added by us)

This is contrasted with the analytic strategy that must be applied if the aim of the analysis is to identify a configuration that is necessary for success. Here, the key task for the researcher is to see if there is some other condition that is causally equivalent to X1 which is found in the cases of the outcome [success] that lack X1. That is, is there a causal condition shared by [these] cases that is substitutable for X1 as a necessary condition? [y] Assume in this example that the researcher [y] concludes that X1 and X2 are causally equivalent as necessary conditions with respect to the outcome in question. [This] results in a recipe for the outcome that is more inclusive than the initial recipe because more cases display X1 or X2 than only X1. (Ragin & Schneider, 2011, p. 161, emphases in the original)

The essential difference between the two analytic strategies is that the search for sufficient configurations requires that conditions are added to the configuration (which makes the configuration more specific and implies that not all successful cases are included in the analysis) and that the search for necessary configurations requires that conditions are substituted (which makes the configuration more general and implies that all successful cases are included in the analysis). It is strange that Ragin and Schneider use the term ‘‘recipe’’ for both types of result. It makes more sense to use this term only for a sufficient configuration and to use an alternative term (e.g., ‘‘list of essential ingredients’’) for a necessary configuration. Drawing an analogy with cooking is illuminating. Take the example of baking grandma’s apple pie. The recipe for this pie is a list of ingredients and a set of instructions. If one wants to find out what is necessary for baking the pie successfully, one can improvise both with the ingredients and the instructions. Some pies will turn out good (like grandma’s own pies) and other pies will be considered failures. Necessary conditions are only those ingredients and actions that occur in all successful pies. If some of these contain no sugar, and other do not contain eggs, then the list of necessary ingredients can be shortened. The remaining list of items is inclusive because it contains the items that are shared by all successful pies.

118

TONY HAK ET AL.

Inevitably, however, some of the pies that contain all necessary ingredients and have been baked according to each of grandma’s instructions turn out to be failures, although grandma never failed to bake a delicious pie herself. In order to find out why grandma’s recipe (which contains, she says, everything that you need to bake a great pie) failed to produce a good pie in some cases we are forced to search for other conditions (actions, utensils, temporal orders) that are not yet specified in grandma’s instructions, but that need to be added to her recipe in order to guarantee success. In fact, as we know from experience, the list of such additional conditions is infinite (‘‘Dear, obviously you also need to make sure that y’’) and some of them are very difficult to specify (‘‘You must have a feeling for it’’). In order to be sufficient, a recipe must be infinitely more specified and, hence, becomes increasingly more exclusive. Our cooking example can be generalized. We know that it is always possible that a goal-oriented action (scoring a goal, passing an exam, winning a war) fails to achieve its goal due to an event or condition that could not have been foreseen and hence could not have been specified before. Every social process can be halted or misdirected at any point (deliberately or unintended) and hence success can never be guaranteed. Therefore, there is no limitation in principle to the number and the type of conditions that must be added to a ‘‘sufficient’’ configuration in order to actually achieve sufficiency. This fact does not fit well with the aim of QCA to generate the most parsimonious explanation that is possible. The more elaborate, exclusive, and ‘‘nuanced’’ the result, the less parsimonious it is. Sufficient configurations, as identified by QCA (or, for that matter, by any other method) must always be expanded with the phrase ‘‘and everything else that is relevant for the outcome.’’ Sufficient configurations are always underspecified in this sense. Let us now return to the question why TQCA has problems with ties and missing events. The key procedure in QCA is Boolean minimization, a procedure that reduces the configurations that generate the outcome to the shortest possible Boolean expression. This Boolean procedure requires an input of a complete set of binary codes (0, 1). It is difficult to meet this requirement in a data set with ties, missing events, and different temporal sequences of the same conditions (e.g., both B-D and D-B). Key to our way of avoiding the limitations of TQCA (discussed below) is that Boolean minimization is used in QCA only to provide for the shortest possible Boolean expression of different equifinal configurations that are sufficient for the outcome, that is, to reduce the inherent exclusivity of sufficient configurations. This minimization is not required for the analysis of

The Analysis of Temporally Ordered Configurations

119

necessary configurations. A necessary configuration always consists of conditions that are necessary themselves. It is a list of ‘‘ingredients,’’ each of which can be discovered separately. In other words, a necessary configuration is not the result of a procedure of elaboration which then requires a procedure of minimization, but is rather the result of combining (or adding) conditions that are already ‘‘minimal.’’ Because really sufficient configurations do not exist and, hence, recipes always need to be made complete by adding the phrase ‘‘and everything else that is relevant for the outcome,’’ it is also more realistic from a practical perspective to try to identify only necessary conditions and pathways, that is, conditions that must be present to allow a desired outcome to emerge. Knowledge about such necessary conditions is of practical value because it allows practitioners to develop policies that avoid a guaranteed failure. The concept of a necessary condition, and by implication of a necessary configuration, is undervalued in research because it is much more difficult to connect this concept (than the concept of a sufficient condition) to the dominant ‘‘variance’’ analytic procedures and their implicit concept of causality. In this common way of thinking a ‘‘cause’’ is seen as a thing, a mechanism or an event that (almost literally) produces the effect. The aim to identify configurations that ‘‘generate’’ an outcome is attractive because it suggests that research findings could be used for the formulation of a recipe for success (‘‘golden bullet’’).

NECESSARY CONDITION ANALYSIS (NCA AND TNCA) A necessary configuration (e.g., configuration ABC) consists of conditions (A, B, and C) that are themselves necessary. In other words, necessary conditions are cumulative. Hence, necessary configurations can be discovered (or ‘‘built’’) by first identifying its building blocks, the individual necessary conditions. Conditions that are necessary for an outcome can be found by identifying conditions that are shared by all cases with that outcome (see Dul et al., 2010, for a justification of this analytic strategy). ‘‘Building’’ a necessary configuration from a data set (only) requires that cases are compared for the occurrence (absence/presence) of single conditions. Such a comparison is rather simple and does not require any technical procedures, of a Boolean or other nature. In a data set as ours, ‘‘manual’’ analysis relying on visual inspection will do the job. When a

120

TONY HAK ET AL.

Table 3. Row

Truth Table for TNCA. Frequency

Outcome=1 (Start-Up)

1 2 3 4 5

6 7 8 9 10

B-D/E-F B/D/F-E D-B/F-E/H F-B/E/H-D B/D/F

1 1 1 1 1

Outcome=0 (Quit) B B-D B-F/E F/E-B D/F

2 1 1 1 1

(smaller or larger) number of necessary conditions have been identified in such a manual analysis, a necessary configuration can be built simply by (cumulatively) bringing the separate necessary conditions together in one configuration. We apply this manual approach to our data set. A necessary temporal sequence (e.g., configuration A-B-C) consists of individual sequences (A-B, and B-C) that are themselves necessary. Hence, in order to find a necessary temporal sequence, we must identify sequences that are shared between all successful cases. Similar to QCA, the first step is building a ‘‘truth table’’ (Ragin, 1987). Table 3 is the truth table derived from Table 2. The second step is identifying individual sequences and coding these in such a way that they can be compared between the (successful) cases. We take here rows 1 and 2 of Table 3 as an example of how this could be done: 1 2

B-D/E-F B/D/F-E

We need to identify a temporal order for six pairs of events (B–D, B–E, B–F, D–E, D–F, and E–F) and to present them in such a way that visual inspection is facilitated: 1 2

B-D B/D

B-E B-E

B-F B/F

D/E D-E

D-F D/F

E-F F-E

121

The Analysis of Temporally Ordered Configurations

As discussed above, co-occurrence of events in one month (as of B and D in row 2) can be consistent with a hypothesis that states that the one must precede the other (B-D) as well as with one that states that the reverse temporal order should occur (D-B). The configurations in rows 1 and 2 share with each other membership of the category ‘‘not in contradiction to B-D,’’ whereas they do not share membership of the category ‘‘not in contradiction to D-B.’’ The configuration in row 1 is not a member of the latter category, and the configuration in row 2 is a member of it. If we use the notation ‘‘B-D’’ not as indicating the actual occurrence of a temporal sequence in which B precedes D in a case but rather as indicating ‘‘not in contradiction to B preceding D,’’ we can recode the two rows as follows: 1

B-D

B-E

B-F

2

B-D D-B

B-E

B-F F-B

D-E E-D D-E

D-F

E-F

D-F F-D

F-E

Shared between these two rows are the following sequences: B-D, B-E, B-F, D-E, and D-F. If this would be the result of the analysis of all five successful cases (which it obviously is not), then this result would indicate that it is necessary for a start-up that the temporal order of the gestation activities does not contradict that B precedes D and does not contradict that D precedes both E and F. This formulation (‘‘does not contradict’’) allows treating the co-occurrence of events in the same month as consistent with the necessary sequence. Another, more convenient way of stating the same result is that it is necessary for a start-up that E and F do not precede D and that D does not precede B. Before this analysis can be conducted in the complete set of five configurations in Table 3 we must also find a solution for the coding of sequences between pairs of conditions of which one is absent in the data set, as is the case with H in rows 1 and 2. In our discussion of TQCA we have shown that in our data set this problem can partly be solved if we assume that events are not missing because of measurement error – they have occurred but have not been recorded – but only because they have not happened yet. Under this assumption, in rows 1 and 2, we can be certain that H will always be preceded by the other four conditions. However, we do not have such a solution for the sequence of E and H in row 5. We need a general solution for missing events, independent of this specific data set. The solution that we propose is to treat missing events in

122

TONY HAK ET AL.

the same way as ties, that is, by coding them as ‘‘not in contradiction to’’ the temporal sequences that could have occurred if the events had happened. This solution implies that we would assign both code E-H and H-E to row 5. We can now code all five configurations that result in a start-up (rows 1–5 in Table 3). We only look at rows 1–5 because we are interested in identifying those temporal configurations that are necessary for success, that is, those factors of which the absence guarantees failure. The result is presented in Table 4. Four sequences are shared between these five configurations: B-E, B-H, F-H, and E-H. Three of these sequences can be combined into a chain: B-E-H. Note that the arrows do not only indicate a temporal order (i.e., a difference in time of occurrence of at least a month) but also includes simultaneousness (as in row 4 of Table 3). Note also that this result does not imply that the sequence must always be present, but only that if two events (e.g., E and H) are present in a case, their temporal sequence must not violate the sequence in the result (e.g., H-E should not occur). Hence, these findings could best be expressed as statements about what should not occur in order to allow a desired outcome: H (if present) never before E or F, and E (if present) never before B. If these statements really express temporal conditions that are necessary for the desired outcome, then a violation of any of them should only occur in cases that have failed to achieve the desired outcome. Moreover, a necessary condition statement is only interesting in theoretical and practical terms (i.e., it is not trivial) if violations actually occur. For instance, if every nascent entrepreneur (i.e., entrepreneurs that eventually succeed as well as those that eventually fail) would start thinking about the hiring of employees Table 4.

Necessary Condition Analysis.

1

B-D

B-F

B-E

B-H

D-F

2

B-H

B-E

B-H

D-F F-D D-F

4

B-D

B-F F-B B-F F-B F-B

B-E

3

B-D D-B D-B

5

B-D D-B —

B-F F-B —

B-E E-B B-E

B-H H-B B-H

B-E

B-H

Shared

D-E E-D D-E

D-H

E-F

F-H

E-H

D-H

F-E

F-H

E-H

D-E

D-H

F-E

F-D

E-D

H-D

F-E

F-H H-F F-H

D-F F-D —

D-E

D-H

F-E

F-H

—

—

—

F-H

E-H H-E E-H H-E E-H H-E E-H

The Analysis of Temporally Ordered Configurations

123

only after (or at the same time as) equipment has been installed, this would be an interesting finding about gestation as a process (irrespective of its outcome), but it would not be informative about how to proceed in order to be successful. Only if violations of the identified necessary sequences occur in practice (and if they are indeed always associated with failure), then it is practically relevant to formulate an advice to avoid them. Hence, it is useful to assess whether such violations occur in the current data set. Inspection of rows 6–10 of Table 3 (‘‘failures’’) shows that H does not occur in cases of quitting without having started a firm. This suggests that H, if occurring at all, is strongly linked to a successful completion of the gestation process and hence will always occur late in the gestation process. Quitting, thus, seems to have the logical implication that the stage of hiring will not be reached. It might still be the case that early hiring (i.e., behaving in contradiction to the sequences that we have found) actually is a guarantee for failure, but there is not a case of early hiring in our data set and, hence, we do not know whether late hiring is a characteristic of all gestation processes (including those that result in success) or is really necessary for success. We could test the hypothesis ‘‘Early hiring guarantees failure’’ in another data set to sort this out. Regarding the remaining result (B-E), there is one case, which is represented by row 9 in Table 3, that is, in a configuration that is associated with quitting the gestation without starting a firm, in which this condition is violated. The fact that a violation of temporal sequence B-E exists in our data set, and that this violation is associated with failure is consistent with the hypothesis that ‘‘equipment (if present) never before business research’’ is a non-trivial necessary condition for a successful outcome of gestation. Obviously, this finding should be formulated as a hypothesis to be tested in other data sets. TNCA, thus, has discovered at least one non-trivial necessary sequence for successful gestation, though negatively formulated: equipment never before business research. Arguably this is a relevant finding, both in theoretical and practical terms, because it suggests that, for a successful gestation of ambitious high-tech start-ups, business research cannot be delayed until after equipment or, in other words, that it is necessary not to install any equipment before business research has begun (although these activities might be started in the same month). This is consistent with the intuitively plausible idea that business research needs to precede decisions about what equipment is needed. Are there also temporal sequences that are ‘‘necessary’’ for quitting the gestation process without having started a firm? Because rows 6–10 of the

124

TONY HAK ET AL.

truth table contain a much smaller number of events, there is not much sequential information available in these rows. Only one necessary sequence for quitting can be identified: ‘‘development (if present) never before business research.’’ If this really is a necessary condition for quitting, then ‘‘development before business research’’ should be a sufficient condition for successfully completing the gestation process with a start-up. The sequence ‘‘development before business research’’ (D-B) is present in row 3 in the truth table (Table 3), which indeed is associated with a successful start-up. It is, however, not immediately clear how this result must be interpreted. It might refer to cases of ambitious high-tech start-ups in which there is, from the outset, so much confidence in the profitability of the intended product or service that this is developed first. If this is a correct interpretation of this result, then it is not the temporal order D-B that is sufficient for the successful outcome but rather the initial confidence of the entrepreneur that induced him/her to develop the product or service before conducting any business research.

CONCLUSION This chapter discussed methods that could identify temporal sequences of events that generate or allow a successful outcome of a process, while an occurrence of the same events in another sequence will not generate or allow that outcome. Various approaches were evaluated using a small set of data (with time stamp) on the gestation activities of ambitious nascent entrepreneurs in the high-tech sector. The aim of the analysis was to identify temporal sequences of gestation activities (e.g., C-A-B) that generate or allow a successful outcome of the gestation process, while an occurrence of the same activities in another temporal order (e.g., A-B-C or B-A-C) will not generate or allow that outcome. The various analytic approaches were evaluated in terms of their ability to achieve this aim. A distinction can be made between different types of temporal sequence (Abbott, 1995). We discuss the analysis of only one such type, the nonrecurrent sequence of events, that is, a temporal sequence of which the (analytical) length cannot be longer than the total number of observed events and in which these events can occur only once. This type of temporal sequence is a configuration if defined as ‘‘multidimensional constellations of conceptually distinct characteristics that occur together’’ (Meyer et al., 1993, p. 1175), provided that ‘‘occur together’’ is interpreted as meaning ‘‘occur in the same case.’’

The Analysis of Temporally Ordered Configurations

125

We have identified three approaches that have been proposed as a tool for the identification of temporal sequences of events that generate or allow a successful outcome of a process, (1) Event Structure Analysis (ESA; Heise, 1989), (2) Optimal Matching (OM; Abbott, 1990, 1995), and (3) Temporal Qualitative Comparative Analysis (TQCA). Event Structure Analysis generates a picture (or ‘‘model’’) of temporal pathways based on a (chronological) narrative. This is very useful as a tool for getting a comprehensive and consolidated overview of all actually occurring pathways in a data set, in particular if events can reoccur. An ESA model is a relatively much less useful tool for representing the set of actually occurring sequences of non-recurrent events. A truth table such as is generated in QCA is at least equally informative. But the main reason why ESA is not the method we are looking for is that its output (the ESA model) still requires the analysis that we are aiming at (as, for that matter, the truth table in QCA, which is the input for the analysis rather than the output). Both the ESA model and the QCA truth table are a summary of the results of a ‘‘within-case’’ analysis, whereas our aim only can be achieved by means of some form of ‘‘cross-case’’ (or ‘‘comparative’’) analysis. Optimal Matching is an approach to comparative analysis. It produces clusters of sequences that differ in the likelihood by which they generate an outcome. This is a ‘‘variance-based’’ approach for which we want to find an alternative in this chapter. Moreover, an important limitation of this approach is that the applied permutation statistics have problems with ties as well as non-occurring events (Abbott, 1990, p. 383). Qualitative Comparative Analysis (QCA), the currently common method used in the analysis of configurations, is able to analyze the causal relevance of the absence or presence of an event (condition), but cannot take the temporal sequence of these events into account. TQCA (Caren & Panofsky, 2005; Ragin & Strand, 2008) provides for a solution of this problem, but only if all events are present in all cases and only if events never tie. The core of our paper is a solution for this ‘‘technical’’ problem. Our solution, however, is not technical but more fundamental. We argue that the technical limitations of TQCA only occur in the process of identifying sufficient configurations, which are discovered by adding ever more specifications to an initial configuration (Ragin & Schneider, 2011). By discussing everyday processes such as baking an apple pie, we demonstrated that this process of adding relevant specifications is infinite in principle and, hence, logically untenable and virtually useless in practice. Having concluded that analysis should focus on searching for necessary configurations instead, we demonstrate that such a necessary condition analysis can

126

TONY HAK ET AL.

easily be performed ‘‘manually,’’3 that is, without making use of QCA software (which, as we argue, serves the analysis of sufficient configurations rather than of necessary configurations).

NOTES 1. Note that the gestational activities are presented in Table 1 in a way (B–D–F– E–H) that already reflects an implicit process theory in which research and development are assumed to take place before finance can be attracted; that seeking finance will precede the purchase of equipment and that employees will be hired for operating the equipment after equipment is purchased; and that then finally the firm can take off, that is generate sales and become profitable. The aim of the analysis is to find out to what extent support for this implicit theory can be found in the data. 2. We will use this language of ‘‘success’’ and ‘‘failure’’ throughout this chapter, although it might be argued that quitting the gestation process without starting a firm can be a successful outcome as well, in particular if gestation is seen as a process of finding out whether a firm can be successful (if the firm is started). Quitting for good reasons can hence be seen as a very welcome outcome of gestation. 3. With more events and more cases, this ‘‘manual’’ procedure can be automated easily with, for example, a macro in a spreadsheet software program.

REFERENCES Abbott, A. (1990). A primer on sequence methods. Organization Science, 1(4), 375–392. Abbott, A. (1995). Sequence analysis. Annual Review of Sociology, 21, 93–113. Caren, N., & Panofsky, A. (2005). TQCA: A technique for adding temporality to qualitative comparative analysis. Sociological Methods & Research, 34, 147–171. Dul, J., Hak, T., Goertz, G., & Voss, C. (2010). Necessary condition hypotheses in operations management. International Journal of Operations and Production Management, 30(11), 1170–1190. Gordon, S. R. (2011). Entrepreneurial discovery and exploitation processes: Sequence or symbiosis? Paper presented at Babson College Entrepreneurial Research Conference, June 9–11 2011, Syracuse, NY. Heise, D. R. (1989). Modeling event structures. Journal of Mathematical Sociology, 14, 139–169. Krook, M. L. (2006). Temporality and causal configurations: combining sequence analysis and fuzzy set/qualitative comparative analysis. Paper presented at the Annual Meeting of the American Political Science Association, Philadelphia, PA, August 31–September 3, 2006. Langley, A. (1999). Strategies for theorizing from process data. The Academy of Management Review, 24(4), 691–710. Markus, M. L., & Robey, D. (1988). Information technology and organizational change: Causal structure in theory and research. Management Science, 43(5), 583–598.

The Analysis of Temporally Ordered Configurations

127

De Meur, G., Rihoux, B., & Yamasaki, S. (2009). Addressing the critiques of QCA. In B. Rihoux & C. R. Ragin (Eds.), Configurational Comparative Methods (pp. 147–165). Los Angeles, CA: SAGE. Meyer, A. D., Tsui, A. S., & Hinings, C. R. (1993). Configurational approaches to organizational analysis. The Academy of Management Journal, 36(6), 1175–1195. Mohr, L. B. (1982). Explaining organizational behavior: The limits and possibilities of theory and research. San Francisco, CA: Jossey-Bass. Morse, E. A. (1998). The temporal dynamics of entrepreneurial growth: An event structure analysis in an entrepreneurial firm. Unpublished doctoral dissertation. Texas Tech University, Lubbock, TX. Pettigrew, A. M. (1999). What is processual analysis? Scandinavian Journal of Management, 13(4), 337–348. Ragin, C. R. (1987). The comparative method. Berkeley, CA: University of California Press. Ragin, C. R., & Schneider, G. A. (2011). Case-oriented theory building and theory testing. In M. Williams & W. P. Vogt (Eds.), The Sage handbook of innovation in social research methods (pp. 150–166). Los Angeles, CA: SAGE. Ragin, C. R., & Strand, S. I. (2008). Using qualitative comparative analysis to study causal order. Sociological Methods & Research, 36(4), 431–441. Reynolds, P. D., & Curtin, R. T. (2008). Business creation in the United States: Panel study of entrepreneurial dynamics II initial assessment. Foundations and Trends in Entrepreneurship, 4(3), 155–307.

CHAPTER 6 UNDERSTANDING COMPLEMENTARITIES AS ORGANIZATIONAL CONFIGURATIONS: USING SET THEORETICAL METHODS Gregory Jackson and Na Ni ABSTRACT The growing literature on complementarities has drawn attention to how the effects of different organizational structures, practices, and institutions are interdependent. Rather than one best way of organizing, complementarities suggest that the effectiveness of one organizational element may be dependent on the presence or absence of another particular element. Consequently, organizational arrangements often display ‘‘multiple equilibria’’ or what is known as equifinality, whereby multiple pathways may lead to the same or similar outcomes. While being a source of theoretical innovation, the configurational nature of complementarities has posed a number of challenges. This chapter reviews the emerging literature on complementarities to identify a series of conceptual challenges related to understanding complementarities as organizational

Configurational Theory and Methods in Organizational Research Research in the Sociology of Organizations, Volume 38, 129–158 Copyright r 2013 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0733-558X/doi:10.1108/S0733-558X(2013)0000038010

129

130

GREGORY JACKSON AND NA NI

configurations, and examines the methodological challenges in studying how such elements combine to produce joint effects on performance. The chapter argues that new set-theoretic methods using Qualitative Comparative Analysis (QCA) may present a very useful methodological alternative to studying complementarities. The chapter illustrates this potential by re-analyzing past work by Aoki, Jackson, and Miyajima (2007) on relationships between ownership structure, board structure, and employment practices of listed firms in Japan to show evidence of complementarities associated with hybrid configurations that combine market and relational forms of organization. Keywords: Corporate governance; Japan; complementarities; Qualitative Comparative Analysis

INTRODUCTION: THE CONCEPT OF COMPLEMENTARITIES The concept of complementarity has been widely applied in the study of organizations. Milgrom and Roberts (1990, 1995) used the concept to investigate the shift from mass production to ‘‘modern manufacturing.’’ Their core proposition was based on Edgeworth’s (1881) notion of complementary goods, whereby doing more of one thing increases the returns to doing more of another. The theoretical model specified complementarities between an organization’s strategy, structure, and managerial process. For example, ownership structure, flexible work rules, and piece rates form a complementary and interdependent whole as ‘‘a system of mutually enhancing elements’’ (Milgrom & Roberts, 1995, p. 204), which may then be difficult for competitors to emulate. The core insight is that certain configurations of organizational structures and practices are associated with a firm’s competitive advantages. Literature on complementarities has renewed attention to how organizational structures, practices, and institutions have interdependent effects. Rather than one best way of organizing, complementarities suggest that the effectiveness of one element may be dependent on the presence or absence of another particular element. Consequently, organizational forms or institutional arrangements often display ‘‘multiple equilibria’’ (Aoki, 2001). One implication is that an organization may become locked-in to certain local solutions, even where these are globally suboptimal, because marginal

Understanding Complementarities as Organizational Configurations

131

changes in one organizational element will lead to worse results without simultaneous changes in other elements. More generically, we can say that the effects of organizational structures and practices should be studied in the context of their overall configuration and implied interdependencies (Aguilera, Filatotchev, Gospel, & Jackson, 2008; Grandori & Furnari, 2008; Ketchen, Thomas, & Snow, 1993). Looking at isolated attributes of organizations as single variables is likely to be misleading, and misses how these attributes form complex types often found in organizational theory. The concept of complementarities has inspired renewed study of the diversity of organizations and gives these studies stronger theoretical grounding. Complementarities require scholars to look beyond the traditional bivariate relationships associated with different organizational elements (Milgrom & Roberts, 1990; Miller & Friesen, 1986). To the extent that complementarities lead to multiple equilibria, organizations are likely to be associated with equifinality (Doty, Glick, & Huber, 1993; Fiss, 2007), whereby multiple pathways may lead to the same or similar outcomes. Complementarities have made contributions to many subfields of organizational theory, including the literatures on organizational resources, strategy, corporate governance, and comparative management (for an excellent overview, see Ennen & Richter, 2010). Nonetheless, the meaning and implications of complementarities remain highly contested. First, the theoretical concept has been used in a variety of different ways (see Crouch, 2005). Better middle-range theories are needed to understand both what and how organizational elements complement one another. Second, the configurational nature of complementarities has also posed a number of methodological challenges. Many studies have tried to capture complementarities by using interaction effects within traditional statistical models such as regression, whereas other studies have used more systemic approaches to descriptively cluster empirical data (Ennen & Richter, 2010). These approaches have not only produced different results, but are not fully satisfactory in linking theory and method (see Fiss, 2007). This chapter analyses these challenges to studying complementarities and argues that new set-theoretic methods, such as fuzzy set Qualitative Comparative Analysis (fs/QCA), may present very useful tools to better link theory and method. The second section will briefly review applications of complementarities in organizational research. The third section analyses the theoretical ambiguities and methodological challenges in studying complementarities, as well as developing the case for applying fs/QCA to study complementarities as configurations. The fourth section develops a brief empirical illustration of using fs/QCA methods to study complementarities

132

GREGORY JACKSON AND NA NI

between corporate governance and human resource management in Japanese firms. The strengths and limitations of this approach will be illustrated by a reanalysis of past work by Aoki, Jackson, and Miyajima (2007) that examined complementarities between ownership structure, board structure, and employment practices. The fifth section discusses the implications of our results for the conceptual and methodological approaches to complementarities, and the sixth section concludes.

COMPLEMENTARITIES IN ORGANIZATIONAL RESEARCH Complementarity refers to situations whereby two activities reinforce each other, such that doing more of one thing increases the value of doing more of the other (Matsuyama, 1995). The concept has been widely applied to study organizational resources, strategies, and various structures or practices – as well as comparing these across institutional contexts. The concept highlights the importance of configurations for understanding organizations. Organizational scholars have long examined ideas related to ‘‘fit,’’ ‘‘congruence,’’ or ‘‘synergies’’ (Nadler & Tushman, 1980). Several studies pioneered notions related to complementarities that were based on configurations and equifinality (Doty et al., 1993; Ketchen et al., 1993). The concept of complementarities formalizes and gives specific theoretical grounding to the idea that different bundles of organizational characteristics may cohere into qualitatively distinct ‘‘types.’’ At the same time, the complementarity falls short of being a fully specified theory. The concept alone does not delineate what features of organizations are likely to complement what else. In their excellent meta-analysis, Ennen and Richter (2010) found that complementarities exist not only among specific practices within single functional areas of organizations, but are also most often evident between different functional areas of the organization. Next we briefly turn to examples from specific literatures that discuss middle-range organizational phenomenon. Many studies have examined complementarities among organizational resources – including knowledge, capabilities, and technology. Drawing on the resource based view (RBV), scholars suggest the heterogeneity and imperfect mobility of resources between firms is related to the particular a configuration of resource factors (Black & Boal, 1994) or resource bundles (Teece, 1986). Rhyne and Teagarden (1997), for example, found that intangible resources, such as human assets, provide competitive advantages

Understanding Complementarities as Organizational Configurations

133

when combined with certain tangible resources (e.g., financial resources). Likewise, other studies found a complementary relationship between information technology and human resources (e.g., Powell & Dent-Micallef, 1997). Complementarities have also been studied in relation to organizational strategies and structures. For example, scholars have suggested potential positive results from hybrid or combination strategies (Hill, 1988; Karnani, 1984), whereby low cost and differentiation strategies are complementary rather than mutually exclusive (e.g., Miller & Friesen, 1986; Wright, Kroll, Tu, & Helms, 1991). Turning to organizational structure, Chandler (1962) found a complementary relationship between structure and corporate strategies. Based on the history of the evolution of four large firms (GM, Du Pont, Standard Oil, and Sears) from 1850 to 1950, Chandler concluded that strategy of diversification works best in divisional structure. Rumelt (1974) later found that firms with related diversification strategy and productdivision structure performed best. Similarly, Williamson (1975) found that firms with M-forms outperformed others with increased product differentiation. Complementarities between strategy and structure also extend to business-level strategies. For example, focus strategy always go together with simple structure, machine bureaucracy is favorable with cost leadership strategy, and organic structures work best for differentiation strategy (Miller, 1986, 1988). Based on related ideas, a large literature makes connections between organizational factors (e.g., culture, organizational design, or process) and factors related to strategy or resources by arguing that complementarities among them may result in different archetypes, gestalts, or configurations (Miller & Friesen, 1986; Miles & Snow, 1978; Siggelkow & Rivkin, 2005). A typical example of complementarities found in organizational contexts concerns strategic human resource management (HRM). Well-designed HRM practices are argued to be positively associated with firm performance (e.g., Bae & Lawler, 2000; Delery & Doty, 1996; Valle, Martin, Romero, & Dolan, 2000). But HRM systems are composed of various subsystems, and these often function together in complex patterns of interdependence (Delery & Doty, 1996; Kang, Morris, & Snell, 2007; Sheppeck & Militello, 2000). For example, Lepak and Snell (2002) developed the four HRM configurations that are commitment-based, productivity-based, compliancebased, and collaborative. They also found that these configurations display fit or complementarities with specific types of work organization (i.e., knowledge work, job-based employment, contract work, and alliance/ partnerships). Thus, the contribution of HRM practices and systems toward

134

GREGORY JACKSON AND NA NI

firm performance may depend on certain boundary conditions such as the wider aspects of corporate and business strategies (see Delery & Doty, 1996; Miles & Snow, 1978; Valle et al., 2000). Likewise, Hitt, Bierman, Shimizu, and Kochhar (2001) found that the positive relationship between human resource system and firm performance depend on the types of firms’ product diversification strategies. Later we will examine how employment practices are linked with corporate governance. Complementarities have been also widely applied in the field of comparative management to understand the comparative institutional advantages of different organizational patterns across countries. This literature often focuses on macro-level bundles of institutions, such as national business systems (Whitley, 2007) or different ‘‘varieties of capitalism’’ (Hall & Soskice, 2001). Here institutions coordinate economic exchange in complementary ways across domains, such as finance, labor, supplier relations, and so forth. These different institutional-level bundles or complementarities have been linked to various macroeconomic outcomes such as growth or innovation (Akkermans, Castaldi, & Los, 2009; Akkermans et al., 2009; Amable, 2003; Boyer, 2004; Hall & Gingerich, 2009; Herrmann, 2005; Schneider & Paunescu, 2012; Schneider, Schulze-Bentrop, & Paunescu, 2009; Taylor, 2004). Complementarities are also seen as a source of path dependence and thus explain why organizational forms fail to converge to a single form, despite pressures from globalized markets. Nonetheless, Ennen and Richter (2010) document that relatively few studies have empiricallystudied complementarities between organizational-level factors and wider features of the institutional environment.

CONCEPTUAL AND METHODOLOGICAL CHALLENGES IN STUDYING COMPLEMENTARITIES The previous section showed the pervasive use complementarities within organizational and institutional research. Many studies use the concept in a theoretically loose or even implicit way. Since complementarities often extend across distinct functional or institutional domains, they are often missed in traditional sub-fields of organizational theory. More elaborate middle-range theories are needed to better explain the mechanisms of interdependence between specific configurational elements and how these as configurations shape certain outcomes in complementary ways. Moreover, researchershave looked at complementarities largely in terms of correlations

Understanding Complementarities as Organizational Configurations

135

or empirical clustering of characteristics, rather than the conjunctural effects of combinations. This section will analyze the very significant conceptual and methodological challenges in studying complementarities. For illustrative purposes, it is helpful to consider the basic functional form of complementarities. We denote two ‘‘types’’ of organization in each domain as X1 or X2 and Y1 and Y2, respectively. Complementarities may be as cases where the two configurations of X1+Y1 or X2+Y2 outperform the mixed combinations of X1+Y2 or X2+Y1. Starting with this basic model, a number of complications can arise. A first set of issues concerns how to identify and model complementary configurations of factors. In identifying the bundles that constitute configurations, one problem is that complementarities may often extend across more than two factors and imply causally complex configurations. A certain complementary pair may only be effective in the presence of additional, contextual factors. The effectiveness of a combination X2+Y2 may depend on other unobserved or contextual conditions Z2. For example, two human resource management practices may only be effective in conjunction with a particular corporate strategy, industry environment, or institutional conditions. In modeling the effects of configurations, we must also recognize that complementarities imply more than simple correlation. The co-occurrence or clustering among two factors does not make them complements. Rather, complementarities imply that the interaction of two or more conditions is related to an outcome defined in terms of performance or effectiveness. Complementarities can be understood here statistically as an interaction effect, but one can go further theoretically to distinguish whether combinations constitute necessary or sufficient conditions for an outcome.1 For example, X1 may be necessary for Y1 to be effective, but not a sufficient condition. Or Y1 may be sufficient but not necessary for the effectiveness of X1. Unlike correlational models, complementarities may also imply asymmetric relationships. For example, X1 may be necessary for the effectiveness of a complement Y1, but not vice versa. Or a configuration may be sufficient for high performance, but the opposite configuration does not necessarily leading to low performance. A second set of conceptual issues concerns the conceptual distinction between coherence and complementarities (see Crouch, 2005). Coherence can be described as forms of organization based on common principles or ‘‘logics,’’ such as strong market-orientation or democratic decision making. While many theoretical applications of complementarities have assumed that similar forms of organizations act as complements, complementarities may exist based on combinations of opposing forms of organization that act

136

GREGORY JACKSON AND NA NI

to counter-balance one another. For example, in a political system, the logic of majority rule is balanced by the rule of law overseen by the judiciary branch that acts independently of the majority. Similarly, organizations may need to combine the logics of the market with other logics based on professionalism or long-term relationships in order to solve particular dilemmas. These distinct logics of complementarity have been document in some fields of management research, such as corporate social responsibility (Kang & Moon, 2012). One cannot merely assume that similar or opposite forms of organization will consistently lead to complementarity – these must be systematically conceptualized and empirical tested. A third conceptual issue concerns the effectiveness or performance outcomes implied when using the concept of complementarities (see Jackson, 2005a). Any middle-range theory of complementarities must define the outcome or dimension of effectiveness. Two elements can only be complements in relation to some outcome. Firm performance is commonly used as the key outcome in business literature, whereas literatures on institutional complementarities have used various macroeconomic indicators such as GDP growth. The larger point is that performance always has multiple dimensions. For example, certain combinations of HRM practices may work together as complements in relation to promoting employee loyalty, but prove to be negative in terms of costs or financial performance. Or two factors that are complementary for GDP growth may also be associated with higher income inequality. Performance metrics often imply give rise to trade-offs or conflicts with other outcomes. It may be very difficult or impossible to specify all the relevant outcomes into a single theoretical model, or develop a meta-notion of complementary that aggregates such trade-offs, including distributional or power struggles. For this reason, it may be highly misleading to assume that complementary elements are necessary stable over time or more broadly that the presence or absence of a factor can be ‘‘explained’’ by complementarities. In sum, complementarities must be understood locally in relation to a particular model or outcome of interest, whereby it is difficult to extrapolate a global view of whether two elements are in a positive equilibrium. These theoretical challenges have posed a number of methodological challenges in understanding complementarities as specific types of configurations associated with positive effects. Both traditional linear statistical models and qualitative case studies have proven unsuitable to fully capture the interdependent nature of these effects or test for these effects across cases. First, complementary relationships are theorized as typologies or complex configurations of multiple elements. These theoretical constructs

Understanding Complementarities as Organizational Configurations

137

are often measured by highly aggregated indicators, such as indices, factor scores, or cluster analyses. In different ways, these methods all impose certain assumptions about how variables combine or treat sub-groups within a certain category as ‘‘intermediate’’ types. Second, standard statistical approaches are based on the net effects of single variables, while holding other factors constant. This type of model is poorly suited to understanding complementarities based on particular combinations of factors. Of course, this problem can be partially overcome by the use of interaction terms within statistical models or using descriptive techniques such as cluster analysis. However, these approaches remain an imperfect substitute for explicitly testing complex combinations of variables as configurations (see Fiss, 2007). Third, as discussed above, the symmetric nature of correlational methods may fail to capture the asymmetric nature of interactions between variables or their effects on performance outcomes. In facing these challenges, we argue for the potential of using set-theoretic approaches based on fs/QCA techniques (Ragin, 2000, 2008) to investigate complementarities as organizational configurations. fs/QCA has been increasingly applied to study configurations in organizations (Crilly, 2011; Fiss, 2007, 2011; Greckhamer, Misangyi, Elms, & Lacey, 2008; Kogut & Ragin, 2006) and macroinstitutional arrangements (Jackson, 2005b). This approach represents a synthesis of qualitative and quantitative methods designed to study complex combinations of factors. In essence, fs/QCA examines whether causal conditions or combinations thereof form consistent necessary or sufficient conditions for a particular outcome. Data regarding each case in the analysis is first calibrated to establish the degree of set membership of that case in the outcome (e.g., cases of high performance) and relevant causal conditions (e.g., organizational characteristics). Calibration is done by establishing qualitative anchors for evaluating the membership of case in the set – whereas crisp sets define membership as being either present or absent, fuzzy set analysis can allow different degrees of membership. Next, fs/QCA tests the set theoretical consistency of relationships, and evaluates to what extent those consistent solutions provide coverage of the outcome.2 A causal condition is ‘‘sufficient’’ if the instances of the causal condition consistently form a subset of the instances of the outcome condition, whereas a condition is ‘‘necessary’’ if the instances of the outcome are a subset of the causal condition (see Ragin, 2008). Consistency of sufficient conditions is calculated as the sum of the membership in the predictor set that are less than or equal to membership in the outcome set (e.g., causal condition or causal combination) divided by the sum of all the membership scores in the predictor set. Consistent set relationships can

138

GREGORY JACKSON AND NA NI

also be evaluated in terms of their coverage of the solution, calculated as the proportion of membership in the outcome set that is also covered by membership in the causal set. Consistency and coverage thus provide the two key numerical benchmarks in evaluating set theoretical relationships. Finally, fs/QCA also allows us to take account of limited diversity, since not all logically possible causal configurations are, in fact, empirically observable. By incorporating counterfactuals, fs/QCA can obtain both a parsimonious and complex result. Counterfactual cases may be treated as being false or inconsistent with the outcome in order to arrive at a conservative or ‘‘complex’’ result. Alternatively, the analysis can also treat specified counterfactuals as being true or consistent with the outcome. This thereof leads to a more parsimonious result, but one that introduces a simplifying assumption, namely that the existence of a nonobserved causal combination would not lead to a contradictory result (Ragin & Sonnett, 2005). Taken together, set theoretical methods distinguish themselves from variable-based quantitative methods in several ways, which are highly relevant for studying complementarities. First, set-theoretic approaches emphasize the notion of causal complexity (Ragin, 2000), where multiple causal factors combine in explaining specific cases rather than cases being disaggregated into average net effects of single variables. Cases can be interpreted as having a certain degree of membership within a specific causal configuration, which can be defined through the set theoretic conjunction of multiple factors. The causal impact of one condition is thus treated in conjunction with a wider configuration of other conditions. This approach can help develop and test theories of complementarity by specifically addressing configurations involving multiple organizational dimensions, and comparing combinations thereof in a systematic fashion. Second, fs/ QCA acknowledges the notion of equifinality (Gresov & Drazin, 1997; Ragin, 2000), which indicates alternative ways to achieve the same final state. Here, different sets of causal factors may lead to the same level of firm performance. Third, the set theoretical nature of fs/QCA helps identify nonlinear and asymmetric relationships by examining sub-set relationships across cases. For example, the conditions leading to low performance are not necessarily the opposite of those for high performance. In particular, fs/QCA is relevant to understand the issue of potential ‘‘hybrid’’ configurations or cases which mix elements with different underlying logics. These types of intermediate cases are not simply theoretical mongrels, but may possess qualitatively distinct characteristics that need to be understood in their own terms (see Aoki, Jackson, & Miyajima, 2007).

Understanding Complementarities as Organizational Configurations

139

COMPLEMENTARITIES AS ORGANIZATIONAL CONFIGURATIONS: THE CASE OF THE JAPANESE FIRM In order to examine the potential of fs/QCA to study complementarities, this section develops a brief empirical application based on complementarities in Japanese firms. We first present some hypothesized complementarities within the so-called J-firm model of the firm, and empirical results of a previous study (Aoki et al., 2007). Next, we seek to extend this analysis be re-analyzing the data used in this study through a set theoretical analysis using fs/QCA.

Complementarities with the J-Firm Model The success of the postwar Japanese firm (J-firm) has often been explained through the complementarities between a number of interrelated elements (Aoki, 1994). In Japan, corporate ownership, board structures, and employment were all organized through long-term relationships. Long-term ownership through cross-shareholdings among corporate groups (keiretsu) and main banks complemented boards dominated by insiders and the use of long-term firm internal labor markets. By contrast, US corporate governance uses market-orientation structures along these three dimensions. Ownership is dispersed among small shareholders, resulting separation of ownership and control. Information disclosure, independent outside directors and equity-based managerial incentives are widely seen as effective remedies to the resulting agency problems, since management will be effectively linked to the stock market. In a somewhat simplified form, we can summarize the intuition behind this model along three theoretical dimensions:  Ownership and finance characteristics: market-oriented (e.g., bond finance and institutional investors) or relational (e.g., bank finance and cross-shareholding).  Board structure characteristics: outsider-oriented (e.g., independent boards and high disclosure) or insider-oriented (e.g., insider boards and private information).  Employment and incentive characteristics: market-oriented (e.g., no lifetime employment, merit pay and use of stock options) or relational (e.g., lifetime employment, seniority pay, and no stock options).

140

GREGORY JACKSON AND NA NI

Looking at ownership and board characteristics, certain complementarities can be hypothesized. As typifies the US mode, market-oriented ownership and outsider-oriented boards complement one another, since dispersed shareholders enjoy the advantages of risk diversification but these advantages will only hold to the extent that complementary structures, including an independent board, exist to limit the ensuing agency costs. By contrast, in Japan, relational owners hold shares to gain strategic or so-called ‘‘private benefits.’’ However, these benefits require complementary board structures that assure relational owners private information and longterm influence over insider-oriented management teams. Taken together, we can stylize this hypothesis in Table 1. While corporate governance is often seen in terms of ownership and management characteristics, this model can be extended in relation to employees (Aguilera & Jackson, 2003; Aoki, 2001; Ho¨pner, 2005). Stakeholder theories suggest complementarities between these corporate governance characteristics of ownership and boards, on one hand, and employment patterns, on the other (see Table 2). For example, commitment by relational investors supports stable long-term employment, investment in worker training, and cooperative industrial relations (Hall & Soskice, 2001). Such firms may achieve dynamic (X-) efficiency in higher-quality product markets which require high skills (Streeck, 1992). However, if employees face potential ‘‘breaches of trust’’ based on the efforts of managers to raise stock market valuations and ward off hostile takeovers, these actions are likely to undermine long-term commitments to employees.

Table 1.

Hypothesized Complementarities among Ownership and Board Structures. Outsider Board

Relational ownership Market ownership

Table 2.

Insider Board J-firm corporate governance

US-firm corporate overnance

Hypothesized Complementarities among Ownership, Boards, and Employment. Market Employment

Relational ownership and insider board Market ownership and outsider board

Relational employment J-firm model

US-firm model

Understanding Complementarities as Organizational Configurations

141

Conversely, market-oriented shareholders move capital more quickly into new areas and are quicker to withdraw capital from declining industries. This market-driven creative destruction of organizations, however, can only likely to be effective in the presence of an active external labor market. Thus, a market-driven ‘‘hire and fire’’ employment system will be more beneficial to the extent that the capital market supports the creation of risky new start-up ventures. Similarly, market-oriented corporate governance may be more effective if managerial and employee incentives are linked to individual performance or share prices. For example, managerial stock options are widely discussed as an effective corporate governance mechanism to ensure the alignment of managerial interests to those of shareholders. The hypothesized relationships in Tables 1 and 2 became broadly accepted in the corporate governance literature on Japan. However, Japanese corporate governance has undergone dramatic institutional change since the mid-1990s. Corporate governance practices have become more diverse, and the effectiveness of older patterns was called into question. Aoki and Jackson (2008) and Aoki et al. (2007) engaged in a comprehensive study that included an empirical typology of Japanese firms based on a cluster analysis using a large scale dataset from 2003. In terms of the complementarities described above, around 55% of firms conformed to the J-firm model characterized by relational ownership (e.g., stable shareholders and main bank lending), insider boards, and long-term employment relations. Meanwhile, two ‘‘hybrid’’ clusters emerged: a first had market-oriented ownership with more relational employment characteristics (24% of sample firms), and a second ‘‘inverse hybrid’’ combining relational ownership or insider boards with more market-oriented employment and incentive patterns (21% of sample firms). Notably, no cluster of Japanese firms adopted a marketoriented US-style pattern of corporate governance. These results were unexpected since prevailing theories suggest that the combination of market and relational characteristics would perform poorly and be unlikely to emerge as a major pattern. Even more astounding, the two hybrid groups of firms significantly outperformed firms with the traditional J-firm characteristics. This stylized fact suggests a need to study the issue of complementarities in greater detail.

A Set Theoretic Approach to the J-Firm Using the 2003 data from the previous study of Aoki et al. (2007), we seek to further explore the hypothesized complementarities in Japanese firms using

142

GREGORY JACKSON AND NA NI

set theoretical methods based on fs/QCA.3 The data is based on a survey of all listed non-financial firms in Japan. The response rate was 34%, resulting in a complete dataset for 756 firms that were supplemented with data from the Nikkei database and Thomson Banker One. Eight variables for understanding configurations leading to firms’ high financial performance are included in the present analysis:  Ownership characteristics:  Foreign ownership: percentage of ownership by financial institutions, nonfinancial corporations, foreign investors and individuals;  Interfirm relationship: percentage of shares owned by other nonfinancial firms.  Corporate governance characteristics of board structure based on three separate index scores:4  Shareholder protection (index of 10 survey items);  Board independence (6 survey items);  Information disclosure (10 survey items).  Employment and incentive characteristics based on the presence or absence of the following:  LTE: lifetime employment norms;  Merit: merit-based pay systems;  Stock options: stock options for top managers. In order to study the potential complementarities among these 8 organizational characteristics, we utilize a novel two-step procedure that combines statistical analysis with set theoretical fs/QCA methods. Since firm performance is influenced by a wide range of factors, in the first step, we estimate financial performance by controlling for a number of organizational- and industry-level variables known to influence performance. In the second step, we utilize fs/QCA to examine the set theoretical relationships between our 8 organizational characteristics and firm’s financial performance. Here we utilize a set theoretical calibration of the unexplained residual performance from the control model in step 1 as the outcome condition in our fs/QCA analysis in step 2. The primary advantage of this approach is that we can utilize statistical controls to net out their effects on the performance outcome, while reducing the number of causal conditions entered into the set theoretical model. Since fs/QCA examines the relationships between all logically possible combinations of conditions, using a large number of causal conditions leads to an exponential number of combinations and results in the problem of logical remainders or limited diversity – configurations that are logically possible, but never observed in the data.

Understanding Complementarities as Organizational Configurations

143

The potential disadvantage of this two-step method is that we assume that no interaction effects or conjunctural causation occurs between factors used in the first step (linear estimation) and second step (set theoretical analysis).5 Specifically, our first stage model uses an ordinary least squares (OLS) regression analysis of our performance variable using STATA 12.0 software: Performance ¼ firm size þ firm age þ export þ Industry Where firm size is equal to the natural log of each firm’s total employee numbers, firm age is equal to the number of years that have passed since the firm’s establishment, and export level is measured by a dummy variable indicating whether the firm operates in an export oriented sector or not. We also included industry dummies representing different industry categories. While many possible measures exist for firm performance, we focus here on return of assets (ROA) due to its wide relevance and comparability as a performance indicator. In the second step of our analysis, we use fs/QCA to assess the set theoretic consistency between configurations of our 8 corporate governance conditions and high ROA outcomes. The data from each item must first be transformed into ‘‘fuzzified’’ set membership scores between zero and one that represent the degree to which each case has membership in the set of cases having a certain property. In terms performance, we utilize the unexplained residual performance of the control model in step 1 as the outcome condition to be explained in our fs/QCA analysis. Following the ‘‘direct method’’ or log odds method (see Ragin, 2008), we calibrated the z-transformed residuals of step 1 our estimations around three anchor points: a z-score of 1 or more being ‘‘fully in’’ with a set membership of 1, a z-score of 1 or less being ‘‘fully out’’ with a set membership of 0, and a z-score of 0 with a set membership of 0.5. For example, if performance was predicted perfectly by the model, the resulting residual of zero would have a set membership of 0.5. Thus, firms with the expected performance level are treated as being neither in, nor out of the set of high performers. Our outcome set is firms who are ‘‘high performing’’ relative to firms of similar size, age and sector.6 The causal conditions based on ownership and board structure were calibrated using an indirect method of a standardized ranking order of cases. Since the raw data on employment characteristics were limited to yes/no indicators, these items received binary set membership scores.

144

GREGORY JACKSON AND NA NI

Performance Based on Single Factors Table 3 shows the results of the performance estimates for ROA. Column A shows a model based on the control variables. As expected, firm size and age were positively related to ROA. Columns B and C present additional

Table 3.

Constant Firm size (log employees) Firm age Export orientation (dummy) Foreign ownership

Regression Estimation of ROA.

Control Variables

Control Variables plus Ownership and Corporate Governance

Control Variables plus Ownership, Corporate Governance, and HRM

0.95 (2.99) 0.33 (0.16) 0.31 (0.14) 2.66 (3.86)

3.73 (3.59) 0.05 (0.22) 0.22 (0.16) 7.02 (3.54) 0.08 (0.03)  0.02 (.01) 0.02 (0.05) 0.14 (0.04) 0.06 (0.04)

Included 823 0.096 0.063

Included 732 0.121 0.078

4.19 (3.78) 0.03 (0.23) 0.19 (0.17) 3.25 (3.65) 0.08 (0.03) 0.02 (0.01) 0.02 (0.05) 0.14 (.04) 0.07 (0.04) 0.46 (0.69) 0.25 (0.51) 0.01 (0.52) Included 703 0.120 0.071

Interfirm ownership Shareholder protection index Board independence index Information disclosure index Lifetime Employment (dummy) Merit pay (dummy) Stock options (dummy) Industry dummies N R2 Adj-R2

Indicates significance at the pr0.05 (pr0.01; pr0.001) level of confidence (two-tailed

test); standard errors are in the parentheses. Note: The total number of observations in each model is less than the total sample due to colinearity between industry variables and either the corporate governance or HRM characteristics.

145

Understanding Complementarities as Organizational Configurations

estimations based on the key ownership, board and employment variables of interest. Specifically, the coefficients of foreign ownership were positive and significantly related to ROA in both models, whereas both coefficients of board independence showed a negative relationship with ROA. Unlike the regression estimates based on statistical correlations, the fs/QCA analysis assesses set theoretical consistency of each condition with an outcome. Here we can distinguish between whether a condition is necessary for an outcome and whether the same condition is sufficient for an outcome. Table 4 presents the set theoretical tests for necessity (e.g., all cases

Table 4.

Set Theoretic Necessity and Sufficiency for High ROA, Single Conditions.

Condition

Foreign ownership BForeign ownership Interfirm ownership BInterfirm ownership Shareholder protection BShareholder protection Board independence BBoard independence Disclosure BDisclosure Lifetime employment BLifetime employment Merit pay BMerit pay Stock options BStock options

Necessity Consistency Score, Benchmark Level 0.9

Sufficiency Consistency Score, Benchmark Level 0.8

Number of Cases with Each Condition (MembershipW0.5)

0.72

0.72

362

0.69

0.65

373

0.69

0.67

364

0.73

0.70

370

0.62

0.69

351

0.74

0.63

381

0.68

0.66

464

0.74

0.72

268

0.71 0.70 0.83

0.71 0.65 0.51

330 402 656

0.20

0.51

137

0.48 0.55 0.29 0.71

0.52 0.51 0.53 0.50

366 427 231 592

146

GREGORY JACKSON AND NA NI

of high performance have a particular organizational characteristic) and for sufficiency (e.g., all cases of an organizational characteristic are associated with the outcome). In evaluating set theoretical consistency, we utilize benchmark levels of consistency 0.90 for necessity and 0.80 for sufficiency, following Ragin (2008). The benchmark level of 0.90 is stricter for necessity, because this leads to more conservative results. For illustrative purposes, we show each condition in isolation. Notably, despite the statistically significant effects on performance in the regression model, the same conditions were neither individually necessary, nor sufficient for high performance. This finding is not surprising, since the theory of complementarities suggest that no organizational factors are best practices in isolation, but will have positive effects only in conjunction with other factors. Next we will more fully exploit the fs/QCA approach to examine more complex configurations.

Performance based on Complementarities Next, we examine complementarities among configurations of the 8 corporate governance conditions based on their set theoretic consistency with membership among high ROA firms. In this specific fs/QCA application, we apply three criteria for set theoretic consistency. First, the consistency between each configuration and the performance outcome is assessed against a benchmark ratio of 0.80 (see Ragin, 2008) using a probabilistic F-test at 0.05 p-value. Second, we have specified a minimum of 2 observed cases for each configuration to avoid drawing conclusions from single cases. Third, we have tested the consistency of each configuration against the negation of the outcome to eliminate potentially trivial conditions, where a condition is identified as being simultaneously a subset of both the outcome (high ROA) and the negative thereof (absence of high ROA). This problem of simultaneous subset relations may occur among cases with low membership scores or a skewed distribution in fuzzy sets (Schneider & Wagemann, 2012). Taken together, these criteria are intended to restrict our analysis to configurations where the set-theoretic relationships are highly consistent and fairly common in our dataset. The resulting configurations are then simplified to eliminate redundant factors using the Quine-McClusky algorithm. When applied to complementarities, however, we note that no easy counterfactuals exist – both the presence and absence of each factor may be hypothetically associated with high ROA, depending on the overall configuration. In our analysis, a parsimonious solution is derived by

Understanding Complementarities as Organizational Configurations

147

introducing counterfactual simplifying assumptions for configurations with two or less cases as counterfactual cases, but we place interpretative emphasis on the complex solution. Following Fiss (2011), we adopt a tabular presentation that uses large symbols to represent the parsimonious solution, and small symbols to represent the more complex solution. While the previous section demonstrated that no conditions are individually sufficient for high performance, here we expand the complexity of our model here stepwise for illustrative purposes.7 First, we test complementarities within each of the three domains of conditions separately – ownership structure (2 conditions), board structure (3 conditions), and employment practices (3 conditions). However, only one configuration was consistent with high performance – firms with high information disclosure, but the absence of board independence.8 Coverage score suggests that these cases constitute 49% of the membership in the outcome, suggesting an important but partial explanation for high ROA. Next, we test for complementarities between the different domains: ownership with board structure, ownership with employment, and board structure with employment. And finally, we present the results using all three domains together for the outcomes of both high ROA and the absence of high ROA. Table 5 presents the results based on configurations of ownership and board independence. Configuration 1 (column C1) is composed of the presence of foreign ownership and absence of interfirm ownership together

Table 5. Set Theoretical Relations between High ROA and Configurations of Ownership and Board Structure. C1 Ownership characteristics Corporate governance characteristics

Consistency Raw coverage Unique coverage Overall solution consistency Overall solution coverage

Foreign ownership Inter-firm ownership Shareholder protection Board independence Information disclosure 0.88 0.37 0.37 0.86 0.37

Note: The complex and parsimonious solution are identical in this model, and we report only parsimonious solutions here using large symbols because they are ‘‘core’’ conditions for differentiating different configurations.

148

GREGORY JACKSON AND NA NI

with the absence of board independence and presence of information disclosure. This result is similar to the configuration based only on board independence, but shows that these two factors need to be complemented by the presence of foreign ownership and absence of horizontal relationship. In theoretical terms, this pattern partially supports the notion of complementarities between market-oriented ownership and outsider-oriented board practices – along the lines of the US model. However, a crucial difference here is the absence of independent members of the board, which suggests something of a ‘‘hybrid’’ configuration where US style and Japanese style corporate governance characteristics combine. Total coverage of this configuration was 0.37, which suggests a very important ‘‘pathway’’ to high ROA. Table 6 presents the results based on configurations of board structure and employment. The set theoretical results show three configurations consistent with high ROA. Configuration 1 (column C1) is characterized by the absence of shareholder protection and board independence, but presence of information disclosure together with merit pay and stock options. This configuration consists of US-style employment characteristics together with strong information disclosure. However, the absence of board independence

Table 6. Set Theoretical Relations between High ROA and Configurations of Board Structure and Employment.

Corporate governance characteristics Employment and incentive characteristics Consistency Raw coverage Unique coverage Overall solution consistency Overall solution coverage Parsimonious solution Overall solution consistency Overall solution coverage

C1

C2

C3

0.90 0.07 0.05 0.86 0.28

0.88 0.03 0.01

0.85 0.20 0.20

Shareholder protection Board independence Information disclosure Lifetime employment Merit pay Stock options

0.83 0.28

Note: The parsimonious solution is shown with large symbols and the complex solution with small symbols.

Understanding Complementarities as Organizational Configurations

149

again suggests a ‘‘hybrid’’ logic, which contains features of insider-boards and limited shareholder protection. Configuration 2 (C2) similarly displays some complementarities based US-style logic, but together with hybrid elements. This configuration combines market-oriented employment (absence of lifetime employment, and presence of merit pay and stock options) together with the presence of board independence but absence of shareholder protection. This configuration, however, has a very low level of coverage (1%), suggesting that it is very rare in Japan. Meanwhile, the most important pattern is provided by Configuration 3 (C3) with a coverage rate of 0.20. This configuration is characterized by more relational or J-firm style employment (life time employment with the absence of merit pay and stock options) together with the presence of information disclosure and absence of board independence. This configuration partially displays the complementarities expected with a more J-style firm. Here relational employment is combined with the absence of board independence, but blends in a ‘‘hybrid’’ fashion with market-oriented elements such as high information disclosure. These firms adopt high levels of information disclosure, but still use insider boards and relational employment system that emphasizes the traditional norm of lifetime employment practices, seniority pay, and no stock options. Taken together, these three configurations have a total coverage of 0.28, suggesting that these patterns are an important explanation for high ROA in Japan. Next, we examined configurations of ownership and employment conditions in combination. Interestingly, none of these configurations passed our combined tests of set theoretical consistency with high ROA. If we relax our F-test to a p-value of 0.10, we find that the presence of both foreign and inter-firm ownership together with the Japanese-style employment practices (lifetime employment plus the absence of merit pay and stock options) has a consistent association to high ROA (consistency of 0.84) and coverage of 0.19. The lower degree of consistency among these factors suggests the potential importance of board structure factors as mediating the relationship between ownership and employment. The last model represents the most complex scenario for understanding complementarities, which consists of all eight ownership, board, and employment conditions. Table 7 reveals a solution with four configurations. For example, Configuration 1 (column C1) comes closest to the logic of US-style corporate governance – the presence of foreign ownership, information disclosure, merit pay, and stock options. Alongside these four market-oriented conditions, the configuration is also characterized by the absence of inter-firm ownership and shareholder protection. The latter factor

150

GREGORY JACKSON AND NA NI

Table 7. Set Theoretical Relations between High ROA and Configurations of Ownership, Board Structure, and Employment.

Ownership characteristics Corporate governance characteristics

Employment and incentive characteristics

C1

C2

C3

C4

0.89 0.08 0.05 0.90 0.30

0.90 0.14 0.06

0.91 0.15 0.02

0.93 0.09 0.01

Foreign ownership Interfirm ownership Shareholder protection Board independence Information disclosure Lifetime employment Merit pay Stock options

Consistency Raw coverage Unique coverage Overall solution consistency Overall solution coverage Note: Only the complex solution is shown.

is puzzling, since we would expect shareholder protection to complement these market-oriented features – an issue that we will return to in the conclusion. The other three configurations (C2, C3, C4) fit more squarely into a ‘‘hybrid’’ logic that combines relational employment patterns with some market-oriented aspects of ownership and a partially outsider-oriented board structure. These three configurations share a common set of core conditions: presence of information disclosure and absence of board independence together with lifetime employment. This core configuration is variously combined with other elements, although the most important configuration (C2) has the absence of merit pay and stock options. This configuration contributes the most to high ROA with a unique coverage score of 0.06. These four configurations have an overall coverage score of 0.30, and maintain very high consistency.9 An interesting aspect of set theoretical analysis is its asymmetry. The conditions leading to low ROA are not always the inverse conditions as those conditions leading to high ROA. Here we conduct a separate analysis of low performance, using the set memberships (1 minus the membership in the set of high ROA firms). The consistent solution configuration has extremely low coverage scores. So for illustrative purposes, we report the solution based on a reduced benchmark level of significance from 0.05 to

151

Understanding Complementarities as Organizational Configurations

0.10. The results in Table 8 are quite suggestive. All three configurations show a combination of board independence and absence of foreign ownership together with lifetime employment and merit pay. These features combine variously with the presence of inter-firm ownership or absence of information disclosure and stock options. Together, these configurations have a total coverage of 0.12. From the perspective of complementarities, we find that the presence of certain practices, such as lifetime employment, is associated with the opposite outcome of low performance when used in conjunction with different corporate governance variables. The empirical results are not intended as a last word on understanding the effectiveness of Japanese corporate governance. Our empirical analysis has important limitations, and intended to be suggestive of a general approach to complementarities. A more comprehensive analysis would require a number of extensions. First, the results need to be confirmed based on other performance measures and a longer time period of observations. Second, additional factors should be introduced. Most theories based on complementarities suggest that different organizational configurations give advantages in relation to particular types of innovation patterns or resources. Thus, the analysis could be extended to take into account a number of additional contextual factors related to industry environment or the competitive strategies of firms. Third, the employment measures in our

Table 8. Set Theoretical Relations between Low ROA and configurations of Ownership, Board Structure and Employment.

Ownership characteristics Corporate governance characteristics

Employment and incentive characteristics

Consistency Raw coverage Unique coverage Overall solution consistency Overall solution coverage Note: Only the complex solution is shown

C1

C2

C3

0.91 0.08 0.01 0.91 0.12

0.90 0.08 0.00

0.92 0.11 0.04

Foreign ownership Interfirm ownership Shareholder protection Board independence Information disclosure Lifetime employment Merit pay Stock options

152

GREGORY JACKSON AND NA NI

current data are binary measures, which limit their application in fuzzy set analysis. We compensated for this limitation by setting consistency thresholds very high, but therefore face well-known trade-off between set theoretic consistency and coverage. Our analysis was only able to explain about 25–30% of set membership in the group of high ROA firms. More fine-grained measures of employment practices may achieve higher coverage rates based on a more finely grained analysis.

IMPLICATIONS FOR THE THEORY AND METHODS OF STUDYING COMPLEMENTARITIES The empirical results based on fs/QCA have suggestive implications for the study of complementarities. First, no organizational element is universally a ‘‘best practice’’ that consistently leads to high performance. Second, a number of complementary configurations exist across the different domains of ownership, board structure, and employment. In particular, the presence and absence of the same features (e.g., shareholder rights, merit pay, or stock options) are associated with high performance when combined with different configurations. Likewise, other factors such as lifetime employment were associated with both high and low ROA, depending on the context. Taken together, this evidence is strongly suggestive that complementarities are at play. Third, the empirical results do not conform to theoretical models of complementarities based on coherent market or relational logics, but are largely consistent with the notion of a ‘‘hybrid’’ model of corporate governance in Japan (Aoki et al., 2007). Complementary configurations contain multiple elements of market-oriented or relational logics, but interestingly these same configurations also contain at least one contradictory element, characterized by an opposite logic. In particular, a hybrid pattern of market-oriented ownership and information disclosure seems to complement more relational aspects of employment within the firm. But even in these cases, certain features of board structure also contain an opposite relational logic – for example, in Table 7, configurations 2, 3, and 4 also contain the absence of board independence, which suggests that a largely insider board is an important mediating factor in stabilizing this ‘‘hybrid’’ configurations. The hybrid logics of complementarities may be surprising at one level, but perhaps less so at another level. This observation returns us to the distinction between complementarities and coherence. While a consistent

Understanding Complementarities as Organizational Configurations

153

single logic of either market or relationships may increase coherence of organizational arrangements, a fully coherent type of configurations does not seem to be complementary. While we can only speculate on the reason here, organizational arrangements based on conflicting or incoherent logics may result in a greater ability to respond to complex or change external conditions or implement a balanced strategic approach (see e.g., Streeck, 1997). Importantly, the empirical analysis does not suggest that any combination of opposites will be effective. Rather, the specific configurations provide guidance for future research and perhaps inspire new or better theoretical models to understand such combinations (see Aoki & Jackson, 2008). The empirical results also highlight the advantages analyzing configurations through set theoretical methods, as discussed in Section 3. The systematic approach to configurations in fs/QCA analysis revealed some unexpected combinations, and thus was able to move beyond the assumptions about complementarity based on coherence. The fs/QCA approach uncovered some suggestive hybrid combinations, and was able to show some asymmetric properties in relation to high and low performance outcomes. At the same time, approaches based on fs/QCA have limitations and need to developed further in terms of approaches to calibration and testing conditional theoretical statements discussed earlier (e.g., X1 is necessary to the effectiveness of Y1 in relation to Z1). As a small step forward, we propose several innovations in relation to previous fs/QCA studies, particularly a new two-step approach combining statistical analysis of net effects with set theoretical analysis. This strategy is potentially useful in future studies, as a way of calibrating outcomes in relation to a baseline model and limiting the number of conditions entered into set theoretical analysis in ways that reduce analytical problems related to limited diversity.

CONCLUSION This study focused on the concept of complementarities as configurations of organizational elements that jointly have positive effects on performance. Using a novel set-theoretic approach, we re-analyze past work by Aoki et al. (2007) and investigate how configurations of ownership, board structure, and employment practices shape firm performance. In particular, the results indicate the presence of multiple pathways that combine various elements of the three categories of organizational factors leading to high ROAs.

154

GREGORY JACKSON AND NA NI

They also suggest that set-theoretic methods may present a very useful methodological alternative to studying complementarities. Overall, we have sought to show how the effectiveness of organizational practices related to corporate governance can be better theorized and explained empirically by taking account of potential complementarities. Complementarity is itself a core concept within the wider effort to construct configurational theories of organizations. An important element of this agenda concerns research methods, and we have sought to make plausible how set theoretical methods such as fs/QCA can help to make more realistic empirical investigations in this direction. In order to further this agenda, we would emphasize that more precise middle-range theories are needed. Our brief look into the existing corporate governance literature shows how dominant theories are based largely on the idea of coherence among organizational elements – meanwhile, our empirical analysis shows a more complex picture, where various hybrid arrangements are effective complements within larger configurations. There is a lot to do in terms of both theory and method, and a huge potential for new contributions to the field based on their effective combination.

NOTES 1. Necessary conditions are more similar to the concept of essentiality rather than complementarity (see Aoki & Jackson, 2008). 2. The set theoretical consistency score for a set of sufficient conditions is given by the following inclusion ratio: IXY ¼ Smin(xi, yi)/Sxi. Here X is equal to the membership in the predictor set and Y is equal to the membership in the outcome set. Consistency can be tested against a benchmark score, such as 0.8. 3. This data was collected by Hideaki Miyajima as part of a research team at the Policy Research Institute, Ministry of Finance. There empirical results of the cluster analysis reflect our joint work and are summarized here with kind acknowledgement and thanks. Further details of the sample, survey construction, and variables are found in Aoki et al. (2007). 4. Each index item takes a value 0 or 1. The index scores were computed by adding up all the survey items in each category. High values indicate a more outsideroriented board structure. 5. To address problems of limited diversity, Schneider and Wagemann (2012) have also proposed a two-step approach using two different stages of fs/QCA to examine more remote and proximate causal conditions. The distinctive feature of the method proposed here is to combine statistical models of net effects in the first stage with a second stage of fs/QCA that utilizes a calibrated set membership for the outcome based on the residuals of this model.

Understanding Complementarities as Organizational Configurations

155

6. Ragin (2008) argues that calibration must theoretically-based (anchored by external criteria that need to be justified on substantive grounds), rather than databased (anchored around the relative variation within a sample). Since we lack a strong theoretical basis for defining high performance, this study makes use of relative variation but anchors these values in relation to the model estimates from step 1. While other approaches or thresholds could be used, the chosen calibration strategy provides an intuitive theoretical interpretation of the cross-over point and includes cases of above expected average performance as being ‘‘more in than out’’ of the set based. 7. This stepwise application is not a normal part of the fs/QCA method, but used here for presentational purposes in order to illustrate the possible complementary interactions based on different configurations. 8. This configuration remains consistent when combined with both the presence or absence of shareholder protection, and consequently this condition can be dropped as logically redundant. 9. No parsimonious model could be calculated within STATA/IC due to technical limitations of the software in handling a large number of macros with this number of variables. Since our theoretical approach does not contain any easy counterfactuals, we concentrate here on the interpretation of the complex results for the full model in Table 5 and the analysis of low firm performance in Table 6.

ACKNOWLEDGMENTS For useful comments and criticisms, the authors would like to thank Peer Hull Christensen, Peer Fiss, Verena Girschik, two anonymous reviewers, and the participants of the VHB conference on organization studies, the workshop ‘‘Institutional Diversity in Europe and Asia’’ at the FU Berlin, and ‘‘Qualitative Comparative Analysis (QCA): Perspectives for Political Sciences, Sociology and Organizational Research’’ conference at the University of Hamburg. All errors remain our own.

REFERENCES Aguilera, R., Filatotchev, I., Gospel, H., & Jackson, G. (2008). An organizational approach to comparative corporate governance: Costs, contingencies, and complementarities. Organization Science, 19(3), 475–492. Aguilera, R., & Jackson, G. (2003). The cross-national diversity of corporate governance: Dimensions and determinants. Academy of Management Review, 28(3), 447–465. Akkermans, D., Castaldi, C., & Los, B. (2009). Do liberal market economies’ really innovate more radically than ‘coordinated market economies’?: Hall and Soskice reconsidered. Research Policy, 38(1), 181–191. Amable, B. (2003). The diversity of modern capitalism. Oxford: Oxford University Press.

156

GREGORY JACKSON AND NA NI

Aoki, M. (1994). The Japanese firm as a system of attributes. In M. Aoki & R. Dore (Eds.), The Japanese firm: Sources of competitive strength (pp. 11–40). Oxford: Oxford University Press. Aoki, M. (2001). Toward a comparative institutional analysis. Cambridge, MA: MIT Press. Aoki, M., & Jackson, G. (2008). Understanding an emergent diversity of corporate governance and organizational architecture: An essentiality-based analysis. Industrial and Corporate Change, 17(1), 1–27. Aoki, M., Jackson, G., & Miyajima, H. (2007). Corporate governance in japan: institutional change and organizational diversity. Oxford: Oxford University Press. Bae, J., & Lawler, J. (2000). Organizational and HRM strategies in Korea: Impact on firm performance in an emerging economy. Academy of Management Journal, 43(3), 502–517. Black, J., & Boal, K. (1994). Strategic resources: Traits, configurations, and paths to sustainable competitive advantage. Strategic Management Journal, 15, 131–148. Boyer, R. (2004). New growth regimes, but still institutional diversity. Socio-Economic Review, 2(1), 1–32. Chandler, A. (1962). Strategy and structure: Chapters in the history of the American industrial enterprise. Cambridge, MA: MIT Press. Crilly, D. (2011). Predicting stakeholder orientation in the multinational corporation: A midrange theory. Journal of International Business Studies, 42, 694–717. Crouch, C. (2005). Capitalist diversity and change. New York, NY: Oxford University Press. Delery, J. E., & Doty, D. (1996). Modes of theorizing in strategic human resource management: Tests of universalistic, contingency, and configurational performance predictors. Academy of Management Journal, 39(4), 802–835. Doty, D., Glick, W., & Huber, G. (1993). Fit, equifinality, and organizational effectiveness: A test of two configurational theories. Academy of Management Journal, 36(6), 1196–1250. Edgeworth, F. Y (1881). Mathematical psychics: An essay on the application of mathematics to the moral sciences. London: C. Kegan Paul & Co. Ennen, E., & Richter, A. (2010). The whole is more than the sum of its parts – Or is it? A review of the empirical literature on complementarities in organizations. Journal of Management, 36(1), 207–233. Fiss, P. (2007). A set-theoretic approach to organizational configurations. Academy of Management Review, 32(4), 1180–1198. Fiss, P. (2011). Building better causal theories: A fuzzy set approach to typologies in organization research. Academy of Management Journal, 54(2), 393–420. Grandori, A., & Furnari, S. (2008). A chemistry of organization: combinatory analysis and design. Organization Studies, 29(3), 459–485. Greckhamer, T., Misangyi, V., Elms, H., & Lacey, R. (2008). Using qualitative comparative analysis in strategic management research: An examination of combinations of industry, corporate, and business-unit effects. Organizational Research Methods, 11, 695–726. Gresov, C., & Drazin, R. (1997). Equifinality: Functional equivalence in organization design. Academy of Management Review, 22(2), 403–428. Hall, P., & Gingerich, D. W. (2009). Varieties of capitalism and institutional complementarities in the political economy: an empirical analysis. British Journal of Political Science, 39(3), 449–482. Hall, P., & Soskice, D. (Eds.). (2001). Varieties of capitalism. Oxford: Oxford University Press.

Understanding Complementarities as Organizational Configurations

157

Herrmann, A. (2005). Converging divergence: How competitive advantages condition institutional change under EMU. Journal of Common Market Studies, 43(2), 287–310. Hill, C. (1988). Differentiation versus low cost or differentiation and low cost: A contingency framework. Academy of Management Review, 13(3), 401–412. Hitt, M., Bierman, L., Shimizu, K., & Kochhar, R. (2001). Direct and moderating effects of human capital on strategy and performance in professional service firms: A resourcebased perspective. Academy of Management Journal, 44(1), 13–28. Ho¨pner, M. (2005). What connects industrial relations and corporate governance? Explaining institutional complementarity. Socio-Economic Review, 3, 331–358. Jackson, G. (2005a). Modeling complementarity: Multiple functions and different levels. SocioEconomic Review, 3, 381–384. Jackson, G. (2005b). Employee representation in the board compared: A fuzzy sets analysis of corporate governance, unionism, and political institutions. Industrielle Beziehungen, 12(3), 1–28. Kang, N., & Moon, J. (2012). Institutional complementarity between corporate governance and corporate social responsibility: A comparative institutional analysis of three capitalisms. Socio-Economic Review, 10(1), 85–108. Kang, S. C., Morris, S., & Snell, S. A. (2007). Relational archetypes, organizational learning, and value creation: Extending the human capital architecture. Academy of Management Review, special issue, 32(1), 236–256. Karnani., A. (1984). Generic competitive strategies: An analytical approach. Strategic Management Journal, 5(4), 367–380. Ketchen, D., Jr., Thomas, J., & Snow, C. (1993). Organizational configurations and performance: A comparison of theoretical approaches. Academy of Management Journal, 36(6), 1278–1313. Kogut, B., & Ragin, C. (2006). Exploring complexity when diversity is limited: Institutional complementarity in theories of rule of law and national systems revisited. European Management Review, 3(1), 44–59. Lepak, D. P., & Snell, S. A. (2002). Examining the human resource architecture: The relationships among human capital, employment, and human resource configurations. Journal of Management, 28(4), 517–543. Matsuyama, K. (1995). Complementarities and cumulative processes in models of monopolistic competition. Journal of Economic Literature, 33(2), 701–729. Miles, R., & Snow, C. (1978). Organizational strategy, structure, and process. New York, NY: McGraw-Hill. Milgrom, P., & Roberts, J. (1990). The economics of modern manufacturing: Technology, strategy and organization. American Economic Review, 80, 511–528. Milgrom, P., & Roberts, J. (1995). Complementarities and fit: Strategy, structure and organizational change in manufacturing. Journal of Accounting and Economics, 19, 179–208. Miller, D. (1986). Configurations of strategy and structure: Towards a synthesis. Strategic Management Journal, 7(3), 223–249. Miller, D. (1988). Relating Porter’s business strategies to environment and structure: Analysis and performance implications. Academy of Management Journal, 31(2), 280–308. Miller, D., & Friesen, P. (1986). Porter’s (1980) generic strategies and performance: An empirical examination with American data. Organizational Studies, 7(1), 37–55. Nadler, D., & Tushman, M. (1980). A model for diagnosing organizational behavior. Organizational Dynamics, 9(2), 35–51.

158

GREGORY JACKSON AND NA NI

Powell, T., & Dent-Micallef, A. (1997). Information technology as competitive advantage: The role of human, business and technology resources. Strategic Management Journal, 18(5), 375–405. Ragin, C. (2000). Fuzzy set social science. Chicago, IL: University of Chicago Press. Ragin, C. (2008). Redesigning social inquiry: Fuzzy sets and beyond. Chicago, IL: University of Chicago Press. Ragin, C., & Sonnett, J. (2005). Between complexity and parsimony: Limited diversity, counterfactual cases, and comparative analysis. In S. Kropp & M. Minkenberg (Eds.), Vergleichen in der Politikwissenschaft (pp. 180–197). Wiesbaden: VS Verlag fu¨r Sozialwissenschaften. Rhyne, L., & Teagarden, M. (1997). Technology-based competitive strategy: An empirical investigation. Journal of High Technology Management Research, 8(2), 187–212. Rumelt, R. (1974). Strategy, structure, and economic performance. Boston, MA: Harvard University. Schneider, C., & Wagemann, C. (2012). Set-theoretic methods for the social sciences: A guide to qualitative comparative analysis. Cambridge, UK: Cambridge University Press. Schneider, M., & Paunescu, M. (2012). Changing varieties of capitalism and revealed comparative advantages from 1990 to 2005: A test of the Hall and Soskice claims. Socio-Economic Review, 10(4), 731–753. Schneider, M., Schulze-Bentrop, C., & Paunescu, M. (2009). Mapping the institutional capital of high-tech firms: A fuzzy-set analysis of capitalist variety and export performance. Journal of International Business Studies, 41(2), 246–266. Sheppeck, M., & Militello, J. (2000). Strategic HR configurations and organizational performance. Human Resource Management, 39(1), 5–16. Siggelkow, N., & Rivkin, J. (2005). Speed and search: Designing organizations for turbulence and complexity. Organization Science, 16(2), 101–122. Streeck, W. (1992). Social institutions and economic performance: Studies of industrial relations in advanced capitalist economies. London: SAGE Publications. Streeck, W. (1997). Beneficial constraints: On the economic limits of rational voluntarism. In J. R. Hollingsworth & R. Boyer (Eds.), Contemporary capitalism: The embeddedness of institutions (pp. 197–219). Cambridge, UK: Cambridge University Press. Taylor, M. (2004). Empirical evidence against varieties of capitalism’s theory of technological innovation. International Organization, 58(3), 601–631. Teece, D. (1986). Profiting from technological innovation: Implications for integration, collaboration, licensing and public policy. Research Policy, 15, 285–305. Valle, R., Martin, F., Romero, P., & Dolan, S. (2000). Business strategy, work processes and human resource training: Are they congruent? Journal of Organizational Behavior, 21(3), 283–297. Whitley, R. (2007). Business systems and organizational capabilities: The institutional structuring of competitive competences. Oxford: Oxford University Press. Williamson, O. (1975). Markets and hierarchies: Analysis and antitrust implications. New York, NY: McMillan Free Press. Wright, P., Kroll, M., Tu, H., & Helms, M. (1991). Generic strategies and business performance: An empirical study of the screw machine products industry. British Journal of Management, 2(1), 57–65.

CHAPTER 7 CORPORATE GOVERNANCE AND CONFIGURATION RESEARCH: THE CASE OF FOREIGN IPOs LISTING IN LONDON R. Greg Bell, Ruth V. Aguilera and Igor Filatotchev ABSTRACT Corporate governance research based on agency theory has been criticized for being ‘‘under-contextualized,’’ and for evaluating various governance practices independently. To address both criticisms, we take a configurational approach and show how firm-level governance practices interact with informational asymmetries associated with a firm’s industry. By examining foreign Initial Public Offerings (IPOs) that have chosen to list on London stock exchanges, we demonstrate that an assessment of the firm-level corporate governance configurations is incomplete without taking into account the firm’s industry affiliation. Our use of fs/QCA underscores the possibilities configurational approaches have in advancing theories of corporate governance. Keywords: Comparative Corporate Governance; Foreign IPO; Fuzzy-set methods

Configurational Theory and Methods in Organizational Research Research in the Sociology of Organizations, Volume 38, 159–180 Copyright r 2013 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0733-558X/doi:10.1108/S0733-558X(2013)0000038011

159

160

R. GREG BELL ET AL.

INTRODUCTION Corporate governance refers to the ‘‘structure of rights and responsibilities among the parties with a stake in the firm’’ (Aoki, 2001). Firms have some degree of freedom regarding the governance practices they decide to adopt, and the analysis of why and what effects these choices will generate is central to corporate governance research (Aguilera & Jackson, 2003). Consequently, a rigorous conceptualization of the context in which organizations operate in is critical since the mechanisms linking the adoption of corporate governance practices and their outcomes (e.g., performance) will depend heavily on the context in which organizations are embedded in (Globerman, Peng, & Shapiro, 2011). Governance theorists, however, have only just begun integrating different strands of organizational theory with economics and finance perspectives on corporate governance. Our understanding of how firm-level governance practices (e.g., board monitoring, executive incentives, share ownership structure) interact with different contexts is relatively limited. The limitations of the mainstream corporate governance research, based largely on agency theory, are related to two inherent shortcomings underpinning the economics and finance perspectives. First, agency-driven research is characterized by its ‘‘under-contextualized’’ nature and inattention to various organizational environments (see Aguilera, Filatotchev, Gospel, & Jackson, 2008). By assuming that the dynamics between principals and agents are universal or axiomatic, the majority of extant research has tended to relegate the patterned variation of organizational forms under different settings as mere distortions or deviations rather than a phenomenon that merits rigorous analysis. But this is a severely limited way of analyzing corporate governance as, for example, evaluations of how the governance practices of firms competing in technology industries differs from more traditional and mature sectors. In addition, although researchers discuss the effectiveness and efficiency of corporate governance as a firm-level mechanism, the most studies look at various governance factors in isolation. Governance factors, however, are linked together in conjunction; therefore, the causal mechanisms linking governance factors with organizational outcomes are conjunctural. Such limitations also stem from a methodological bias in corporate governance research which is dominated by regression analyses. An overwhelming majority of corporate governance research applies various regression techniques to model the association between governance practices (e.g., board independence, executive share options, ownership patterns) with

Corporate Governance and Configuration Research

161

a range of organizational outcomes (e.g., performance). By its very design, researchers assume that firms can use governance at no costs, and the practical implications are often that adding more or less governance practices should lead to better (or worse) organizational outcomes. However, there is a growing understanding that changes in a firm’s strategic positioning and its industry affiliation may be associated with re-balancing between the wealth-protection and wealth-creation functions of governance (Filatotchev, Wright & Toms, 2006), and that the same organizational outcomes may be achieved by different constellations of governance factors. Unfortunately, traditional multivariate econometric techniques are not able to fully address these theoretical assumptions. New methodological approaches are needed to investigate patterned variations of corporate governance mechanisms as well as infuse more existential realism to the analysis. In this chapter, we discuss the advantages of a configurational approach to corporate governance research and explain how this approach can fill the caveats left by extant research dominated by agency theory and regressionbased analysis. We begin by developing theoretical arguments and assumptions justifying the configurational approach. Afterwards, we demonstrate an application of the said approach through an analysis of the bundles of corporate governance practices necessary to achieve a successful foreign Initial Public Offerings (IPOs) – firms that bypass their home country stock exchanges to make their first public equity offers on foreign capital markets. The context of our research is the London stock exchange where non-British companies raised $22.7 billion through initial public offers on London’s Main Market and Alternative Investment Market (AIM) exchanges in 2007. We conclude the chapter with a discussion of how investigations of governance practices as bundles or configurations could advance governance research and suggest a number of future research possibilities.

CONFIGURATIONAL RESEARCH OF CORPORATE GOVERNANCE Effective corporate governance implies mechanisms to ensure executives respect the rights and interests of company stakeholders, as well as guarantee that stakeholders act responsibly with regard to the generation, protection, and distribution of wealth invested in the firm (Aguilera et al., 2008). The empirical literature on corporate governance has been mostly rooted in

162

R. GREG BELL ET AL.

agency theory, assuming that by managing the principal-agency problem between shareholders and managers, firms will operate more efficiently and perform better. This stream of research identifies situations in which shareholders’ and managers’ interests are likely to diverge and proposes mechanisms that can mitigate managers’ self-serving behavior (Shleifer & Vishny, 1997), such as the board of directors, shareholder involvement, information disclosure, auditing, the market for corporate control, executive pay, and stakeholder involvement (Filatotchev et al., 2006). Despite the large body of research, the empirical findings on the link between governance practices and firm outcomes (e.g., firm performance) are mixed and inconclusive. As a reaction to these mixed findings from the agency theory, configurational research has been proposed as an alternative theoretical framework capable of introducing some flexibility in the firm governance–outcome relationships as well as accounting for the broader environmental-firm contingencies. Specifically, the configurational perspective takes into account four main premises (Aguilera & Jackson, 2010; Aguilera, Desdender, & Kabbach de Castro, 2012). First, firms are embedded in different industry settings that either enable or constrain their strategic choices. Second, firms in different settings have some degree of freedom to decide which practices to adopt as well as the extent to which the choices are implemented. Third, governance practices do not act in isolation from each other and must be considered as an interdependent whole. Finally, there is no ‘‘one best fit all’’ set of governance practices leading to effective corporate governance. In fact, multiple combinations or constellation of governance practices may produce the same outcome; conversely, the same combinations may generate different outcomes due to differences in contextual factors. We discuss each of these in turn. The first premise is fairly well-established in the comparative corporate governance literature and claims that governance practices are embedded in the firm’s environment, with the disclaimer that structural forces are neither all-powerful nor completely intractable. For example, Filatotchev et al. (2006) attribute the mixed empirical results concerning the effectiveness of various governance mechanisms to the extant literature’s failure to incorporate the patterned variations in corporate governance contingent to the organizational environments. Similarly, Aguilera and Jackson (2003) claim that the ‘‘under-contextualized’’ approach of agency theory restricts the analysis of corporate governance to agents and principles, abstracting away other aspects of the organizational context such as diverse task environments, the life-cycle of organizations, or legal constraints which influence governance patterns.

Corporate Governance and Configuration Research

163

Second premise concerns the interpretation of agency. Emirbayer and Mische (1998) define organizational agency as ‘‘the temporally constructed engagement by actors of different structural environments – the temporalrelational contexts of action – which, through the interplay of habit, imagination, and judgment, both reproduces and transforms those structures in interactive response to the problems posed by changing historical situations’’ (p. 970). Actors within firms not only reproduce existing practices but also respond by developing new strategies to influence the environment, as delineated in Oliver’s (1991, 1992) pioneering research. Firms, therefore, have significant degrees of freedom to configure their corporate governance practices, that is, firms often elect to either fully endorse a particular practice or simply seek to comply with the minimum requirements without truly internalizing the governance practice. An illustrative example would be the considerable variation that exists in terms of how different firms define director independence or disclose compensation systems. Third, a growing number of corporate finance scholars and organizational theorists argue that corporate governance practices should not be considered in isolation from each other but rather as ‘‘bundles’’ of corporate governance practices that are aligned with one another and mutually enhance the ability of those practices to achieve efficiency outcomes (Rediker & Seth, 1995). A key dimension of the configurative approach focuses on identifying complementarities and substitutive effects among governance practices (Aguilera et al., 2008). For example, empirical research demonstrates that the simultaneous operation of multiple corporate governance practices is important in limiting managerial opportunism (Rediker & Seth, 1995). Furthermore, Desdender et al. (2011) illustrate that ownership concentration and board monitoring become substitutes when it comes to monitoring management. Specifically, the results show that while the board of directors complements its monitoring role through the higher use of external audit services when ownership is dispersed, this is not the case when ownership is concentrated. In the following section, we apply the configurational approach on an exploratory study of foreign IPOs. Our example is built from a sample of firm undergoing an IPO on U.K. stock exchanges. Foreign IPOs represent a unique laboratory for theory building related to this complex interplay between industry factors and firm-level governance since these firms originate in countries with different governance regulations and different degrees of protection of minority public market investors long before they come for a listing in the United States. As a result, they may suffer from significant information asymmetries between insiders and overseas public

164

R. GREG BELL ET AL.

market investors that make corporate governance factors particularly salient. Yet there is little research examining the relationships between foreign IPO’s governance parameters and their performance outcomes. The specific implications we derive from the study are as follows: first, we suggest that strategic choices and their effects on performance are a function of the interaction between organizations and their industry environment (Aguilera et al., 2008). That is, foreign IPO firms may substitute one governance practice for another to achieve similar performance outcomes but the extent of governance ‘‘substitutability’’ is contingent on the characteristics of the foreign IPO firm’s industry. Second, we introduce the idea of equifinality and show that this concept better explains the outcomes of foreign IPOs. Specifically, if corporate governance is contingent on the alignment of interdependent organizational and environmental characteristics, then it logically follows that there no universal ‘‘one best way’’ of corporate governance. Rather, particular practices will be effective only under certain combinations and may result in different patterns of corporate governance under different contexts (Aguilera et al., 2008). Lastly, we make a methodological contribution through our study by applying a relatively novel methodology called fs/QCA. While the idea behind configurational approach is simple and powerful, realizing it through empirical research is challenging due to methodological constraint. Because the goal of it is to account for as many interdependencies as possible, it is important that researchers use research designs and methodological techniques that enable them to (1) capture interrelationships among multiple explanatory factors – which are nearly impossible to incorporate into linear logic, and (2) can also incorporate equifinal solutions. We solve this issue through use of one such configurational tool, fs/QCA, and demonstrate its merits through the current study. By providing this example, we hope to not merely demonstrate the usefulness of this methodological tool, but to underscore the possibilities configurational approaches have in advancing our collective understanding and theories of corporate governance.

EXPLORING GOVERNANCE CONFIGURATIONS WITH FOREIGN INITIAL PUBLIC OFFERINGS Corporate Governance and Initial Public Offers IPOs have received increasing attention among scholars in a range of business disciplines from strategy (Carpenter, Pollock, & Leary, 2003;

Corporate Governance and Configuration Research

165

Higgins & Gulati, 2003), entrepreneurship (Florin, 2005; Mudambi & Treichel, 2005) and finance (Brau & Fawcett, 2006) as well as practitioners from many countries. Much of the interest is due to the fact that IPO represents a pivotal event in the history of firms in that it firms gain access to large sums of equity capital that can in turn enhance their survival chances. By undertaking an IPO, firms can accelerate their growth, launch new products, enter new markets, and attract employees. Signaling research argues that public market investors pay particular attention to governance characteristics of IPO firms because they capture the firms’ ability to deal with information asymmetries and associated agency costs (Certo, Daily, & Dalton, 2001; Filatotchev & Bishop, 2002; Sanders & Boivie, 2004). A number of IPO studies focuses on the governance roles of board characteristics and ownership patterns as internal governance solutions to the agency problem (Filatotchev & Bishop, 2002). Others show that the firm’s ‘‘professionalization’’ process manifested in significant governance changes (e.g., a low percentage of founders on board, separation of CEO and Chairman, etc.) can be important determinants of the success of firms in capital markets (Sanders & Boivie, 2004). Researchers often focus on the benefits of ‘‘good’’ governance and overlook potential negative side-effects. Indeed, despite the value IPO investors place in strong governance, neither internal nor external governance exist without costs or friction (Aguilera et al., 2008). For example, in their research on signaling through governance, Sanders and Boivie (2004) emphasize that the same governance factors that reduce investor uncertainty may also impose monitoring costs on firms which may more than offset the marginal benefits. Governance creates a trade-off between the benefit of reducing information asymmetry, and the costs associated with introducing incentives and monitoring. As a result, an IPO firm undergoes a complex process of evaluating costs and benefits of various signaling mechanisms in search of an optimal combination that minimizes both information asymmetry and costs of signaling. The costs related to each governance practice and the substitution effects between them suggest that governance practices do not operate independently (i.e., their effects are not additive). Consequently, IPO firms try to find optimal configurations that minimize costs while simultaneously maximize investor sentiments. Not surprisingly, both academics and managers contemplating an IPO are interested in understanding the make-up and number of optimal governance bundles available to IPO firms. Since the late 1990s, there have been a growing number of foreign IPOs choosing to bypass their local stock exchanges in favor of making their

166

R. GREG BELL ET AL.

capital market debut on developed country stock exchanges (Bell, Moore, & Filatotchev, 2012; Bell, Moore, & Al-Shammari, 2008). These firms usually seek financial resources in capital markets outside their home country in order to fulfill their high growth potential. However, foreign IPO firms represent untested management, and it is quite likely that investors will have concerns about protecting their investment. In addition to ‘‘liabilities of newness,’’ these firms are exposed to ‘‘liability of foreignness’’ because of information asymmetries and potential agency costs (Bell, Filatotchev, & Rasheed, 2012). To date there has been very little attention paid to the study of foreign IPOs and the factors which impact the benefits of international listings. Foreign IPOs present a unique context in which to examine how different combinations of governance practices support the ability of firms to reduce informational asymmetries and to achieve high levels of performance at IPO. For example, these firms are not only making their capital market debut but also doing so in foreign markets where they are usually unknown to investors. Some foreign IPO firms come from traditional industries, such as retailing or banking, whereas others represent a population of highgrowth/high-risk technology firms without long trading histories or stable cash flows. Our main argument is that foreign IPOs coming from hi-tech industrial sectors, such as IT, biotechnology, and Internet, may particularly suffer from information asymmetries between insiders and public market investors. Such asymmetries, in turn, will have a negative effect on the firms’ stock-market performance. As a result, they should be limited in the extent of substitution between their governance factors and may be forced to rely more extensively on governance ‘‘bundles’’ to achieve the same level of performance as IPOs coming from more traditional industrial sectors. At present, we know little on how the industry context shapes the optimal bundles, or combinations, of governance practices that determine investors’ evaluation of the foreign IPO’s quality. In the following sections we explore these possibilities with the fs/QCA method.

Fs/QCA and Foreign IPOs Fs/QCA is unique in that it allows researchers to explore the multiple alternate internal and external governance combinations leading to high IPO performance, and how features associated with their country of origin impact corporate governance decisions. Fs/QCA is based in set-theory and causal claims are developed by means of superset and subsets (Ragin, 2008)

Corporate Governance and Configuration Research

167

and is useful when researchers argue that a combination, or bundle, of factors work in concert with one another to be a sufficient cause for an outcome (Mahoney & Goertz, 2006). The number of scholars investigating economic and organizational phenomena with set-theoretic methods has risen considerably in the last few years (Fiss, 2011, 2007; Grandori & Furnari, 2008; Greckhamer, Misangyi, Elms, & Lacey, 2008; Jackson, 2005; Kogut, MacDuffie, & Ragin, 2004; Pajunen, 2008; Schneider, SchulzeBentrop, & Paunescu, 2010). Fs/QCA offers a number of advantages over regression analysis, a method often used to evaluate IPO performance. For example, assessing how three or more factors interact to produce on outcome can be quite challenging with regression given that if a factor does influence the outcome in only a handful of cases the effects can become masked or invisible. Fs/QCA helps overcome this caveat by ignoring variation and distribution in variables, and by not isolating the net independent effect of each variable on an outcome. Furthermore, fs/QCA is not centered upon variable distributions and the search for patterns of covariation, difference, or frequency clustering. Rather, the technique is effective in evaluating both the number and complexity of alternative paths leading to a desired outcome (Ragin, 2008). Fs/QCA’s approach to causality, referred to as ‘‘multiple conjunctural causation,’’ has three important implications. First, an outcome can be produced by multiple conditions. Second, there can be more than one combination of conditions leading to the outcome under investigation, a condition known as equifinality. Third, fs/QCA allows for outcomes to occur as a result of the presence (e.g., high levels of monitoring) or absence of a condition (e.g., absence of incentive alignment). Hence, the configurational approach relaxes some of the assumptions normally associated with many quantitative techniques, such as permanent causality, additivity, and causal symmetry. We use fs/QCA in an exploratory manner in the following sections to identify how governance mechanisms combine in unique and multiple ways to bring about high levels of foreign IPO performance for firms choosing to list on U.K. stock exchanges.

Sample and Initial Steps Our evaluation of governance configurations focuses exclusively on foreign issuers which are not listed on any exchange prior to their U.K. initial public offer. We used Thomson Financial’s Security Data Corporation (SDC) New Issues database to identify all foreign firms that made IPOs on U.K.

168

R. GREG BELL ET AL.

stock exchanges, both the London Stock Exchange and the AIM, between 2002 and 2006. Other studies of foreign IPOs have also relied upon SDC to identify samples of these unique firms (Bell et al., 2008; Bruner, Chaplinsky, & Ramchand, 2006). We classify ‘‘foreign’’ to be those firms incorporated and whose primary executive offices are located outside the U.K. Firms whose public offers were the result of mergers or acquisitions, spin-offs of publicly listed firms, and units, warrants, and rights offerings were eliminated from the sample. Following Bruner et al. (2006), we removed all new issues of foreign utility firms from consideration and financial service firms incorporated in tax havens because these firms often choose to incorporate in these countries for tax purposes alone. After identifying the sample of U.K. listed foreign IPOs, we referred to each offering firm’s prospectus to acquire the information needed to build the governance conditions. Our final sample includes 99 foreign IPOs listing on London’s Main Market and AIM exchanges from 2002 to 2006. We begin our analysis by identifying the various factors that influence foreign IPO performance and then we proceed to calibrate our raw data into crisp sets and fuzzy sets based on the identified factors (Ragin, 2008). To clarify the methodological procedure, a couple of definitions need to be put in place: first, calibration refers to the process of assigning the degree of membership into a given set (membership to a defined group). This process is used extensively in chemistry, astronomy, and physics (Ragin, 2008), but has yet found widespread use social sciences despite its usefulness. The transformation into sets is driven by theoretical grounds. Next, crisp sets are the simpler type of sets and a given case can either belong or not belong to the set. Fuzzy sets offer important qualitative content to the analysis in that researchers can, based on theoretical reasoning, decide whether a given case belongs ‘‘fully’’ into the membership set in which case the case receives the value of 1 or, when a given case has no membership into a set, a 0. In addition, researchers can identify a turning point or point of maximum ambiguity of 0.5 in which the case is in the intermediate stage between full and no membership, the point of inflexion. It is important to note that without careful consideration of the extant theoretical knowledge and empirical works, the assigned calibrations and transformation into set membership scores maybe critically flawed and could ultimately undermine the validity of the interpretation derived from the results. Therefore, in the following section, we describe how we captured each of the variables of interest in our study, and how we arrived at the breakpoints for set membership.

Corporate Governance and Configuration Research

169

Outcome Condition Our outcome condition, ‘‘high foreign IPO performance,’’ is based upon the IPO’s first day underpricing level (e.g., difference between offer and first day of trading prices). To specify what constitutes low underpricing (our measure for high foreign IPO performance), we consider Brau and Fawcett’s (2006) comprehensive survey that revealed that industry executives generally expect IPO underpricing to average approximately 10%. We then conducted separate interviews with two investment bankers, both knowledgeable in the international IPO process, at Citigroup and Morgan Stanley. Both representatives suggested that IPO underpricing levels have historically averaged between 10% and 13%. With these information in hand, we established our baseline, or floor level, of membership in the set of ‘‘high foreign IPO performance’’ to be those firms that achieved average (20% underpricing) or below average performance. In other words, firms that experienced underpricing levels above 20% were considered ‘‘fully out’’ of the set of high performing firms. We then considered firms to be ‘‘fully in’’ the set of high performing firms to be those that experienced10% underpricing or less. We defined the crossover point to be approximate the midpoint of about 15% underpricing.

Predictor Conditions Top Management Team Incentive Equity-based compensation is often used as a proxy for managerial incentives as it has become an important element of the compensation packages paid to top managers. The pay-scheme is a useful tool to align the interests of top managers and that of the shareholders. Following previous IPO research (Beatty & Zajac, 1994; Certo, Daily, Cannella, & Dalton, 2003) we built the Top Management Team Incentives set as a crisp-set by accounting for those firms whose entire top management team owned stock options. United Kingdom listed foreign IPO firms were coded fully in this set if stock options were owned by all of the top management team, and 0 otherwise. High external links among board member Previous IPO studies suggest that external affiliations of the firm’s board directors may play critical resource and strategy roles and improve performance (Filatotchev & Bishop, 2002). Following Higgins and Gulati

170

R. GREG BELL ET AL.

(2003), we assessed board interests by summing the employment and board membership affiliations of board insiders and independent members listed in the firms’ final prospectuses. We classified boards that had over fifteen external board affiliations per member, or the 90th percentile of our study sample, to be fully in the set of high external linkages. Those boards whose members averaged 8 external board affiliations, or the 50th percentile, are considered to be fully out of the set. The cross-over point was the midpoint between these two extremes. Separated CEO and Board Chair Researchers consider CEO duality to be a conflict of interest in corporate governance, (Coles & Hesterly, 2000; Daily & Dalton, 1994; Finkelstein & Hambrick, 1996). Board members may be less effective in governing firms when power is consolidated in the hands of a single person. Indeed, CEOs who serve as board chairs gain influence over board member nominations, board agendas, and compensation setting. Hence, foreign IPOs whose CEO and Board Chair are different individuals are likely to send positive signals to U.K. investors. We built a crisp set and defined firms who had Separated CEO-Board Chairs to be fully in the set, and those who did not (i.e., duality) to be fully out of the set. High retained ownership of blockholder Blockholders who sell a large percentage of their ownership at the time a firm goes public may be a sign of a firm with a short term orientation and limited growth prospects (Pound, 1988; Stiglitz, 1985). Hence, blockholders who retain a high percentage of the firm at IPO are likely to be a positive signal to external investors. We defined firms whose blockholders retained 70% of the firm to be fully in the set of high blockholder retained ownership, and those owned less than 50% of the firm to be fully out of the set. Low percentage of founders on board Foreign IPOs that have a high percentage of founding members on their board will likely experience considerable difficulties adjusting to the shortterm performance expectations of public market investors, while simultaneously adhering to the heightened transparency requirements imposed on public firms. Indeed, once foreign IPOs becomes public in the United Kingdom, they must confront different laws, regulations, and press scrutiny than what they are accustomed to in their home market. Foreign IPOs with a high percentage of founders may be burdened by their lack of experience

Corporate Governance and Configuration Research

171

and understanding of the U.K. governance processes and performance expectations of U.K. investors. As a result, U.K. investors may perceive that the governance oversight of the board may suffer, especially during the critical early periods when the firm is transitioning to public company status. Indeed, investors will likely prefer firms with boards are not composed of founding members. To capture the effects of board composition, we built a fuzzy set in which firms that have less than one tenth of its board members composed of founders are classified as fully in the set whereas those with more than one third of its board members composed of founders as fully out of the set. The breakpoint is the midpoint between these two extremes. Technology industry One of the most common ways to differentiate between firms in traditional and fast-growing sectors is to isolate whether the IPO operates in a hightech industry or not (Daily, Certo, & Dalton, 2005; Loughran & Ritter, 2004). We categorized all internet-related, electronics, and software firms as fully in the set of ‘‘Technology’’ foreign IPO firms.

Necessary Conditions The next step in operationalizing fs/QCA is to test whether any predictor or contextual condition is necessary for the high foreign IPO performance outcome. An argument for causal necessity can be supported when it can be demonstrated that instances of an outcome (dependent variable) constitute a subset of instances of a causal condition (independent variable). A consistency score of 1 indicates that the combination of causal conditions fulfills the criterion across all the cases. The more cases fail to meet the consistency criterion and the larger the distance from meeting the criterion, the further the consistency score will fall below 1. Conventionally, a fuzzy set variable, or crisp set variable, or a combination of two or more of these variables, is called ‘‘necessary’’ or ‘‘almost always necessary’’ if the consistency score meets or exceeds the threshold of 0.90 (Ragin, 2006). The measure to evaluate the relevance of a necessary condition is the coverage rate. Trivially necessary conditions would yield a coverage rate near 0 (Ragin, 2006). Table 1 reports the results pertaining to whether the variables we chose meet the standard for necessity upon the presence of our conditions (capitalized) and their negation (not capitalized). None of our conditions exceeds the 0.90 threshold for the high foreign IPO performance outcome

172

R. GREG BELL ET AL.

Table 1. Analysis of Necessary Conditions. Condition Contextual condition TECHNOLOGY INDUSTRY technology industry Corporate governance conditions TOP MANAGEMENT TEAM INCENTIVES top management team incentives HIGH EXTERNAL LINKAGES AMONG BOARD MEMBERS high external linkages among board members SEPARATED CEO AND BOARD CHAIR separated CEO and board Chair HIGH RETAINED OWNERSHIP OF BLOCKHOLDERS high retained ownership of blockholders LOW PERCENTAGE OF FOUNDERS ON BOARD low percentage of founders on board

Consistency

Coverage

0.37 0.62

0.68 0.59

0.68 0.31 0.34

0.61 0.64 0.7

0.69 0.78 0.21 0.32 0.72 0.28 0.75

0.62 0.63 0.59 0.67 0.64 0.69 0.63

Capitalized conditions indicate the presence of a condition. Lower case conditions indicate the negation of a condition. Calculations based on fsQCA2.0 software (www.fsqca.com).

condition. These outcomes indicate that no single condition can be regarded as necessary for high foreign IPO performance to occur. Among the conditions, separated CEO and Board Chairs assume the highest consistency value of 0.78. The absence of a high retained ownership of blockholders, and the absence of a low percentage of founders on board also assume high consistency values of 0.72 and 0.75, respectively.

Sufficient Conditions for High Foreign IPO Performance The next step involves an evaluation of the extent to which industry factors impact the combinations of governance conditions that lead to high performance (i.e., low underpricing) for foreign IPOs listing in the United Kingdom. Sufficiency of causal combinations is assessed through the use of fs/QCA’s Truth Table Algorithm. Fs/QCA’s truth table function generates a list of different combinations of governance and contextual conditions that are sufficient for a particular outcome to occur (Ragin, 2008). Initially, our analysis involves the six governance and industry conditions. This results in

Corporate Governance and Configuration Research

173

26 or 64 potential combinations of these causal conditions. Due to the size of the truth table, we do not reproduce it here but is available upon request. We observe 34 out of 64 logically possible governance and industry combinations are present for high performance to occur among foreign IPOs listing in the United Kingdom. Reducing the truth tables requires an evaluation of both the consistency levels across the configurations and establishing a frequency threshold that will be applied to the data listed in the ‘‘number’’ column. Rihoux and Ragin (2009) advise that consistency levels be a minimum of 80%. In addition, Ragin (2008) suggests that when establishing a frequency threshold ‘‘the issue is not which combinations have instances, but which combinations have enough instances to warrant conducting as assessment of the subset relationship’’ (p. 133). Using these guidelines, we adopt a consistency cut-off value of 80% and a minimum acceptable solution frequency of two cases to reduce the truth tables. Ragin (2008) advises that the selected configurations should capture at least 75–80% of the cases. Our adoption of a frequency threshold of two enables us to perform our analysis on 85% of the cases. The next step is to reduce the truth table algorithm rows into more simplified combinations. It is at this point that researchers should address limited diversity and logical remainders. Limited diversity refers to instances where the configuration of conditions and outcome across the empirical cases is not very diverse and therefore leaves a large portion of combinations ‘‘blank’’ (without empirical referents). Logical remainders are those logical configurations of conditions which are not empirically present in the dataset in relation to the presence or absence of the outcome of interest. Ragin (2008) suggests the use of intermediate solutions because logical remainders can be restricted to those that are the most plausible. The intermediate solution makes it possible for the researcher to choose three different options: (i) the presence of the conditions, (ii) the absence of the conditions, and (iii) the inclusion of either presence or absence of conditions. Our discussion throughout our theory development section makes clear that the presence of governance conditions is relevant for high foreign IPO performance. Therefore, we have included every governance condition as present. Reduction of the truth table reveals several useful statistics. The values reported in Table 2 as ‘‘solution consistency’’ and ‘‘solution coverage’’ offer a means of assessing the degree of fit of the solution with the fuzzy-set scores for each condition. Consistency indicates the degree to which the subset relationship holds for sufficiency. The overall coverage refers to the joint

174

Table 2.

R. GREG BELL ET AL.

High Performance Configurations for Foreign IPOs listing on London stock exchanges (2002–2006). Solutions 1

2

3

4

0.95 0.07 0.06

0.85 0.03 0.03

0.85 0.05 0.05

Industry Technology Industry Strategic, Incentive, and Monitoring Governance Conditions Top management team incentives High external linkages among board members CEO and Board Chair roles are separated High retained ownership of blockholders Low percentage of founders on board Consistency Raw Coverage Unique Coverage

0.81 0.11 0.11

Overall Solution Consistency Overall Solution Coverage

0.85 0.27

Full circles indicate the presence of a condition. Crossed-out circles indicate the absence of a condition. Large circles indicate conditions that are part of the parsimonious and intermediate solutions/ Small circles indicate conditions that are only part of the intermediate solutions.

importance of all causal paths (Schneider et al., 2010). Unique coverage is useful because it illustrates the relative weight of each path in leading to high foreign IPO performance by measuring the degree of empirical relevance of a certain cause or causal combination to explain the outcome. Unique coverage of causal conditions is similar to R-square calculations in regression analysis (Fiss, 2009). The results in Table 2 show four combinations leading to high performance for foreign IPO firms listing in the United Kingdom. Full circles indicate the presence of a condition, while crossed-out circles indicate the absence of a condition. Large circles designate conditions that are part of both parsimonious and the intermediate solutions, whereas smaller circles only occur in intermediates solutions. The unique coverage ranges from 0.03 to 0.11. Therefore, each of these four combinations provides a unique contribution to the explanation of high foreign IPO performance. The first two solutions in Table 2 are for foreign IPOs which compete in technology-related industries and solutions 3 and 4 refer to nontechnology related solutions. Solution 1 indicates that high-tech foreign IPOs achieve

Corporate Governance and Configuration Research

175

high performance when all of the firm’s top managers own stock options, along with a separated CEO and board chair, low presence of founders on their board of directors, and boards with few external director interlocks. The results show that combination of these factors lead to high IPO performance (i.e., low underpricing). Blockholders are not relevant to solution 1 while solution 2 shows that technology foreign IPOs that have a relatively high percentage of founders on its board must also have a high percentage of retained ownership among block holders, a separated CEO and board chair, a high percentage of board affiliations among board members, and also all of the top managers with stock options. Indeed, solution 2 indicates that high-tech foreign IPOs listing in the United Kingdom must introduce considerable governance practices in order to achieve high IPO performance. Interestingly, a key difference between solutions 1 and 2 stems from the degree to which founders comprise a technology foreign IPO’s board. Solution 1 confirms that high-tech firms that have a low percentage of founders on their boards are better able to achieve higher levels of performance. In addition, a comparison between solutions 1 and 2 indicates that in order for technology related foreign IPOs listing in the United Kingdom to achieve high IPO performance, they must issue stock options to all of the firm’s top managers and have dual leadership. Solutions 3 and 4 illustrate governance configurations that lead to high performance for foreign IPOs competing in nontechnology related industries. Interestingly, both of these configurations confirm that nontechnology foreign IPOs can achieve comparably high levels of performance (low underpricing) with fewer corporate governance practices in place than technology industry foreign IPOs (solutions 1 and 2). The comparison between solutions 3 and 4 in Table 2 shows that low percentage of founders on the board is necessary for nontechnology foreign IPOs to achieve success upon listing on U.K. exchanges. Also, duality does not appear to be an issue for nontechnology foreign IPOs to the extent it is for technology foreign IPOs. Finally, results show that blockholder retained ownership and stock option incentives paid to top managers could substitute for one another.

Modeling the Inverse of the Outcome When examining the configurational logic, it is important to note that the reverse of a given configuration will not necessarily lead to the opposite outcome. Therefore, we explore what configurations might consistently lead

176

R. GREG BELL ET AL.

to the absence of high performance. The ability to conduct such an analysis is an important advantage of QCA, because it relates to causal asymmetry (Ragin, 2008), in that the factors that lead to the presence of an outcome in question may in fact be different from those conditions that lead to the absence of the outcome. We examined the absence of high foreign IPO performance and the presence of poor foreign IPO performance. This analysis identified only one configuration among high tech foreign IPOs which indicated that these firms would suffer poor performance (high underpricing) if they had only one governance mechanism, stock options owned by all of the firm’s top managers. Similarly, nontechnology related foreign IPOs would suffer poor performance if they went public with four governance mechanisms, stock options to top managers, a board that was highly connected to other boards, and a CEO who also served as the board chairperson.

DISCUSSION AND CONCLUSION In this chapter, our primary purpose was to demonstrate, first hand, some of the advantages in adopting the configurational logic when conducting corporate governance research. Configurational logic allows researchers to explore complex interdependencies between firm-level governance factors, industry parameters, and firm performance. Extant corporate governance research and IPO studies in particular assume that governance factors act independently with respect to their effects on performance. Indeed, variables related to monitoring and incentive alignment mechanisms are seen as separate causes of IPO success and that each governance practice independently influence the level of equity resources firms will receive. This is a rather simplistic and problematic notion as governance practices are interpreted as having linear and additive effects and will produce the same effect regardless of the level of adoption, presence other governance practices, or even the industry conditions surrounding the firm leading up to the IPO. The inconsistency of evidence across studies suggests that the performance implications of governance factors associated with firms leading up to their first equity offers is a significantly more complex phenomenon than previously understood. The collective evidence of Table 2 demonstrates that foreign IPOs that list on London exchanges competing in hi-tech industries must implement a greater range of signals of ‘‘good’’ governance mechanisms to achieve

Corporate Governance and Configuration Research

177

comparable levels of high performance than foreign IPOs competing in low tech industries. Our data show that there is an element of substitution between governance practices in this subsample of firms (e.g., firms may not need a high level of retained ownership by blockholders if they have a low percentage of founders on the board), but the extent of this substitution is rather limited compared to firms from mature industries. In contrast, firms from mature industries need to use rather limited governance bundles to achieve a high level of performance. Interestingly, solution 2 in Table 2 suggests that firms from hi-tech sectors may benefit from high external linkages among board members, or so-called external ‘‘board interlocks.’’ The resource provision role of directors is generally based on the work by Pfeffer and Salancik (1978, p. 163), which states that ‘‘when an organization appoints an individual to a board, it expects the individual will come to support the organization, will concern himself with its problems, will variably present it to others and will try to aid it.’’ In fulfilling their strategic role, scholars point not only to the resources, knowledge, and experiences that both board insiders and independent directors can bring to firms, but they also account for the external relationships that board members establish with constituent actors. Recent studies show that the external relationships that directors have outside of the board are important predictors of social influence (Stevenson & Radin, 2009). Indeed, the external ties of board members (i.e., social capital) are an important resource to the firm (Stevenson & Radin, 2009). Certainly, both scholars and practitioners alike consider affiliations of board insiders and independent directors to be a vital means in which to develop organizational legitimacy (Higgins & Gulati, 2006; Mizruchi, 1996; DiMaggio & Powell, 1983). Our research, however, indicates that board ties do play a critical but only in combination with other governance factors. More significantly, they seem to affect performance only in a specific industry context. Taken together, our findings in this chapter indicate that the effectiveness and efficiency of governance practices are not context free, and they should be considered in conjunction with the firm’s external environment, such as its industry or economic sector. More generally, our efforts represent an important step in realizing the ‘‘contextualization’’ of corporate governance research (Aguilera et al., 2008) through the use of QCA, a powerful tool that helps explore the configurational logic underlying our approach. We believe that our findings from foreign IPOs listed in London can be generalized to other contexts where firms, investors and other stakeholders face different types of information asymmetries and associated costs. By applying configurational thinking to other contexts, researchers may discover new

178

R. GREG BELL ET AL.

patterned variations of corporate governance and suggest solutions to governance problems that affect companies around the world.

REFERENCES Aguilera, R. V., Desdender, K., & Kabbach de Castro, L. R. (2012). A Bundle perspective on comparative corporate governance. In T. Clarke & D. Branson (Eds.), SAGE handbook of corporate governance (pp. 379–405). New York, NY: Sage. Aguilera, R. V., Filatotchev, I., Gospel, H., & Jackson, G. (2008). An organizational approach to comparative corporate governance: Costs, contingencies, and complementarities. Organization Science, 19(3), 475–492. Aguilera, R. V., & Jackson, G. (2003). The cross-national diversity of corporate governance: Dimensions and determinants. Academy of Management Review, 28(3), 447–465. Aguilera, R. V., & Jackson, G. (2010). Comparative and international corporate governance. Annals of the Academy of Management, 4(1), 485–556. Aoki, M. (2001). Toward a comparative institutional analysis. Cambridge, MA: MIT Press. Beatty, R. P., & Zajac, E. J. (1994). Managerial incentives, monitoring and risk bearing: A study of executive compensation, ownership, and board structure in initial public offerings. Administrative Science Quarterly, 39(2), 313–335. Bell, R. G., Filatotchev, I., & Rasheed, A. (2012). The liability of foreignness in capital markets: Sources and remedies. Journal of International Business Studies, 43(2), 107–122. Bell, R. G., Moore, C., & Al-Shammari, H. (2008), Country of origin and foreign IPO legitimacy: Understanding the role of geographic scope and insider ownership. Entrepreneurship: Theory & Practice, 32(1), 185–202. Bell, R. G., Moore, C. B., & Filatotchev, I. (2012). Strategic and institutional effects on foreign IPO performance: Examining the impact of country of origin, corporate governance, and host country effects. Journal of Business Venturing, 27(2), 197–216. Brau, J., & Fawcett, S. (2006). Initial public offerings: An analysis of theory and practice. The Journal of Finance, 61(1), 399–436. Bruner, R., Chaplinsky, S., & Ramchand, L. (2006). Coming to America: IPOs from emerging market issuers. Emerging Markets Review, 7(3), 191–212. Carpenter, M., Pollock, T., & Leary, M. (2003). Testing a model of reasoned risk-taking: Governance, the experience of principals and agents, and global strategy in hightechnology IPO firms. Strategic Management Journal, 24(9), 803–820. Certo, S., Daily, C., Cannella, A., Jr., & Dalton, D. (2003). Giving money to get money: How CEO stock options and CEO equity enhance IPO valuations. Academy of Management Journal, 46(5), 643–653. Certo, S. T., Daily, C. M., & Dalton, D. R. (2001). Signaling firm value through board structure: An investigation of initial public offerings. Entrepreneurship Theory and Practice, 26(2), 33–50. Coles, J., & Hesterly, W. (2000). Independence of the chairman and board composition: Firm choices and shareholder value. Journal of Management, 26(2), 195–214. Daily, C., & Dalton, D. (1994). Corporate governance and the bankrupt firm: An empirical assessment. Strategic Management Journal, 15(8), 643–654.

Corporate Governance and Configuration Research

179

Daily, C., Certo, S., & Dalton, D. (2005). Investment bankers and IPO pricing: Does prospectus information matter. Journal of Business Venturing, 20(1), 93–112. Desdender, K., Aguilera, R. V., Crespi-Cladera, R., & Garcia-Cestona, M. A. (2011), Board characteristics and audit fees: Why ownership structure matters. Working Paper No. 09-0107. College of Business, University of Illinois, Urbana-Champaign. DiMaggio, P. J., & Powell, W. W. (1983). The iron cage revisited: Institutional isomorphism and collective rationality in organizational fields. American Sociological Review, 48, 147–160. Emirbayer, M., & Mische, A. (1998). What is agency. American Journal of Sociology, 103(4), 962–1023. Filatotchev, I., & Bishop, K. (2002). Board composition, share ownership, and ‘underpricing’ of U.K. IPO firms. Strategic Management Journal, 23(10), 941–955. Filatotchev, I., Wright, M., & Toms, S. (2006). The firm’s strategic dynamics and corporate governance life-cycle. International Journal of Managerial Finance, 2(4), 256–279. Finkelstein, S., & Hambrick, D. (1996). Strategic leadership: Top executives and their effects on organizations. New York, NY: West Publishing Company. Fiss, P. C. (2009). Case studies and the configurational analysis of organizational phenomena. In C. Ragin & D. Byrne (Eds.), Handbook of Case Study Methods (pp. 424–440). Thousand Oaks, CA: Sage. Fiss, P. C. (2007). A set-theoretic approach to organizational configurations. Academy of Management Review, 32(4), 1180–1198. Fiss, P. C. (2011). Building better causal theories: a fuzzy set approach to typologies in organization research. Academy of Management Journal, 54(2), 393–420. Florin, J. (2005). Is venture capital worth it? Effects on firm performance and founder returns. Journal of Business Venturing, 20(1), 113–136. Globerman, S., Peng, M. W., & Shapiro, D. (2011). Corporate governance and Asian companies. Asia Pacific Journal of Management, 28(1), 1–14. Grandori, A., & Furnari, S. (2008). A chemistry of organization: Combinatory analysis and design. Organization Studies, 29, 459–485. Greckhamer, T., Misangyi, V. F., Elms, H., & Lacey, R. (2008). Using qualitative comparative analysis in strategic management research: An examination of combinations of industry, corporate, and business-unit effects. Organizational Research Methods, 11(4), 695–726. Higgins, M. C., & Gulati, R. (2003). Getting off to a good start: The effects of upper echelon affiliations on underwriter prestige. Organization Science, 14(3), 244–263. Higgins, M. C., & Gulati, R. (2006). Stacking the deck: The effect of upper echelon affiliations for entrepreneurial firms. Strategic Management Journal, 27, 1–26. Jackson, G. (2005). Employee representation in the board compared: A fuzzy sets analysis of corporate governance, unionism, and political institutions. Industrielle Beziehungen, 12(3), 1–28. Kogut, B., MacDuffie, J. P., & Ragin, C. C. (2004). Prototypes and strategy: Assigning causal credit using fuzzy sets. European Management Review, 1(2), 114–131. Loughran, T., & Ritter, J. R. (2004). Why has IPO underpricing changed over time. Financial Management, 33(3), 5–37. Mahoney, J., & Goertz, G. (2006). A tale of two cultures: Contrasting qualitative and quantitative research. Political Analysis, 14(3), 227–249. Mizruchi, M. S. (1996). What do interlocks do? An analysis, critique, and assessment of research on interlocking directorates. Annual Review of Sociology, 22, 271–298.

180

R. GREG BELL ET AL.

Mudambi, R., & Treichel, M. Z. (2005). Cash crisis in newly public Internet-based firms: An empirical analysis. Journal of Business Venturing, 20(4), 543–571. Oliver, C. (1991). Strategic responses to institutional processes. Academy of Management Review, 16(1), 145–179. Oliver, C. (1992). The antecedents of deinstitutionalization. Organization Studies, 13(4), 563–588. Pajunen, K. (2008). Institutions and inflows of foreign direct investment: A fuzzy-set analysis. Journal of International Business Studies, 39(4), 652–669. Pfeffer, J., & Salancik, G. (1978). The external control of organizations: A resource dependency perspective. New York, NY: Harper and Row. Pound, J. (1988). Proxy contests and the efficiency of shareholder oversight. Journal of Financial Economics, 20, 237–265. Ragin, C. C. (2006). Set relations in social research: Evaluating their consistency and courage. Political Analysis, 14, 291–310. Ragin, C. C. (2008). Redesigning social inquiry: Fuzzy sets and beyond. Chicago, IL: University of Chicago Press. Rediker, K., & Seth, A. (1995). Boards of directors and substitution effects of alternative governance mechanisms. Strategic Management Journal, 16(2), 85–99. Rihoux, B., & Ragin, C. C. (2009). Configurational comparative methods: Qualitative Comparative Analysis (QCA) and related techniques. Thousand Oaks, CA: Sage. Sanders, G. W., & Boivie, S. (2004). Sorting things out: valuation of new firms in uncertain markets. Strategic Management Journal, 25(2), 167–186. Schneider, M. R., Schulze-Bentrop, C., & Paunescu, M. (2010). Mapping the institutional capital of high-tech firms: A fuzzy-set analysis of capitalist variety and export performance. Journal of International Business Studies, 41(2), 246–266. Shleifer, A., & Vishny, R. (1997). A survey of corporate governance. Journal of Finance, 52(2), 737–783. Stevenson, W. B., & Radin, R. F. (2009). Social capital and social influence on the board of directors. Journal of Management Studies, 46(1), 16–44. Stiglitz, J. (1985). Credit markets and the control of capital. Journal of Money, Credit and Banking, 17, 133–152.

CHAPTER 8 CORPORATE SOCIAL RESPONSIBILITY: A MULTILEVEL EXPLANATION OF WHY MANAGERS DO GOOD Donal Crilly ABSTRACT This chapter explores the integrative effects of individual psychology and social context in explaining why managers would behave in socially responsible ways. To identify how factors at different levels of analysis combine to shape attitudes toward social responsibility, I apply fuzzy-set qualitative comparative analysis (fsQCA) to survey and archival data from 335 managers of overseas subsidiaries of three Dutch corporations. Attention to the simultaneous effects of individual psychological factors, the organizational context, and the broader social context offers a configurational perspective on the micro and macrofoundations of social responsibility. Keywords: Social responsibility; cognition; affect; norms; configurations

Configurational Theory and Methods in Organizational Research Research in the Sociology of Organizations, Volume 38, 181–204 Copyright r 2013 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0733-558X/doi:10.1108/S0733-558X(2013)0000038012

181

182

DONAL CRILLY

INTRODUCTION Some managers allocate resources to social initiatives that do not directly promote economic performance, whereas others eschew discretionary social engagement altogether. Why would managers perceive the role of business in society in such different ways? Most answers to this question center on a sole level of analysis. Microscholars attend to individual psychology, whereas macroscholars attend to features of the organizational context or society. This separation comes at a cost of overlooking the complex interplay between individual psychology and social context. Managers situate their perceptions within a social context, and attention to both psychology and context can produce richer insights than attention to either by itself. The premise of the present chapter is that the integrative effects of individual psychology and social context matter in explaining managers’ decisions when concern for social welfare is incompatible with shareholder wealth maximization. This approach builds on existing multilevel approaches to social responsibility (Aguilera, Rupp, Williams, & Ganapthi, 2007). However, it goes one step further by recognizing that the influence of effects at any single level of analysis could depend on effects at other levels. Specifically, the influence of individual psychology on how managers perceive their responsibility toward society might depend on organizational incentives and social sanctions. Much evidence linking managerial psychology to socially responsible behavior is inconclusive, and one explanation for this is the neglect of the social context in which managers find themselves. Attention to the integrative effects of psychology and context requires an appropriate empirical approach. Although regression techniques such as hierarchical linear modeling (HLM) shed light on effects at multiple levels of analysis, they are less appropriate for understanding the combined influence of effects at different levels of analysis. Configurational comparative methods potentially permit greater insight. Rather than disaggregate variance to identify the contribution of effects at each individual level of analysis, these methods are explicitly focused on the interplay between factors at different levels of analysis. To illustrate the application of configurational thinking in explaining socially responsible behavior, I apply fuzzy-set qualitative comparative analysis (fsQCA) to survey and archival data from a random sample of 335 overseas-based managers of three Dutch corporations. Attention to the simultaneous effects of individual psychology, the organizational context, and the broader societal context offers a configurational perspective on the micro and macrofoundations of social responsibility.

Multilevel Explanation of Why Managers Do Good

183

The chapter is structured as follows. I first provide an overview of the challenges of studying the foundations of socially responsible behavior. Next, I compare and contrast two potential methods to address these challenges: HLM, and fsQCA. I then turn to an empirical analysis to illustrate the application of fsQCA, before discussing the strengths and weaknesses of this particular approach for developing multilevel theory in social responsibility research.

THEORETICAL AND METHODOLOGICAL CHALLENGES IN STUDYING SOCIAL RESPONSIBILITY Despite the longstanding recognition that managers’ responses to ethical and moral dilemmas depend on both their individual characteristics and the contexts in which they interact (Trevin˜o, 1986), many theories developed to explain social responsibility focus on one level of analysis. Further, most theories ignore the possibility of competing explanations for the same behaviors. Developing theory in this domain hence faces two challenges relating to complex causality: the interaction of variables at distinct levels of analysis, and the existence of competing explanations for the same phenomenon. Responding to these challenges requires an appropriate empirical approach.

Theoretical Challenges: The Need for a Multilevel, Integrative Approach Thus far, most research has privileged explanations at a single level of analysis. Much work examines the impact of social forces on individuals. Scholars in the sociology of knowledge tradition emphasize the embeddedness of the individual in social structures (Garnier, 1972, 1973) and highlight cultural and institutional influences on beliefs about the economic system (Somers & Block, 2005). For example, national regulations and educational systems sway managers’ definitions of their social responsibilities (Matten & Moon, 2008). Similarly, organizational theorists emphasize the influence of the organizational context in prompting understandings of social responsibility. However, explanations relying exclusively on social structures and norms do not adequately explain variation within organizations (Simons & Ingram, 1997) or across organizations (Crilly & Sloan, 2012).

184

DONAL CRILLY

In contrast, psychologists center their explanations on mental processes that account for variation across individuals. Moral psychologists have paid attention to the role of cognition (reasoning) in shaping responses to social dilemmas (Kohlberg, 1984). How managers weigh concerns such as costs, benefits, legality, and morality can explain different stances on social responsibility. Contemporary psychologists also emphasize the role of intuition and affect on how individuals make decisions when confronted with social dilemmas (Haidt, 2001; Sonenshein, 2007). However, any explanation focused exclusively on the individual level of analysis is likely to be incomplete; managers take decisions within an organizational and social context (Elsbach, Barr, & Hargadon, 2005). Given that social responsibility is an inherently multilevel issue, involving individuals, organizations, and societies, it seems fruitful to bring together insights from multiple levels of analysis (Aguilera et al., 2007). A truly integrative explanation goes beyond merely identifying effects at different levels of analysis to consider how influences at different levels combine to shape orientations toward social responsibility. Integrating insights from different levels of analysis can help to resolve some inconsistent findings from prior research linking psychological variables to social responsibility. The relationship between affect and behavior serves as a case in point. Affect can predict prosocial decisions and behavior (Crilly, Schneider, & Zollo, 2008). Positive affect – the tendency to experience pleasant feelings – is linked to cooperative (Crilly et al., 2008) and prosocial behavior. In contrast, negative affect – the tendency to experience unpleasant feelings – is linked to egoism and, on average, less concern for others’ well-being (Mor & Winquist, 2002). However, these relationships do not hold in all cases. For example, negative affect can also predict the propensity to help other people and to act altruistically (Cialdini, Darby, & Vincent, 1973). One explanation for this inconsistency lies in the social context facing the decision-maker. Negative-affect individuals appear sensitive to rewards (Weyant, 1978) and arguably may behave socially responsibly when their organization rewards socially responsible behavior or when norms sanction irresponsible behavior. An integrative approach has also to acknowledge the co-occurrence of effects at the same level of analysis. Though psychologists increasingly eschew predominantly rational explanations of prosocial decisions and behaviors (Haidt, 2001), cognition – how individuals think and reason – does play a role in how individuals respond to social dilemmas (Kohlberg, 1984). Cognition and affect are not only independently important for individuals’ responses, they are important together. Different forms of

Multilevel Explanation of Why Managers Do Good

185

cognition and affect cooccur (Chau, Morris, & Ingram, 2009), however competing theoretical perspectives make the direction of any causal relationship between cognition and affect difficult to specify a priori. Appraisal theory proposes that cognitive processes influence the nature and strength of affective reactions (Lazarus, 1991). According to this theory, individuals assess events and situations, and their initial assessment provokes an affective reaction. In contrast, affective primacy theory proposes that affective reactions can form the basis of judgments (Zajonc, 1980). According to this theory, the initial reaction to an event is affective (e.g., people react with disgust, shame, or happiness), and reasoning follows as a post-hoc justification of the initial reaction (Sonenshein, 2007). Understanding the interplay between individual psychology and social forces holds the promise to resolve some of these inconsistent relationships between psychological antecedents and socially responsible behavior. Though experimental research has identified many psychological mechanisms, managers make decisions outside of the laboratory, and the influence of social context is crucial. At the macrolevel, managers are exposed to social norms and sanctions. At the meso level, senior executives put in place control systems and incentives, and, in doing so, they potentially influence behavior and, even, human nature (Ferraro, Pfeffer, & Sutton, 2005). This complexity is reinforced by the potential reciprocal influence between social contexts because individuals can shape the contexts in which they interact (Bandura, 1978; Chatman, 1989). A second challenge in identifying why managers respond to social dilemmas in different ways concerns the existence of competing explanations for social responsibility. Just as affect, cognition and social influence combine to influence behavior, it is unlikely that any individual theory can adequately explain all instances of socially responsible behavior. Motivations for adopting social responsibility policy differ (Bansal & Roth, 2000). Rather than construct a grand theory of social responsibility around a single variable, researchers might seek nuanced, mid-range explanations that specify the kinds of contexts in which competing explanations are likely to be valid.

Addressing the Methodological Challenges The theoretical challenges described above have implications for research design in the domain of social responsibility. There is a need to move away from thinking about net effects, and move toward the integrative effects of

186

DONAL CRILLY

variables at the same and different levels of analysis. Further, any design should leave open the possibility of multiple competing explanations. Microscholars have embraced multilevel regression techniques, such as HLM, as a solution to these challenges. These techniques offer advantages overordinary least squares (OLS) when introducing contextual variables into microlevel theories. Multilevel data typically violate the assumption of OLS that observations are independent of one another. HLM makes an adjustment for the nonindependence of error terms. In so doing, it accounts for the possibility that individuals are nested in groups and societies. HLM involves two processes.1 In the first stage, the relationships among variables at different levels of analysis are analyzed. Subsequently, there is an analysis of withingroup slopes as an indicator of moderation and variance in within-group intercepts as an indicator of mediation. HLM is thus suitable for investigating cross-level interactions (Davison, Kwak, Seo, & Choi, 2002). Hence, in the present context, HLM might be useful for assessing whether the influence of managerial psychology on socially responsible behavior differs according to characteristics of the firms or societies in which managers are located. Despite the advantages of HLM over conventional regression, its features limit its application in the present context. Like other forms of regression, HLM relies on a number of assumptions about the data for analysis to be valid. Error terms at higher levels of analysis are assumed to be independent of predictors at the same level and to adhere to a multivariate normal distribution (Griffin, 1997). Similarly, multicollinearity causes estimation problems. Multicollinearity is probable when assessing multiple psychological factors that influence social responsibility behaviors because affect and cognition are likely to be highly correlated (Isen, 2001). Additionally, statistical power is likely to be a concern when assessing managers in multiple organizations and working in multiple countries. As each new level of analysis is added, the sample size necessary in order to have the statistical power to explain variation within and across groups increases quickly (Raudenbush & Liu, 2000; Snijders & Bosker, 2000). Apart from the data requirements to conduct a valid HLM analysis, a more fundamental concern is whether HLM is really appropriate for addressing the question at hand. In practice, the aim of much work using HLM has been to partition variance between levels of analysis, and this work has employed additive models to identify the unique contribution of each level (Lau & Nie, 2008). This aim focuses attention on net effects rather than the kinds of interactive effects that would provide insight into the causal complexity inherent in an adequate explanation of socially responsible behavior.

Multilevel Explanation of Why Managers Do Good

187

Configurational methods, in particular set-theoretic methods such as qualitative comparative analysis (QCA), represent a distinct set of techniques for multilevel analysis (Lacey & Fiss, 2009). These techniques differ in some important ways from regression-based approaches such as HLM and make them amenable to identifying why managers might perceive the role of business in society in different ways. Crucially, whereas multicollinearity is a problem in HLM, from the perspective of configurational thinking, collinearity can provide useful information. Configurational approaches embrace the idea that causes combine to bring about outcomes. In the present context, the cooccurrence of variables such as affect and cognition is expected and may be inherent in any causal explanation. Similarly, how managers’ psychology influences their understanding of social responsibility could depend on features of their firms and the societies in which they work. Hence, rather than control for effects at other levels to isolate the net effects at a given level of analysis, QCA researchers are explicitly interested in the combined effects. Such configurational methods are especially apt to analyze the interplay between factors at the same and different levels of analysis. Specifically, QCA uses Boolean algebra to identify combinations of factors associated with an outcome. As we will see subsequently, Boolean expressions are not linear. Further, Boolean expressions are not unifinal, permitting the identification of competing causal paths linked to the outcome of interest. This property of QCA is important because, as the above discussion on the theoretical challenges suggests, the search for a single, all-encompassing theory of social responsibility is likely to prove futile. Rather than prove or disprove prior explanations of socially responsible behavior, QCA holds the potential to shed light on the circumstances under which competing explanations for socially responsible behavior are likely to hold.

EMPIRICAL APPLICATION As discussed above, a central challenge in conducting research into social responsibility is how the effect of individual psychology (affect, cognition) might depend on the organizational and societal context facing managers. In the present study, I use fsQCA to understand how individual, firm-level, and societal factors combine to influence why managers would behave in socially responsible ways at work. Individual managers are nested within business units that follow distinct strategies and implement distinct policies, and within societies with distinct social norms about the role of business in

188

DONAL CRILLY

society. Whereas crisp-set comparative analysis (csQCA) identifies binary conditions associated with an outcome, fsQCA recognizes that membership in a condition is not always binary. Applying fsQCA rather than csQCA is appropriate because some of the conditions included in the analysis are not easily reduced to dichotomous measures. Specifically, I apply fsQCA to survey and archival data from a sample of 335 managers of overseas subsidiaries of three multinational firms headquartered in the Netherlands. Managers at this level interpret situations and allocate resources, and hence the survey data focus on individual level measures. These managers situate their personal perceptions and actions within an organizational and societal context, and the archival data include measures of the organizational context facing each manager and societal context. Conditions at these three levels of analysis allow investigation of how the business unit and societal context combine with individual psychology to influence these orientations. To capture individual psychology, I attend to managers’ affect and reasoning. Both feature prominently in psychological explanations of ethical behaviors (Crilly et al., 2008; Kohlberg, 1984). As affect and reasoning reinforce each other, they are central to a configurational investigation of socially responsible behavior. By themselves, however, they are insufficient because context influences the effect of psychology on behavior. To capture the key dimensions of the organizational context, I attend to performance appraisal (whether adherence to ethical principles is taken account of in annual performance appraisals) and resource availability (whether the business unit allocates resources to CSR initiatives). Performance appraisal incentivizes behavior, potentially influencing affect and reasoning styles (Stone & Ziebart, 1995). Resource availability shapes the reasoning that people bring to bear in making difficult choices (Zauberman & Lynch, 2005). At the societal level, I attend to the collectivism of the host country. Collectivism provides a normative foundation for CSR (Crilly, 2011).

Sample Selection The sample is composed of 335 randomly sampled managers in the overseas subsidiaries of three Dutch multinational firms. The managers represent 71 different countries, broadly reflecting the geographic dispersion of the three firms. The average age is 41 years. The focus on managers of overseas business units of three multinational Dutch firms embeds variance in business unit policy and host country characteristics into the sample whilst

Multilevel Explanation of Why Managers Do Good

189

controlling for home country characteristics. Respondents are middle managers and junior executives with some discretion over local resource allocation.

Calibration of Conditions The first step in performing fsQCA is to calibrate set membership. To do so, consistent with recent large-N studies (Fiss, 2011), I employ the direct method (Ragin, 2008) with the exception of performance assessment which is a dichotomous condition. The direct method calculates set membership on a range of 0–1 based on three anchors: the point of full inclusion in the set, the point of full exclusion, and the crossover point at which there is greatest ambiguity surrounding the observation’s member in the set. Three conditions involve factor scores based on scales used in prior research. To calibrate each of these conditions, I used regression factor scores. The regression method of obtaining factor scores produces scores with a mean of 0 and variance equivalent to the square of the correlation between the factor estimates and the true factor values. In calibrating these three conditions, I chose the mean (0) as the cross-over point and the lower- and upperquartiles as the points where observations were entirely nonmembers (0) or complete members (1) of the relevant set.

Socially Responsible Behavior The outcome of interest is how managers would likely respond to pressures for social responsibility. To measure managers’ behavioral intentions, I use four vignettes from the Reidenbach and Robin (1990) Multidimensional Ethics Scale (Table 1). Managers were asked how likely they were on a fourlevel Likert scale to behave in the same manner as the protagonist. Further, managers were also asked how likely their local peers were to behave in the same manner as the protagonist. Individuals frequently describe their actual behavior when asked to characterize the behavior of their peers (Watkins & Cheung, 1995). Replicating the analysis with this alternative outcome condition produced substantially similar results, providing confidence that social desirability bias did not drive the results. Factor analysis revealed two dimensions underlying the scenarios: one factor loaded on two scenarios that involved the protagonist causing potential harm (e.g., by cutting costs and avoiding onerous product quality

190

DONAL CRILLY

Table 1. Vignettes Used to Measure Respondents’ Social Responsibility Orientation. Scenario

Action

1. A company has just introduced a highly successful new kitchen appliance. The sales manager, who is paid partly on a commission basis, discovers that there has been insufficient product testing to meet government guidelines. These tests so far indicate no likelihood of a safety problem.

The sales manager continues to promote and sell the product.

2. A large manufacturer is considering the outsourcing of production of their main product to a supplier in a low cost third world country. The move will significantly improve the cost structure of the company due to lower labor costs. It will also increase the risks of violating the company’s socially advanced principles for their labor practices, as it will be difficult to monitor the work conditions at the supplier’s plant.

The CEO decides to proceed with the outsourcing arrangement.

3. The CEO of a small pharmaceutical company specializing in developing medicine for infectious diseases has been told by the head of R&D that the lab has just found, by accident, a treatment that may cure a serious debilitating illness that affects millions of people in Africa. Developing and distributing this drug will prove extremely costly, however. Given the increasingly competitive business environment, the company is under growing pressure to improve financial performance.

The CEO gives the go ahead to develop and distribute the drug to African countries at a small fraction of the full price.

4. The plant manager of a precision instruments company is concerned about increasing productivity and cost control. The HR manager has just proposed a program which would pay workers for spending one day a month providing community service (e.g., helping handicapped children, visits of elderly people).

The plant manager agrees with plan.

191

Multilevel Explanation of Why Managers Do Good

Table 2.

Overview of Explanatory Conditions.

Level of Analysis

Explanatory Condition

Individual

Reasoning Positive affect

Organizational

Societal

Resource constraints Ethical component in performance assessment Collectivism

Operationalization

Six-item scale based on World Values Survey (Inglehart & Baker, 2000) Positive and negative affect schedule (Watson et al., 1988) Single survey item Data sourced from interviews with senior executives in the firms, and confirmed by single survey item National measure of collectivism (Hofstede, 1984)

checks), and two scenarios involved the protagonist doing good beyond any legal requirement and without any promise of economic return (e.g., facilitating product access to those without adequate means to pay, and supporting a volunteering scheme in the local community). To operationalize socially responsible behavior in the present analysis, I used factor scores representing the intention to do good beyond any legal requirement or promise of economic return. This represents an expansive definition of the role of business in society. I calibrated membership in the manner described above; full membership (1) was allocated to respondents in the upper quartile. Above this threshold, most respondents gave the highest possible score to both the two ‘‘do good’’ scenarios. Nonmembership (0) was allocated to respondents in the lower quartile who were the most likely to eschew an expansive role for the firm in society. The cross-over point was the mean across all respondents. I include five explanatory conditions, on overview of which is provided in Table 2. At the individual level of analysis, I draw on insights from moral psychology to take account of managers’ affect and reasoning. I assess two conditions that reflect how the local business unit rewards and sanctions socially responsible behavior, and one condition at the societal level that reflects norms relevant to social responsibility in the host country.

Positive Affect Affect was measured using the Positive and Negative Affect Schedule (PANAS) scale (Watson, Clark, & Tellegen, 1988). As I was interested to

192

DONAL CRILLY

understand the influence of managers’ dispositional affect rather than temporary mood, respondents indicated the extent to which they had experienced 28 sentiments over the prior six weeks on a six-point scale. I applied factor analysis and varimax rotation to examine the underlying dimensions. The items loading highly on the first factor were consistent with those for positive affect (e.g., happy, cheerful, excited, and enthusiastic) (Watson & Clark, 1994). Factor scores were calibrated using the direct method described above. Full membership (1) was allocated to respondents in the upper quartile, who tended to score themselves at the highpoint of the scale on at least three positive-affect items. Nonmembership (0) was allocated to respondents in the lower quartile who were the least likely to report feeling happy, cheerful, excited, and enthusiastic. The cross-over point was the mean across all respondents. The present analysis does not include a separate condition for negative affect. Though negative affect is not strictly the absence of positive affect, in the present sample low positive affect scores are overwhelmingly matched by the presence of high negative affect scores.

Ethical Reasoning For each of six pairs of decision-making criteria based on the World Values Survey (Inglehart & Baker, 2000), respondents were asked to indicate on a five-point scale which of the two criteria they were more likely to employ in their daily decision-making. Each pair of criteria reflected a common trade-off in decisions (e.g., ‘‘attending to long-term versus short-term consequences’’). The six items loaded on two factors. One factor represented priority given to higher social and environmental impacts in daily decision-making, consistent with higher levels of moral reasoning (Kohlberg, 1984). Factor scores were calibrated in the manner described above. Full membership (1) was allocated to respondents in the upper quartile. Most respondents above this threshold reported prioritizing social and environmental criteria across all of the six items. Nonmembership (0) was allocated to respondents in the lower quartile who overwhelmingly reported prioritizing economic criteria in their daily decision-making. The cross-over point was the mean across all respondents. Respondents who scored close to the mean balanced social and economic criteria in their decision-making or responded inconsistently to the six items.

Multilevel Explanation of Why Managers Do Good

193

Ethical Component in Performance Evaluation The nature of performance goals is liable to influence managers’ behavior (Barsky, 2008). I include a condition reflecting whether, at the business unit level, there is formal account taken of managers’ adherence to the respect of ethical principles in their annual performance evaluation. Business units even within the same firm vary in their performance assessment policies, and these policies could act as incentives for managers to behave responsibly and as sanctions when managers behave irresponsibly. Performance assessment is measured as a dichotomous condition (0/1).

Resource Constraints Resource constraints in the firm might limit managers’ pursuit of social initiatives (Crilly, 2011) as well as sway how managers approach dilemmas (Zauberman & Lynch, 2005). Respondents reported on a four-point scale whether limited resources for social initiatives in their business unit impeded the pursuit of CSR policy. In calibrating this condition, I chose values of 0 (nonmembership) and 1 (full membership) to represent responses at the two endpoints of the scale, and values of 0.33 and 0.66 to represent responses between the two endpoints, where 0.33 implies ‘‘more out than in’’ the set of resource-constrained business units. 63% of respondents were scored either 0.66 or 1.

Collectivism Collectivistic societies might be especially likely to call for managers to balance interests between shareholders and nonshareholding stakeholders, and hence social norms for social responsibility might be salient in these contexts. Collectivism could influence managers’ behavior directly, and it could also activate the effects of managers’ psychology by altering the perceived pay-offs of complying with local norms. To capture this dimension, I used Hofstede’s (1984) collectivism-individualism measure. Countries scoring at or below 0.2 on collectivism were coded as complete nonmembers of the set of collectivistic countries. This threshold includes all large Anglo-Saxon countries, Canada being at this threshold point. Countries scoring at or above 0.7 on Hofstede’s measure, largely East Asian

194

DONAL CRILLY

and Latin American societies, are considered full members in the set of collectivist societies. I chose 0.49 on the collectivism scale as the cross-over point because this represented the highest score among the large Western economies (Spain). Together these five explanatory conditions at different levels of analysis create substantial causal complexity. Though any single condition could exert a direct effect on managers’ propensity to act in a socially responsible manner, there are possible interactions between them. As the calibration of some conditions relies on self-reports, I took steps to reduce the potential for common method bias. The survey was designed to mix the questions, and scale end points differed across questions (Podsakoff, MacKenzie, & Lee, 2003). Harman’s (1967) single factor test confirmed that the majority of the covariance between the outcome and the predictors was not accounted for by one factor, providing confidence that common method problems in the data are not likely to be serious. Finally, the results (Table 1) do not show the kinds of relationships that would be expected in the presence of common method bias.

Analysis Approach I created a truth table to identify combinations of causal conditions associated with socially responsible behavior. After deleting the combinations not associated with any observation, I specified a minimum acceptable frequency of five observations for any combination and a consistency cut-off of 0.78. This consistency cut-off is above the recommend minimum of 0.75 (Ragin, 2006) and corresponded to a gap in the distribution of consistency scores. I repeated the analysis to explain the absence of socially responsible behavior. In doing so, I specified a minimum acceptable frequency of four observations for any combination and a consistency cut-off of 0.82, which likewise corresponded to a gap in the distribution of consistency scores. To simplify the causal combinations, I employed the truth table algorithm (cf., Ragin, 2008). This algorithm produces a range of solutions. The parsimonious solution includes all possible simplifying assumptions, whereas the intermediate solution, which I also report in the results, is more conservative and includes only the most plausible assumptions to simplify the solutions.

Multilevel Explanation of Why Managers Do Good

195

RESULTS I report the results in Table 3, showing the combinations of core and peripheral conditions, associated with socially responsible behavior and its absence. The overall solution that explains socially responsible behavior has a consistency of 0.77, and the solution that explains the absence of a socially responsible orientation has a consistency of 0.79. However, solutions can be consistent without being empirically important. Coverage, which ranges from 0 to 1 (Ragin, 2008), measures the extent to which the solutions explain all cases in the data set. The coverage of the solution explaining socially responsible behavior is 0.49, whereas that of the solution explaining its absence is 0.47. Hence, the solutions account for a substantial proportion of the cases of socially responsible behavior (and its absence), though neither comes close to explaining the totality of cases. I also report measures of consistency and coverage for each individual configuration. The Boolean equation linked to socially responsible behavior is: Socially responsible behavior ¼ positive affect n ethical reasoning n collectivism þ  positive affect n  ethical reasoning n  collectivism n performance appraisal þ  positive affect n  resource constraints n collectivism þ ethical n reasoning n  resource constraints

Each line represents a configuration of core conditions associated with socially responsible behavior. The star () represents the Boolean logic term ‘‘AND.’’ The plus sign (+) represents the Boolean term ‘‘OR.’’ The tilde (B) represents the Boolean logic term ‘‘NOT.’’ These configurations shed light on the interplay between causes at different levels of analysis and help to resolve some of theoretical puzzles mentioned above. Each configuration comprises conditions at two or three distinct levels, implying that explanations of socially responsible behavior focused exclusively on a single level of analysis are insufficient. With the exception of collectivism, the organizational and societal conditions are consistently positively or negatively associated with socially responsible behavior and concur with the findings of prior research. In contrast, the sign of the relationship between the two individual-level conditions – affect and reasoning – and socially responsible behavior is sensitive to the social context in which managers find themselves. This finding would not be revealed by conventional regression techniques that isolate the net effects of variables.

0.774 0.283 0.125

1

0.489 0.769

0.797 0.200 0.028

3

0.783 0.239 0.031

4

0.744 0.105 0.023

1a

0.841 0.099 0.036

1b

0.470 0.787

0.816 0.278 0.049

2

0.815 0.334 0.099

3

) core causal condition (absent); ( ) peripheral causal condition (absent); ( ) peripheral causal

0.801 0.211 0.061

2

Configurations Explaining Absence of Socially Responsible Behavior

Configurations Predicting Socially Responsible Behavior.

Configurations Explaining Socially Responsible Behavior

( ) Core causal condition (present); ( condition (present).

Individual Positive affect Ethical reasoning Organizational Ethical component in performance appraisal Resource constraints Societal Collectivism Consistency Coverage Unique coverage Solution coverage Solution consistency

Condition

Table 3.

196 DONAL CRILLY

Multilevel Explanation of Why Managers Do Good

197

Taken together, the four configurations also demonstrate equifinality. That multiple pathways to socially responsible behavior exist is especially evident when comparing the first two configurations. The first represents an apparently altruistic or ethical pathway, whereas the second represents an apparently reward-focused pathway to socially responsible behavior. The combination of positive affect and ethical reasoning (configuration 1) is associated with socially responsible behavior in collectivistic societies. The positive interaction between affect and reasoning is consistent with work in moral psychology that shows that positive affect can encourage individuals to attend to the ethical dimensions of problems, but notably this positive interaction occurs in collectivistic societies. In contrast, in individualistic societies, low levels of positive affect and limited ethical reasoning (configuration 2) are associated with socially responsible behavior subject to the presence of performance assessment that rewards compliance with ethical principles. Arguably, low positive affect increases sensitivity to rewards, and hence individuals might engage in responsible behavior to avoid punishment or to enhance self-interest. Further, avoiding punishment and enhancing one’s self-interest are associated with low levels of ethical reasoning focused on the external consequences of behavior (Kohlberg, 1984). The analysis reinforces the idea of a synergistic effect between affect, reasoning, and company incentive policies. Configurations 3 and 4 also demonstrate distinct pathways linking the provision of organizational resources to socially responsible behavior. For this relationship to be salient, either managers find themselves in collectivistic societies (configuration 3), or they engage in ethical reasoning (configuration 4). Similar to configuration 2, the absence of positive affect appears to prompt socially responsible behavior when there are norms and sanctions supporting such behavior (configuration 3). The Boolean equation associated with the absence of socially responsible behavior is: Absence of socially responsible behavior ¼  performance appraisal n  resource constraints þ  ethical reasoning n  resource constraints n  collectivism þ positive affect n  ethical reasoning n  collectivism Though the first configuration implies the sufficiency of business unit policy in predicting the absence of socially responsible behavior, peripheral conditions associated with this configuration are either the lack of collectivism or the combined absence of positive affect and ethical reasoning.

198

DONAL CRILLY

These peripheral conditions imply that business unit policies can negatively influence managers’ disposition to socially responsible behavior when these policies are reinforced by social norms against collectivism or by managers’ individual characteristics (a lack of ethical reasoning, and a low degree of positive affect). The second and third configurations imply that the absence of ethical reasoning and collectivism is associated with the absence of socially responsible behavior – either when the individual manager is low on positive affect or when he or she faces resource constraints to implement CSR policy. Notably, ethical reasoning cooccurs alongside collectivism in multiple configurations, whereas its absence cooccurs alongside individualism. This provides fairly strong support that the guidance that managers use in responding to social dilemmas is shaped by the contexts in which managers interact. It is also noteworthy that positive affect can be associated with either socially responsible behavior or the absence thereof depending on the dominant societal norm. Without wishing to infer unwarranted causality, a possible explanation is that positive affect can sometimes prompt less careful reflection about problems (Forgas, 2002), and hence positive-affect managers might be prone to accept as appropriate the kinds of behaviors that they most frequently observe locally.

CONCLUDING THOUGHTS: IMPLICATIONS FOR CONFIGURATIONAL THEORY AND METHODS The analysis in the present chapter illustrates the application of comparative, configurational methods to research in social responsibility. Specifically, it demonstrates the application of fsQCA in a context where HLM appears to be an obvious alternative. Social responsibility involves the intentions and behaviors of individual actors, organizations, and broader society. In responding to social dilemmas, managers are subject to personal preferences and influences in the organizational and external environment. Scholars from distinct disciplines – economics, sociology, and psychology – address the topic. Gaining greater insight into social responsibility thus frequently demands a multilevel and multidisciplinary perspective. Configurational approaches seem apt both as a tool and research paradigm for providing this perspective. Specifically, the analysis underlines the application of configurational approaches to multilevel research in social responsibility. The analysis

Multilevel Explanation of Why Managers Do Good

199

demonstrates that the cooccurrence of conditions at various levels of analysis matters more than any single condition by itself. For example, positive affect can be associated with both the presence and the absence of socially responsible behavior. Hence, it is insufficient to study psychological variables to the exclusion of the social context, and vice versa. Even more importantly, multilevel research can go beyond explaining the net effects of distinct levels of analysis. Rather, configurational approaches to multilevel research focus on the combined effects of influences at different levels of analysis (Lacey and Fiss, 2009). Thus far, configurational approaches such as fsQCA have been used in multilevel research but predominantly to investigate the interplay of causes at meso and macrolevels of analysis. Configurations are certainly relevant to the microlevel of analysis. For example, scholars have identified the importance of fit between the individual and the organization (Chatman, 1989), implying distinct configurations of managers and firms. The analysis in the present chapter also provides strong support for configurations of affect, reasoning, and firms’ incentive policies. Yet, there has been relatively little research exploring configurations generally at the microlevel, and certainly minimal research at the microlevel specifically using variants of QCA such as csQCA and fsQCA. The focus of csQCA and fsQCA on meso and macrolevels of analysis is consistent with their application in sociology and political science where their use is more advanced than in organization science. In contrast, microlevel researchers have embraced techniques such as HLM. Configurational approaches have a role to play in introducing behavioral variables into macrolevel theories. Organizational scholars are increasingly interested in microfoundations. One of the challenges in studying microfoundations is to make the bridge between levels of analysis (Felin & Foss, 2005) and to go beyond simple aggregation. csQCA and fsQCA provide relevant tools for analyzing how causes at the microlevel combine to influence organizational outcomes. Similarly, scholars of the microfoundations of organizing have to understand the interplay between the organizational context and these microfoundations in shaping behavior. The present chapter illustrates one way to achieve this. More broadly, configurational approaches offer the prospect of bridging the divide between social psychology and organizational theory to questions of organizational design. In the specific domain of social responsibility, configurational thinking is likely to spur multilevel research investigating how individuals, firms, and societies combine to influence firms’ behaviors. A relevant example would be to connect perspectives from social psychology and organizational

200

DONAL CRILLY

theory to explain variation in response to institutional pressures for social responsibility. Further, given the inconclusive link between social and financial performance, configurational methods hold the prospect of shedding light on the boundary conditions that determine when firms create or destroy shareholder value from their social responsibility initiatives. Despite the promising applications of configurational thinking to organization science, in general, and social responsibility research, in particular, the present study also highlights a number of challenges for researchers: the application of configurational thinking to microlevel research, the definition of standards for calibration, the definition of standards for selecting causal conditions, and causal inference. Exploring individuals in configurations opens up a potential problem. A core premise of configurational thinking is the expected cooccurrence of attributes or elements and, hence, limited variety. Organizational configurations might exist in many circumstances because environmental selection pressures weed out inappropriate organizational forms. In contrast, though individuals do self-select into organizations (Chatman, 1989), such selfselection is likely to be weaker than the selection pressures in the environment on organizations. This potentially narrows the range of topics for which configurational approaches are appropriate. The coverage scores in the present analysis (0.49 in predicting socially responsible behavior, and 0.47 in predicting its absence) are reasonable by the standards of prior fsQCA research, but many managers cannot be adequately allocated to any configuration. Before automatically applying these methods to a given research problem, it makes sense to consider the likely importance of configurations based on existing theoretical knowledge. Additionally, though set-theoretic methods are becoming increasingly sophisticated, standards for calibrating set membership are likely to be a particular concern to microorganizational scholars. Microorganizational scholars have developed well-established scales to measure latent constructs that matter in their theories. Though survey research could lend itself readily to set calibration if researchers used a single item (i.e., a single Likert-scale response could be calibrated directly with either end of the scale representing the points of maximum exclusion and maximum inclusion in the set), the use of a single item to measure a latent construct would raise questions of validity and reliability. In turn, after combining items to measure a single construct, calibration is no longer as straightforward. In the present analysis I used factor scores, that take account of the factor loading of each item. Though these are superior measures than scores based on single items, they do not lend themselves as easily to calibration. Choosing the mean as the

Multilevel Explanation of Why Managers Do Good

201

crossover point and the first and third quartiles as the points of exclusion and inclusion is convenient but these anchors would ideally be theoretically justified. Further, the number of conditions that can be included in configurational analysis is limited. As there is no way of handling missing conditions, there is a need to select causal conditions carefully. For example, many more individual characteristics than in the present study could matter in explaining socially responsible behavior, as could many more features of the organizational context. However, including many conditions would be infeasible given that the number of theoretical combinations of conditions rises exponentially with each new condition. In the present chapter, I identified core drivers from theory, focusing on those that seemed to matter and that might interact with drivers at other levels of analysis. Also, though adding more conditions as quasicontrols is often unfeasible because of a restricted number of cases, a suitable approach would be to split the sample on the basis of the characteristic of interest and see whether similar configurations seem to explain the outcome across each subsample (cf., Greckhamer, Misangyi, Elms, & Lacey, 2008). Notably, similar concerns apply to multilevel regression models. Though regression could control for the effects of many variables, adding new parameters or levels of analysis to an HLM analysis substantially increases the difficulty of interpreting results and places more demands on the data. Finally, scholars have demonstrated in HLM studies that nonsensical contextual effects can be found when group-level coefficients are interpreted causally (Gelman, 2006). Though set-theoretic methods are useful in identifying patterns of association and causality, researchers must similarly be wary of unwarranted causal inference. In small-N and intermediate-N studies, researchers might have substantial knowledge of the cases to guide inference. In large-N studies, this in-depth knowledge of individual cases is usually lacking, and there is a risk of attributing causality to any configuration. Yet, merely identifying a combination of conditions associated with an outcome does not necessarily imply a causal relationship. Careful research design – for example, judicious selection of the causal conditions based on existing theoretical knowledge, and, where applicable, taking measures to avoid common method bias in survey research – can increase confidence in the meaningfulness of the results. Because longitudinal studies provide a strong basis for causal inference, the difficulty of combining set-theoretic methods and panel data reinforces this concern. Future development of these methods to allow a more nuanced understanding of causal processes would be welcome.

202

DONAL CRILLY

NOTE 1. Readers are referred to Bryk and Raudenbush (1992) and Griffin (1997) for a more complete discussion of this method.

ACKNOWLEDGMENTS This research was funded by a grant provided by the 6th Framework Programme of the European Commission (Directorate-General for Research). I am indebted to Lourdes Casanova, Kai Hockerts, Mario Minoja, Peter Neergaard, Esben Pedersen, Francesco Perrini, Susan Schneider, Pamela Sloan, Antonio Tencati, and Maurizio Zollo for contributing to the data collection.

REFERENCES Aguilera, R. V., Rupp, D. E., Williams, C. A., & Ganapthi, J. (2007). Putting the S back in corporate social responsibility: A multilevel theory of social change in organizations. Academy of Management Review, 32(3), 836–863. Bandura, A. (1978). The self system in reciprocal determinism. American Psychologist, 33(4), 344–358. Bansal, P., & Roth, K. (2000). Why companies go green: A model of ecological response. Academy of Management Journal, 43(4), 717–737. Barsky, A. (2008). Understanding the ethical cost of organizational goal-setting: A review and theory development. Journal of Business Ethics, 81(1), 63–81. Bryk, A. S., & Raudenbush, S. W. (1992). Hierarchical linear models: Applications and data analysis methods. Newbury Park, CA: Sage. Chatman, J. A. (1989). Improving interactional organizational research: A model of personorganization fit. Academy of Management Review, 14(3), 333–349. Chau, R. Y. J., Morris, M. W., & Ingram, P. (2009). Guanxi versus networking: Distinctive configurations of affect- and cognition-based trust in the networks of Chinese and American managers. Journal of International Business Studies, 40, 480–508. Cialdini, R. B., Darby, B. L., & Vincent, J. E. (1973). Transgression and altruism: A case for hedonism. Journal of Experimental Social Psychology, 9(6), 502–516. Crilly, D. (2011). Predicting stakeholder orientation in the multinational enterprise: A midrange theory. Journal of International Business Studies, 42, 694–717. Crilly, D., & Sloan, P. (2012). Enterprise logic: Explaining corporate attention to stakeholders from the ‘inside-out’. Strategic Management Journal, 33(10), 1174–1193. Crilly, S., Schneider, S. C., & Zollo, M. (2008). The psychological antecedents to socially responsible behavior. European Management Review, 5(3), 175–190.

Multilevel Explanation of Why Managers Do Good

203

Davison, M. L., Kwak, N., Seo, Y. S., & Choi, J. (2002). Using hierarchical linear models to examine moderator effects: Person-by-organization interactions. Organizational Research Methods, 5(3), 231–254. Elsbach, K. D., Barr, P. S., & Hargadon, A. B. (2005). Identifying situated cognition in organizations. Organization Science, 16(4), 422–433. Felin, T., & Foss, N. J. (2005). Strategic organization: A field in search of micro-foundations. Strategic Organization, 3(4), 441–455. Ferraro, F., Pfeffer, J., & Sutton, R. I. (2005). Economics language and assumptions: How theories can become self-fulfilling. Academy of Management Review, 30(1), 8–24. Fiss, P. C. (2011). Building better causal theories: A fuzzy set approach to typologies in organizational research. Academy of Management Journal, 54(2), 393–420. Forgas, J. P. (2002). Feeling and doing: Affective influences on interpersonal behavior. Psychological Inquiry, 13, 1–28. Garnier, M. A. (1972). Changing recruitment patterns and organizational ideology: The case of a British military academy. Administrative Science Quarterly, 17(4), 499–507. Garnier, M. A. (1973). Power and ideological conformity: A case study. American Journal of Sociology, 79(2), 343–363. Gelman, A. (2006). Multilevel (hierarchical) modeling: What it can and cannot do. Technometrics, 48(3), 432–435. Greckhamer, T., Misangyi, V., Elms, H., & Lacey, R. (2008). Using qualitative comparative analysis in strategic management research: An examination of combinations of industry, corporate, and business-unit effects. Organizational Research Methods, 11(4), 695–726. Griffin, M. A. (1997). Interaction between individuals and situations: Using HLM to estimate reciprocal relationships. Journal of Management, 23(6), 759–773. Haidt, J. (2001). The emotional dog and its rational tail: A social intuitionist approach to moral judgment. Psychological Review, 108(4), 814–834. Harman, H. (1967). Modern factor analysis. Chicago, IL: Chicago University Press. Hofstede, G. (1984). Culture’s consequences: International differences in work-related values. Beverly Hills, CA: Sage. Inglehart, R., & Baker, W. E. (2000). Modernization, cultural change, and the persistence of traditional values. American Sociological Review, 65(1), 19–51. Isen, A. M. (2001). An influence of positive affect on decision making in complex situations: theoretical issues with practical implications. Journal of Consumer Psychology, 11(2), 75–85. Kohlberg, L. (1984). Essays in moral development: The psychology of moral development. New York, NY: Harper Row. Lacey, R., & Fiss, P. C. (2009). Comparative organizational analysis across multiple levels: A set-theoretic approach, in Studying differences between organizations: Comparative approaches to organizational research. In B. King, T. Felin & D. Whetten (Eds.), Research in the Sociology of Organizations (Vol. 26, pp. 91–116). Lau, S., & Nie, Y. (2008). Interplay between personal goals and classroom goal structures in predicting student outcomes: A multilevel analysis of person-context interactions. Journal of Educational Psychology, 100(1), 15–29. Lazarus, R. S. (1991). Emotion and adaptation. Oxford: Oxford University Press. Matten, D., & Moon, J. (2008). ‘‘Implicit’’ and ‘‘explicit’’ CSR: A conceptual framework for a comparative understanding of corporate social responsibility. Academy of Management Review, 33(2), 404–424.

204

DONAL CRILLY

Mor, N., & Winquist, J. (2002). Self-focused attention and negative affect: A meta-analysis. Psychological Bulletin, 128, 638–662. Podsakoff, P. M., MacKenzie, S. B., & Lee, J. Y. (2003). Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology, 88(5), 879–903. Ragin, C. C. (2006). User’s guide to fuzzy-set/qualitative comparative analysis 2.0. Tucson, AZ: University of Arizona. Ragin, C. C. (2008). Redesigning social inquiry: Fuzzy sets and beyond. Chicago, IL: University of Chicago Press. Raudenbush, S. W., & Liu, X. (2000). Statistical power and optimal design for multisite randomized trials. Psychological Methods, 5(2), 199–213. Reidenbach, R. E., & Robin, D. P. (1990). Toward the development of a multidimensional scale for improving evaluations of business ethics. Journal of Business Ethics, 9(8), 639–653. Simons, T., & Ingram, P. (1997). Organization and ideology: Kibbutzim and hired labor, 1951–1965. Administrative Science Quarterly, 42(4), 784–813. Snijders, T., & Bosker, R. (2000). Multilevel analysis. Thousand Oaks, CA: Sage. Somers, M. R., & Block, F. (2005). From poverty to perversity: Ideas, markets, and institutions over 200 years of welfare debate. American Sociological Review, 70(2), 260–287. Sonenshein, S. (2007). The role of construction, intuition, and justification in responding to ethical issues at work: The sensemaking-intuition model. Academy of Management Review, 32(4), 1022–1040. Stone, D. N., & Ziebart, D. A. (1995). A model of financial incentive effects in decision making. Organizational Behavior and Human Decision Processes, 61(3), 250–261. Trevin˜o, L. K. (1986). Ethical decision making in organizations: A person-situation interactionist model. Academy of Management Review, 11(3), 601–617. Watkins, D., & Cheung, S. (1995). Culture, gender, and response bias. Journal of Cross-Cultural Psychology, 26(5), 490–504. Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief measures of positive and negative affect: The PANAS scales. Journal of Personality and Social Psychology, 54(6), 1063–1070. Watson, J. P., & Clark, L. A. (1994). The PANAS-X. Manual for the positive and negative affect schedule – Expanded form. Iowa, IA: University of Iowa. Weyant, J. M. (1978). Effects of mood states, costs, and benefits on helping. Journal of Personality and Social Psychology, 5(5), 1167–1169. Zajonc, R. B. (1980). Feelings and thinking: Preferences need no inferences. American Psychologist, 35(2), 151–175. Zauberman, G., & Lynch, J. G. (2005). Resource slack and propensity to discount delayed investments of time versus money. Journal of Experimental Psychology, 134(1), 23–37.

CHAPTER 9 THE VALUE OF CONFIGURATIONAL APPROACHES FOR STUDYING DIGITAL BUSINESS STRATEGY YoungKi Park and Omar A. El Sawy ABSTRACT This chapter shows how configurational approaches can be a valuable inquiring system for examining and understanding complex messy phenomena in the area of digital business strategy in turbulent environments such as digital ecodynamics. Digital ecodynamics is defined as the holistic confluence among environmental turbulence, dynamic capabilities, and IT systems – and their fused dynamic interactions unfolding as an ecosystem. With fuzzy-set qualitative comparative analysis (fsQCA) we analyze firm-level field survey data and describe how IT systems, organizational dynamic capability and environmental turbulence simultaneously combine to result in multiple configurations, which have different causal structures to produce competitive firm performance. This equifinality shows how configurational approaches can create new practical insights in digital ecodynamics by suggesting multiple strategic options from which organizations can choose the best solution that fits their context. Keywords: Digital ecodynamics; digital business strategy; IT systems; fsQCA; equifinality

Configurational Theory and Methods in Organizational Research Research in the Sociology of Organizations, Volume 38, 205–224 Copyright r 2013 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0733-558X/doi:10.1108/S0733-558X(2013)0000038013

205

206

YOUNGKI PARK AND OMAR A. EL SAWY

THE COMPLEX PHENOMENON OF DIGITAL ECODYNAMICS Digital ecodynamics represents a complex phenomenon of fused, dynamic interactions among information technology, organizational dynamic capabilities, and environmental turbulence, unfolding as an ecosystem (El Sawy, Malhotra, Park, & Pavlou, 2010). Information technologies are becoming fused with business processes and it is almost impossible to separate information technologies from business processes (El Sawy, 2003). New advances in information technologies make business environments change more rapidly and unpredictably. Simultaneously, information technologies enhance organizational dynamic capabilities (Pavlou & El Sawy, 2006). On the other hand, more frequent and rapid introduction of new innovations to markets by IT-enabled dynamic capability makes environments more turbulent (Brown & Eisenhardt, 1997; Davis, Eisenhardt, & Bingham, 2009; Mendelson & Pillai, 1998). Thus, it is becoming one interdependent fusion with entangled messiness and complexity. Digital ecodynamics are disrupting industry boundaries (Burgelman & Grove, 2007) and creating a new business ecosystem (e.g., a smart phone ecosystem) in which some companies suddenly achieve a great success while some quickly fade away. Since the fast cycle of success and failure is the result of the entangled messiness and complexity of digital dynamics,

Fig. 1. Digital Ecodynamics. Source: El Sawy et al., (2010).

Value of Configurational Approaches for Studying Digital Business Strategy

207

competitive advantage in digital ecodynamics cannot be effectively explained by a single factor but rather a combination of many key factors. Fig. 1 depicts the entanglements of digital ecodynamicsthrough intersecting orbit of elements, but also signals the fusion quality of digital ecodynamics through an entangled Gordian knot at the center. Thus digital ecodynamics has no separations among its core elements, but it is the wholeness of the fused interactions among the three elements.

THE NEED FOR CONFIGURATIONAL APPROACHES TO STUDY DIGITAL ECODYNAMICS Digital ecodynamics as fused dynamic interactions of the key elements often creates synergetic effects and nonlinear change, which can be more effectively captured from a holistic perspective and at the system level (El Sawy et al., 2010; Meyer, Gaba, & Colwell, 2005). We argue that configurational approaches are especially helpful as an inquiring system (Churchman, 1971) for exploring the holistic nature of digital ecodynamics. By the holistic nature in digital ecodynamics, we mean ‘‘how information technologies, organizational dynamic capabilities, and environmental turbulence simultaneously and systemically combine to produce competitive firm performance.’’ Studies in information technologies have mainly espoused the paradigmatic lenses of two types of theories: variance theories and process theories (Markus & Robey, 1988). While many research advances in IS studies came about through variance theories, the pristine mental frames of variance theories may be insufficient for studying the holistic complexity of digital ecodynamics. Nor is the additive relationship in variance theories wellsuited to disequilibrium conditions in digital ecodynamics. Process theories are good at tracing and explaining the unfolding ‘‘how’’ of a phenomenon based on triggers and manifestations over time and excel at capturing contextual detail. While many advances in IS strategy have relied on process theories, they too are not best-suited for capturing holistic systemic effects where the interplay of variables spawns emergent properties only evident at the level of the whole (El Sawy et al., 2010). Although we do not argue configurational theory approaches replace variance theories and process theories, we argue that configuration theories can more appropriately investigate digital ecodynamics by especially discovering its holistic systemic features that cannot be effectively captured by variance and process theory

208

YOUNGKI PARK AND OMAR A. EL SAWY

approach. Configuration theories emanate from understanding patterns and combinations of elements, and how they, as configurations, cause certain outcomes to occur. A configuration is a specific combination of elements or conditions that generate an outcome of interest (Rihoux & Ragin, 2009). Configuration theories view phenomena as clusters of interconnected elements that must be simultaneously understood as a holistic integrated pattern (vs. individual elements separately in variance theories), thus overcoming the traditional reductionism problem in structural contingency theory (Meyer, Tsui, & Hinings, 1993). In configuration theories, elements are viewed together in combinations, and theoretical models can parsimoniously capture complexity at the system level. Besides, configurational theories allow us to have the intuitive simplicity of typologies (Fiss, 2011), which is heuristically attractive to practitioners (Mintzberg, 1979). Similar to variance theories and process theories, configuration theories can be applied at many levels of analysis to understand phenomena in organizations, supply chains, and ecosystems. Configuration theories also accommodate complex inter-connectedness of multiple elements, nonlinearities, and discontinuities (Meyer et. al., 1993). Therefore, they are well-suited to the realities of mutual causality, synergetic effects, and nonlinear change associated with digital ecodynamics. Configuration theories accommodate equifinality with a flexible analytical frame (variance theories only allow unifinality with a fixed analytical frame) where a different set of elements can produce the same outcome (Fiss, 2007; Ragin & Amoroso, 2011); this is a property that typically characterizes digital ecodynamics. Finally, their inherent structure keeps the notion of mutual causality connected to the context, making them wellsuited to middle range, context-sensitive (rather than universal) theories – which are appropriate for the field of digital business strategy in general (Rihoux & Ragin, 2009). Thus, configurational theory approaches have the potential to render the next quantum leap in advancing digital business strategy research by complementing variance theories and process theories. However, until recently, configurational approaches have had limitations, especially in terms of the dealing with complex causality. Theorizing typically ended when an effective configuration was identified (Fiss, 2007, 2011), and there was little insight as to which element in the configuration was more critical, and why or how various elements interacted to produce an outcome (Ragin, 2008). Most recently, advances in configurational methodologies have exposed the inner workings of the ‘‘black box’’ of configurations and developed associated methodologies for data analysis, such as fuzzy set qualitative configurational analysis (fsQCA). These advances have provided more fine-grained understanding of complex causality, while also developing

Value of Configurational Approaches for Studying Digital Business Strategy

209

advanced methodologies that remain connected to the holistic perspective (e.g., Fiss, 2007; Ragin, 2008; Rihoux & Ragin, 2009), thus ‘‘not throwing out the baby with the bathwater.’’ These theoretical and methodological advances are allowing us to move beyond many of the limitations of ‘‘conventional’’ configuration theories, and we are heading into what we might conceive as a ‘‘second generation’’ of configuration theory approaches. In the following section, based on the findings of an empirical field study, we show how a set-theoretic configurational approach, more specifically, fsQCA can effectively describe the holistic nature of digital ecodynamics. First, we articulate key constructs and their validation, and then show some examples of configurations found by fsQCA. Lastly, we explain the implication of configuration approaches to study digital business strategy.

THE FIELD STUDY WITH FSQCA We support our argument of the need for configurational approach for studying the holistic nature of digital ecodynamics with the findings from an empirical field study. We focus on explaining why configurational approaches can appropriately investigate the complex phenomenon of digital ecodynamics. We especially investigate the different roles of information technologies in achieving competitive firm performance: information technologies as either enablers or inhibitors for competitive firm performance. While, many IS studies have argued one of these two opposing roles of information technologies, we show that configurational approaches can effectively capture such opposing roles of IT simultaneously, which could not be done with traditional approaches. Articulating the Key Constructs and Measurement Validity We define key constructs of digital ecodynamics based on existing studies on organizational sensemaking and responding to environmental change and studies on IT-enabled organizational dynamic capabilities to cope with turbulent environments. First, among several possible forms of organizational dynamic capability (Eisenhardt & Martin, 2000), we choose organizational agility. It has been defined as the organizational ability to sense and respond to important market opportunities and threats in a timely manner (Sambamurthy, Bharadwaj, & Grover, 2003; Tallon & Pinsonneault, 2011). In complex rapid changes in digital ecodynamics, organizational agility

210

YOUNGKI PARK AND OMAR A. EL SAWY

enables organizations to successfully sense and respond to important environmental change and achieve competitive advantage. We define and operationalize organizational agility in a way that represents the unique aspects of agility in sensing, decision-making, and acting. Sensing agility is the ability to detect in a timely manner important business events such as customers preference change, competitors new strategic moves, and emergence of new products, technologies and regulations. Decision-making and acting agility are the ability to make a timely decision and action respectively to respond to captured business events. As such, each type of agility represents the unique aspect of organizational agility, and the three together can capture the full aspects of organizational level agility. Therefore, we define an organizational agility as a second-order formative construct consisting of these three first-order constructs: sensing, decision-making, and acting agility (Diamantopoulos & Winklhofer, 2001). We measure organizational agility using 15 survey items, which are presented in the appendix. To more effectively represent the role of information technology in digital ecodynamics, we define the IT capability construct, which is defined as an organizational ability to mobilize, reconfigure and deploy IT resources to support work processes and tasks (Bharadwaj, 2000; Pavlou & El Sawy, 2006). Based on task-technology fit theory (Goodhue & Thompson, 1995), we define three types of IT systems that can support organizations to effectively sense and respond to environmental change: business intelligence systems, communication and collaboration systems, and business process and resource management systems. Then, we define IT capability as a second-order construct that can represent all the unique capabilities provided by these three types of information technologies. We measure IT capability using 18 survey items presented in the appendix. We define environmental turbulence in a way to represent two dimensions: the speed and the direction of change (McCarthy, Lawrence, Wixted, & Gordon, 2010). The speed of change is the rate at which new opportunities emerge and the rate at which new products and services are introduced. Unpredictability as the direction of environmental change is the amount of disorder, showing no consistent similarity or pattern (Davis et al., 2009). We measure these two variables regarding changes in customers’ preference, competitors’ competitive action, and technologies. In addition to these three elements, we consider top management team (TMT) energy defined as top managers’ commitment to change. Existing studies explain its important role in organizational sensemaking and responding to changing environments, such as collecting important

Value of Configurational Approaches for Studying Digital Business Strategy

211

information about business environmental change and make a decision to adapt to change (Hambrick, Cho, & Chen, 1996). TMT has also proved one of the most important factors for information systems success. Top managers’ energetic initiatives for changing their organizations can help employees overcome resistance to change, and successfully drive employees to adopt and use information systems for their business event management (Markus, 1983; Wixom & Watson, 2001). Lastly, we consider firm size, since it is the well-known organizational factor that influences organizational dynamic capability and firm performance. To decide a firm’s size, we follow the definition provided by the Korean Government agency for administering small and medium companies,1 which considers the number of employees, sales revenue, assets, industry type and other factors simultaneously. This study administered survey questionnaires to managers in Korean companies in diverse industries differing in the level of changing velocity. Korea is famous for its advanced information technologies; for example, it ranks first in high-speed internet coverage in the world and its economy relies heavily on the high tech industry. Therefore, this data set is relevant for this study, allowing us to explore the dynamics between IT capability, organizational agility, and environmental turbulence that produce competitive firm performance. A total of 106 firms responded to our survey. Composite reliabilities were greater than 0.7 for all constructs, indicating internal consistency (Nunnally, 1978). The square root of average variance extracted (AVE) for individual constructs was greater than its correlations with other constructs and it was greater than 0.5 (Table 1). Further, all standardized item loadings were

Table 1.

IT Agility PERF ENV TMT

Correlation and Composite Reliability for the Main Constructs.

Mean

STD

Reliability

IT

Agility

PERF

ENV

TMT

4.1 4.4 4.5 4.5 5.1

0.9 0.7 1.1 0.8 1.1

0.90 0.74 0.93 0.79 0.93

0.86 0.48 0.36 0.26 0.45

0.70 0.40 0.22 0.43

0.90 –0.07 0.38

0.81 0.16

0.93

 Square roots of average variances extracted (AVE’s) shown on diagonal.  Correlations greater than 0.25 are significant at the 0.01 level; greater than 0.19 are significant at the 0.05 level.

212

YOUNGKI PARK AND OMAR A. EL SAWY

greater than 0.7 and loaded on their corresponding factors. Thus, constructs have discriminant and convergent validity (Fornell & Larcker, 1981). Further, Harman’s single-factor test showed no common method bias (Podsakoff & Organ, 1986).

The Findings with fsQCA Using fsQCA software, we calibrate an interval scale to a fuzzy membership score ranging from 0.0 to 1.0. Calibration is a process of transforming interval scale values to fuzzy set membership scores based on three qualitative anchors: full membership, full non-membership, and the crossover point of maximum ambiguity regarding membership in the set of interest (Ragin, 2008). The set membership score represents the extent to which each case is a member of, for example, a high level of organizational agility. This study uses a 7-point Likert scale: 1 ¼ lowest, 4 ¼ ambiguous (crossover), 7 ¼ highest level. This study defines the interval scale 2 as the anchor for full nonmembership, 4 as the crossover point, and 6 for the full membership anchor for the set of high level of constructs, including organizational agility, IT capability, environmental turbulence, firm performance, and TMT energy. Fig. 2 depicts configurations for achieving high firm performance resulted from fsQCA. The configurations are expressed by the notation systems from Ragin and Fiss (2008). The dark shaded circles indicate the presence of an element, crossed-out circles indicate the absence of an element, large circles indicate core elements, and small circles indicate peripheral elements. Blank spaces indicate a ‘‘don’t care situation,’’ in which the causal element may be either present or absent. For example, the dark shaded circle of IT capability means that a high level of IT capability should exist, while the crossed-out circle of IT capability means that a high level of IT capability should not exist in order for the configuration to result in the outcome of interest. This study set the minimum acceptable frequency of cases for solutions at 3, and the lowest acceptable consistency cutoff at 0.9, which is above the minimum recommended threshold of 0.75 (Ragin, 2008). Overall, 66 cases fell into configurations exceeding the minimum solution frequency. Of these cases, 57 also exceeded the minimum consistency threshold of 0.9 for higher performance. Five different configurations were found to result in high performance, meaning five different paths to the same outcome (i.e., equifinality).

Value of Configurational Approaches for Studying Digital Business Strategy

Fig. 2.

213

Configurations for Achieving High Performance.

Consistency for five configurations ranges from 0.88 to 0.93, which well exceeds the usually acceptable level of 0.80. By comparing similarities and differences of these five configurations, we can find several patterns. Among them, we articulate some interesting patterns that describe the different roles of information technology represented by IT capability as follows: IT matters versus IT does not matter  In turbulent environments, organizations can achieve high performance by either IT-enabled agility (see configuration 1) or non-IT-enabled agility (configuration 4, 5). Both have a high-level of coverage. Opposing roles of IT: Contingency effect  In turbulent environments, a high level of IT capability should exist and play a core role in achieving high performance (see configurations 1, 2). In stable environments, a high level of IT capability should be absent for a configuration to result in high performance (see configurations 3).

214

YOUNGKI PARK AND OMAR A. EL SAWY

The second pattern explains how configurational approaches simultaneously capture two opposing roles of information technologies. In rapidly and unpredictably changing environments, information technologies can help organizations effectively sense and respond to critical events in a timely manner and achieve high performance (Pavlou & El Sawy, 2006, 2010; Tallon & Pinsonneault, 2011). In slowly and predictably changing environments, a high-level of IT capability possibly inhibits firms from achieving high performance. In stable environments, there could be multiple alternative paths to competitive advantage (Davis et al., 2009; Fine, 1998). In stable environments in which change is relatively slow and predictable, there is enough time to gather and process relevant information and do rational analysis (Fine, 1998). Therefore without much help from information technologies, organizations are less likely to fail in sensing and responding to environmental change (Davis et al., 2009). Therefore, too much investment in information technologies can be costly. These findings from configurational approaches can effectively explain such seemingly opposite roles of information technologies with the rich combinatorial expression of configurations. On the other hand, since TME energy exists for all the five configurations as mostly a core element, we can assume it is a necessary condition for organizations to achieve competitive performance. With the membership plot that fsQCA provides, we examine whether TMT energy is a necessary condition for high performance. The plot depicts the membership distribution of cases in terms of TMT energy and firm performance. According to the plot, a high percentage of the cases are located below the diagonal, with some cases being just a little above the diagonal. Therefore, the plot also provides the evidence of TMT energy as a necessary condition for high performance. Further, the result of a necessary condition test with fsQCA shows that the consistency of TMT energy for performance is 0.91 and coverage is 0.76. Based on all the evidence, we define TMT energy as a necessary condition for high performance. Treating TMT energy as a necessary condition, we then executed another fsQCA without TMT energy, although TMT energy will still be considered a necessary condition for high performance when we interpret the results. Overall, 78 cases fell into configurations exceeding the predefined frequency cutoff of 3. Of these cases, 60 exceeded the predefined consistency threshold of 0.9 for high performance. Fig. 3 shows the resulting configurations. For this configurational analysis, we set the minimum acceptable frequency of cases for solutions at 3, and the lowest acceptable consistency cutoff at 0.9. Overall, 78 cases fell into configurations exceeding the

Value of Configurational Approaches for Studying Digital Business Strategy

Fig. 3.

215

Configurations for Achieving High Performance with TMT Energy as a Necessary Condition.

minimum solution frequency. Of these cases, 60 also exceeded the minimum consistency threshold of 0.9 for higher performance. Four different configurations were found to result in high performance, meaning four different paths to competitive firm performance. Consistency for five configurations ranges from 0.86 to 0.93, which well exceeds the usually acceptable level of 0.80. By comparing these configurations, we can extract a clearer pattern that explains the opposing roles of information technologies: In turbulent environments, IT capability plays a critical role in achieving high firm performance (1a, 1b), while in stable environments, a high level of IT capability should be absent for organizations to achieve high performance.

Configurations (1a, 1b) are examples of second-order equifinality, which means that multiple configurations with same core elements and different peripheral elements can result in the same outcome (Fiss, 2011). In this case, configurations 1a and 1b have the same core element of IT capability. However, they have different causal structures with different peripheral elements to produce the same outcome. Further, the first configuration consisting of a high level of IT capability and agility as core elements has the highest coverage, meaning that

216

YOUNGKI PARK AND OMAR A. EL SAWY

organizational agility and IT capability together can most effectively achieve high performance in turbulent environments, on condition of TMT energy. In sum, configurational approaches enable a researcher to more clearly investigate and describe the complex dynamics and the roles of elements at the system level, compared to other approaches. In the following, we describe insights that we could develop with all the patterns, and explain implications for digital business strategy.

INSIGHTS AND IMPLICATIONS This chapter has suggested a configurational approach for developing a better understanding of the holistic nature of digital ecodynamics. With the findings from an empirical field study we support this argument. The empirical findings demonstrate how a configurational approach effectively can capture the holistic systemic patterns from complex phenomenon of digital ecodynamics. We summarize the insights we have gained in this instance, outline some shortcomings, and point to future opportunities for research.

Insights Uniquely Captured by a Configurational Approach Configurational approaches assume that the causal structure of a configuration is flexible and thus allows multiple configurations with different causal structures to result in the same outcome. This means that configurational approaches can express equifinality that explains a system can reach the same final state from different initial conditions through different paths (Fiss, 2007). Dynamic interactions of information technologies, organizational agility and environmental turbulence can results in multiple different configurations that have different causal structures. Fuzzy-set qualitative comparative analysis (fsQCA) can effectively express such complex dynamics in a rich combinatorial way with core/peripheral and present/absent elements. The same element that is core and present in some configurations can be a peripheral and absent in other configurations. For example, information technologies in turbulent environments play a core role in achieving high performance, but in stable environments become a peripheral and absent element. Thus, configurational approaches can effectively capture the multifaceted roles of information technologies in digital ecodynamics. The opposing roles of IT as either an enabler or an inhibitor for competitive

Value of Configurational Approaches for Studying Digital Business Strategy

217

firm performance have been an important topic in the IS literature. Traditional approaches based on a fixed causal structure cannot effectively explain such inconsistent roles of information technologies. Further, configurational approaches can better explain nonlinear change in digital ecodynamics. The findings of this study imply that IT-enabled organizational transformation is not a continuous linear change, but rather a discontinuous nonlinear shift from one state to another. The fsQCA results describe how multiple configurations could achieve a similar level of performance. By comparing multiple configurations that have possibly different causal recipes with different roles of elements, we could suggest how one configuration can move from one state to another by changing the structure of configurational elements.

Practical Implications for Managers This chapter shows how configurational approaches create new practical insights in the context of digital ecodynamics. Taking advantage of such properties as equifinality and causal asymmetry, a configurational approach provides organizations with multiple strategic options from which they can choose the best solution to competitive advantage that fits their unique contexts. Organizations can choose the best solution among the multiple configurations by considering their own organizational characteristics and environment (e.g., size and environmental velocity). For example, according to the findings of this study, organizations in high velocity industries need to invest more in information technologies than organizations in low velocity industries. Actually, too much investment in information technologies for sensing threats and opportunities in stable environments can be a misplaced investment.

Shortcomings Configuration theories are not without limitations, even with recent advancement in theories and methodologies. Configuration theories have many good features that can effectively discover holistic patterns of complex systems, but they are unable to explain the longitudinal ‘‘how’’ of causal configurations – and process theories are better there. They also have a ‘‘temporality’’ problem (Rihoux & Ragin, 2009) in being unable to inherently track shifts over time.

218

YOUNGKI PARK AND OMAR A. EL SAWY

On the other hand, while configurational approaches are appropriate for building new middle-range theories, variance theories are better suited for building and testing universal and generalizable theories. Configuration theories can be best suited to ‘‘modest generalization’’ (Rihoux & Ragin, 2009) based on the nature of causality being both context-specific and conjecture-specific; always keeping the connection between context and causal inference. This is a double-edged sword because the researcher needs to be engaged and cognizant of the study’s context, thus not making the interpretation of the data ‘‘automatic’’ but requiring to identify plausible configurations that are applicable in practice. As such, configurational approaches are not proposed to supplant variance theories or process theories, but are complementary for explaining more completely the complex aspects of digital ecodynamics. Further, it may also take a mind-shift and a learning effort for researchers who are still naturally ingrained in variance and process theories to employ a configurational lens in theory building and testing. Thinking holistically in terms of configurations and fuzzy sets takes cognitive effort and strain that seems unnatural at first, and much more work is needed to educate researchers.

Unexplored Arenas This study is just one example that shows how a configurational approach can effectively describe the diverse holistic features of digital ecodynamics with some empirical findings regarding competitive firm performance. There are many interesting and important areas that are still understudied. Configuration approaches can be especially helpful for developing a better understanding of many other emerging issues around digital ecodynamics. For example, configurational approaches can more appropriately explain how digital platforms interact with organizational cultures and change the way of communicating with key customers and strategic partners, and eventually creating new competitive digital ecosystems. There are many such vivid examples around us which a holistic perspective would better explain. For example, as the digital business ecosystem recently unfolded due to changes in customer preferences and changes in social interaction, coupled with new advances in smartphone software, some key players such as Apple and Google successfully created new competitive ecosystems with their digital platforms and key stakeholders. But, previously successful Nokia failed to respond quickly enough with an appropriate configuration and

Value of Configurational Approaches for Studying Digital Business Strategy

219

floundered. Existing theories that are based on linear, additive relationships between elements and assume equilibrium status cannot effectively explain such dynamically changing punctuated disequilibrium in digital ecodynamics (Meyer et al., 2005). A configurational approach accompanied by strong methods such as fsQCA is one of the best ways to build new theories that can effectively explain such nonlinear discontinuous changes driven by dynamic interactions of digital technologies, organizational and social factors. It has opened up new research vistas and opportunities for understanding digital ecodynamics. We believe that configurational theories and their associated methods will trigger a transformational change in how research is carried out, and how theories are built and tested in the area of digital business strategy. We look forward to being part of that exciting intellectual journey.

NOTE 1. http://eng.smba.go.kr/pub/kore/kore020101.jsp

ACKNOWLEDGMENTS We would like to thank two anonymous reviewers and Peer Fiss for their very helpful comments.

REFERENCES Bharadwaj, A. (2000). A resource-based perspective on information technology capability and firm performance: An empirical investigation. MIS Quarterly, 24(1), 169–196. Brown, S. L., & Eisenhardt, K. M. (1997). The art of continuous change: Linking complexity theory and time-paced evolution in relentlessly shifting organizations. Administrative Science Quarterly, 42, 1–34. Burgelman, R. A., & Grove, A. S. (2007). Cross-boundary disrupters: Powerful interindustry entrepreneurial change agents. Strategic Entrepreneurial Journal, 1, 315–327. Churchman, C. W. (1971). The design of inquiring systems. New York, NY: Basic Books. Davis, J. P., Eisenhardt, K. M., & Bingham, C. B. (2009). Optimal structure, market dynamism, and the strategy of simple rules. Administrative Science Quarterly, 54, 413–452. Diamantopoulos, A., & Winklhofer, H. M. (2001). Index construction with formative indicators: An alternative to scale development. Journal of Marketing Research, 38(2), 269–277.

220

YOUNGKI PARK AND OMAR A. EL SAWY

Eisenhardt, K. M., & Martin, J. (2000). Dynamic capabilities: What are they? Strategic Management Journal, 21, 1105–1121. El Sawy, O. A. (2003). The IS core IX: The 3 faces of IS identity: Connection, immersion, and fusion. Communication of the Association for Information Systems, 12(1), 588–598. El Sawy, O. A., Malhotra, A., Park, Y., & Pavlou, P. A. (2010). Seeking the configurations of digital ecodynamics: It takes three to tango. Information Systems Research, 21(4), 835–848. Fine, C. (1998). Clockspeed: Winning industry control in the age of temporary advantage. New York, NY: Perseus Books. Fiss, P. (2007). A set-theoretic approach to organizational configurations. Academy of Management Review, 32(4), 1180–1198. Fiss, P. (2011). Building better causal theories: A fuzzy set approach to typologies in organization research. Academy of Management Journal, 54, 393–420. Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18, 39–50. Goodhue, D. L., & Thompson, R. L. (1995). Task-technology fit and individual performance. MIS Quarterly, 19(2), 213–236. Hambrick, D. C., Cho, T. S., & Chen, M. (1996). The influence of top management team heterogeneity on firms competitive moves. Administrative Science Quarterly, 41(4), 659–684. Markus, M. L. (1983). Power, politics, and MIS implementation. Communications of the ACM, 26(6), 430–444. Markus, L. M., & Robey, D. (1988). Information technology and organizational change: Causal structure in theory and research. Management Science, 34(5), 583–598. McCarthy, I. P., Lawrence, T. B., Wixted, B., & Gordon, B. R. (2010). A multidimensional conceptualization of environmental velocity. Academy of Management Review, 35(4), 604–626. Mendelson, H., & Pillai, R. (1998). Clockspeed and informational response: Evidence from the information technology industry. Information Systems Research, 9(4), 415–433. Meyer, A. D., Gaba, V., & Colwell, K. A. (2005). Organizing far from equilibrium: Nonlinear change in organizational fields. Organization Science, 16(5), 456–473. Meyer, A. D., Tsui, A. S., & Hinings, C. R. (1993). Configurational approaches to organizational analysis. Academy of Management Journal, 36(6), 1175–1195. Mintzberg, H. (1979). The structuring of organizations: A synthesis of the research. Englewood Cliffs, NJ: Prentice Hall. Nunnally, J. C. (1978). Psychometric theory. New York, NY: McGraw-Hill. Pavlou, P. A., & El Sawy, O. A. (2006). From IT leveraging competence to competitive advantage in turbulent environments: The case of new product development. Information Systems Research, 17(3), 198–227. Podsakoff, P. M., & Organ, D. W. (1986). Self-reports in organizational research: Problems and prospects. Journal of Management, 12(4), 531–544. Ragin, C. C. (2008). Redesigning social inquiry: Fuzzy set and beyond. Chicago, IL: University of Chicago Press. Ragin, C. C., & Amoroso, L. M. (2011). Constructing social research. Los Angeles, CA: Sage. Rihoux, B., & Ragin, C. C. (2009). Configurational comparative methods: Qualitative comparative analysis (QCA) and related techniques. Thousand Oaks, CA: Sage.

Value of Configurational Approaches for Studying Digital Business Strategy

221

Sambamurthy, V., Bharadwaj, A., & Grover, V. (2003). Shaping agility through digital options: Reconceptualizing the role of IT in contemporary firms. MIS Quarterly, 27(2), 237–263. Tallon, P. P., & Pinsonneault, A. (2011). Competing perspectives on the link between strategic information technology alignment and organizational agility: Insights from a mediation model. MIS Quarterly, 36(2), 463–486. Wixom, B. H., & Watson, H. J. (2001). An empirical investigation of the factors affecting data warehousing success. MIS Quarterly, 25(1), 17–41.

222

YOUNGKI PARK AND OMAR A. EL SAWY

APPENDIX: MEASUREMENT ITEMS Constructs Firm performance (PERF)

Measures Compared to competitors, our company: is more successful. has a greater market share. is growing faster. is more profitable.

Organizational agility Sensing agility

Our organization: is slow to detect changes in our customers’ preferences on products. is slow to detect changes in our competitors’ moves. is slow to detect changes in technologies.

Decision-making agility

analyzes important events about customer/ competitor/ technology without delay. finds out opportunities and threats from changes in customer/competitor/ technology in a timely manner. makes an action plan to meet customers’ needs without delay. makes an action plan to react to competitors’ strategic moves without delay. makes an action plan on how to use new technology without delay.

Acting agility

can reconfigure our resources in a timely manner. can modify/restructure processes in a timely manner. can adopt new technologies in a timely manner. can introduce new products in a timely manner. can change price quickly.

Value of Configurational Approaches for Studying Digital Business Strategy

223

Appendix. (Continued ) Constructs

Measures can change strategic partnerships in a timely manner. can solve our customers’ changing needs and complaints without delay.

IT Capability (IT) Business intelligence

Information systems in our organization: support to acquire information from diverse sources about changes in customers, competitors, and technologies. filter out unimportant information of customers, competitive actions, and technology change based on predefined rules. help appropriate managers to know important events about customers, competitors, and technologies in a timely manner. support to access to relevant data in a timely manner. provide enterprise-wide integrated, consistent data. support what-if analysis which shows ‘‘how the outcomes can change when the situations change.’’

Communication and collaboration

support disseminating relevant information to people who need it. support information sharing within the company. support exchanging relevant information with key partner companies and customers. support virtual conferences with real-time video and audio. support effective collaboration between employees. support effective collaboration with key partner companies and customers.

224

YOUNGKI PARK AND OMAR A. EL SAWY

Appendix. (Continued ) Constructs Business process and resource management

Environmental turbulence (ENV) Speed of environmental change 1 ¼ very low/slow 7 ¼ very high/fast Unpredictability of environmental change 1 ¼ very unpredictable 7 ¼ very predictable Top management team energy (TMT)

Measures visually present business processes. support the design and creation of new business processes. support streamlining and scheduling processes. automate business processes. provide information about what human and other resources are needed for business processes. provide real-time information about resource availability.

The speed of change in our customers’ product preferences isy The speed of change in competitors’ moves isy The speed of change in the technology in our industry isy The direction of change in our customers’ product preferences isy The direction of change in competitors’ moves isy The direction of change in the technology in our industry isy Our top management team is energetic. Our top management team drives dynamic change.

CHAPTER 10 THE CONFIGURATIONAL APPROACH IN ORGANIZATIONAL NETWORK RESEARCH Jo¨rg Raab, Robin H. Lemaire and Keith G. Provan ABSTRACT This chapter explores how a configurational approach and set-theoretic methods can contribute to a deeper and more nuanced understanding of organizational networks and network relations. This is especially true for the study of ‘‘whole networks’’ of organizations where data collection difficulties often limit the sample size (Provan, Fish, & Sydow, 2007). We present two empirical examples of current research on whole networks, demonstrating how qualitative comparative analysis (QCA) can be used to analyze organizational networks. We then discuss the methodological and theoretical implications of the configurational approach for future organizational network research. Keywords: Configurational approach; set-theoretic methods; QCA; networks; organization; interorganizational relations

Configurational Theory and Methods in Organizational Research Research in the Sociology of Organizations, Volume 38, 225–253 Copyright r 2013 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0733-558X/doi:10.1108/S0733-558X(2013)0000038014

225

JO¨RG RAAB ET AL.

226

INTRODUCTION One of the major trends that have emerged in the study of organizations over the past 20 years has been a focus on networks and network relationships (see Brass, Galaskiewicz, Greve, & Tsai, 2004; Kilduff & Brass, 2010; Provan, Fish, & Sydow, 2007 for reviews). Studies that use network data to explain organizational outcomes, as well as attempts to explain the formation, structure, and functioning of multilateral organizational networks, now regularly appear in the major journals in management and organization studies, sociology, political science, and public administration. In the late 1980s and early 1990s, scholars first attempted to describe the ‘‘nature of the beast,’’ explaining why networks are different from markets or hierarchical organizations (Powell, 1990). Scholars also described the characteristics of connected systems and demonstrated their impact on policy formation (Laumann & Knoke, 1987), and showed how network ties go beyond formal contracts to facilitate business relationships (Larson, 1991; Nohria & Eccles, 1992; Uzzi, 1997). While much of this early work was qualitative and descriptive, developments in the study of social networks among individuals were soon applied, allowing for quantitative analysis of network relations among organizations (Borgatti, Everett, & Freeman, 2002; Wasserman & Faust, 1994), including visualization tools (Freeman, 2004). More recently, there have been significant advances in the methods and software available to analyze network data and in modeling network effects directly; most notably, Multiple Regression Quadratic Assignment Procedure (MRQAP) (Dekker, Krackhardt, & Snijders, 2007; Krackhardt, 1988), Exponential Random Graph Models (ERGM) (Robins, Pattison, Kalish, & Lusher, 2007), and longitudinal network analysis tools like SIENA (Snijders, Steglich, & van de Bunt, 2010). Despite these advances in the quantitative analysis of networks, research on organizational, rather than interpersonal networks, has tended to underutilize these methods, focusing mainly on traditional linear statistical methods and procedures when it comes to testing hypotheses and developing theories. There may be good reasons for this focus, especially when examining large numbers of dyadic interorganizational relations, often referred to as ‘‘alliance networks.’’ To expand a scholarly understanding of organizational networks we believe that an alternative analytical approach is needed, especially one that allows for a more nuanced understanding of networks and network relations. Specifically, we argue that qualitative comparative analysis (QCA) can and should be utilized. As we discuss below, this is

The Configurational Approach in Organizational Network Research

227

especially true for the study of ‘‘whole networks’’ of organizations where data collection difficulties often limit the sample size (Provan et al., 2007).1 We envision at least three main reasons why the configurational approach and the application of QCA can help in further advancing the field of interorganizational networks, not only analytically, but also in generating new theoretical perspectives and insights. First, following the general understanding in this volume and similar to Fiss (2007, p. 1180), we believe that organizational network research will benefit considerably if organizations, relations, and networks are understood as ‘‘clusters of interconnected structures and practices’’ that should be analyzed from a systemic and holistic view rather than in terms of individual independent variables (see also Meyer, Tsui, & Hinings, 1993). As we demonstrate below in our second empirical example, application of QCA is especially advantageous when examining organizational or network outcomes such as performance, effectiveness, or knowledge transfer.2 These outcomes very likely are best explained not through models of singular causation, but rather through configurations of variables, which assumes complex causation. Second, interorganizational relations and networks comprise nested social entities that jointly contribute to the production of outcomes. For instance, individuals make up organizations and organizations forge relations among each other which are often initiated or sustained by individuals. Thus, ideally, an analysis of outcomes of interorganizational relations and networks should comprise complex configurations that also encompass characteristics of social entities at different levels of analysis. This sort of analytical complexity has rarely been applied in empirical studies of networks (Moliterno & Mahony, 2011). As we demonstrate below in our first empirical example, the configurational approach is well suited to address complex causalities with regard to the multilevel issues that underlie interorganizational relations and networks. Third, one of the significant gaps in network research in our view is that core attributes and the context of interorganizational relations and networks have received scant attention. This may be due to a strong emphasis on structural and relational embeddedness that is the basis for the network paradigm. We believe that the institutional, political, or cultural contexts of the networks being studied must also be considered, especially when explaining the outcomes of interorganizational relations and networks. When considering context, complex constellations of conditions typically emerge. The configurational approach is ideally suited for addressing these issues, because assumptions about the fit between various combinations

228

JO¨RG RAAB ET AL.

of network characteristics and certain external institutional, political, or cultural conditions can be tested. It is important to recognize, however, that the number of internal and external factors that can be considered is limited, since QCA as a technique only provides meaningful results when relatively few conditions are included in the analysis at the same time, because every additional variable increases the number of possible configurations. Our discussion begins with the presentation of a 9-cell typology of organizational network research. We first briefly discuss the different levels of analysis addressed in past studies of organizational networks, including the major theoretical perspectives that have been developed at each level of analysis. The purpose of the typology is to discuss which specific methods have typically been used in past network research and for which combinations the configurational approach holds the most promise. Following discussion of our typology, we then present two empirical studies that utilize QCA to explicate our ideas. Finally, we will discuss what methodological and theoretical implications might follow from the application of the configurational approach for future organizational network research.

ANALYTICAL PERSPECTIVES AND METHODS IN ORGANIZATIONAL NETWORK RESEARCH In recent years, researchers have started to look more closely at what exactly the conditions are under which networks might produce more effective outcomes, or the conditions under which organizations might be able to reap more benefits from their network involvement. In these studies, outcomes of interorganizational relations have been explored at three levels of analysis. The first is the individual node; in this case, the focus is on examining how an organization’s activities, structure, and outcomes are affected by its ties to other organizations. The second level of analysis is the dyadic ties these nodes form, focusing on the properties of each relationship rather than on characteristics of the node/organization itself. The third level of analysis is at the network level, which includes all relationships larger than a dyad such as triads (three actors), various subnetwork structures like cliques and clusters, and most broadly, a whole network. At the whole network level the focus is on all the actors in a relatively bounded set of relationships, examining their interactions or the lack of interactions among all the actors, the overall structural properties of the network, and its

The Configurational Approach in Organizational Network Research

229

effectiveness. This last category also includes research on consciously created, goal-directed organizational networks, including examination of their effectiveness and governance (Provan & Kenis, 2008). Matching the attributes of individual organizations, dyads, and whole networks with possible outcomes at each of these three levels of analysis allows for the construction of a 3  3 table that helps to categorize the various approaches to organizational network research (see Table 1). We do not claim that the cells of this table represent exclusive analytical perspectives. In fact, combinations of at least two levels of analysis (cells) are in evidence in many organizational network studies. This is not unusual in network research, where different levels of analysis are combined and ‘‘micro and macro can be very similar theoretically and methodologically’’ (Borgatti & Foster, 2003, p. 1001). In fact, as mentioned above, analyzing organizational networks across levels of analysis may be an important path forward for understanding the complexity of organizational networks. For example, Gulati and Gargiulo (1999) combined relational attributes of individual organizations with dyadic ties and network characteristics for explaining the formation of interorganizational relationships. Moreover, it is not our intention to provide an exhaustive overview of the field according to the suggested typology, but rather to categorize and illustrate key analytical perspectives in organizational network research, and then, to discuss and evaluate the potential of the configurational approach for analyzing each different perspective. In Table 1, cell 1 represents research that attempts to explain the structure, functioning, and performance of individual organizations based on their ties to other organizations. This might include organizational outcomes that are derived from its relationships with others, like the effect of an organization’s centrality in a network on its innovative capacity (Owen-Smith & Powell, 2004) or the effect of external search depth and breadth, through use of network ties, on innovative organizational performance (Laursen & Salter, 2006). The perspective represented by cell 1 is especially prominent in business network research and most likely represents the largest single category of organizational network studies. The main focus of research in this cell is the investigation of the benefits of cooperation and involvement in networks for a focal organization. Focus on the value of ties to other organizations regarding firm-level outcomes is especially relevant from a business management perspective since this provides the rationale for why ties with other organizations are formed and maintained. Research in cell 2 is also concerned with the impact of connections and cooperation for individual organizations but from an explicitly dyadic

Individual Organizations

Cell 1 Impact of structural position, linking strategy of org. on its functioning and performance (ex. learning, innovation). Laursen & Salter, 2006; Owen-Smith & Powell, 2004

Cell 2 Impact of critical dyadic ties on org. performance (i.e., tie to high reputation/centrality/high resource org.). Levin & Cross, 2004; Zaheer et al., 1998

Cell 3 Impact of network structural features on characteristics or performance of lead organization or other individual orgs. Gilsing et al., 2008; Powell et al., 1996

Individual Organizations (egocentric ties, alliance portfolio, node centrality)

Dyads (resource dependencies, strategic alliances, embedded ties, etc.)

Subnetworks (Zthree organizations) and whole networks

Cell 8 Impact of existing or missing ties (nature of ties) between crucial actors on network governance, functioning, effectiveness.

Cell 9 Impact of network governance or structure on network effectiveness Provan & Milward, 1995

Cell 6 Impact of network governance/ structure on dyads (i.e., trust between members, formation of dyadic ties). Gulati & Gargiulo, 1999

Cell 7 Impact of lead org. on network governance, functioning, or effectiveness. Human & Provan, 2000; Jarillo, 1988; Sydow & Windeler, 1998

Cell 4 Formation of dyadic IORs (i.e., alliances) on the basis of homophily. Bakker et al., 2010

Cell 5 Formation and dissolution of ties dependent on relationship characteristics, change in neighboring dyads or over time, effectiveness of dyadic ties (alliances) Gulati & Gargiulo, 1999; Lincoln et al., 1992

Subnetwork (Z Three Organizations) and Whole Networks

Dyads

Level of Analysis (Outcomes)

A Typology of Organizational Network Research.

Level of Analysis (Attributes)

Table 1.

230 JO¨RG RAAB ET AL.

The Configurational Approach in Organizational Network Research

231

perspective. Here, interest is not simply on the outcomes of an organization’s network ties, but on the attributes and properties of specific dyadic connections and their impact on the performance or other outcomes, such as the adoption of practices, of the organizations involved in the dyad. Thus, even though the outcome is at the organization level, it is the dyadic relationship itself that is affecting that outcome, which may differ for each partner in the dyad. An example of this would be the impact of the level of trust between two firms on the performance of each partner organization (Zaheer, McEvily, & Perrone, 1998). Another study that falls into this cell is the study by Levin and Cross (2004) in which the focus is on the strength of dyadic ties between a focal organization and a partner, and the receipt of useful knowledge by the focal organization. The distinction between cells 1 and 2 (and between cells 4 and 5 and between 7 and 8) is somewhat blurry, and depends on whether one is conceptualizing a tie solely from the perspective of an individual organization (cell 1), or based on properties of the dyad (cell 2), which typically requires data from both partners. Cell 3 corresponds to research that examines the effect of whole network characteristics or network components (triads, cliques, subnetworks) on individual organizations (Powell, Koput, & Smith-Doerr, 1996). Gilsing, Nooteboom, Vanhaverbeke, Duysters, and Van den Oord (2008), for example, tested the effect of technological distance, network position, and overall network density on the explorative innovation performance of companies in three different industries. While technological distance and network position are organizational attributes, overall network density is a network characteristic, which makes the study a combination of research in cells 1 and 3. As for all research that includes whole network characteristics, the data requirements for this line of research are significant, since researchers need different networks in order to achieve variation on the independent variable. This can be achieved, however, either by examining multiple networks or by studying a single network at several points in time and exploring the impact of different network structures, for instance, on organizational outcomes. Moving from organizational outcomes to dyadic outcomes, research in cell 4 attempts to explain the formation, existence, quality, characteristics, or dissolution of dyadic ties based on the attributes of one of the organizations that comprise the dyad. For example, Bakker et al. (2010) have argued that the quality of knowledge transfer between partners is based either on the motivation of each partner in the dyad to transfer knowledge to the other, the absorptive capacity of the receiving organization, or both.

232

JO¨RG RAAB ET AL.

Cell 5 characterizes research that focuses on the explanation of the formation, existence, characteristics, dissolution, and consequences of dyadic relationships on the basis of the characteristics of these relationships (their trust, for instance), the existence of prior ties, or the joint characteristics and actions of neighboring dyads (Gulati & Gargiulo, 1999). For example, in her review and discussion of determinants of interorganizational relationships, Oliver (1990) concluded that the existence of specific types of dyadic relationships between organizations, like joint ventures or corporate– financial interlocks, could be explained, at least in part, by the properties of the dyad, such as its asymmetry or reciprocity. Lincoln, Gerlach, and Takahashi (1992) tested a host of hypotheses explaining dyadic intercorporate ties in Japanese Keiretsu networks based on a number of dyadic covariates. They demonstrated, for example, that directors are regularly dispatched from the owning company to the owned company, not the other way round, indicating that shareholding is a necessary but not a sufficient condition for director transfer. Studies in cell 6 represent research that tries to explain the impact of network characteristics, like the form of governance or its structure (density, centralization) on dyadic relationships, such as the level of trust between members or the probability of tie formation. Gulati and Gargiulo (1999, p. 1451), for example, hypothesized and confirmed that ‘‘the probability of a new alliance between any two organizations increases with the level of structural differentiation in the interorganizational network.’’ The last column of the table is based on outcomes at the network level. Cell 7 encompasses studies in which researchers use organization-level factors to explain how individual organizations and their activities might affect outcomes at the network level, such as network structure, stability, legitimacy, or performance. This approach is most likely to be found in studies of interorganizational networks led by a hub firm or lead organization, as for example in studies by Jarillo (1988) or Sydow and Windeler (1998) and as discussed by Provan et al. (2007, p. 483) in their review of the literature on whole networks. In these studies, the attributes, strategies, and behavior of the network’s lead organization had an impact on the overall governance and performance of the network as a whole. For example, Human and Provan (2000, p. 353), in their study of legitimacy building in multilateral networks, found that the legitimacy of the network as a whole depended on the ability of its network administrative organization (NAO) to work closely with member firms to establish a recognizable network identity. Studies that fall into cell 8 represent research that attempts to explain outcomes at the network level by focusing on the nature of relations between

The Configurational Approach in Organizational Network Research

233

crucial actors in the network. For example, studies in this cell might examine the nature of the dyadic relationship (level of trust, multiplexity) between a lead organization and other important actors in the network on overall network governance and performance, which to our knowledge has not yet been explored. Finally, cell 9 represents research that examines subnetwork or whole network properties on outcomes that are also at the subnetwork or network level, that is, three or more organizations. Outcomes would typically be explained using characteristics at the same level of analysis, so that outcomes at the clique level would be based on characteristics of the clique, outcomes at the level of the subnetwork cluster on cluster characteristics, and outcomes at the network level on network-level characteristics. This includes what Provan et al. (2007) labeled ‘‘whole network research.’’ Here, network-level outcomes like network effectiveness or the specific mode of governance used (Provan & Kenis, 2008) might be explained by the overall task the network is trying to accomplish, the number of network members involved, the density and strength of prior or current network ties, or the stability of the network over time (Provan & Milward, 1995). While we have cited empirical and conceptual research to explicate the perspective and nature of research for each of the nine cells of our table, it is important to recognize that much of the research cited is not limited to one cell and may encompass perspectives from at least two cells. For example, we categorized Gulati and Gargiulo (1999) as research that would fall into cell 6 because dyadic alliance formation was explained by network characteristics. But in this work, the probability of new alliance formation between two organizations was also explained by the level of interdependence between those organizations (dyadic perspective, cell 5). Human and Provan (2000) found not only that the NAO was critical for building the legitimacy of the full network (cell 7), but also that the subsequent legitimacy (both internal and external) of the network as a whole was essential for explaining the success of one network and the failure of the other (whole network perspective, cell 9). When a study includes multiple levels of analysis, focus is typically on the same outcome level,3 and thus, falls within a single column of our table. Research on networks often does, however, include at least two levels regarding the independent variables (the rows of our table). This is due to the nature of organizational network research, where different levels are nested in each other. Specifically, dyads are made up of two nodes/ organizations, triads are made of three nodes and three dyads, and networks are made up of many nodes (n) and dyads (n(n1)/2). Taking this into

234

JO¨RG RAAB ET AL.

account, complexity is important and should be addressed more explicitly, a key point made recently by Moliterno & Mahony, 2011. The configurational approach and QCA are well suited to this task for a number of reasons. First, QCA can regularly be used in multilevel situations, since it is not susceptible to the problems that come with multilevel issues as in regular linear regression. Second, while hierarchical clustering is a way to deal with multilevel data, the need for a large sample size limits its use with network data, especially when considering subnetworks and larger network units. The configurational approach in general and QCA in particular enable the researcher to collect in-depth qualitative and quantitative data and still examine the additional complexity created by the multilevel nesting. Third, from a conceptual standpoint, it should be highly attractive to look at complex configurations of conditions that are located at different levels of analysis and jointly produce an outcome (see for example the study by Bensaou & Venkatraman (1995) on the configurations of interorganizational relationships in the United States and Japanese car industries). In addition, Meyer et al. (1993) have already argued that ‘‘the configuration perspective evokes rich opportunities for cross level theorizing and research’’ (p. 1191). We suspect that one reason why organizational network scholars have not yet been supportive of the approach may be due to a lack of suitable methods and technological support. However, recent developments in QCA (Ragin, 2000, 2008; Ragin & Strand, 2008; Rihoux & De Meur, 2009) have been substantial, including the computer programs fs/QCA and Tosmana.4

THE ANALYSIS OF NETWORK DATA The standard methods used to analyze network data and related effects have almost exclusively either been qualitative, based on case studies (sometimes comparative), or quantitative, using standard statistical analyses, such as correlation or regression. The former has mainly been used in exploratory network research done in the 1990s or in studies in which the whole network is the unit of analysis (cells 7–9 in Table 1). With the network as the unit of analysis, the researcher typically must collect data on several dozens of organizations for each network, which makes it very difficult to reach a sample size of more than 4–6 networks. Thus, due to the challenging data requirements that this type of research demands, the qualitative approach is used almost exclusively.

The Configurational Approach in Organizational Network Research

235

Statistical methods based on standard covariation techniques have mostly been applied in studies in which the unit of analysis is the individual organization. This is because it is easier to reach a larger N, which makes it possible to use standard statistical tools (cell 1 in Table 1). Much, if not most of this research, has relied on secondary data, where large numbers of organizations and their ties, typically formalized through contract or equity stakes, can be collected and analyzed. This statistical approach is also attractive for analysis at the dyad level (cells 4 and 5) and the node level with dyads being the independent variable (cell 2). Even in relatively small networks, many dyads may occur (a maximum of n(n1)/2 ties for undirected relationships, where n is the number of actors in the network). However, a problem here is that for some research questions and operationalizations, the assumption of independence of observations is violated, as the existence or nonexistence of one dyad potentially has an influence on other dyadic relationships. Recently, novel statistical models have been developed (MRQAP, SIENA, and ERGM) to address this and other problems associated with network data. These models have been applied in organizational network research to explain characteristics and behavior of nodes (cell 2), the existence of dyads (cell 5) (Bell & Zaheer, 2007), as well as the evolution of networks through the existence, formation, and dissolution of ties over time (cells 4 and 5 in Table 1). Thus, these models are especially valuable for research in cells 2, 4, and 5. When it comes to the explanation of network evolution, these models also clearly hold an advantage compared to covariate methods and even QCA despite recent attempts to make QCA available to the analysis of temporal processes (see contribution by Hak, Jaspers, and Dul in this volume).5 Considering the many contributions network researchers have made in recent years using traditional statistical methods, we are not suggesting that these methods be abandoned. Rather, our aim is to demonstrate that a configurational approach and set-theoretic methods (Ragin, 1987, 2008; Fiss, 2007, 2009) can be extremely valuable for studying networks. In particular, we believe that these methods have the most added values when used in studies that examine network-level characteristics or focus on network-level outcomes (cells 3 and 6–9 in Table 1). This is due to both theoretical demands of the methods, which require a complex configuration of multilevel factors to explain outcomes, and small and medium sample sizes on which this type of research is typically based. For example, in one of the most widely cited empirical studies comparing whole organizational networks, Provan and Milward (1995) suggest a theoretical framework to explain network effectiveness. In their four propositions, which were

236

JO¨RG RAAB ET AL.

developed on the basis of a systematic comparison of four networks in mental health care, they formulated arguments that are of a configurational nature, as they state necessary conditions that in combination with certain other factors might lead to network effectiveness. Yet given the methodological development at the time, they treated these relationships as linear associations. In this chapter, we argue that while propositions like these have been very helpful for advancing theory about networks, they can best be tested and further advanced theoretically and empirically using settheoretic methods. We do not claim that a particular method is exclusive to a certain perspective (cell) within the realm of organizational network research. To the contrary, it can be very fruitful to investigate research questions with regard to a particular perspective from different theoretical and methodological angles and combine the insights. For example, the research on organizational configurations as generally understood (Short et al., 2008) falls completely into cell 1 and profits greatly from the application of settheoretic methods as recently demonstrated by Fiss (2007, 2011), although such research does not qualify as organizational network research. Depending on the extent to which the independence of observations is violated, stochastic models are probably most appropriate for research at the dyadic level. However, since set-theoretic methods are not based on an assumption of the independence of observations, we regard them also as suitable for research that corresponds to the perspectives of cells 2, 4, and 5. To further demonstrate the value of the configurational approach and set-theoretic methods for the understanding of organizational networks, we discuss two empirical applications using QCA. Both examples represent original research by the authors and both involve the collection of whole network data; however, the studies involve different analytical perspectives and can be categorized as falling into different cells of the typology presented in Table 1.

EMPIRICAL APPLICATION We will present two recent studies as in-depth empirical applications: one on a child and youth health network in Canada that fits in cell 3, and the other on the effectiveness of whole networks in the field of crime prevention in the Netherlands that represents cell 9. For both studies we will present the research goal, the research question, the basic theoretical concepts,

The Configurational Approach in Organizational Network Research

237

operationalization, and data collection, and briefly describe some key findings that were achieved by using the set-theoretic method. The goal is to give the reader a more concrete impression of the type of research and the added value of the configurational approach rather than present a fullfledged research report.

A CONFIGURATIONAL APPROACH FOR EXPLAINING THE IMPACT OF NETWORK STRUCTURE ON ORGANIZATIONAL OUTCOMES In the first empirical piece demonstrating the use of QCA in the study of networks, we provide an overview of a study of a child and youth health network (Lemaire & Provan, 2010). In this example, fuzzy-set QCA (fs-QCA) was used to examine the relationship between tie patterns of the network member organizations and their perceptions of the impact of the network on their organizations. The whole network case, the Southern Alberta Child and Youth Health Network (SACYHN), was a large (nearly 50 organizations), publicly funded, goal-directed network with the goal of improving services for children, youth, and their families. The network was governed through what Provan and Kenis (2008) have described as an NAO. The general question guiding the research was: What structural paths, or configurations, based on an organization’s ties to the network, best explain the positive impact of network involvement on organizational performance? What is known from past research, which has taken a traditional linear and statistical perspective, is that organizations having many ties to other organizations, or being central in the network, are more likely to be affected by the network as a whole (O’Toole & Meier, 2004; Powell et al., 1996). It has also been proposed, though, that cohesive subgroup membership is a better predictor of organizational performance than degree centrality (Schalk, Torenvlied, & Allen, 2010). In this empirical study we hypothesized that centrality is a sufficient but not a necessary condition for organizational performance, since other tie structures may substitute for high centrality. However, we also hypothesized that cohesive subgroup membership alone is not sufficient, and instead, building on small world theory (Watts, 1999), the combination of being a member of a cohesive subnetwork and having a strong direct tie to the NAO, will result in a structure that can substitute for high centrality. The outcome in this study is the positive impact of the network on organizational performance, depending on the way an organization is

JO¨RG RAAB ET AL.

238

embedded in and connected to the whole network. The research is thus an example of taking multiple attributes of an organization, based on the ties it has to other member organizations in a whole network, as well as characteristics of the network itself, to explain an outcome at the organizational level. Therefore, this work can be placed in cell 3 of Table 1.

Data Collection and Measures Data on SACYHN were collected from key informants at each organization in the network using a survey. The main components of the questionnaire were organizational demographics, organizational ties (i.e., network relationships), and perspectives of the impact of SACYHN on organizational performance. To determine network ties, which were the causal factors, respondents were provided with a matrix listing all organizations in SACYHN and asked to identify which of six types of links (if any) their organization had with the other organizations in the network ‘‘over the past year.’’ Responses to this question were only counted if confirmed by both organizations. The network member organizations were the cases used in the fs-QCA analysis (Ragin, 2008) and the outcome, impact of the network on the organization, was based on the perceptions of the impact of SACYHN on the respondent’s organization. These perceptions were collected through a multiitem question where respondents scored each item on a five-item Likert scale (1 ¼ significant negative impact, to 5 ¼ significant positive impact). A performance scale was developed using exploratory factor analysis (orthogonal, varimax rotation) and factors were then combined into a single scale of organizational performance (12 items, reliability alpha ¼ .81). The original response scale from the questionnaire was then used to calibrate the outcome for the fs-QCA analysis. Since a value of 3 in the original response scale indicated performance that was ‘‘pretty much neutral’’ and 4 indicated an impact level that was ‘‘mostly positive,’’ response scores for the 12-item performance scale were calibrated into fuzzy sets using a three-point scale. The ‘‘fully in’’ value was set to 4, the cutoff value set to 3.5, and ‘‘fully out’’ set to 3.0. The three causal factors we examined were having a central network position, being a member of a cohesive subnetwork, and having a strong tie to the NAO. Central network position was measured by degree centrality, which is the total number of organizations with which an organization has a confirmed relationship. Organization degree centrality was calibrated to

The Configurational Approach in Organizational Network Research

239

reflect membership in the set of high-degree centrality, with the cutoff point between ‘‘in’’ and ‘‘out’’ being the mean of the raw degree centrality scores (mean=14), and ‘‘fully in’’ membership reflecting a degree centrality score equal to or greater than 1 standard deviation higher than the mean. The subnetworks were formal smaller networks based on geographical regions, thus membership in a subnetwork was defined by the formal network structure. The ties between subnetwork members, however, were based on the confirmed relationships. To measure the cohesiveness of the subnetworks, we used the ratio of within-group ties to between-group ties (Wasserman & Faust, 1994). Unlike the measure of centrality, this cohesiveness measure included the strength of ties to reflect relationship intensity. Specifically, we calculated densities using the total number of types of activities reported and confirmed divided by the total number of relationships possible ¼ 6(n(n1))/2. The cohesiveness of each of the subnetworks was calibrated to reflect highly cohesive subgroups. Two of the subnetworks were much more highly cohesive than the other two subnetworks, so this gap between the two groups was used as the cutoff point and the highest cohesiveness score was used to set ‘‘fully in’’ membership. The final causal condition was the strength of an organization’s connection to the SACYHN NAO. This value represented the total number of confirmed activities each organization had (again, 0 to 6) with the NAO, thereby representing not only the existence, but also the strength of the relationship. This causal condition was calibrated so that membership in the set reflected a strong tie. A strong tie was defined as having a link to the NAO through at least five of the six activities, with three being the cutoff between ‘‘in’’ and ‘‘out’’ membership.

Results Plotting the various conditions against the outcome, as discussed by Ragin (2000), no necessary conditions were found. The data were then analyzed using fs-QCA software which uses ‘‘truth tables,’’ showing all theoretically possible combinations of conditions that relate to the outcome. The truth table has 2k rows, with k representing the number of causal conditions, reflecting all possible combinations of causal conditions. In this case, there are 16 (24) possible combinations. The full truth table is presented in Appendix A. The table was minimized using the process described by Ragin (2008). A consistency cutoff of 84% was used for coding the outcome as present or absent. This consistency threshold was chosen based on an examination of the gaps in the distribution of the consistency scores and

JO¨RG RAAB ET AL.

240

Table 2.

fs-QCA Results (Outcome=Positive Network Impact on Organizational Performance).

Causal Recipes

DEGREE COHESIVENAO

Solution coverage Solution consistency

Description

Raw Coverage

Unique Coverage

Consistency

Number of Cases

High-degree centrality Cohesive subnetwork and strong tie to NAO

0.631

0.377

0.809

22

0.334

0.079

0.881

5

0.711 0.813

based on what is reasonable given the n of 47 (Ragin, 2009). The actual analysis was then performed on the minimized truth table to find the solutions for when the outcome was present. The result of the analysis (Table 2), which is the intermediate solution, indicates that there are two paths associated with the outcome. The first causal path shows that as hypothesized, organizations centrally positioned in the network viewed the network as having a positive impact on their organization. This causal path has the highest coverage with 22 of the cases fitting this combination, mostly large public organizations; however, the consistency is not as high as the consistency for the second path. The second path supports the second hypothesis, where organizations connected to a cohesive subnetwork and with a strong tie to the SACYHN NAO indicated the network’s impact on their organization was positive. This combination is highly consistent, though its unique coverage is small, explaining only five of the cases, which were mostly smaller rural organizations. Though this path does not explain many of the cases uniquely, the fact that the causal path emerges from the analysis suggests this way of connecting to a whole network is worthy of further examination, especially for considering how small rural organizations with less capacity may be able to sustain their participation in a network. Discussion This work demonstrated that in addition to having members that are highly central in the whole network, the network may also be able to affect

The Configurational Approach in Organizational Network Research

241

organizational performance positively through a structure based on cohesive subnetworks, but only as long as a broker, such as the NAO, creates a short path connecting the organizations in these subnetworks to the whole network. Previous network research has focused on the importance of centrality, which was also demonstrated here since most of the cases fit the first solution. What is lost with traditional statistics though is the alternative path. Prior work has found subgroup cohesiveness to be important as well, but our work and theorizing suggests cohesive subgroup membership is the so-called ‘‘INUS condition,’’ which means that it is only part of the solution. For an organization to reap the benefits of the whole network and not just the benefits of its local network, and to prevent network degeneration (Gulati, Sytch, & Tatarynowicz, 2012), cohesive subgroup membership alone is not a sufficient condition. A tie connecting an organization to the whole network is also needed, such as a strong tie to the NAO. By examining the relationship between network structural properties and the impact of the network on its member organizations using a configurational approach, we were able to compare alternative measures of embeddedness and connectedness (i.e., the organization’s ties to the network as a whole and its involvement in a cohesive subnetwork) within a single network. Rather than doing this in a traditional, linear fashion, through use of QCA we were able to test for equifinality (Fiss, 2007), and thus, determine that different combinations of tie patterns were associated with the same outcome; in this case, having a positive assessment of the impact of the network and its activities on organizational performance. Unlike the results that would have been found using traditional statistics (i.e., the significance of degree centrality), the configurational approach allowed us to determine the structural properties that can substitute for high-degree centrality, which is important for theory and practice. While all organizations seek high performance, not all have the capacity or desire to be centrally embedded in a whole network.

A CONFIGURATIONAL APPROACH TO WHOLE NETWORKS: STRUCTURE, CONTEXT, GOVERNANCE, AND EFFECTIVENESS In the second empirical example, we summarize a recently completed research project by Cambre´, Mannak, and Raab (2011) on the effectiveness of 39 interorganizational networks in the area of crime prevention in the

JO¨RG RAAB ET AL.

242

Netherlands, for which a crisp-set qualitative comparative analysis (cs-QCA) was conducted. The networks consisted of representatives of organizations in the field of crime prevention (like the police, the public prosecutor’s office, and rehabilitation agencies) and partners in the field of welfare services (like mental health care agencies, municipalities, and housing cooperatives). The general aim of these networks was to reduce crime and increase public safety. This was to be achieved by regularly exchanging information and integrating the services of the participating organizations. The research goal was to determine the configurations of factors that could explain the effectiveness of these crime prevention networks. The study was based primarily on prior work by Provan and Milward (1995) who emphasized structural and contextual factors in the explanation of network effectiveness, and work by Provan and Kenis (2008), who developed a conceptual framework to formulate propositions linking three modes (forms) of network governance with several crucial contingency factors to explain network effectiveness. A meta-analysis of the literature on network effectiveness by Turrini, Cristofoli, Frosini, and Nasi (2009) also helped guide the research. These authors grouped the explanatory factors for network effectiveness into categories, including network structural characteristics like integration mechanisms, network functioning characteristics like steering network processes, and network contextual characteristics such as system stability and resource munificence. The research question therefore was: What is the effect on network effectiveness of configurations based on network governance, network structure, system stability, network age, and resource munificence?

Data Collection and Measures Cross-sectional data were collected from the 39 networks using surveys and phone interviews focused on key network informants. Centralized integration, an aspect of network structure, was defined as the extent to which a network is integrated through one or a small number of highly central organizations, but not (simultaneously) through density. Network age, a factor not considered by Provan and Milward (1995), was measured as the elapsed time between the start-up of the network and the time of data collection. System stability was determined as the fluctuation of important organizations in and out of the network, the continuity of network coordination, and the impact of (internal or external) changes in the network. Resource munificence was measured as per capita resource availability

The Configurational Approach in Organizational Network Research

243

within each network’s geographical area (financial, material, and personnel input, provided by members or external parties). The form of network governance was determined by investigating whether control of decision making, supervision, and management of the network was formally exercised by one or two member organizations jointly (lead organization mode) or by a separate coordination agency (NAO mode). Data on all 39 networks were coded either 0 or 1. For instance, it became apparent that by three years of age a network could be regarded as fully operational. Networks younger than 3 years, therefore, were coded 0, and those three years and older, as 1. Networks that were rated as unstable systems were coded 0, and stable ones as 1. The same coding strategy was used for resource munificence, centralized integration, and effectiveness (low ¼ 0, high ¼ 1). Networks were coded as effective (the outcome variable) if they had operational procedures in place, achieved local network goals, and met or exceeded the national reduction rate in recidivism, as defined by the Dutch government. The 39 networks were governed either by a lead organization or by an NAO and were coded 0 or 1, respectively. The data were subsequently analyzed with cs-QCA (Rihoux & De Meur, 2009) as implemented in the computer program Tosmana 1.3.1.6 In brief, cs-QCA requires the construction of a dichotomous data table based on within-case knowledge from various sources. The procedure then deduces a set of necessary and sufficient conditions from this table leading to a certain outcome; in this case network effectiveness (Fiss, 2007).

Results The model that was analyzed contains five binary conditions. This leads to 32 (=25) possible configurations. Typically, however, not all configurations exist in empirical reality. The truth table (Appendix B) shows that 15 out of 32 configurations (47% of all possible configurations) were observed (17 theoretically possible configurations were not observed). The results reported in Table 3 demonstrate that networks that combine age (existing for at least three years) with high levels of system stability, high levels of centralized integration, and high resource munificence will be effective. In addition, networks existing for at least three years, with high levels of system stability, high levels of centralized integration, and NAO governance also will be effective, thus representing a second alternative pathway to network effectiveness. However, the first recipe has a higher coverage than the second one (raw 80%, unique 60% versus 40%/20%),

JO¨RG RAAB ET AL.

244

Table 3. Causal Recipes Effectiveness ASIR ASIG

Solution coverage

cs-QCA Results (Outcome=Network is Effective). Description

Raw Coverage

Unique Coverage

Number of Cases

Age combined with stability, centralized integration, and high resource munificence Age combined with stability, centralized integration, and NAO as governance mode

80%

60%

10 8

40%

20%

4

Solution Consistency 1.0

which indicates that the first recipe represents more cases, and thus, from an empirical perspective, has more explanatory power. Discussion The results offer some interesting insights based on the different configurations that lead to network effectiveness and provide a new and potentially valuable perspective on network governance. The results demonstrate, among other things, that despite the dominant claim in the literature about the flexibility of networks, it appears that at least for consciously created goal-directed networks, development time and stability are important necessary conditions for the effectiveness of networks. QCA was especially useful in this case, because it permitted the comparison and analysis of a relatively small number of cases, which is often the situation in whole network research; second, QCA made it possible to compare configurations of variables (network age, system stability, network structure, resource munificence, and the form of network governance), while taking equifinality into account (recipes which might lead to the same outcome) (Fiss, 2007); third, QCA helped to distinguish between necessary and sufficient factors (Fiss, 2007).

IMPLICATIONS In the chapter’s introduction, we stated that we see at least three main reasons why the configurational approach and QCA are of added value for

The Configurational Approach in Organizational Network Research

245

the research on interorganizational relations and networks. First, the configurational approach in general and QCA in particular can help researchers apply a holistic view of interorganizational networks while analyzing their data accordingly. The second empirical example we presented, in particular, demonstrates that interesting new theoretical insights can be gained by exploring organizational networks using complex recipes derived by the combination of causal conditions from a holistic perspective. In both of the examples presented, it was shown that outcomes occur not by single necessary factors, but by a combination of factors that lead to sufficient pathways, or INUS conditions. It is also important to note that with a configurational approach, it is not a matter of one combination leading to the outcome of interest, but rather, that multiple combinations can lead to the same outcome (i.e., equifinality). In addition, the first example demonstrated that the configurational approach enables researchers to compare possible alternative pathways from the outset, if theory suggests that certain combinations of factors might form a sufficient but not a necessary condition. These insights are highly relevant for management practitioners. One pathway may not always be possible so managers must look for alternatives. Second, we argued that the configurational approach and QCA are well suited for examining nested entities and for explaining outcomes with various factors that cross different levels of analysis – characteristics that are common to most interorganizational networks. The first example we presented, on the SACYHN network, shows the great potential of the configurational approach for theory development when attempting to integrate multiple levels of analysis (Moliterno & Mahony, 2011). Organizational networks consist of a series of nested routines, processes, and structures ranging from interpersonal relations to dyadic organizational ties, to triads and subnetworks, to entire networks of multilateral relations among organizations. In view of this complexity, the configurational approach offers a valuable way of exploring the combined effects of structural characteristics at different levels of analysis. This is not to say that the same cannot be accomplished with traditional statistical techniques, but combining the causal complexity of networks with the small N problem typical of most whole network studies limits the statistical possibilities. Thus, for organizational network research, the configurational approach provides a great opportunity to engage in more systematic theory building, especially when the research is characterized by one or a small number of cases. Third, we argued that QCA and the configurational approach can help fill one of the significant gaps in this line of research – the fact that certain core attributes of interorganizational networks and the context in which they

246

JO¨RG RAAB ET AL.

operate have only received scant attention. The second empirical case, for example, is a first attempt to combine attributes of networks such as age or the governance mode with structural characteristics like centralization and context variables like system stability for explaining network outcomes. Thus, broad contextual factors, like the network’s institutional environment, may now be examined empirically when comparing networks across countries or cultural areas. The configurational approach seems especially promising in the area of consciously created goal-directed networks, since one of the key aims is to explain network-level outcomes, such as network effectiveness, using configurations of factors that closely relate to the management of these networks. Scholars and practitioners alike are eager to learn more about how goal-directed networks function and what outcomes they can achieve under what circumstances. In particular, the configurational approach enables researchers to explore the possibility that pathways for network effectiveness and ineffectiveness are asymmetrical. That is, the pathway for network ineffectiveness may not simply be the opposite of the pathway for network effectiveness, but rather, involves a different combination of factors. This possible conclusion cannot be reached on the basis of analysis using traditional linear statistics, but it has considerable value both for building network theory and from a policy perspective. Under some conditions, avoiding being ineffective might be the best that can be achieved in highly complex networks, especially when network conditions or structures make significant change difficult. Recent calls for the reintegration of design questions in organization studies (Greenwood & Miller, 2010) and for greater emphasis on the practical relevance of management theories (Corley & Gioia, 2011) underscore the potential for the application of a configurational approach for the analysis of organizational networks. The strength of the configurational approach is in examining combinations of variables in connection with specific conditions and certain outcomes as cause–effect relationships (Fiss, 2011). Thus, theories based on this approach should help to answer important questions about the design, management, and governance of goal-directed interorganizational networks, which will have important implications for the formulation of network policy. From a methods point of view, we have argued that set-theoretic methods can be especially valuable for research that falls into cells 3, 6, 7, 8, and 9 of Table 1. These are cells where either the explanatory factor or the outcome is at the subnetwork or network level where the small N problem is likely to occur most often. We, therefore, believe that especially for research in which

The Configurational Approach in Organizational Network Research

247

the outcome is at the network level of analysis (Provan et al., 2007; Raab & Kenis, 2009; Turrini et al., 2009), the configurational approach has great potential in moving forward the current research agenda, both theoretically and methodologically. However, we would like to emphasize that settheoretic methods are not and should not be restricted to research in single cells, as for example demonstrated by Bensaou and Venkatraman (1995) and Bakker, Cambre´, Korlaar, and Raab (2011) who address outcomes in cells 4 and 5. Rather, it is research falling in cells 3, 6, 7, 8, and 9, for which QCA has the greatest added value compared to traditional statistical methods or recent stochastic models. Despite the fact that the configurational approach is only in its infancy in organizational network research, in terms of theory development, we do not see any restrictions for its application. Rather than suggesting, though, that use of more traditional techniques of analysis should be discontinued, we believe that the configurational approach and set-theoretic methods should be viewed as a systematic extension of qualitative comparative research and as a complement to standard statistical methods. In part, this is because, like other methods, there are limitations to QCA. There is some debate, for instance, whether QCA can be used to test theory since its purpose is not inference but causal description. Also, QCA was designed for, and has mostly been utilized in small to intermediate N studies using cross-sectional data. In this regard, it is important to note that QCA has recently been applied in large N contexts (Fiss, 2011) and there have been recent efforts to include temporal aspects in the analysis (Caren & Panofsky, 2005) (see the contribution by Greckhamer, Misangyi, and Fiss in this volume). We view the configurational approach and QCA as a stand-alone way of building theory on organizational networks, especially for research examining ‘‘whole’’ networks. The configurational approach and QCA can also be beneficial for examining interorganizational relations from egocentric and dyadic perspectives (cells 1, 2, 4, and 5). However, the value added by using QCA for these types of studies depends on the outcome of interest. For instance, if the research interest is on changes in network structure over time focusing on the evolution of dyadic ties, then using SIENA is a superior approach, as that is the type of research for which the program was designed. If, however, the goal is to explain overall network performance, the configurational approach and QCA may be a better way to build theory since it enhances an understanding of causal complexity and allows consideration of different possible solutions (equifinality). Having said that, we would like to emphasize that given the progress in the development of fs-QCA the combination of quantitative network analysis

JO¨RG RAAB ET AL.

248

and therefore measures like node centrality, network density, or centralization with set-theoretic methods are now possible. Even more, it opens up the possibility to jointly analyze numeric network indices and qualitative information on actors, relationships, and networks. Despite some limitations, adoption of a configurational approach using methods like QCA to study organizational networks has considerable appeal and can be an important method for analyzing network data and building new theory. Developing more methods alone is not necessarily what will move the study of networks forward, as Salancik (1995) noted. More than simply being an alternative method, the configurational approach and settheoretic methods like QCA provide the tools to examine important new research questions. Rather than trying to simplify networks, configurational approaches embrace their complexity and allow researchers to find the causal combinations that may more consistently explain network outcomes, taking into consideration variation in network contexts and nested levels of analysis. For many types of network research, especially when focusing on cliques, subnetworks, and whole networks, case or comparative case research may be all that is possible, given time and cost constraints. In these cases, the configurational approach provides the tools needed to advance theory on a range of organizational network topics and levels of analysis, thus contributing significantly to a broader and a deeper understanding of this important and sometimes under-theorized field of study.

NOTES 1. Whole networks are understood here as groups of three or more organizations that are consciously created and goal directed, and where multilateral, rather than solely egocentric, ties are critical for goal accomplishment (Provan et al., 2007). 2. The configurational approach is not to be confused with the use of organizational configurations (Short, Payne, & Ketchen, 2008), which have prominently figured in organization theory and strategic management as a way to create typologies that have helped to understand organizational structure and their performance since the 1960s. Even though the configurational approach lends itself to the analysis of organizational configurations, it has rarely been applied to the study of organizations until very recently (Fiss, 2007, 2011). 3. An exception to this rule is Uzzi’s (1997) excellent qualitative study on the paradox of embeddedness, in which he mainly focuses on the firm as the unit of analysis but also addresses network-level effects. 4. See http://www.compasss.org/pages/resources/software.html for an overview of software implementations. 5. We would like to emphasize that all the more traditional analytical methods mentioned above (both qualitative and quantitative) as well as set-theoretic methods

The Configurational Approach in Organizational Network Research

249

can be combined with techniques now available for the analysis of social network data (Wasserman & Faust, 1994). 6. http://www.tosmana.net/

REFERENCES Bakker, R. M., Cambre´, B., Korlaar, L., & Raab, J. (2011). Managing the project learning paradox: A set-theoretic approach toward project knowledge transfer. International Journal of Project Management, 29(5), 494–503. Bell, G. G., & Zaheer, A. (2007). Geography, networks, and knowledge flow. Organization Science, 18(6), 955–972. Bensaou, M., & Venkatraman, N. (1995). Configurations of interorganizational relationships – A comparison between US and Japanese automakers. Management Science, 41(9), 1471–1492. Borgatti, S. P., Everett, M. G., & Freeman, L. C. (2002). UCINet for Windows. Harvard: Analytic Technologies. Borgatti, S. P., & Foster, P. C. (2003). The network paradigm in organizational research: A review and typology. Journal of Management, 29(6), 991–1013. Brass, D. J., Galaskiewicz, J., Greve, H. R., & Tsai, W. (2004). Taking stock of networks and organizations: A multilevel perspective. Academy of Management Journal, 47(6), 795–817. Cambre´, B., Mannak, R., & Raab, J. (2011). Combining structure, governance and context: A configurational approach in explaining the effectiveness of Dutch safety houses. Paper presented at the 27th EGOS Colloquium Gothenburg, Sweden, July 7–9, 2011. Caren, N., & Panofsky, A. (2005). TQCA: A technique for adding temporality to qualitative comparative analysis. Sociological Methods & Research, 34(2), 147–171. Corley, K. G., & Gioia, D. A. (2011). Building theory about theory building: What constitutes a theoretical contribution. Academy of Management Review, 36(1), 12–32. Dekker, D., Krackhardt, D., & Snijders, T. A. B. (2007). Sensitivity of MRQAP tests to collinearity and autocorrelation conditions. Psychometrika, 72(4), 563–581. Fiss, P. C. (2007). A set-theoretic approach to organizational configurations. Academy of Management Review, 32(4), 1180–1198. Fiss, P. C. (2009). Case studies and the configurational analysis of organizational phenomena. In C. C. Ragin & D. Byrne (Eds.), The Handbook of Case Study Methods (pp. 424–440). Thousand Oaks, CA: Sage. Fiss, P. C. (2011). Building better causal theories: A fuzzy set approach to typologies in organization research. Academy of Management Journal, 54(2), 393–420. Freeman, L. C. (2004). The development of social network analysis. A study in the sociology of science. Vancouver, BC. Empirical Press. Gilsing, V., Nooteboom, B., Vanhaverbeke, W., Duysters, G., & Van den Oord, A. (2008). Network embeddedness and the exploration of novel technologies: Technological distance, betweenness centrality and density. Research Policy, 37(10), 1717–1731. Greenwood, R., & Miller, D. (2010). Tackling design anew. Getting back to the heart of organizational theory. Academy of Management Perspectives, 24(4), 78–88. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104(5), 1439–1493.

250

JO¨RG RAAB ET AL.

Gulati, R., Sytch, M., & Tatarynowicz, A. (2012). The rise and fall of small worlds: Exploring the dynamics of social structure. Organization Science, 23(2), 449–471. Human, S. E., & Provan, K. G. (2000). Legitimacy building in the evolution of small-firm multilateral networks: A comparative study of success and demise. Administrative Science Quarterly, 45(2), 327–365. Jarillo, J. C. (1988). On strategic networks. Strategic Management Journal, 9(1), 31–41. Kilduff, M., & Brass, D. J. (2010). Organizational social network research: Core ideas and key debates. Academy of Management Annals, 4, 317–357. Krackhardt, D. (1988). Predicting with networks: Nonparametric multiple regression analysis of dyadic data. Social Networks, 10(4), 359–381. Larson, A. (1991). Partner networks: Leveraging external ties to improve entrepreneurial performance. Journal of Business Venturing, 6(3), 173–188. Laumann, E. O., & Knoke, D. (1987). The organizational state: A perspective on the social organization of national energy and health policy domains. Madison, WI: University of Wisconsin Press. Laursen, K., & Salter, A. (2006). Open for innovation: The role of openness in explaining innovation performance among UK manufacturing firms. Strategic Management Journal, 27(2), 131–150. Lemaire, R. H., & Provan, K. G. (2010). Network governance, structure, and performance in a publicly funded child and youth health network. Paper presented at the 2010 Academy of Management Annual Meeting Montreal, Canada, August 6–10, 2010. Levin, D. Z., & Cross, R. (2004). The strength of weak ties you can trust: The mediating role of trust in effective knowledge transfer. Management Science, 50(11), 1477–1490. Lincoln, J. R., Gerlach, M. L., & Takahashi, P. (1992). Keiretsu networks in the Japanese economy – A dyad analysis of intercorporate ties. American Sociological Review, 57(5), 561–585. Meyer, A. D., Tsui, A. S., & Hinings, C. R. (1993). Configurational approaches to organizational analysis. Academy of Management Journal, 36(6), 1175–1195. Moliterno, T. P., & Mahony, D. M. (2011). Network theory of organization: A multilevel approach. Journal of Management, 37(2), 443–467. Nohria, N., & Eccles, R. G. (Eds.). (1992). Networks and organizations: Structure, form, and action. Boston, MA: Harvard Business School Press. Oliver, C. (1990). Determinants of interorganizational relationships – Integration and future directions. Academy of Management Review, 15(2), 241–265. Owen-Smith, & Powell, W. W. (2004). Knowledge networks as channels and conduits: The effects of spillovers in the Boston Biotechnology Community. Organization Science, 15(1), 5–21. Provan, K. G., Fish, M. A., & Sydow, J. (2007). Interorganizational networks at the network level: A review and commentary on the empirical literature. Journal of Management, 33(3), 479–516. Provan, K. G., & Kenis, P. N. (2008). Modes of network governance: Structure, management, and effectiveness. Journal of Public Administration Research and Theory, 18(2), 229–252. Provan, K. G., & Milward, H. B. (1995). A preliminary theory of interorganizational network effectiveness: A comparative study of four community mental health systems. Administrative Science Quarterly, 40(1), 1–33. Powell, W. W. (1990). Neither market nor hierarchy: Network forms of organization. In B. M. Staw & L. I. Cummings (Eds.), Research in organizational behaviour (Vol. 12, pp. 295–336). Greenwich, CT: JAI Press.

The Configurational Approach in Organizational Network Research

251

Powell, W. W., Koput, K. W., & Smith-Doerr, L. (1996). Interorganizational collaboration and the locus of innovation: Networks of learning in biotechnology. Administrative Science Quarterly, 41(1), 116–145. Raab, J., & Kenis, P. (2009). Heading towards a society of networks. Journal of Management Inquiry, 18(3), 198–210. Ragin, C. C. (1987). The comparative method: Moving beyond qualitative and quantitative strategies. Berkeley, CA: University of California Press. Ragin, C. C. (2000). Fuzzy-set social science. Chicago, IL: University of Chicago Press. Ragin, C. C. (2008). Redesigning social inquiry: Fuzzy sets and beyond. Chicago, IL: University of Chicago Press. Ragin, C. C. (2009). Qualitative comparative analysis using fuzzy sets (fs-QCA). In B. Rihoux & C. C. Ragin (Eds.), Configurational comparative methods: Qualitative comparative analysis (QCA) and related techniques (pp. 87–121). Thousand Oaks, CA: Sage. Ragin, C. C., & Strand, S. I. (2008). Using qualitative comparative analysis to study causal order. Sociological Methods & Research, 36(4), 431–441. Rihoux, B., & De Meur, G. (2009). Crisp-set qualitative comparative analysis (Cs-QCA). In B. Rihoux & C. C. Ragin (Eds.), Configurational comparative methods: Qualitative comparative analysis (QCA) and related techniques (pp. 33–68). Thousand Oaks: Sage. Robins, G. L., Pattison, P., Kalish, Y., & Lusher, D. (2007). An introduction to exponential random graph (p) models for social networks. Social Networks, 29(2), 173–191. Salancik, G. (1995). Wanted: A good network theory of organization. Administrative Science Quarterly, 40(2), 345–349. Schalk, J., Torenvlied, R., & Allen, J. (2010). Network embeddedness and public agency performance: The strength of strong ties in Dutch higher education. Journal of Public Administration Research and Theory, 20(3), 629–653. Short, J. C., Payne, G. T., & Ketchen, D. J. (2008). Research on organizational configurations: Past accomplishments and future directions. Journal of Management, 34(6), 1053–1079. Snijders, T. A. B., Steglich, C. E. G., & van de Bunt, G. G. (2010). Introduction to actor-based models for network dynamics. Social Networks, 32(1), 44–60. Sydow, J., & Windeler, A. (1998). Organizing and evaluating interfirm networks: A structurationist perspective on network processes and network effectiveness. Organization Science, 9(3), 265–284. O’Toole, L. J., & Meier, K. J. (2004). Public management in intergovernmental networks: Matching structural networks and managerial networking. Journal of Public Administration Research and Theory, 14(4), 469–494. Turrini, A., Cristofoli, D., Frosini, F., & Nasi, G. (2009). Networking literature about determinants of network effectiveness. Public Administration, 88(2), 528–550. Uzzi, B. (1997). Social structure and competition in interfirm networks: The paradox of embeddedness. Administrative Science Quarterly, 42(1), 35–67. Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications. Cambridge, MA: Cambridge University Press. Watts, D. J. (1999). Networks, dynamics, and the small world phenomenon. American Journal of Sociology, 105(2), 493–527. Zaheer, A., McEvily, B., & Perrone, V. (1998). Does trust matter? Exploring the effects of interorganizational and interpersonal trust on performance. Organization Science, 9(2), 141–159.

0 1 1 1 1 0 0 0

Degree

1 1 0 1 0 0 0 1

Cohesive

1 0 0 1 1 0 1 0

NAO 2 0 0 3 19 4 9 7

Number 1 1 1 1 1 0 0 0

OrgImp 0.936842 0.924812 0.894737 0.894242 0.842865 0.767059 0.694672 0.407273

Consist 0.896806 0.473685 0.675676 0.813664 0.783612 0.698171 0.582633 0.189055

Pre

APPENDIX A: TRUTH TABLE FOR SACYHN NETWORK STUDY

0.840166 0.438069 0.604552 0.727613 0.660479 0.535538 0.404739 0.076997

Product

252 JO¨RG RAAB ET AL.

253

The Configurational Approach in Organizational Network Research

APPENDIX B: TRUTH TABLE FOR CRIME PREVENTION NETWORKS IN THE NETHERLANDS 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1

0 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1

0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1

0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1

0 0 0 ? 0 ? 0 0 0 ? 0 ? 0 ? 0 ? ? ? ? ? ? ? ? ? ? ? 0 ? 0 1 1 1

CHAPTER 11 COUNTRY-SPECIFICITY AND INDUSTRY PERFORMANCE: A CONFIGURATIONAL ANALYSIS OF THE EUROPEAN GENERIC MEDICINES INDUSTRY Kalle Pajunen and Ville Airo ABSTRACT The identification of country-specific advantages for business activities is one of the most crucial issues of strategic management and international business literatures. We address this issue by examining location-specific conditions for a successful generic medicines industry within 24 European countries. The findings of our fuzzy-set qualitative comparative analysis show that there are no necessary conditions for the high performance or absence of the high-performance industry. By revealing the causal complexity related to the issue, however, we show that the high performance or lack of it results from a configuration of conditions. Specifically, we identify two sufficient causal configurations to both outcomes. These findings provide clear implications for generic medicines industry firms who are planning location choices of their operations. In addition, this study provides a methodological advancement to explain and understand

Configurational Theory and Methods in Organizational Research Research in the Sociology of Organizations, Volume 38, 255–278 Copyright r 2013 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0733-558X/doi:10.1108/S0733-558X(2013)0000038015

255

256

KALLE PAJUNEN AND VILLE AIRO

which country elements matter more, for what outcomes, and under what conditions. Keywords: Countries and locations; institutions; international strategy; fsQCA; generic medicines industry; industry performance

INTRODUCTION Recent research on international business and strategic management has argued for the importance of institutional and country contexts of business activities. Ghemawat (2003) claimed that gross-border integration of markets is far from complete and that more attention should be focused on the location-specifics of different countries as a core content of international business research (see also Dunning, 1998; Ricart, Enright, Ghemawat, Hart, & Khanna, 2004). In the same spirit, Peng, Wang, and Jiang (2008) underlined the need for institution-based view to explain what drives firm strategy and performance in international context. In addition, Jackson and Deeg (2008), by drawing from the comparative capitalism research, underlined how different institutional arrangements of countries have distinct strengths and weaknesses for different kinds of economic activity. Accordingly, it seems obvious that the competitive advantage of firms depends on how they are able to adapt to different institutional environments (cf. Schneider, Schulze-Bentrop, & Paunescu, 2010). Empirical studies of the country, institution, or context-based effects on firm or industry performance, however, are still rare. Yet, the existing evidence emphasizes the important role of these conditions. Christmann, Day, and Yip (1999) found that country conditions are the most important determinant of subsidiary performance, followed by industry structure, subsidiary strategy, and corporate characteristics. The findings of Makino, Isobe, and Chan (2004) also stress the importance of country effects. In addition, Chan, Isobe, and Makino (2008) showed that the different levels of institutional development of host countries have a different influence on the performance of foreign affiliates. Importantly, while it seems evident that the country context influences firm performance, the findings of earlier research indicate that this issue is far from an unambiguous. Indeed, a notable methodological problem in these earlier studies has been their inability to capture the causal complexity related to effects of location-specific conditions. That is, usually, no single country condition has

A Configurational Analysis of the European Generic Medicines Industry

257

an independent effect on performance, investments, and growth of firms and industries. Instead, the effect is conditional on the presence or absence of other conditions (Pajunen, 2008). Therefore, in order to understand how country conditions influence firm and industry performance, we cannot only examine the independent effects of single conditions (e.g., Christmann et al., 1999; Makino et al., 2004). Likewise, if we want to capture the effects of different country conditions, we cannot use a proxy or index that tries to capture all these different effects in a single measure (e.g., Chan et al., 2008) – reality is by far more complex. In this chapter, we seek to address this issue by examining how the country-specific conditions influence the performance of the generic medicines industry within 24 European countries. We do this by applying a set-theoretic approach based on fuzzy-set qualitative comparative analysis (fsQCA) (e.g., Ragin, 2000, 2008b). Specifically, fsQCA provides a methodological approach that enables the systematic comparison of different national contexts as configurations of causal conditions. Such examination enables us to better understand which country elements matter, for what outcomes, and under what conditions; an issue which is acknowledged as a challenging task for future international strategy research (Tong, Alessandri, Reuer, & Chintakananda, 2008). We focus on a set of European countries wherein the generic medicine industry firms operate under the guidance of European Medicines Agency (EMEA) (incl. Switzerland). By excluding for example Russia and Turkey, we can focus on countries that share a similar, basic regulatory environment for generic medicines. This case selection strategy enables clearer and more focused implications for policy makers and firms operating in these countries. In addition, it provides a scoping condition that aids in the selection of country conditions and their calibration into fuzzy-set scores. While the political rhetoric and public discussions related to generic medicines sometimes underline the dominant role of single conditions of countries, we show in this chapter that the location-specific advantages of the generic medicines industry are defined by the configurations of causally relevant conditions and that there can be different paths to the same outcome. That is, because of the interaction effects, equifinality, and the fact that the effect of a condition is often conditional on the presence or absence of other conditions, the causal relationships between country contexts and the industry performance are fairly complex. In particular, we identify that there are no necessary conditions for the high performance or absence of the high-performance industry and that there are two sufficient causal configurations to both outcomes. Altogether, our study

258

KALLE PAJUNEN AND VILLE AIRO

contributes to international business and strategy literature by showing how the configurational approach provides a needed and useful way to explain how location-specific conditions influence industry performance and opportunities.

COUNTRY-SPECIFIC CONDITIONS AND GENERIC MEDICINES INDUSTRY Generic medicines1 are growing in importance in industrialized countries simultaneously with the increasing overall usage of medicines. One general reason for this is the cost cutting benefits and savings that generic substitution offers. In other words, generic medicines may offer a way for governments to cut down their health expenses (DiMasi & Grabowski, 2007; European Commission, 2009; Hoffman, 2005; Karwal, 2009; Sheppard, 2009). However, most suggestions concerning the success factors of the generic medicines industry refer directly to the country-specific conditions of the activity. In the selection of these causal conditions for our analysis, we consulted, first, earlier research on institutions and comparative capitalism that has identified country-specific conditions that are likely to influence firm and industry performance. Hall and Soskice (2001), for example, show that coordinated and liberal market economies are supportive for different kinds of industries. Second, and more importantly, we identified conditions that have been acknowledged in earlier literature to be important from the perspective of European generic medicine industry. For example, in terms of economic legislation and regulation, we specifically focus on price regulation that is argued to be the most central form of nonmarket regulation in this industry in Europe (e.g., Economic Policy Committee, 2001; Kjoenniksen, Lindbaek, & Granas, 2006; Mrazek & Frank, 2004; Redwood, 2004). Building on this substantive knowledge, we excluded some important institutional conditions such as political stability since there are no major differences among this condition in our set of European countries. While we acknowledge that our selection strategy may not be totally flawless, we believe that the chosen five causal conditions are the most central country-specific conditions related to the performance of the generic medicine industry in Europe. Next, we discuss more specifically about these conditions and their selection criteria. We start from the broader institutional framework and consider the effects of nonmarket coordination

A Configurational Analysis of the European Generic Medicines Industry

259

and regulation. Thereafter, we turn to the roles of the public financing of health care, demographics of nations, and the wealth of the economies.

Causal Conditions Related to Institutional Framework The discussion of varieties of capitalism (Hall & Soskice, 2001) can be seen as a part of comparative capitalism research that refers to the diverse set of approaches and frameworks, which have a common concern in understanding the institutional foundations that affect the formation of diverse business organizations within different nations (Deeg & Jackson, 2007). In their original framework, Hall and Soskice (2001) distinguish two ideal types of political economies: liberal market economies (LMEs) and coordinated market economies (CMEs). This division is first and foremost based on the differences observed in the developed market economies of Western Europe and Northern America. In LMEs, companies coordinate their activities via hierarchies and competitive market arrangements. The market relationships are characterized by the arm’s-length exchange of products or services. In CMEs, companies are more dependable on nonmarket relationships and they use these relationships to conduct business and build their own competitiveness. According to Hall and Soskice (2001), superior macroeconomic performance is a product of institutional coherence of the nation. This means that both LMEs and CMEs can provide an institutional framework for creating comparative advantages for industries and firms, but a nation’s position in the middle of these frameworks, or having a mix of both coordinated and liberal market institutions, is likely to lead to underperformance (Hall & Gingerich, 2004; Hall & Soskice, 2001; Jackson & Deeg, 2008; Kenworthy, 2006). One of the key arguments of Hall and Soskice (2001) is that because of the aspects related to fluid decision making systems, job markets, and financing systems, LMEs provide comparative advantage for industries that depend on radical innovations. CMEs, in turn, are advantageous for industries that are depending more on incremental innovation. Consequently, we may assume that the performance of the generic medicines industry that is not directly dependent on radical innovations but instead benefits from continuous small scale improvements in production processes, is likely to be more successful in the institutional framework of coordinated than LMEs. Legislation and regulation also are institutional factors that are likely to have a notable influence on the generic medicines industry development.

260

KALLE PAJUNEN AND VILLE AIRO

Indeed, patent and IP-related legislation are central to the foundations of the generic medicine industry. Major changes in European legislation since the 1980s have essentially simplified the process of getting generic medicines to the markets. At the same time, homogeneous patent protection including patenting, supplementary protection certificates, data and market exclusivity, as well as marketing authorization gives all the actors in the market a similar operative environment (European Commission, 2009; Redwood, 2004). As a result, market access and market protection in Europe are predictable and do not provide any notable opportunity for comparative advantages. In short, the current state of European patent and IP regulation enables the activities of the generics medicines industry, but does not define its success. Differing from patent legislation, however, the rules and procedures regarding promotion of generic medicines are typically settled at the national level. In fact, there are not many policies or even guidelines that would be unified across Europe. The central regulative aspects that relate to the promotion of generic medicines include generic prescribing, generic substitution, and price regulation (e.g., Mrazek & Frank, 2004). Generic prescribing provides the opportunity for a physician to write prescriptions using generic or international nonproprietary names. Generic substitution, in turn, allows a pharmacist to change prescribed medicine to the generic equivalent. Generic substitution policies vary between European countries, but typically require the agreement from a prescribing physician at the very least (Mestre-Ferrandiz, 2003). Finally, price regulation includes two main aspects: reference price systems (RPS) and reimbursements with patient copayments. In RPS, health authorities set a maximum reimbursement level for a given medicine. This level is called the reference price. Usually reference prices are not set for individual medicines directly, but medicines are categorized in groups based on similarity in active ingredients, usage, dosage, or any combination of these (Mestre-Ferrandiz, 2003; Vogler et al., 2008). Reimbursement is the fraction of the actual medicine price that is paid by a third-party payer, namely, governments and insurance companies. When RPS is in use, a patient has the possibility to choose a product that is under or exactly the reference price, which requires no payment to be made by the patient. While other regulative aspects, generic prescribing and substitution, are usually necessary conditions for the use of RPS, patients’ acceptance of generic medicines is more directly related to financial benefits they get when substituting for generic medicines via RPS. Thus, price regulation via RPS

A Configurational Analysis of the European Generic Medicines Industry

261

can be seen as the foremost regulatory condition influencing on the success of the generic medicines industry. Possible variations in RPS systems between the countries are not considered that significant (Andersson, Sonesson, Petzold, Carlsten, & Lonnroth, 2005; Kjoenniksen et al., 2006; Lund-Jacobsen, 1992).

Causal Conditions Related to Public Health Care Expenditures, Ageing, and the Wealth of Nations One of the main reasons why generic medicines are important in Europe is the rising expenditures on pharmaceuticals. Pharmaceutical companies’ relative expenses per new medicine have risen because the research and development costs have increased and the number of new medicines or new molecular entities brought to the markets has declined. This is illustrated by the fact that the total costs of developing a new molecule have increased at an annual rate of 7.4% above the general price inflation from 1987 to 2001 (DiMasi, Hansen, & Grabowski, 2003). Subsequently, this trend has intensified (DiMasi & Grabowski, 2007). This development is problematic for many European countries as the majority of health care costs are paid by their national governments. Generic medicines, as cheaper alternatives to original trademarked medicines, are seen as a viable policy option to keep health care expenditures under control. Indeed, according to World Bank Health, Nutrition and Population (HNP) statistics, the public health expenditure (PHE) as a share of total health expenditures has increased from 75.7% in 2002 to 76.9% in 2006 in the group of EUR countries. Similarly, the share of PHE as a share of GDP increased from 7.2% to 7.5%. Another driving force behind the rising public expenditures on pharmaceuticals in several European countries is the ageing of the population. In several European countries, as well as in any other industrialized area of the world, the demographic structures of the societies are biased toward older age groups. One implication of this is that elderly people consume more health care resources than any of the younger age groups. For example, in EU member states health expenditures for citizens over 65 years are 30–40% of their total health expenditures. Consequently, it is likely that this demographic factor influences the demand of generic medicines and the success of the industry (Hoffman, 2005). Finally, we may consider that the level of national income affects the generic medicine industry success. In general, wealthier nations are found to

262

KALLE PAJUNEN AND VILLE AIRO

be healthier nations (Pritchett & Summers, 1996). One obvious reason for this is that wealthier countries can use more resources on health care and pharmaceuticals than poorer countries (Well, 2007). This general demand can also positively influence the demand of generic medicines. Yet, there is also evidence of the reverse causality. That is, earlier research has found that the health of the population can contribute positively on the wealth of the nation, for example, in terms of earnings and labor supply (e.g., Bloom, Canning, & Sevilla, 2004). In this sense, the relative demand for cheaper medicines in poorer countries that seek economic growth may even be higher than in wealthier countries. Altogether, the level of national income can be seen as a relevant factor that can influence the demand of generic medicines.

METHODS AND DATA As discussed above, we identified five central location-specific factors that may have an influence on the performance of the generic medicines industry in Europe. Some of the factors may have a stronger influence than the others. Yet, considering the findings of earlier research that emphasize the combinational effect of institutional conditions (Pajunen, 2008), we suggest that the influence of the location on industry performance is likely to be a joint effect of a combination of conditions. Inability to examine the effects of configurations of causal conditions, however, is one of the main challenges when using conventional statistical interaction models. Therefore, in order to examine this issue, we turn to settheoretic methods and in particular to fsQCA, which explicitly conceptualizes cases as combinations of attributes and emphasizes that it is these very combinations that give cases their unique nature (Fiss, 2011; Pajunen, 2008; Ragin, 2000, 2008b). Next, we discuss the analytic approach of fsQCA in the context of our study and explain how we operationalized the outcome factor (performance) and the five country-specific causal conditions.

FsQCA as an Analytic Approach The idea of fsQCA is to make comparisons between cases on a set of causal conditions to determine if their presence or absence is necessary or sufficient for a particular outcome. By definition, a causal condition is necessary if it is present in all instances of an outcome. In fsQCA, the causal condition X is

A Configurational Analysis of the European Generic Medicines Industry

263

necessary if the outcome Y can be considered a subset of the causal condition X. Sufficiency, in turn, implies that causal condition by itself can produce a certain outcome. That is, the causal condition X is a subset of the outcome Y. Altogether, as a notable advantage, fsQCA, by allowing the examination of the effects of combinations of conditions, takes into consideration those factors as causally relevant that alone are neither necessary nor sufficient. Regarding the analysis of country-level data, these features of fsQCA become especially advantageous as it allows conducting an analysis with fewer data points than with more traditional methods such as regression analysis and, in particular, it can handle multiple conjunctural causation (Fiss, 2007). Indeed, we may assume that there can be more than one ideal setting for the successful generic medicines industry: that is, the same outcome may be produced by different kinds of combinations of causal factors that are sufficient to the outcome (Ragin, 2008b). The analysis begins with selecting causal conditions to explain an outcome. In our study we selected five conditions. This number of causal conditions is consistent with the practicalities suggested for QCA (Marx, 2010). Thereafter, the values of these conditions should be calibrated into fuzzy-set membership scores ranging from 0 to 1 (see Table 1). In our study, we apply both direct and indirect methods of calibration described in detail by Ragin (2008b). In the direct method of calibration, the conditions need the specification of three anchor points of full membership (1), full nonmembership (0), and the crossover point (0.5). Thereafter, the values of conditions can be converted into a metric of log odds. For example, the full membership with set membership scores Z0.95 get the log odds of membership Z3.0 and full nonmembership with set membership scores r0.05 get the log odds of membership r–3.0. After the log odds transformation the membership scores are calculated using the formula where exponentiated log odds are divided by the unity plus the exponentiated log odds (Fiss, 2011; Ragin, 2008b). If finding the three important thresholds is problematic, the indirect method of calibration is used. It relies on the broad grouping of the cases (e.g., 1 fully in, 0.8 mostly but not fully in, 0.6 more or less in, 0.4 more or less out, 0.2 mostly but not fully out, and 0 fully out). The indirect method uses a fractional logit model to estimate the predicted set membership scores. The original data is used as the independent variable and the qualitative codings from the grouping as the dependent variable. This is implemented in STATA5 in the FRACPOLY procedure, which is indicated to be the simplest way to construct this transformation (Ragin, 2008b).

264

KALLE PAJUNEN AND VILLE AIRO

Table 1.

Fuzzy-Set Membership Scores of Causal Conditions.

Case

SUR

PHE

GNI

O65

CME

RPS

Austria Belgium Switzerland Czech Republic Germany Denmark Spain Finland France Greece Hungary Ireland Iceland Italy Lithuania Netherlands Norway Poland Portugal Romania Sweden Slovenia Slovak Republic United Kingdom

0.001 0.202 0.307 0.975 0.817 0.202 0.618 0.975 0.975 0.975 0.602 0.307 0.975 0.899 0.001 0.389 0.001 0.602 0.917 0.26 0.975 0.26 0.001 0.389

0.471 0.311 0.031 1 0.561 1 0.261 0.471 0.881 0.051 0.241 0.751 0.971 0.611 0.211 0.901 0.991 0.211 0.231 0.561 0.961 0.301 0.231 1

0.831 0.781 1 0.061 0.621 1 0.281 0.921 0.611 0.201 0.051 0.971 1 0.381 0.041 0.951 1 0.041 0.101 0.031 0.971 0.131 0.051 0.751

0.871 0.941 0.761 0.411 0.991 0.681 0.901 0.821 0.821 0.981 0.661 0.011 0.031 1 0.701 0.411 0.471 0.141 0.911 0.501 0.951 0.761 0.031 0.801

1 0.801 0.601 0.601 1 0.801 0.601 0.801 0.601 0.801 0.601 0.201 0.801 0.801 0.601 0.601 0.801 0.601 0.801 0.601 0.601 0.601 0.601 0.001

0.001 1 1 1 1 1 1 0.001 1 1 1 0.001 1 1 1 1 0.001 1 1 1 0.001 1 1 0.001

Note: Outcome factor: SUR, survival rate. Causal conditions: PHE, high public health expenditure; GNI, high gross national income; O65, high share of elderly people; CME, coordinated market economy; RPS, reference price system.

These fuzzy-set membership scores of causal conditions together construct a multidimensional vector space; in our case with 25 corners (5 denotes the number of causal conditions). Each of these corners represents one line in the truth table. A case is regarded as a member of the corner when it has a fuzzy-set membership score of more than 0.5 in that corner. After the truth table is constructed, the relevance of each possible configuration (i.e., each row) should be assessed with the empirical evidence. That is, the empirical cases are placed into the rows of this truth table based on their scores on various causal conditions. Some rows may contain many cases while some rows may not contain cases at all. In our study, configurations that have membership frequencies below the threshold of one are called logical

A Configurational Analysis of the European Generic Medicines Industry

265

remainders and are removed from the table because of lack of adequate empirical evidence.2 Thereafter, the necessity and sufficiency of the remaining configurations on the outcome can be assessed by using the measurements consistency and coverage. As defined by Ragin (2006), set-theoretic consistency measures the degree to which the cases sharing a given condition or configuration of conditions agree in displaying the outcome in question. Consistency scores can range from zero to one, where one indicates full consistency. In the analysis of necessity, it is recommended to use a demanding threshold for the level of consistency (e.g., Skaaning, 2011). Therefore, we set it to 0.90. Regarding the analysis of sufficiency, Ragin (2008a) recommends that a threshold value should be 0.75 or higher. In our study we use the value of W0.75. Coverage, in turn, measures the generality of the solution. Typically, in studies with a small number of cases and several causal combinations found, the coverage of a single combination is low (Ragin, 2006). We conducted the analysis of necessary and sufficient conditions by using the software package intended for fsQCA (www.fsqca.com). Regarding the analysis of sufficiency, the analysis produces two main solutions called ‘‘complex’’ and ‘‘parsimonious’’ based on the acceptance of ‘‘easy’’ and ‘‘difficult’’ counterfactuals. The complex solution aims to capture the full complexity and diversity which is present in the data. The parsimonious solution is a streamlined version of the complex solution (Ragin, 2008a). However, this reduction in complexity requires the incorporation of simplifying assumptions that entail ‘‘difficult’’ counterfactuals, as does also the possible ‘‘intermediate’’ solutions for this analysis (Ragin & Sonnett, 2004). A researcher may have a reason to provide these assumptions if complex solutions are too complex (Fiss, 2011).

Measurement of Outcome Factor The outcome factor, industry performance, is operationalized by using survival rate data of the generic medicines companies in a given country. Survival rate (SUR) as a measurement is able to capture the long-term success of the industry as it calculates all the entries and exits. It is also one of the most used measurements of firm performance (Klepper, 2002). Regarding other potential industry performance measurements, an entry rate would have been largely biased because several companies have origins long before the start of the modern generic medicines industry. They are in a sense de alio or diversifying entrants. The number of entries also has

266

KALLE PAJUNEN AND VILLE AIRO

diminished in recent years. Thus, entry rate-based metrics would not describe the current situation that well. Furthermore, density is a valid measurement at the European level, but when certain countries (like Finland) have only one entrant, using a density measurement at the country level was not justified. Survival data is based on a database conducted as part of the GloStraresearch program.3 The database is not publicly available. Company data was originally extracted from a number of sources. The main source was the yearly member information of European Generic medicines Association (EGA). Another major source for company information was the LexisNexis newsfeed, which was partly processed with the custom made software. This information was supplemented from company websites. After the companies were identified, the database was supplemented with the incorporation dates as well as the exit dates of the applicable companies. The main source for this data was the AMADEUS database. In addition, company websites were used to fill some data gaps. Some concerns regarding the reliability of this data may rise due to the use of multiple sources as well as the manual work required to build this database. Altogether, to conduct the survival rate for the companies in the country, the following formula was used:   Exits Survival rate % ¼ 12  100% Entries The highest score is 100%, which is obtained with at least one entry and no exits. The lowest score is 0%, which is obtained if exits are equal to entries. If there are no entries, this formula cannot be used. However, zero entry basically means that there is no generic medicines industry and a country is fully out of the set representing a successful industry. This data was encoded in fuzzy sets by using the indirect method. A country is considered fully in a set if the survival rate is 100%. A case is fully out of the set if the survival rate is 20% or below or there have been no entries. In these cases, the majority of companies have exited or there is no generic industry. Therefore, in qualitative coding we used the following six categories: 100% survival rate scores 1; 80% and above scores 0.8; 60% and above scores 0.6; 40% and above scores 0.4; 20% and above scores 0.2; and below 20% as well as countries with no entries scores 0. For the STATA analysis, countries with no entries were encoded with 0% survival rate. After that, in the above described way, the interval-scale survival data and qualitatively coded groupings were processed in STATA, which gave the predicted set membership scores that were used in the analysis.

A Configurational Analysis of the European Generic Medicines Industry

267

Measurements of Causal Conditions The first causal condition is the fuzzy set of countries with a CME reflecting the effect of the coordination in the national economy on the performance of the industry. The majority of the CME scores come from coordination index by Hall and Gingerich (2004) from 1990 to 1995. They give coordination scores for these countries in the closed interval from 0 to 1, where 1 is totally coordinated and 0 is totally uncoordinated or liberal. Countries are set at this interval indicating the level of the coordination they have in their economic system. The fuzzy-set calibration of the CME was obtained using the indirect method, but with certain modifications. First, countries with a coordination index were transformed to six fuzzy-set classes. In this division, 1 indicates a CME with Germany and Austria having full scores. For uncoordinated or a liberal market economy, only the United Kingdom scores 0. Other scores were linearly transformed. At this point, the indirect method was terminated and these scores were used as the fuzzy-set scores. Termination was done because there was not enough statistical data to construct the final scores. However, these types of scores without any transfer procedure have been used previously in fsQCA studies (e.g., Katz, Vom Hau, & Mahoney, 2005). Therefore, this is not seen as a major limitation. Hall and Gingerich (2004) exclude nine case countries: Czech Republic, Greece, Hungary, Iceland, Poland, Slovak Republic, Lithuania, Slovenia, and Romania. Of the unanalyzed countries, Central and Eastern European countries were set to 0.6 based on Lane’s (2005) suggestion that indicates that these countries have relatively high state coordination.4 Iceland was set to 0.8 to correspond with the majority of the Nordic countries. Furthermore, Greece was set to 0.8 based on the analysis by Knell and Srholec (2006). Their analysis indicates Greece to have the highest level of coordination of all European countries with the index value of 11.6. Germany and Austria, that score the highest in the index of Hall and Gingerich, score now 4.8 and 3.8 respectively. Knell and Srholec analysis would lead to Greece getting a fuzzy score of 1, but as the main data is from Hall and Gingerich and we want to follow that as closely as possible, the full score of 1 to Greece is discarded. However, it is understood that Greece has a high level of coordination and gets a score of 0.8. The second causal factor reflects the effect of the regulation in the national economy on the performance of the industry and is based on the data whether an RPS is in place. RPS data is mainly from Pharmaceutical Pricing and Reimbursement Information (PPRI) from the year 2007

268

KALLE PAJUNEN AND VILLE AIRO

(Vogler et al., 2008; Vogler, Espin, & Habl, 2009). RPS is will be used as a crisp set, where 1 indicates the presence of an RPS and 0 indicates absence of it. In a fuzzy-set analysis, the use of crisp sets is possible, as crisp sets are a subset of fuzzy sets. The third causal factor reflects the effect of the PHE on the industry performance. PHE scores are based on the data set of PHE (% of total health expenditure) in 2006 provided by the World Bank Health, Nutrition and Population statistics. PHE was configured using the direct method. The threshold for being completely out of the set is set to 61.6%. This is the average share of expenditure in the reference group of high-income countries. Below the average there cannot be a high share of PHE. A country was encoded to be fully in the set if its share was above 81.3%, which is Japan’s share; considered to be an example of a country with a high share of public expenditure in health care. The share of 81.3% is also close to the 75th percentile of the case countries (75th percentile is between the values of 80% and 81.7%). The crossover point was set to about 76%, which is the 50th percentile of case countries. The fourth causal condition is the fuzzy set of countries with a high share of elderly people (hereafter O65). It reflects the effect that elderly people have on the performance of the industry. It is based on the data set of the share of the population aged 65 years and over in 2007 provided by the World Bank. The fuzzy set of O65 is configured using the direct method similar to PHE. A country is considered fully in the set if the share of the elderly people exceeds 17.7%. This was the average share of the population aged 65 years and over in the area of countries using the Euro in 2007 and sets in the case countries to about the 90th percentile. A country is considered fully out of the set if the share falls below 12.3%. This is the value of the United States and it is categorically low for an industrialized economy. The crossover point is set to 14.9%, which was the average share of elderly people in the high-income countries of the world in 2007. The final causal condition is the fuzzy set of countries with high gross national income (GNI). It reflects the effect that the level of the income has on the performance of the industry and is based on the data set of gross national income per capita in current US dollars provided by the World Bank. The data is from the year 2007 except in the case of the United Kingdom, which is from 2006. The GNI is configured using the direct method. A case is considered to be completely out of the set if the GNI per capita is below $11,906. This is the threshold that World Bank uses for highincome countries. A case is considered to be fully in the set if the GNI per capita exceeds $46,040, which was the GNI per capita in the United States in

A Configurational Analysis of the European Generic Medicines Industry

269

2007. This value is close to the 75th percentile of the case countries. The crossover point is set to $37,572, which was the average GNI per capita for high-income countries in 2007.

RESULTS We report our analysis and results in two sections. We start by considering whether any of the five causal conditions, and their negations, is necessary to account for a high-performance generic medicines industry. We also conduct analyses considering if there are necessary conditions for the absence of a high-performance generic medicines industry. Thereafter, we examine potential causal combinations of conditions that can be considered sufficient for the presence and absence of a high-performance industry.

Analysis of Necessary Conditions As noted earlier, a causal condition is ‘‘necessary’’ if the instances of the outcome form a subset of the instances of the causal condition. We follow here Ragin’s (2008a) measure of consistency of necessary condition and use a threshold consistency score of 0.90 to assess whether the observed proportion significantly exceeds this benchmark. Thus, we consider that causal condition is ‘‘almost always necessary’’ if the consistency score exceeds the score of 0.90. As the results in Table 2 clearly show, we find no condition for a highperformance industry that would exceed the threshold of 0.90. Similarly, as indicated by Table 3, there are no necessary conditions for the absence of a high-performance industry. These findings are not surprising. In our initial assumption we did not expect that a single location-specific condition would be necessary for high performance. Therefore, to explain and understand how different locations influence generic medicine industry performance, we need fuzzy-set analysis of sufficient conditions.

Analysis of Sufficient Conditions Table 4 shows the results of a fuzzy-set analysis for a high-performance industry. The analysis results in two complex solutions and one parsimonious solution for high-performance industry that exhibit acceptable

270

KALLE PAJUNEN AND VILLE AIRO

Table 2. Condition PHE phe GNI gni O65 o65 CME cme RPS rps

Table 3. Condition PHE phe GNI gni O65 o65 CME cme RPS rps

Necessary Conditions for High Performance. Consistency

Coverage

0.67 0.51 0.57 0.55 0.79 0.40 0.81 0.45 0.79 0.21

0.64 0.60 0.56 0.62 0.64 0.59 0.65 0.70 0.55 0.44

Necessary Conditions for Absence of High Performance. Consistency

Coverage

0.62 0.58 0.63 0.51 0.70 0.51 0.78 0.51 0.71 0.29

0.53 0.61 0.56 0.51 0.51 0.69 0.56 0.71 0.45 0.56

consistency (i.e., consistency Z0.75) and one parsimonious solution with consistency o0.75. Also, the overall consistency score of the parsimonious solutions can be considered to be a bit too low (i.e., 0.72). Altogether, there seem to be two sufficient configurations of location-specific conditions that are favorable for the generic medicines industry. Specifically, solution 1 indicates that CME combining a high level of public health care expenditure and price regulation via price reference system is sufficient for achieving high performance. Denmark and Iceland are typical examples of such locations. Solution 2, however, indicates that there is another sufficient path to high performance via the configuration of CME, price reference system, high share of elderly people, and the absence of high national income. Greece, Italy, and Portugal are typical examples of

271

A Configurational Analysis of the European Generic Medicines Industry

Table 4.

Configurations for High Performance.

Complex Solution 1: CME  RPS  PHE 2: gni  O65  CME  RPS

Raw Coverage

Unique Coverage

Consistency

0.44 0.44

0.16 0.16

0.76 0.82

Raw Coverage

Unique Coverage

Consistency

0.20 0.23

0.77 0.74

Overall coverage: 0.60 Overall consistency: 0.76 Parsimonious Solution 3: gni  O65 4: PHE  RPS

0.46 0.50

Overall coverage: 0.70 Overall consistency: 0.72

Table 5.

Configurations for Absence of High Performance.

Complex Solution

Raw Coverage

Unique Coverage

Consistency

0.14 0.23

0.14 0.23

0.87 0.90

Raw Coverage

Unique Coverage

Consistency

0.10 0.28

0.77 0.84

1: PHE  GNI  o65  rps 2: phe  gni  o65  CME  RPS Overall coverage: 0.37 Overall consistency: 0.89 Parsimonious Solution 3: o65  rps 4: phe  o65

0.14 0.31

Overall coverage: 0.42 Overall consistency: 0.83

such locations. The only acceptable parsimonious solution emphasizes the importance of solution 2. The overall solution coverage that indicates the percentage of cases that take these paths to the outcome is 60% for complex solutions and 70% for parsimonious solutions, suggesting a notable effect on performance. Table 5 shows the results of a fuzzy-set analysis regarding the absence of a high-performance industry. The analysis results again in two complex and parsimonious solutions to the outcome that exhibit acceptable consistency. Parsimonious solution 3 indicates that the absence of a high share of elderly people and the absence of an RPS is a sufficient configuration for the

272

KALLE PAJUNEN AND VILLE AIRO

absence of a high-performance industry. Complex solution 1 further specifies that this is the case especially in high-income countries with a high level of public health care expenditure. Ireland is an example of such a location. Parsimonious solution 4 provides another configuration for the absence of high-performance including the absence of a high level of public health care expenditure and the absence of high share of elderly people. Complex solution 2 further specifies that this is the case especially in CMEs with an RPS and the absence of high income. Slovakia is an example of such a location. The overall solution coverage for complex solutions is 37% and for parsimonious solutions 42% indicating that the idiosyncrasy within configurations that lead to the absence of high performance is somewhat higher than within configurations that lead to high performance.

DISCUSSION AND CONCLUSIONS Even if the speed of global market integration has been notable during the last decades, we are still living in a world of various cultures, institutional frameworks, and economic conditions (Ghemawat, 2003). This heterogeneity essentially explains why some locations are attracting industrial activities related to basic production while, at the same time, other locations are occupied by advanced services or high-tech R&D firms. Multinational firms are also taking advantage of these differences by separating their diverse activities and affiliates to most appropriate geographic locations (Jackson & Deeg, 2008; Makino et al., 2004). The existence of some location-specific conditions may be necessary for the success of all kinds of business activities. However, most conditions are likely to be related to the more specific attributes and requirements of specific industries. Accordingly, the understanding of which kind of location is advantageous for a specific kind of business activity is one of the most crucial questions in international business and strategic management. This chapter has addressed this issue by examining country-specific conditions for a successful generic medicine industry in the context of 24 European countries.

Implications for Generic Medicines Industry Building on the specific attributes of this industry, and in particular on the location-specific factors that have been connected to the generic medicines industry in earlier literature (e.g., Economic Policy Committee, 2001;

A Configurational Analysis of the European Generic Medicines Industry

273

Mrazek & Frank, 2004), we identified five location-specific conditions that may have an influence on industry performance. The findings of fuzzy-set analysis show that none of these conditions alone is necessary for high performance or absence of high performance. Addressing the causal complexity related to the issue, even in the simplest logically possible form, we found that high performance or lack of it results from a configuration of conditions. Specifically, we identified two sufficient paths to both outcomes. Considering the configurations for a high-performance generic medicines industry, both complex solutions include the existence of an RPS and CME. The parsimonious solution, in turn, underlines the importance of the high share of elderly people together with the absence of high national income. The role of demographic factors seems to be particularly significant as all of the configurations leading to the absence of high industry performance also include the absence of a high share of elderly people. Because Europe is ageing fairly rapidly (e.g., Carone & Costello, 2006), the overall prospects for the generic medicines industry can be seen as advantageous. While the existence of the RPS and CME are conditions included in both of the complex solutions, they do not guarantee the existence of a highperformance industry. As the analysis of sufficient conditions shows, together with the absences of a high share of elderly people, high national income, and a high share of public health care expenditure, these conditions may also lead to the absence of a successful industry. The situation is basically comparable also concerning the absence of high national income. Yet, as the parsimonious solution for high performance indicates, a relatively poorer country can provide an advantageous environment for generic medicines. This finding supports the reasoning that especially in the relatively poorer countries, the increasing group of elderly people both need more medicines and prefer the low-cost alternatives. We argue that the understanding of the complex nature of locationspecificity is important for the medicine industry companies that operate in both areas of innovative and generic medicines and especially for those who are planning location choices of their operations. In fact, our study indicates that medicine firms have arbitrage possibilities – that is, exploiting differences among countries – even in the context of Europe. The situation regarding other geographical and economic contexts (e.g., Asian and African countries) may, of course, be different. In addition, while we believe that the chosen five conditions are the most central country-specific conditions related to performance of the generic medicine industry in Europe, these conditions may not be equally relevant in other contexts.

274

KALLE PAJUNEN AND VILLE AIRO

Implications for Research on International Business and Strategy The findings of this study also show that the fsQCA, and configurational methods in general, provide a needed and useful methodological advancement for international business and strategy research to explain how location-specific conditions influence firm performance and opportunities. Specifically, earlier research has examined this issue by using regression models (Chan et al., 2008; Christmann et al., 1999). However, as indicated in earlier research, country contexts consist of intertwined institutional conditions, where the effect of one condition is likely to depend on the presence or absence of the effects of other conditions (Pajunen, 2008). Thus, examining the net effects of conditions may give a significantly distorted view of whether or not a particular location is appropriate for a particular type of firms. Indeed, Tong et al. (2008) noted that an essential and challenging task for future research on firm growth and performance would be to understand which country elements matter more, for what outcomes, and under what conditions. These are exactly the questions that fsQCA is able to address in a way that acknowledges the inherent causal complexity and equifinality of the issue. For example, the findings of this study indicate that there is more than one optimal country environment for generic medicine firms in Europe: both Mediterranean and Northern European countries can provide comparative advantages for generic medicines industry firms and the existence of the RPS is not a necessary condition for a good location. International business and strategy research is full of causally complex questions that are difficult to examine by using regression models and estimations of interaction effects of independent variables. As countries are not alike, nor are industries and firms. A specific country context may support the performance of some industries while at the same time induce an adverse effect regarding some other industry (cf. Porter, 1990; Ricart et al., 2004). Therefore, in terms of the future research, configurational methods can be suggested to provide a useful way to examine varieties in performance within and between industries by focusing on the effects of arrangements of causal conditions rather than on any individual factors or variables. We also consider that it would be useful for future research to apply configurational theory and methods to deepen our understanding of the interaction effects between country, industry, and firm conditions relative to firm performance (cf. Tong et al., 2008). However, while fsQCA seems to be a promising methodological approach in explicating the complex causal

A Configurational Analysis of the European Generic Medicines Industry

275

mechanisms behind firm and industry performance, it is not intended to substitute statistical methodologies; examination of independent effects of causal conditions remains as an important issue in international business and strategy research and this can be better implemented by using statistical approaches.

NOTES 1. A generic medicine is produced without a patent protection as the original medicine’s patent has expired. It contains the same quantity of active substances as the original medicine, differing only in terms of the name, appearance, and packaging (EMEA, 2007). In this study, we focus on the producers of generic prescription medicines that have a marketing authorization for their generic prescription medicines in the European Union. This means that companies focusing solely on over-the-counter and other self-treatment medicines as well as hospital medicines are excluded. 2. Studies with small number of cases, as our study, start with the threshold of one and when the number of cases notably increases (e.g., hundreds of cases) a higher frequency threshold should be established (for further discussion see Ragin, 2008a). 3. See more about the GloStra-research program at www.glostra.fi. 4. We acknowledge that the score of 0.6 is a conservative estimate of the level of coordination in these countries.

REFERENCES Andersson, K., Sonesson, C., Petzold, M., Carlsten, A., & Lonnroth, K. (2005). What are the obstacles to generic substitution? An assessment of the behaviour of prescribers, patients and pharmacies during the first year of generic substitution in Sweden. Pharmacoepidemiology and Drug Safety, 14(5), 341–348. Bloom, D. E., Canning, D., & Sevilla, J. (2004). The effect of health on economic growth: A production function approach. World Development, 32(1), 1–13. Carone, G., & Costello, D. (2006). Can Europe Afford to Grow Old? Finance and Development, 43(3), 8–31. Chan, C. M., Isobe, T., & Makino, S. (2008). Which country matters? Institutional development and foreign affiliate performance. Strategic Management Journal, 29(11), 1179–1205. Christmann, P., Day, D., & Yip, G. S. (1999). The relative influence of country conditions, industry structure, and business strategy on multinational corporation subsidiary performance. Journal of International Management, 5, 241–265. Deeg, R., & Jackson, G. (2007). Towards a more dynamic theory of capitalist variety. SocioEconomic Review, 5(1), 149. DiMasi, J., & Grabowski, H. (2007). The cost of biopharmaceutical R&D: Is biotech different? Managerial and Decision Economics, 28(4–5), 469–479.

276

KALLE PAJUNEN AND VILLE AIRO

DiMasi, J., Hansen, R., & Grabowski, H. (2003). The price of innovation: New estimates of drug development costs. Journal of Health Economics, 22(2), 151–185. Dunning, J. H. (1998). Location and multinational enterprise: A neglected factor? Journal of International Business Studies, 29(1), 45–66. Economic Policy Committee. (2001). Budgetary challenges posed by aging populations: The impact on public spending on pensions, health and long-term care for the elderly and possible indicators of the long term sustainability of public finances. Executive Summary of a report of Economic Policy Committee of the European Commission to European Council No. EPC/ECFIN/630-EN final. EMEA. (2007). Questions and answers on generic medicines. European Medicines Agency. Retrieved from http://www.emea.europa.eu/pdfs/human/pcwp/39390506en.pdf European Commission. (2009). Pharmaceutical sector inquiry. Final Report. Retrieved from http://ec.europa.eu/competition/sectors/pharmaceuticals/inquiry/index.html Fiss, P. C. (2007). A set-theoretic approach to organizational configurations. Academy of Management Review, 32(4), 1180–1198. Fiss, P. C. (2011). Building better causal theories: A fuzzy set approach to typologies in organization research. Academy of Management Journal, 54(2), 393–420. Ghemawat, P. (2003). Semiglobalization and international business strategy. Journal of International Business Studies, 34(2), 138–152. Hall, P., & Gingerich, D. (2004). Varieties of capitalism and institutional complementarity in the macroeconomy. Discussion Paper No. 04/5, MPlfG Discussion Paper, Max Planck Institute for the Study of Societies, Cologne. Hall, P., & Soskice, D. (2001). Varieties of capitalism: The institutional foundations of comparative advantage. New York, NY: Oxford University Press. Hoffman, J. (2005). Generics growth in the USA and the EU: The role of India. Journal of Generic Medicines, 3(1), 7–19. Jackson, G., & Deeg, R. (2008). Comparing capitalisms: Understanding institutional diversity and its implications for international business. Journal of International Business Studies, 39(4), 540. Karwal, V. (2009). Mergers acquisitions update: Changing the strategic paradigm in the global generics market. Journal of Generic Medicines, 6(4), 315–322. Katz, A., Vom Hau, M., & Mahoney, J. (2005). Explaining the great reversal in Spanish America: Fuzzy-set analysis versus regression analysis. Sociological Methods & Research, 33(4), 539. Kenworthy, L. (2006). Institutional coherence and macroeconomic performance. SocioEconomic Review, 4(1), 69. Kjoenniksen, I., Lindbaek, M., & Granas, A. (2006). Patients’ attitudes towards and experiences of generic drug substitution in Norway. Pharmacy World & Science, 28(5), 284–289. Klepper, S. (2002). Firm survival and the evolution of oligopoly. RAND Journal of Economics, 33(1), 37–61. Knell, M., & Srholec, M. (2006). Emerging varieties of capitalism in central and eastern Europe. In D. Lane & M. Myant (Eds.), Varieties of capitalism in post-communist countries. Basingstoke: Palgrave. Lane, D. (2005). Emerging varieties of capitalism in former state socialist societies. Competition and Change, 9(3), 227–247.

A Configurational Analysis of the European Generic Medicines Industry

277

Lund-Jacobsen, B. (1992). Generic substitution – Patients’ incentives are too low. Ugeskrift for laeger, 154(46), 3215. Makino, S., Isobe, T., & Chan, C. M. (2004). Does country matter? Strategic Management Journal, 25(10), 1027–1043. Marx, A. (2010). Crisp-set qualitative comparative analysis (csQCA) and model specification: Benchmarks for future csQCA applications. International Journal of Multiple Research Approaches, 4(2), 138–158. Mestre-Ferrandiz, J. (2003). Reference prices and generic medicines: What can we expect? Journal of Generic Medicines, 1(1), 31–38. Mrazek, M., & Frank, R. (2004). The off-patent pharmaceutical market. In E. Mossialos, M. Mrazek & T. Walley (Eds.), Regulating pharmaceuticals in Europe: Striving for efficiency, equity and quality (Vol. 1, pp. 245–259). Berkshire: Open University Press. Pajunen, K. (2008). Institutions and inflows of foreign direct investment: A fuzzy-set analysis. Journal of International Business Studies, 39(4), 652–669. Peng, M. W., Wang, D. Y. L., & Jiang, Y. (2008). An institution-based view of international business strategy: A focus on emerging economies. Journal of International Business Studies, 39, 920–936. Porter, M. E. (1990). The competitive advantage of nations. New York, NY: Free Press. Pritchett, L., & Summers, L. H. (1996). Wealthier is healthier. Journal of Human Resources, 31(4), 841–868. Ragin, C. C. (2000). Fuzzy-set social science. Chicago, IL: University of Chicago Press. Ragin, C. C. (2006). Set relations in social research: Evaluating their consistency and coverage. Political Analysis, 14(3), 291–310. Ragin, C. C. (2008a). Qualitative comparative analysis using fuzzy sets (fsQCA). In B. Rihoux & C. C. Ragin (Eds.), Configurational comparative methods: Qualitative comparative analysis (QCA) and related techniques. Thousand Oaks, CA: Sage. Ragin, C. C. (2008b). Redesigning social inquiry: Fuzzy sets and beyond. Chicago, IL: University of Chicago Press. Ragin, C. C., & Sonnett, J. (2004). Between complexity and parsimony: Limited diversity, counterfactual cases, and comparative analysis. In S. Kropp & M. Minkenberg (Eds.), Vergleichen in der Politikwissenschaft. Wiesbaden: VS Verlag fur Sozialwissenschaft. Redwood, H. (2004). Innovation patents, generics and payers: The political dynamics of interaction. Journal of Generic Medicines, 1(3), 249–258. Ricart, J. E., Enright, M. J., Ghemawat, P., Hart, S. L., & Khanna, T. (2004). New frontiers in international strategy. Journal of International Business Studies, 35(3), 175–200. Schneider, M. R., Schulze-Bentrop, C., & Paunescu, M. (2010). Mapping the institutional capital of high-tech firms: A fuzzy-set analysis of capitalist variety and export performance. Journal of International Business Studies, 41, 246–266. Sheppard, A. (2009). Opportunities and challenges in an ever generic world. Journal of Generic Medicines, 6(4), 303–313. Skaaning, S. E. (2011). Assessing the robustness of crisp-set and fuzzy-set QCA results. Sociological Methods & Research, 40(2), 391–408. Tong, T. W., Alessandri, T. M., Reuer, J. J., & Chintakananda, A. (2008). How much does country matter? An analysis of firms’ growth options. Journal of International Business Studies, 39, 387–405.

278

KALLE PAJUNEN AND VILLE AIRO

Vogler, S., Espin, J., & Habl, C. (2009). Pharmaceutical pricing and reimbursement information (PPRI) – New PPRI analysis including Spain. Pharmaceuticals Policy and Law, 11(3), 213–234. Vogler, S., Habl, C., Leopold, C., Rosian-Schikuta, I., de Joncheere, K., & Lyager Thomsen, T. (2008). PPRI report. Directorate-General Health and Consumer Protection, Austrian Federal Ministry of Health, Family and Youth, European Commission, Vienna. Well, D. N. (2007). Accounting for the effect of health on economic growth. Quarterly Journal of Economics, 122(3), 1265–1306.

CHAPTER 12 APPLYING FUZZY SET METHODOLOGY TO EVALUATE SUBSTITUTES FOR LEADERSHIP J. Lee Whittington, Victoria McKee, Vicki L. Goodwin and R. Greg Bell ABSTRACT Transformational leadership has been found to positively influence employee attitudes and behaviors. However, research also has shown that a variety of task and motivational factors lead to similar outcomes. Yet, little research has explored the potential interaction of transformational leadership with these other factors. We utilize fuzzy-set/qualitative comparative analysis to explore the ways these factors may interact to produce positive employee outcomes. Specifically, we found that high levels of employee commitment and performance can be achieved in the absence of a transformational leader through various ‘‘bundles’’ of enriched jobs, challenging goals, and high quality leader–follower relationships. Keywords: Transformational leadership; substitutes for leadership; fuzzy set

Configurational Theory and Methods in Organizational Research Research in the Sociology of Organizations, Volume 38, 279–302 Copyright r 2013 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0733-558X/doi:10.1108/S0733-558X(2013)0000038016

279

280

J. LEE WHITTINGTON ET AL.

INTRODUCTION Over the past three decades, both empirical research and the popular press have championed transformational leadership as a key to high levels of employee performance  and for good reason. Transformational leadership (TFL) has consistently been linked to high levels of in-role performance (Howell & Avolio, 1993; Whittington, Goodwin, & Murray, 2004) organizational citizenship behaviors (OCBs) (Whittington et al., 2004) and employees’ affective commitment to the organization (Bycio, Hackett, & Allen, 1995; Whittington et al., 2004). Yet, research has also shown that task and motivational factors lead to similar outcomes. Regardless, scholars have yet to fully investigate the variety of ways in which transformational leadership, task, and motivational factors could potentially complement the effectiveness of each other. Moreover, leadership researchers have not uncovered how various combinations, or ‘‘bundles,’’ of task and motivational factors could produce high levels of employee commitment and performance in the absence of a transformational leader. Despite its growing use in organizational research, Short and colleagues (2008) suggest that ‘‘the use of configurational logic is noticeably absent in leadership research to date’’ (p. 1070). In this chapter, we provide a framework and offer a methodological tool that researchers can use to address this void. Specifically, we assess how ‘‘bundles’’ of leadership, task, and motivational factors interact with one another to lead to positive employee attitude and performance outcomes. Previous research has evaluated each of these factors and demonstrated that in certain circumstances, task and organizational factors may moderate, or may, in fact, substitute for transformational leadership (Podsakoff, MacKenzie, & Bommer, 1996; Whittington et al., 2004). Yet, due to the reliance on hierarchal regression techniques, published research does not reflect results that describe how certain fully, interactive combinations of leadership, task, and motivational factors impact employee attitudes and behaviors differently, nor how they relate to one another. In this chapter, we draw upon the substitutes for leadership model as a comprehensive framework to ground our illustrative example. Specifically, we review prior research on job characteristics and goal difficulty because of the similarity in outcomes associated with these motivational strategies and those associated with transformational leadership. In addition, we consider LMX theory (e.g., Liden, Wayne, & Stilwell, 1993) as a basis for examining the leader–follower relationship and its influence on critical

Applying Fuzzy Set Methodology to Evaluate Substitutes for Leadership

281

follower outcomes, such as performance, OCBs, and affective commitment. Prior studies demonstrate that high quality leader–follower relationships have very similar follower outcomes to those that occur when the leader is transformational. We then use the fuzzy-set qualitative comparative analysis (fs/QCA) technique to identify the combinations of task and motivational factors that lead to high levels of employee performance, citizenship behaviors, and affective commitment. Our example not only helps illustrate the equifinality of various leadership, task, and organizational characteristic combinations, it also suggests a number of directions for future research. Indeed, the results demonstrate the usefulness of configurational methods in examining the complexity of leadership in organizations.

THE SUBSTITUTES FOR LEADERSHIP MODEL The substitutes for leadership model (Howell, Dorfman, & Kerr, 1986; Kerr & Jermier, 1978) provide a useful framework for identifying potential alternative paths to the positive outcomes associated with transformational leadership. The model was developed to describe conditions under which the relationships between leadership and a variety of outcomes may be modified in some way (neutralized or enhanced) or may be duplicated by substituting another variable in place of leadership (Howell et al., 1986; Keller, 2006; Podsakoff, Niehoff, MacKenzie, & Williams, 1993). In their original conception of the substitutes for leadership model, Kerr and Jermier (1978) suggested that certain task characteristics (e.g., those that provide feedback and that are intrinsically satisfying) and organizational characteristics (e.g., formalization in terms of explicit goals) may substitute for the effects of a leader or perhaps neutralize the impact of a leader’s behavior. Furthermore, Griffin (1982) suggests that the quality of the leader–follower relationship is also a factor that impacts employee attitudes and performance. In support of these suggestions, job enrichment, challenging goals, and the quality of the relationship between leaders and followers have been shown to have relationships with employee attitudes and behaviors similar to those associated with transformational leadership (Dienesch & Liden, 1986; Griffin & McMahan, 1993; Hackman & Oldham, 1976; Locke & Latham, 1990). In the following sections we discuss how each of these factors has been shown to produce high levels of employee performance, OCB, and affective commitment.

282

J. LEE WHITTINGTON ET AL.

TASK DESIGN Enriched jobs have been found to provide intrinsically satisfying tasks (Griffin, 1991) which were identified in the substitutes for leadership model. The most extensive treatment of these job characteristics was developed by Hackman and Oldham (1976). According to their model, there are five core job characteristics (task variety, task significance, task identity, autonomy, and feedback) that, if present, can produce a variety of positive results through their effect on critical psychological states. These results include high levels of performance, high internal motivation, low levels of absenteeism, and low turnover rates. Using the Hackman and Oldham (1976) model, Griffin (1991) found enriched jobs to be positively related to each of these outcomes. Recently, Piccolo, Greenbaum, Den Hartog, and Folger (2010) found task significance and autonomy to be positively related to OCB. According to Podsakoff et al. (1996), the relationship between task characteristics and OCBs is positive. Furthermore, they demonstrated that both task characteristics and transformational leadership behaviors are positively related to performance, organizational commitment, and OCB. Their results led them to conclude that models that do not include both task characteristics and transformational leadership are misspecified. To address this criticism, Whittington et al. (2004) examined job enrichment and transformational leadership together in a field study. These authors found that transformational leadership and enriched jobs were each positively related to employee in-role performance, OCB, and affective organizational commitment. Furthermore, they found that job enrichment substituted for the effects of transformational leadership on affective commitment. In sum, considerable research indicates that task design leads to similar outcomes as transformational leadership. Therefore, it may provide an alternative to transformational leadership for obtaining positive outcomes in organizations.

GOAL SETTING Goal setting is identified as an aspect of organizational formalization in the substitutes model. The impact of goal setting on employee performance has been documented on a wide variety of tasks in both laboratory and field

Applying Fuzzy Set Methodology to Evaluate Substitutes for Leadership

283

settings (Locke & Latham, 1990; Mento, Steel, & Karren, 1987). Latham and Locke (1991) found that specific, difficult goals lead to high performance creating a cascading effect such that employees are more satisfied on the job and less likely to leave the organization. This ‘‘high performance cycle’’ may lead to performance beyond expectations, extrarole behaviors, and commitment to the organization as does transformational leadership. More recently, research has shown that goal setting may achieve similar results as transformational leadership for both affective commitment and performance. For example, Whittington et al. (2004) found that goal difficulty enhanced the relationship between transformational leader behavior and both affective commitment and performance. They concluded that difficult goals may provide a clarification for the abstraction and longer-term perspective that is associated with transformational behavior, thus increasing its link to these outcomes.

THE QUALITY OF THE LEADER–FOLLOWER RELATIONSHIP Leadership is a process that takes place in the context of a relationship between the leader and the follower. No theory of leadership addresses the relational nature of leadership more directly than Leader–Member Exchange (LMX) theory (Dienesch & Liden, 1986; Gerstner & Day, 1997; Graen & Uhl-Bien, 1995; Liden et al., 1993; Liden, Sparrowe, & Wayne, 1997; Schriesheim, Castro, & Cogliser, 1999; Scott & Bruce, 1998; Zhou & Schriesheim, 2009). LMX theory focuses on differences in the quality of the relationship between a leader and his/her followers. Followers in highquality relationships are likely to receive assignments to interesting and desirable tasks, have greater responsibility and authority delegated to them, more information shared with them, participate in making some of the leader’s decisions, and receive personal support and approval (Yukl, 2012). Research shows that followers who have high-quality relationships with their leaders tend to perform well (Bauer, Erdogan, Liden, & Wayne, 2006; Ilies, Nahrgang, & Morgeson, 2007; Walumbwa, Cropanzano, & Hartnell, 2009; Wang, Law, Hackett, Wang, & Chen, 2005), are more committed to the organization (Bauer et al., 2006), and perform more extra-role behaviors (i.e., OCBs; Hackett, Farh, Sung, & Lapierre, 2003; Ilies et al., 2007; Uhl-Bien, 2006; Wang et al., 2005).

284

J. LEE WHITTINGTON ET AL.

Furthermore research supports a positive relationship between LMX and transformational leadership (Howell & Hall-Merenda, 1999; Wang et al., 2005). Consequently, a high-quality leader–follower relationship may provide for the same positive follower outcomes that transformational leadership does. As this review has demonstrated, there are multiple avenues for achieving high levels of in-role and extra-role performance, as well as high levels of affective commitment. Unfortunately, research has not explored how these factors may operate in conjunction with one another, which may be due to methodological limitations. In the following section, we investigate this issue and introduce the potential for studying leadership and its substitutes from a configurational perspective.

A CONFIGURATIONAL APPROACH TO SUBSTITUTES FOR LEADERSHIP There is a growing recognition among organizational researchers that many factors contribute to performance and they should not be evaluated in isolation from each other. Unfortunately, nonsupportive results in studies that have attempted to evaluate these factors in combination (e.g., Avolio, Walumbwa, & Weber, 2009; Dionne, Yammarino, Atwater, & James, 2002) may be due to limitations of moderated multiple regression (Villa, Howell, Dorfman, & Daniel, 2003). An examination of these variables as ‘‘bundles’’ may provide a better opportunity for the identification of substitutes among them (Aguilera, Filatotchev, Gospel, & Jackson, 2008; Hoskisson, Castleton, & Withers, 2009; Ward, Brown, & Rodriguez, 2009). Furthermore, understanding the simultaneous operation of multiple factors has potential significance for managers’ day-to-day responsibilities (Rediker & Seth, 1995; Walsh & Seward, 1990). A configurational approach provides a deductive method for addressing these issues (Greckhamer, Misangyi, & Fiss, 2013) and may be used to answer two important related questions. First, how do organizational, task, employee–supervisor relationship characteristics, and transformational leadership combine with one another to achieve high levels of employee performance, commitment, and OCB? Second, are there configurations of task design, goal setting, and leader–follower relationships that can enable firms to achieve these behavioral and attitudinal outcomes in the absence of transformational leadership?

Applying Fuzzy Set Methodology to Evaluate Substitutes for Leadership

285

Fs/QCA is quite helpful in addressing these two questions. Fs/QCA’s approach to causality, referred to as multiple conjunctural causation, has three important implications for leadership research. First, an outcome can be produced by multiple conditions. Second, fs/QCA recognizes there can be more than one combination of conditions that lead to the outcome under investigation. Indeed, equifinality is a central element to fs/QCA. Third, fs/QCA allows for outcomes to occur as a result of the presence of a condition (e.g., the presence of transformational leader behavior), or because of the absence of a condition (e.g., the absence of leader–member exchange). Hence, configurational approaches relax some of the assumptions normally associated with many quantitative techniques, such as permanent causality, additivity, and causal symmetry. Conjunctural causation is particularly important when it is likely that there can be multiple reasons to bring about an outcome, and when causal conditions could combine in unique and multiple ways to bring about an outcome. Fs/QCA also offers a number of advantages over regression analysis, a method scholars often rely upon to evaluate leadership outcomes (Avolio, 2010; Bass & Riggio, 2006). First, evaluating how two or more factors interact to produce an outcome can be quite challenging with regression (Braumoeller, 2004). Second, if a factor does influence an outcome in only a handful of cases, it can become masked or invisible because of the overlap in variance explained by two or more variables. Fs/QCA helps overcome these issues by ignoring variation and distribution in variables, and by not isolating the net independent effect of each variable on an outcome (Ragin, 2006). Indeed, fs/QCA is not centered upon variable distributions and the search for patterns of covariation, difference, or frequency clusterings (Ragin, 2006). Rather, the technique is quite helpful in evaluating both the number and complexity of alternative paths leading to a desired outcome. In the remainder of this chapter we utilize fs/QCA to determine whether high levels of employee commitment and performance can be achieved in the absence of a transformational leader through various ‘‘bundles’’ of enriched jobs, challenging goals, and high-quality leader–follower relationships (LMX).

A FIELD STUDY APPLICATION OF FS/QCA TO LEADERSHIP SUBSTITUTES These subjects for this example were drawn from 12 different organizations representing a variety of industries (e.g., manufacturing, governmental

286

J. LEE WHITTINGTON ET AL.

agencies and departments, and health care), departments (e.g., production, accounting, and personnel), and organizational levels (ranging from firstline supervisors to company president). We examined data from 209 leader– follower dyads. No assumptions were made relative to their leadership skills; rather we chose them for participation in our study as ‘‘leaders’’ based on their formal positions. Our analysis begins by identifying the various factors that work in combination to influence citizenship behaviors, affective commitment, and employee performance. We then calibrate our raw data into crisp sets and fuzzy sets (Ragin, 2008). Calibrated measures are often found in fields like chemistry, astronomy, and physics (Ragin, 2008). However, it is still uncommon to see calibrated measures in social sciences despite their usefulness. Crisp sets assign membership into either ‘‘fully in’’ or ‘‘fully out’’ status. The presence or absence of affective commitment would be an example of a crisp set. Fuzzy sets, however, allow researchers to account for the varying degree of membership of cases to a particular set by using the anchors of 1 to designate ‘‘fully in’’ a particular set, 0 for nonmembership, and 0.5 as the point of maximum ambiguity to mean neither in, nor out, of a particular set. The crossover point (or the point of maximum ambiguity) designates when a case is more in or more out of the set. Because of how fuzzy sets are calibrated, they are more than an ordinal scale that lists rank order. That is, a fuzzy set ‘‘can be seen as a continuous variable that has been purposefully calibrated to indicate degree of membership in a well defined and specified set’’ (Ragin, 2008, p. 26).1 After specifying full membership, full nonmembership, and the crossover point of maximum ambiguity regarding membership, the transformation of a variable into a set measure is relatively simple. Fs/QCA (2.0) automates the calibration process and produces rescaled measures ranging from 0 to 1 with the converted scores tied to the thresholds of full membership, full nonmembership, and the crossover point. To accomplish meaningful calibrations, Ragin (2008) emphasizes the importance of both substantive and theoretical knowledge when calibrating measures and translating them into set membership scores (Ragin, 2000). In the following section, we describe how we captured each of the variables of interest in our study, and how we arrived at the breakpoints for set membership.

OUTCOME CONDITIONS Subordinate performance was measured using a fixed-sum-weighted, Likerttype interval scale on three dimensions: quality of work, quantity of work,

Applying Fuzzy Set Methodology to Evaluate Substitutes for Leadership

287

and promotability to the next level. For each of the items, managers were asked to indicate the percentage of time the subordinate’s performance fell into one of four categories (Category 1 – ‘‘Unsatisfactory: Needs to improve substantially’’; Category 2 – ‘‘Questionable: Needs some improvement’’; Category 3 – ‘‘Satisfactory: Meets normal expectations’’; Category 4 – ‘‘Outstanding: Substantially exceeds normal performance)’’. We coded Category 1 as 0, Category 2 as 1, Category 3 as 2, and Category 4 as 3, weighting each category by the percent of time reported by the manager (accounting for 100% of the subordinate’s performance time). The resulting score for the employee on each dimension was the weighted average category level. Summing the weighted evaluations of the three performance dimensions created a composite score for performance (a=0.92). Based on the four-category scale of subordinate performance, we created a measure of membership in the set of managers reporting their subordinates exhibited a high degree of subordinate in-role performance. We designated managers’ responses with ‘‘Questionable: Needs some improvement’’ (Category 2) to be fully out of the set. Those managers’ responses with ‘‘Outstanding: Substantially exceeds normal performance’’ (Category 4) were considered fully in the set. We defined the crossover point in this set to correspond to managers’ responses with ‘‘Satisfactory: Meets normal expectations’’ (Category 3). Manager evaluations of subordinates’ OCB were obtained from a 24-item social report scale adapted from Podsakoff, MacKenzie, Moorman, and Fetter (1990). Each of these items was measured using a seven-point Likert scale ranging from (1) Strongly Disagree to (7) Strongly Agree. Total score measures were used in the analysis, and were obtained by averaging responses across all items (a=0.95). These scores ranged from a minimum of 10.8 to a maximum of 34.75, with a median of 24.2. We utilized these points to create breakpoints for membership in the set of subordinates exhibiting high citizenship behaviors. Those scores in the 75th percentile were coded as ‘‘fully in,’’ whereas those in the 25th percentile were coded as ‘‘fully out.’’ As a crossover point, we chose the 50th percentile. Affective organizational commitment was measured using subordinate responses to the eight-item affective commitment dimension in the organizational commitment scale developed by Allen and Meyer (1990). Each item was measured on a 7-point Likert scale ranging from 1 (Strongly Disagree) to 7 (Strongly Agree) (a=0.83). Based on the 7-point affective commitment scale, we created a measure of membership in the set of individuals with high degree of affective commitment in which a response of (2) was coded as fully out, and a response of (6) considered fully in the set. The crossover point for this set corresponded to the middle of the scale (4).

288

J. LEE WHITTINGTON ET AL.

PREDICTOR CONDITIONS Transformational leader behavior was assessed by subordinates’ responses to the subscales in the Multifactor Leadership Questionnaire (MLQ-5X; Bass & Avolio, 1994). We incorporated the modified scale definitions validated by Goodwin, Wofford, and Whittington (2001), which assigns implicit psychological construct items from the contingent reward subscale to transformational leadership. Transformational leadership scores were summations from responses to 42 items (a ¼ 0.97). The scale ranged from a minimum of 9.4 to a maximum of 44.56, with a median of 29.57. We utilized these points to create breakpoints for membership in the set of leaders exhibiting high transformational leadership. Scores in the 75th percentile were coded as ‘‘fully in,’’ whereas those in 25th percentile were coded as ‘‘fully out.’’ As a crossover point, we chose the 50th percentile. Subordinate perceptions of LMX were measured using a six-item scale to assess the subordinates’ perceptions of their relationship with their manager. Each of these items was measured using a 7-point Likert scale ranging from (1) Strongly Disagree to (7) Strongly Agree. This scale was developed by Liden et al. (1993). A sample item from the LMX scale is ‘‘I can count on my supervisor to ‘bail me out,’ even at his or her own expense, when I really need it.’’ A total score for LMX was derived by averaging across the six items (a ¼ 0.91). Based on the 7-point LMX scale, we created a measure of membership in the set of subordinates perceiving a high degree of leader–member exchange with their supervisors. Responses of (2) were designated as ‘‘fully out’’ of this set, while responses of (6) were considered ‘‘fully in’’ the set. The crossover point corresponded to the middle of the scale (4). Goal difficulty was measured using a four-item scale developed by Steers (1973). A sample item from this scale is ‘‘My work objectives will require a great deal of effort from me to complete them.’’ Subordinate responses to these items were measured on a 7-point Likert scale ranging from (1) ‘‘Strongly Disagree’’ to (7) ‘‘Strongly Agree.’’ A total score for goal difficulty was derived by averaging across the four items (a ¼ 0.83). Using these responses we created a measure of membership in the set of individuals reporting a high goal difficulty. Responses of (2) were designated as ‘‘fully out’’ of this set, whereas responses of (6) were considered ‘‘fully in’’ the set. The crossover point corresponded to the middle of the scale (4). Task characteristics were measured by subordinate responses to items from the Job Diagnostic Survey (JDS; Hackman & Oldham, 1976). The JDS

Applying Fuzzy Set Methodology to Evaluate Substitutes for Leadership

289

measures task variety, autonomy, feedback, significance, and identity. Each task characteristic is measured using three items. Following Griffin (1991), the subscales of the JDS were combined into an overall motivating potential score (MPS). MPS was calculated as: (Skill Variety+Task Identity+Task Significance/3)  Autonomy  Feedback). The higher the MPS score, the more enriched the job is. The scale ranged from a minimum of 14 to a maximum of 117, with a median of 63.56. We utilized these scores to create breakpoints for membership in the set of subordinates reporting highly enriched jobs (MPS). Those scores in the 75th percentile were coded as ‘‘fully in’’ the set of subordinates with enriched jobs, whereas those in the 25th percentile were coded as ‘‘fully out.’’ As a crossover point, we assigned the 50th percentile.

ASSESSING NECESSARY CONDITIONS After identifying and calibrating our variables of interest, the next step is to test whether any predictor or contextual condition is necessary for high levels of employee outcomes. An argument for causal necessity can be supported when it can be demonstrated that instances of an outcome (dependent variable) constitute a subset of instances of a causal condition (independent variable). A consistency score of 1 indicates that the combination of causal conditions fulfills the criterion across all the cases. The more cases fail to meet the consistency criterion and the larger the distance from meeting the criterion, the further the consistency score will fall below 1. Conventionally, a fuzzy set variable, or crisp set variable, is called ‘‘necessary’’ or ‘‘almost always necessary’’ if the consistency score meets or exceeds the threshold of 0.90 (Ragin, 2006). The measure to evaluate the relevance of a necessary condition is the coverage rate. Trivially necessary conditions would yield a coverage rate near 0 (Ragin, 2006). Table 1 contains the results of our tests of necessity upon the presence of our conditions (capitalized) and their negation (not capitalized). None of our conditions exceeds the 0.90 threshold for the subordinate performance, citizenship behavior, and affective commitment outcome conditions. Therefore, no single leadership condition can be regarded as necessary for these performance outcomes to occur. The presence of LMX assumes the highest consistency values of 0.75 for the subordinate performance and 0.77 for the affective commitment outcome.

290

J. LEE WHITTINGTON ET AL.

Table 1. Analysis of Necessary Conditions. Condition

TFL tfl MPS mps LMX lmx GOAL goal

Performance

OCBs

Affective Commitment

Consistency

Coverage

Consistency

Coverage

Consistency

Coverage

0.51 0.63 0.46 0.68 0.75 0.44 0.53 0.59

0.42 0.57 0.40 0.59 0.44 0.72 0.46 0.51

0.48 0.60 0.45 0.64 0.71 0.42 0.53 0.55

0.42 0.57 0.40 0.58 0.43 0.72 0.47 0.49

0.50 0.71 0.49 0.76 0.77 0.57 0.60 0.62

0.30 0.47 0.30 0.47 0.32 0.68 0.37 0.38

ASSESSING SUFFICIENT CONDITIONS The next step in fs/QCA analysis involves analyzing whether any combination of causal conditions are sufficient for our three outcome conditions. The assessment of the sufficiency of causal combinations is carried out with fs/QCA’s Truth Table Algorithm. Fs/QCA’s truth table function generates a list of different combinations of leadership conditions that are sufficient for a particular outcome to occur (Ragin, 2008). Our analysis involves four leadership conditions which results in 24 or 16 potential combinations of these causal conditions. We generate truth tables and observe 12 out of 16 logically possible causal combinations for each of our outcome conditions. The program works by using fuzzy membership scores to weigh the relevance of each case with the result being an index of consistency. The truth table’s measure of consistency is similar to significance in that it provides a signal that an empirical connection exists and that it merits additional attention by the researcher (Ragin, 2008). We adopt Ragin’s recommendations of a consistency cutoff value of 0.75 when evaluating sufficiency solutions. Consistency scores below 0.75 are considered too low and inconsistent to draw any meaningful conclusions (Rihoux & Ragin, 2009). We assigned a value of 1 to those combinations in the truth table above the 0.75 consistency score, whereas those combinations below this level are assigned 0. We then reduce the truth table rows into more simplified combinations. It is at this point that researchers should address limited diversity and logical remainders. Limited diversity refers to instances where the configuration of conditions and outcomes across the empirical cases is not very diverse and,

Applying Fuzzy Set Methodology to Evaluate Substitutes for Leadership

291

therefore, leaves a large proportion of combinations ‘‘blank’’ (without empirical referents). Logical remainders are those logical configurations of conditions that are not empirically present in the dataset in relation to the presence or absence of the outcome of interest. Fs/QCA provides three different approaches to deal with logical remainders: a ‘‘complex solution,’’ ‘‘a parsimonious solution,’’ and an ‘‘intermediate solution.’’ All three solutions are valid; however, the difference between each solution lies in the simplifying assumptions researchers make about the hypothetical outcome of the configurations that do not have any corresponding empirical cases. Ragin (2008) suggests that parsimonious solutions are too parsimonious in that they make whatever counterfactual assumptions are necessary to produce the simplest solutions. Unfortunately, these solutions can occur at the expense of important conditions (e.g., necessary conditions). In other words, any remainder that will help generate a logically simpler solution is used without evaluating their plausibility. On the other hand, the complex solution does not incorporate any simplifying assumptions in the final solution because all remainders are set to false, which leads to no counterfactuals being allowed. Hence, no remainders are used. We follow Ragin (2008) who advocates the use of the intermediate solutions because logical remainders can be restricted to those that are the most plausible. Specifically, by using the intermediate solution, researcher can choose three different options, the first being the presence of the conditions, the second being the absence of the conditions, and the third being the inclusion of either presence or absence of conditions. The choice among these three options is driven by the researcher’s knowledge and theory regarding the conditions under investigation. Considerable research has shown that the presence of leadership factors is relevant for achieving high subordinate performance, high OCBs, and high affective commitment. Therefore, we have included each leadership condition as present. Hence, we do not incorporate into the reduction process so-called ‘‘difficult counterfactuals.’’ Difficult counterfactuals would be those that would lead to the elimination of leadership conditions from the solution. The final step in fs/QCA is interpreting the results. Reduction of the truth table shows several useful statistics. Overall solution consistency indicates the degree to which the subset relationship holds for sufficiency. The overall solution coverage refers to the joint importance of all causal paths (Schneider, Schulze-Bentrop, & Paunescu, 2010). Unique coverage is useful because it illustrates the relative weight of each path by measuring the degree of empirical relevance of a certain cause or causal combination to explain the outcome. The unique coverage is derived because outcomes are

292

J. LEE WHITTINGTON ET AL.

usually explained by more than one expression or causal path. Furthermore, the unique coverage controls for overlapping explanations by partitioning the raw coverage. It is calculated for a certain causal condition by subtracting the joint raw coverage of all the remaining causal paths from the joint raw coverage of all causal paths including the one of interest (Schneider et al., 2010). Unique coverage of causal conditions is similar to unique R2 calculations in regression analysis (Fiss, 2009). Our results in Tables 2–4 are depicted in the manner that Fiss (2011) and Ragin and Fiss (2008) advocate. Full circles indicate the presence of a condition, while crossed-out circles indicate the absence of a condition. The solution tables only list configurations that consistently led to the outcome of interest. Blank conditions are those that are not relevant to a solution configuration. In Table 2, there are three combinations for high subordinate performance. The raw coverage for the configurations ranges from 0.46 to 0.14. The unique coverages range from 0.20 to 0.07 indicating that each solution configuration contributes to the explanation of high subordinate performance. The first empirically important combination, with a unique coverage rate of 0.20 indicates that high levels of performance can be achieved through the combination of transformational leadership, the presence of enriched jobs, and when subordinates perceive a high-quality relationship with their managers. This combination of conditions leads to high subordinate performance independent of the goal difficulty condition. While configuration 1 reveals that performance can be achieved through

Table 2.

Configurations for Achieving High Subordinate Performance. Solution 1

TFL Goal MPS LMX Consistency Raw coverage Unique coverage Overall solution consistency Overall solution coverage

2

3

 K K 0.80 0.32 0.06

 K  K 0.81 0.14 0.07

K K K 0.83 0.46 0.20 0.80 0.62

Full circles indicate the presence of a condition; crossed-out circles indicate the absence of a condition.

Applying Fuzzy Set Methodology to Evaluate Substitutes for Leadership

293

Table 3. Configurations for Achieving High Subordinate Organizational Citizenship Behaviors (OCBs). Solution 1 TFL Goal MPS LMX Consistency Raw coverage Unique coverage Overall solution consistency Overall solution coverage

2

K  K K 0.74 0.30 0.08

K K 0.76 0.43 0.22 0.75 0.53

Full circles indicate the presence of a condition; crossed-out circles indicate the absence of a condition.

Table 4.

Configurations for Achieving High Affective Commitment. Solution

TFL Goal MPS LMX Consistency Raw coverage Unique coverage Overall solution consistency Overall solution coverage

1

2

K

   K 0.93 0.16 0.08

K 0.90 0.42 0.10 0.88 0.65

3

K K 0.93 0.38 0.06

4

K  0.92 0.17 0.03

Full circles indicate the presence of a condition; crossed-out circles indicate the absence of a condition.

transformational leadership, importantly configurations 2 and 3 do not. Indeed, the second empirically important configuration, with a unique coverage rate of 0.06, demonstrates that performance can be realized through the presence of highly enriched jobs and LMX, and the absence of goal difficulty. Solution 2 occurs independent of the transformational leadership condition. Finally, solution 3 reveals the combination of all four

294

J. LEE WHITTINGTON ET AL.

conditions can lead to high subordinate performance. In this solution, the presence of goal difficulty, and the presence of high LMX, along with both the absence of transformational leadership and the absence of an enriched job can lead to high subordinate performance. Table 3 reveals two combinations that lead to high levels of employee citizenship behaviors. The raw coverage for the configurations in Table 3 ranges from 0.43 to 0.30. The unique coverage ranges from 0.22 to 0.08. Therefore, both of these configurations provide a unique contribution to the explanation of high subordinate citizenship behaviors (OCBs). The first combination indicates that citizenship behaviors can be achieved through the presence of high transformational leadership, the presence of enriched jobs, and when subordinates perceive a high-quality relationship with their managers. In the second configuration solution, high levels of OCB can be achieved through the combination of highly enriched jobs, strong LMX with supervisors, and the absence of difficult goals. Solution 2 shows that the presence of enriched jobs, the presence of high quality leader– follower relationships, and the absence of difficult goals lead to high OCB independent of transformational leadership. Finally, we sought the combination of leadership conditions that lead to high levels of affective commitment. The results found in Table 4 reveal four solutions with raw coverages for ranging from 0.42 to 0.16. The unique coverage ranges from 0.10 to 0.03. Therefore, each of these configurations provides a unique contribution to the explanation of high subordinate affective commitment. The first most relevant configuration with a unique coverage of 0.10 indicates that affective commitment is achieved through the combination of transformational leadership and enriched jobs. The presence of these two conditions leads to high levels of employee affective commitment, independent of the remaining two conditions. The second most relevant configuration for high affective commitment, with a unique coverage of 0.08, is the presence of a high quality leader–follower relationship combined with the absence of transformational leadership, the absence of difficult goals, and the absence of enriched jobs. The third configuration, with a unique coverage of 0.06, reveals the presence of difficult goals in combination with the presence of enriched jobs can lead to high affective commitment among subordinates. This combination is independent of the transformational leadership and leader–follower relationship conditions. Finally, the fourth most relevant configuration, with a unique configuration of 0.03, indicates that high affective can be achieved through the combination of an enriched job and low quality leader–follower relationships. This configuration is independent of the transformational leadership and goal

Applying Fuzzy Set Methodology to Evaluate Substitutes for Leadership

295

difficulty conditions. Hence, the Solutions 2, 3, and 4 suggest that in the absence of transformational leadership, there can be multiple paths to attaining high affective commitment.

IMPLICATIONS FOR THE APPLICATION OF CONFIGURATIONAL METHODS IN LEADERSHIP RESEARCH Our study builds on the growing body of work by others applying a settheoretic approach in organizational settings (Fiss, 2007, 2009, 2011; Grandori & Furnari, 2008; Greckhamer et al., 2008; Jackson, 2005; Kogut, MacDuffie, & Ragin, 2004; Pajunen, 2008). This study makes four important contributions. First, our study introduces the fs/QCA methodology to leadership scholars. To date, most researchers utilizing fs/QCA are in the fields of sociology, political science, international business, and strategic management. Whether utilized as an exploratory tool, or in hypothesis testing, we believe that fs/QCA offers considerable potential to leadership researchers. Second, our approach can prompt leadership theorists to more thoroughly evaluate why the presence, and in certain circumstances, the absence, of certain task and motivational constructs combine together enabling firms to achieve performance outcomes similar to those achieved through transformational leadership. Third, our study helps demonstrate that certain combinations of task and organizational characteristics do not substitute for transformational leadership in the same manner across our three performance outcome measures (employee in-role performance, OCB, and affective commitment). Finally, this study makes an important contribution to the leadership literature by providing a method for addressing the criticisms of previous substitutes for leadership-based research. Villa et al. (2003) argued that many of the factors examined in earlier substitutes for leadership research had not been supported by a strong theoretical rationale. The ‘‘substitutes’’ examined here do have a strong theoretical rationale. Furthermore, by adopting a set-theoretic approach, our examination answers calls for research that includes ‘‘multiple moderators that may interact with each other to impact performance that might be erroneously attributed to the leader’’ (Avolio et al., 2009, p. 436). Recent critics of the substitutes model (Avolio et al., 2009; Dionne, Yammarino, Howell, & Villa, 2005) have called for research that explores

296

J. LEE WHITTINGTON ET AL.

the possible conditions that link leadership behavior, outcomes associated with that behavior, and the situational variables (e.g., substitutes) that may impact the relationships between leader behavior and employee outcomes. These conditions include (a) a leadership main effects model, (b) a substitutes main effect model, and (c) an interactive effects model. The fs/QCA methodology employed here provides a positive response to that call. Fs/QCA offers a valuable alternative to researchers interested in using a method that is not centered upon variable distributions and the search for patterns of covariation, difference, or frequency clusterings (Ragin, 2006). It ignores variation and distribution in variables, and also ignores any measurable assessments of the relative contribution of the variables to a particular outcome. As a result, fuzzy set methods cannot be used to detect standard linear correlations even in the presence of those associations (Mahoney, 2001). As discussed in Chapter 3 of this volume, it is important to point out that fs/QCA techniques are not meant to supplant regression as these two methods have very different logic and underlying goals. The intent of fs/QCA is not to isolate the net independent effect of each condition variable on an outcome (Ragin, 2006). At first glance, this departure may suggest the inferiority of this method to other data analytic methods. However, calculating the net effects of independent variables in linear models should not be the only reason to conduct research on the factors that motivate employees. Fs/QCA does have its limitations. For example, only one outcome variable or factor and roughly 10 potential ‘‘causes’’ may be considered in a single analysis. In addition, interpreting the output from fs/QCA is not necessarily straightforward and results are subject to criticism because they are sensitive to the calibrations researchers employ. Also, some have suggested that QCA is probably best used in conjunction with traditional analyses to locate patterns that the latter might miss. Despite the limitations, fs/QCA is quite valuable to researchers interested in studying combinations of effects. Indeed, rather than isolating independent effects, fs/QCA helps to identify and investigate interdependencies and causal complexity among causal factors that researchers may otherwise overlook. Our investigation in this chapter yielded some interesting results. First, our results confirmed that the presence of transformational leadership does contribute to high levels of in-role performance, OCB, and affective commitment. However, none of our solution configurations indicate that transformational leadership will singularly explain our performance outcomes. Indeed, high levels of in-role and extra-role performance can occur through the presence of transformational leaders in conjunction with both

Applying Fuzzy Set Methodology to Evaluate Substitutes for Leadership

297

enriched job design and a quality leader–follower relationship (LMX). However, a quality leader–follower relationship is less relevant in achieving high levels of affective commitment if both transformational leadership and enriched jobs are present. The results of configurations leading to high subordinate in-role performance were also instructive. Most importantly, the third configuration supports the idea that even if transformational leadership is low, there is some configuration of these factors that still may produce high levels of follower in-role performance. Specifically, this configuration indicates that when both transformational leadership and enriched jobs are absent, high levels of in-role performance can still be attained if a high quality leader– follower relationship exists along with challenging goals. We also looked for combinations of enriched jobs, challenging goals, and high LMX that would be associated with high levels of OCB in the absence of transformational leadership. We found that the combination of enriched jobs, a high quality leader–follower relationship, and goals that are not challenging is also associated with high levels of OCB. Perhaps individuals with enriched jobs feel as though they are able to more easily handle the challenges associated with their jobs when they have a good relationship with their leader. As a result, they may have more opportunity to spend time on extra-role behaviors. Finally, we sought combinations associated with high levels of affective commitment. The first configuration provided results similar to previous research suggesting that the combination of high levels of transformational leadership and enriched jobs is associated with high levels of affective commitment. Interestingly, we also found that firms can still experience high levels of affective commitment from their employees despite low levels of transformational leadership, difficult goals, and job enrichment if there is a high-quality relationship between the leader and his or her employees. The third configuration shows that high levels of affective commitment can also be attained when the combination of challenging goals and enriched jobs are present. An enriched job is of particular importance when there isn’t a high quality leader–follower relationship. Taken as a whole, the results of this study indicate that there are multiple alternative paths to high levels of in-role performance, extra-role performance (OCBs), and affective commitment. While transformational leadership is certainly one of those paths, high-quality leader–follower relationships, enriched jobs, and challenging goals have also been shown to work independently and in conjunction with each other to attain results similar to those associated with transformational leadership.

298

J. LEE WHITTINGTON ET AL.

IMPLICATIONS FOR FUTURE RESEARCH Previous organizational research that has utilized the fs/QCA methodology has focused almost exclusively on country and organizational level topics (Fiss, 2011; Grandori & Furnari, 2008; Greckhamer et al., 2008; Jackson, 2005; Kogut et al., 2004; Pajunen, 2008). In contrast, to the best of our knowledge, our study is the first application of the fs/QCA methodology to employee-level topics. Indeed, we believe there is a great deal of opportunity for additional investigations at the individual and team levels of analysis. In this chapter we focused on only three potential ‘‘substitutes’’ for transformational leadership. However, there are many others that should also be investigated. For instance, the substitutes for leadership framework suggests that in addition to goal difficulty, other characteristics of the organization may substitute for leadership including formalization, group cohesiveness, inflexible rules, and organizational rewards not under the control of the leader. Additional task characteristics that should be investigated include routine and repetitive tasks. Likewise, characteristics of subordinates may also be substitutes for leadership. These include ability, experience, training, and job-related knowledge. In addition, individual differences among employees have been shown to impact a variety of outcomes. Therefore, future configurational studies should examine the impact of personality and other stable traits, such as need for achievement, in a fully interactive model. In this study we focused on in-role and extra-role aspects of performance, along with affective commitment. Future research should also investigate the various bundles that lead to other important dimensions of employee engagement with the organization, such as job satisfaction and intention to quit. An important aspect of any investigation of job satisfaction should include the subscales of satisfaction. The various substitutes we have identified may impact the different dimensions of satisfaction in unique ways. Transformational leadership has been associated with high levels of performance, commitment, trust, and OCBs, but few researchers have questioned what should be done if a transformational leader is not in place in an organization. By using the set-theoretic approach, researchers are able to help managers better understand how they can achieve success in performance and work attitudes if they have not developed transformational skills. From the perspective of the practicing manager, our results suggest that effective managers should carry a wide arsenal of motivational tools. Greater levels of in-role and extra-role performance, and affective

Applying Fuzzy Set Methodology to Evaluate Substitutes for Leadership

299

commitment to the organization can be achieved if the manager engages in setting challenging goals or designing intrinsically motivating tasks for their followers. In addition, development of quality relationships with followers provides additional opportunity to achieve outcomes that benefit both employees and the organization.

NOTE 1. For an introduction to fs/QCA as well as tutorials and empirical examples, please see Fiss (2007), Greckhamer, Misangyi, Elms, and Lacey (2008), Herrmann and Cronqvist (2009), and Jackson (2005).

REFERENCES Aguilera, R. A., Filatotchev, I., Gospel, H., & Jackson, G. (2008). An organizational approach to comparative corporate governance: Costs, contingencies, and complementarities. Organization Science, 19(3), 475–492. Allen, N., & Meyer, J. (1990). The measurement and antecedents of affective, continuance and normative commitment to the organization. Journal of Occupational Psychology, 63(1), 1–18. Avolio, B. (2010). Full leadership development (2nd edn.). Thousand Oaks, CA: Sage. Avolio, B., Walumbwa, F., & Weber, T. (2009). Leadership: Current theories, research, and future directions. Annual Review of Psychology, 60, 421–449. Bass, B., & Avolio, B. (1994). Improving organizational effectiveness through transformational leadership. Thousand Oaks, CA: Sage. Bass, B., & Riggio, R. (2006). Transformational leadership (2nd edn.). Mahwah, NJ: Lawrence Erlbaum and Associates. Bauer, T. N., Erdogan, B., Liden, R. C., & Wayne, S. J. (2006). A longitudinal study of the moderating role of extraversion: Leader-member exchange, performance, and turnover during new executive development. Journal of Applied Psychology, 91(2), 298–310. Braumoeller, B. F. (2004). Hypothesis testing and multiplicative interaction terms. International Organization, 58(4), 807–882. Bycio, P., Hackett, R., & Allen, J. (1995). Further assessments of Bass’s (1985) conceptualization of transactional and transformational leadership. Journal of Applied Psychology, 80(4), 468–478. Dienesch, R., & Liden, R. (1986). Leader-member exchange model of leadership: A critique and further development. Academy of Management Review, 11(3), 618–634. Dionne, S. D., Yammarino, F. J., Atwater, L. E., & James, L. R. (2002). Neutralizing substitutes for leadership theory: Leadership effects and common source bias. Journal of Applied Psychology, 87(3), 454–464. Dionne, S. D., Yammarino, F. J., Howell, J. P., & Villa, J. (2005). Substitutes for leadership, or not. The Leadership Quarterly, 16(1), 169–193.

300

J. LEE WHITTINGTON ET AL.

Fiss, P. C. (2007). A set-theoretic approach to organizational configurations. Academy of Management Review, 32(4), 1180–1198. Fiss, P. C. (2009). Case studies and the configurational analysis of organizational phenomena. In C. Ragin & D. Byrne (Eds.), Handbook of case study methods (pp. 424–440). Thousand Oaks, CA: Sage. Fiss, P. C. (2011). Building better causal theories: A fuzzy set approach to typologies in organization research. Academy of Management Journal, 54(2), 393–420. Gerstner, C. R., & Day, D. V. (1997). Meta-analytic review of leader-member exchange theory: Correlates and construct issues. Journal of Applied Psychology, 82(6), 827–844. Goodwin, V. L., Wofford, J. C., & Whittington, J. L. (2001). A theoretical and empirical extension to the transformational leadership construct. Journal of Organizational Behavior, 22(7), 759–774. Graen, G. B., & Uhl-Bien, M. (1995). Relationship-based approach to leadership: Development of leader-member exchange (LMX) theory of leadership over 25 years: Applying a multilevel multi-domain perspective. The Leadership Quarterly, 6(2), 219–247. Grandori, A., & Furnari, S. (2008). A chemistry of organization: Combinatory analysis and design. Organization Studies, 29(03), 459–485. Greckhamer, T., Misangyi, V., & Fiss, P. C. (2013). Two QCAs: From a small-n to a large-n settheoretic approach. In B. Cambre, P. Fiss & A. Marx (Eds.), Research in the sociology of organizations (RSO): Configurational theory and methods in organizational research. Bingley, UK: Emerald Publishing Group. Greckhamer, T., Misangyi, V. F., Elms, H., & Lacey, R. (2008). Using qualitative comparative analysis in strategic management research: An examination of combinations of industry, corporate, and business-unit effects. Organizational Research Methods, 11(4), 695–726. Griffin, R. (1982). Task design: An integrative approach, Scott. Glenview, IL: Foresman and Company. Griffin, R. (1991). Effects of work redesign on employee perceptions, attitudes and behaviors: A long-term investigation. Academy of Management Journal, 34(2), 425–435. Griffin, R., & McMahan, G. (1993). Motivation through job design. In J. Greenberg (Ed.), Organizational behavior: State of the science (pp. 23–43). Hillsdale, NJ: Lawrence Erlbaum. Hackett, R., Farh, J., Sung, L., & Lapierre, L. (2003). LMX and organizational citizenship behavior: Examining the links within and across Western and Chinese samples. In G. B. Graen (Ed.), Dealing with diversity (pp. 219–264). Charlotte, NC: Information Age Publishing. Hackman, J., & Oldham, G. (1976). Motivation through the design of work: Test of a theory. Organizational Behavior and Human Performance, 16(2), 250–279. Herrmann, A. M., & Cronqvist, L. (2009). When dichotomization becomes a problem for the analysis of middle-sized datasets. International Journal of Social Research Methodology, 12(1), 33–50. Hoskisson, R. E., Castleton, M. W., & Withers, M. C. (2009). Complementarity in monitoring and bonding: More intense monitoring leads to higher executive compensation. Academy of Management Perspectives, 23(2), 57–74. Howell, J., & Avolio, B. (1993). Transformational leadership, transactional leadership, locus of control, and support for innovation: Key predictors of consolidated-business-unit performance. Journal of Applied Psychology, 78(6), 891–902. Howell, J., Dorfman, P., & Kerr, S. (1986). Moderator variables in leadership research. Academy of Management Review, 11(1), 88–102.

Applying Fuzzy Set Methodology to Evaluate Substitutes for Leadership

301

Howell, J. M., & Hall-Merenda, K. (1999). The ties that bind: The impact of leader-member exchange, transformational and transactional leadership, and distance on predicting follower performance. Journal of Applied Psychology, 84(5), 680–694. Ilies, R., Nahrgang, J. D., & Morgeson, F. P. (2007). Leader-member exchange and citizenship behaviors: A meta-analysis. Journal of Applied Psychology, 92(1), 269–277. Jackson, G. (2005). Employee representation in the board compared: A fuzzy sets analysis of corporate governance, unionism and political institutions. Industrielle Beziehungen, 12(3), 252–279. Keller, R. (2006). Transformational leadership, initiating structure, and substitutes for leadership: A longitudinal study of research and development project team performance. Journal of Applied Psychology, 91(1), 202–210. Kerr, S., & Jermier, J. (1978). Substitutes for leadership: their meaning and measurement. Organizational Behavior and Human Performance, 22(3), 375–403. Kogut, B., MacDuffie, J. P., & Ragin, C. C. (2004). Prototypes and strategy: Assigning causal credit using fuzzy sets. European Management Review, 1(2), 114–131. Latham, G., & Locke, E. (1991). Self-regulation through goal setting. Organizational Behavior and Human Decision Processes, 50(2), 212–247. Liden, R., Wayne, S., & Stilwell, D. (1993). A longitudinal study on the early development of leader-member exchanges. Journal of Applied Psychology, 78(4), 662–674. Liden, R. C., Sparrowe, R. T., & Wayne, S. J. (1997). Leader-member exchange theory: The past and potential for the future. Research in Personnel and Human Resources Management, 15, 47–119. Locke, E., & Latham, G. (1990). A theory of goal setting and task performance. Englewood Cliffs, NJ: Prentice-Hall. Mahoney, J. (2001). Beyond correlational analysis: Recent innovations in theory and method. Sociological Forum, 16(3), 575–593. Mento, A., Steel, R., & Karren, R. (1987). A meta-analytic study of the effects of goal setting on task performance: 1966-1984. Organizational Behavior and Human Decision Processes, 39(1), 52–83. Pajunen, K. (2008). Institutions and inflows of foreign direct investment: A fuzzy-set analysis. Journal of International Business Studies, 39(4), 652–669. Piccolo, R., Greenbaum, R., Den Hartog, D., & Folger, R. (2010). The relation between ethical leadership and core job characteristics. Journal of Organizational Behavior, 31(2–3), 259–278. Podsakoff, P., MacKenzie, R., & Bommer, W. (1996). Transformational leader behaviors and substitutes for leadership as determinants of employee satisfaction, commitment, trust, and organizational citizenship behaviors. Journal of Management, 22(2), 259–298. Podsakoff, P., MacKenzie, S., Moorman, R., & Fetter, R. (1990). Transformational leader behaviors and their effects on followers’ trust in leader, satisfaction, and organizational citizenship behaviors. The Leadership Quarterly, 1(2), 107–142. Podsakoff, P., Niehoff, B., MacKenzie, S., & Williams, M. (1993). Do substitutes for leadership really substitute for leadership? An empirical examination of Kerr and Jermier’s situational leadership model. Organizational Behavior and Human Decision Processes, 54(1), 1–44. Ragin, C. C. (2000). Fuzzy set social science. Chicago, IL: University of Chicago Press. Ragin, C. C. (2006). Set relations in social research: Evaluating their consistency and courage. Political Analysis, 14(3), 291–310.

302

J. LEE WHITTINGTON ET AL.

Ragin, C. C. (2008). Redesigning social inquiry: Fuzzy sets and beyond. Chicago, IL: University of Chicago Press. Ragin, C. C., & Fiss, P. (2008). Net effects analysis versus configurational analysis: An empirical demonstration. In C. C. Ragin (Ed.), Redesigning social inquiry: Fuzzy sets and beyond (pp. 190–212). Chicago, IL: University of Chicago Press. Rediker, K., & Seth, A. (1995). Boards of directors and substitution effects of alternative governance mechanisms. Strategic Management Journal, 16(2), 85–99. Rihoux, B., & Ragin, C. C. (2009). Configurational comparative methods: Qualitative comparative analysis (QCA) and related techniques. Thousand Oaks, CA: Sage. Schneider, M. R., Schulze-Bentrop, C., & Paunescu, M. (2010). Mapping the institutional capital of high-tech firms: A fuzzy-set analysis of capitalist variety and export performance. Journal of International Business Studies, 41(2), 246–266. Schriesheim, C. A., Castro, S. L., & Cogliser, C. C. (1999). Leader-member exchange (LMX) research: A comprehensive review of theory, measurement, and data-analytic practices. The Leadership Quarterly, 10(1), 63–113. Scott, S. G., & Bruce, R. A. (1998). Following the leader in R&D: The joint effect of subordinate problem-solving style and leader-member relations on innovative behavior. IEEE Transactions on Engineering Management, 45(1), 3–10. Short, J. C., Payne, G. T., & Ketchen, D. J. (2008). Research on organizational configurations: Past accomplishments and future challenges. Journal of Management, 34(6), 1053–1079. Steers, R. (1973). Task goals, individual need strength, and supervisory performance. Unpublished doctoral dissertation, University of California at Irvine. Uhl-Bien, M. (2006). Relational leadership theory: Exploring the social processes of leadership and organizing. The Leadership Quarterly, 17(6), 654–676. Villa, J. R., Howell, J. P., Dorfman, P. W., & Daniel, D. L. (2003). Problems with detecting moderators in leadership research using moderated multiple regression. The Leadership Quarterly, 14(1), 3–23. Walsh, J. P., & Seward, J. K. (1990). On the efficiency of internal and external corporate control mechanisms. Academy of Management Review, 15(3), 421–458. Walumbwa, F. O., Cropanzano, R., & Hartnell, C. A. (2009). Organizational justice, voluntary learning behavior, and job performance: A test of mediating effects of identification and leader-member exchange. Journal of Organizational Behavior, 30(8), 1103–1126. Wang, H., Law, K. S., Hackett, R. D., Wang, D., & Chen, Z. X. (2005). Leader-member exchange as a mediator of the relationship between transformational leadership and follower’s performance and organizational citizenship behavior. Academy of Management Journal, 48(3), 420–432. Ward, A., Brown, J., & Rodriguez, D. (2009). Governance bundles, firm performance and the substitutability and complementarity of governance mechanisms. Corporate Governance, An International Review, 17(5), 646–660. Whittington, J. L., Goodwin, V. L., & Murray, B. (2004). Transformational leadership, goal difficulty, and task design: Independent and interactive effects on employee outcomes. The Leadership Quarterly, 15(5), 593–606. Yukl, G. (2012). Leadership in organizations (8th edn.). Englewood Cliffs, NJ: Prentice-Hall. Zhou, X., & Schriesheim, C. A. (2009). Supervisor-subordinate convergence in descriptions of leader-member exchange (LMX) quality: Review and testable propositions. The Leadership Quarterly, 20(6), 920–932.

CHAPTER 13 WE TRY HARDER: SOME REFLECTIONS ON CONFIGURATIONAL THEORY AND METHODS David J. Ketchen, Jr. Configurational research has a long and distinguished history within organizational inquiry. For example, the basic configurations of ‘‘organic’’ and ‘‘mechanistic’’ organizations that were described more than 50 years ago by Burns and Stalker (1961) continue to have relevance today (e.g., Sine, Mitsuhashi, & Kirsch, 2006). More than three decades after its creation, the influence of Miles and Snow’s (1978) typology of prospector, defender, and analyzer organizations remains very strong. One implication of how these concepts have endured is that, when done well, configurational research is a compelling approach to understanding organizations (Short, Payne, & Ketchen, 2008). In recent years, some of the strongest contributions under the banner of configurational research have been offered by authors who have used set-theoretic approaches (e.g., Fiss, 2007, 2011). The chapters contained in this volume offer a variety of contributions that promise to further propel configurational research forward. My hope in this brief commentary is to highlight a few (but certainly not all) of the ways that the chapters

Configurational Theory and Methods in Organizational Research Research in the Sociology of Organizations, Volume 38, 303–309 Copyright r 2013 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0733-558X/doi:10.1108/S0733-558X(2013)0000038017

303

304

DAVID J. KETCHEN

advance configurational thinking and to offer some related ideas for future research.

KEY CONTRIBUTIONS OF THE CHAPTERS I have always felt that one of the most important roles that a book chapter can fulfill is effectively setting the stage for future scholars to build on the content of the chapter. This is a particularly important task where doctoral students are concerned, given their status as newcomers to organizational research. Ideally a doctoral student can pick up tips in a book chapter that will facilitate her or him conducting better studies than if s/he had never read the chapter. Newcomers to the field who are interested in configurational research need to read Chapter 2. Marx, Cambre´, and Rihoux provide a valuable service to future scholars by laying out an easy to follow, step-by-step approach to executing crisp-set Qualitative Comparative Analysis (csQCA). Further, by articulating the strengths and criticism of csQCA, the authors have provided a basis for scholars to determine whether csQCA offers a good fit with the needs of particular studies. In short, Marx, Cambre´, and Rihoux have put future scholars in a position to succeed via their skillful summary of csQCA. The third chapter was written by three individuals – Greckhamer, Misangyi, and Fiss – that I know personally and for whom I have enormous respect. Thomas, Vilmos, and Peer have not only introduced significant innovations into organizational research, they are genuinely nice people. Their chapter is notable for accomplishing what many of their colleagues would have thought impossible: Offering a new twist on one of the decadesold mainstay approaches to organizational research. The use of large samples has played a dominant role within organizational studies since the 1970s. The introduction of datasets such as PIMS and Compustat allowed researchers to test hypotheses using high degrees of statistical power. As studies that drew on large databases proliferated in the 1980s and 1990s, important advances were made in understanding organizations. Simultaneously, however, many scholars became skeptical about such studies. The use of large samples led to seemingly minor linear relationships between variables being declared as meaningful because they were statistically significant. The reliance on archival proxies fostered concerns about what exactly was being assessed when performing tests involving measures such as research and development intensity. Encouragingly, the use

Some Reflections on Configurational Theory and Methods

305

of Qualitative Comparative Analysis as explained by Greckhamer, Misangyi, and Fiss offers the potential to infuse richness into large N studies that otherwise would be characterized by coarseness. The promise of studies that maintain the important advantages of large samples while escaping a key weakness is cause for significant optimism. In the fourth chapter, Grandori and Furnari provide at least two important contributions. The first contribution that attracted my attention is highlighting some of the subtle ways that configurational thinking has been captured and represented over time within organizational theory. Perhaps because the search for linear relationships (including links that are direct, moderated, and mediated) has been a more dominant approach within organizational research, the role of configurational ideas is sometimes overlooked or downplayed. The second contribution is the presentation of a two-by-two matrix that classifies configurational approaches in terms of internal heterogeneity and external heterogeneity. Given the ubiquity of two-by-two matrices within organizational research, it is easy to become cynical about their use. Indeed, some of them are not very useful. In contrast, Grandori and Furnari’s framework provides an intuitively pleasing and conceptually rich means of classifying different approaches. Future scholars can leverage this framework to diagnose which type of approach they are using, assess the advantages and disadvantages of their approach, and possibly incorporate one of the other approaches in order to remedy the weakness of their original approach. Hak, Jaspers, and Dul’s chapter is notable for their emphasis on the passage of time. Advocates of configurational research often point to the ability of configurational thinking and methods to provide enhanced richness in theorizing and findings. Yet too often configurational studies ignore the temporal dimension and how configurations can evolve in form and substance over time. A good (and possibly painful) example of this is my own dissertation and a 1993 Academy of Management Journal article that is based on the dissertation (i.e., Ketchen, Thomas, & Snow, 1993). In these works, I relied on five individual snapshots of single years without consideration of the temporal dimension.1 Using data from the Panel Study of Entrepreneurial Dynamics, Hak, Jaspers, and Dul illustrate how to use Temporal Qualitative Comparative Analysis (TQCA). Their ideas should prove to be very useful to configurational researchers who want to account for the passage of time within their empirical models. To their credit, Hak, Jaspers, and Dul highlight not only the advantages of TQCA but also its pitfalls and

306

DAVID J. KETCHEN

limitations. Thus, researchers who are contemplating the use of TQCA are offered a realistic preview of what they will encounter. Reanalyzing data from a past study using more advanced methodology than was used in the initial study is a very effective way to advance knowledge. This approach can generate both theoretical and methodological contributions. Theoretically, revisiting past data can highlight relationships that previously went undetected. Methodologically, revisiting past data provides tangible evidence regarding the advantages of the newly deployed technique. This is the type of twofold contribution offered by Jackson and Ni’s chapter on complementarities. Specifically, Jackson and Ni take a second look at Japanese data that were used in a previous study to examine relationships among ownership structure, board structure, and employment practices. Using Qualitative Comparative Analysis, the authors find that interesting complementarities appear to exist within the data. Viewed more broadly, I believe the authors have developed an idea that could be quite useful to other scholars. I would be very interested to learn what new light QCA could shed on datasets that have appeared in the configurational literature, especially those that have been analyzed using controversial methods such as cluster analysis. Configurational approaches have been largely absent from research on corporate governance. Meanwhile, a fair amount of corporate governance studies appear to have generated inconclusive and ambiguous findings. The thoughtful work by Bell, Aguilera, and Filatotchev suggests that this may not be a coincidence. Using fuzzy set Qualitative Comparative Analysis, the authors find that accounting for the regulatory and legal factors that protect minority shareholders are essential to capturing corporate governance. Although the dominant approach to corporate governance research seems to implicitly assume that various aspect of governance have independent effects on performance, Bell, Aguilera, and Filatotchev provide a more realistic empirical depiction by examining the complex relationships among governance factors, institutional parameters, and performance. It would be difficult to identify a topic in greater need of more configurational thinking than corporate social responsibility. Any given firm’s approach to a social issue is driven by a complex milieu of organizational culture, managerial priorities, stakeholder pressures, industry norms, and other important factors. Meanwhile, a firm’s approach may vary across different issues. Configurational theorizing and methods are well suited to capture this complexity, but their application within the corporate social responsibility literature is limited.

Some Reflections on Configurational Theory and Methods

307

The chapter by Crilly offers important advances toward infusing corporate social responsibility research with configurational thinking. Crilly seeks to shed new light on instances where shareholder value and social good are in conflict. In contrast to the more traditional approach of using individual levels of analysis to examine such dilemmas, Crilly relies on fuzzy set Qualitative Comparative Analysis to uncover how individual, organizational, and societal factors coalesce in the corporate social responsibility context. As interest in describing, explaining, and predicting corporate social responsibility continues to grow, scholars would be well served by following Crilly’s lead and applying configurational theory and methods. Like Bell, Aguilera, and Filatotchev’s application of configurational thinking to corporate governance research and Crilly’s application of configurational thinking to corporate social responsibility inquiry, the chapter by Park and El Sawy involves delving into an interesting topic via fuzzy set Qualitative Comparative Analysis. In particular, Park and El Sawy consider the value of configurational approaches for studying digital business strategy. The authors focus their chapter on digital ecodynamics, which they define as ‘‘a complex phenomenon of fused, dynamic interactions among information technology, organizational dynamic capabilities, and environmental turbulence, unfolding as an ecosystem.’’ Part of the value of this concept is derived from the facts that its origins come from different fields, including management information systems, strategy, and organization theory. Further advances might be achieved by incorporating notions from fields such as marketing and finance. At one time or another, most of us have run across a work that causes a reaction of ‘‘why didn’t I think of that?!’’ The chapter by Raab, Lemaire, and Provan caused me to have this reaction. The value of bringing together configurational thinking and organizational network thinking, as they have done, seems so obvious. Of course, the best ideas often seem obvious in retrospect. Both configurational research and network research center on the relationships between entities, thus exploring the possible links between them has immense merit. Raab, Lemaire, and Provan take important initial steps by developing a nine-part typology of organizational network research approaches and then applying Qualitative Comparative Analysis within two of the nine cells. Researchers who seek to further integrate configurational and network thinking could examine additional cells. Important steps could also be taken by viewing topics such as corporate governance, corporate social responsibility, and digital ecodynamics through a network lens.

308

DAVID J. KETCHEN

One regrettable aspect of past configurational research is that it has almost always been set within domestic contexts. The chapter by Pajunen and Airo begins to remedy this shortcoming by applying fuzzy set Qualitative Comparative Analysis within the domain international business. Specifically, Pajunen and Airo leverage data from the generic medicines industry within 24 European countries in order to discover how locationspecific conditions influence firm performance. It seems likely that configurational thinking could enhance understanding within a variety of international business research streams. Despite a great numbers of relevant studies, much remains unknown about the contribution of international joint ventures to organizational performance. Similarly, knowledge is limited regarding how the use of global supply chains influences important outcomes. Meanwhile, much of the past inquiry into these issues has relied upon traditional linear methods. Given the interconnectedness of the international arena, it seems likely that configurational thinking in general and Qualitative Comparative Analysis in particular can help resolve unanswered questions surrounding international joint ventures and global supply chains. Just as configurational thinking is arguably underutilized within international business research, such thinking seems to offer a great deal of unexplored potential within the field of organizational behavior. The scant use of configurational thinking within organizational behavior is perhaps surprising in that a key element within psychology – personality – is often conceptualized via typologies such as the 16 types that can be revealed using the Myers–Briggs survey as well as the more simplistic Type A versus Type B dichotomy. The chapter by Whittington, McKee, Goodwin, and Bell makes important strides toward leveraging configurational thinking within the organizational behavior field. These authors examine how configurations among leadership, task factors, and motivational factors coalesce and influence employee attitudes and outcomes. Whittington, McKee, Goodwin, and Bell discover equifinal conditions – there is more than one arrangement of leadership, tasks, and motivation that can enhance outcomes. Of particular note is that the chapter sheds light on how managers can create a situation that is conducive to success even if managers lack transformational leadership skills. This highlights a key source of value offered by configurational thinking to organizational behavior research – the absence of a seemingly desirable trait does not ensure suboptimal performance. Instead, such a shortcoming can be overcome via arrangements of other factors.

Some Reflections on Configurational Theory and Methods

309

CONCLUSION Despite the many merits of configurational research, such inquiry seems unlikely to supplant linear approaches as the dominant way to examine organizations. While this situation is arguably regrettable, I prefer to take a positive perspective. In a famous advertising campaign, rental car company Avis proclaimed that it had to work more diligently than industry leader Hertz did in order to be successful. Avis’ tagline of ‘‘we try harder because we have to’’ could serve as an effective rallying cry for configurational researchers. It may be more difficult for configurational research to gain acceptance among our colleagues, but this has the positive effect of forcing us to work harder to build compelling insights. The authors of the chapters in this book have taken up this challenge. They have tried harder and the result is a series of novel ideas that can effectively guide future research.

NOTE 1. To his credit, doctoral committee member Don Bergh presciently lamented in 1993 the absence of the temporal dimension in my study.

REFERENCES Burns, T., & Stalker, G. M. (1961). The management of innovation. London: Tavistock. Fiss, P. (2007). Towards a set-theoretic approach for studying organizational configurations. Academy of Management Review, 32, 1180–1198. Fiss, P. (2011). Building better causal theories: A fuzzy set approach to typologies in organization research. Academy of Management Journal, 54, 393–420. Ketchen, D. J., Thomas, J. B., & Snow, C. C. (1993). Organizational configurations and performance: A comparison of theoretical approaches. Academy of Management Journal, 36, 1278–1313. Miles, R. E., & Snow, C. C. (1978). Organizational strategy, structure, and process. New York, NY: McGraw Hill. Short, J. C., Payne, G. T., & Ketchen, D. J. (2008). Research on organizational configurations: Past accomplishments and future challenges. Journal of Management, 34, 1053–1079. Sine, W. D., Mitsuhashi, H., & Kirsch, D. A. (2006). Revisiting Burns and Stalker: Formal structure and new venture performance in emerging economic sectors. Academy of Management Journal, 49, 121–132.

CHAPTER 14 CONCLUSION: THE PATH FORWARD Bart Cambre´, Peer C. Fiss and Axel Marx ABSTRACT In this concluding chapter, we look ahead to future theoretical and methodological directions that emerge from the contributions in this volume and that carry the potential to enrich contemporary organizational research. We furthermore point to some issues that remain unsolved and need to be addressed in future research to further establish the configurational approach in the field of organizational studies, such as the growing need for homogeneity in how the analysis is conducted and results are presented. We argue that the momentum of the configurational approach in organizational research is strong, but that important challenges remain. Keywords: Configurational theory; configurational methods; settheory; Qualitative Comparative Analysis (QCA); management; organization studies.

In our introduction to this volume, we noted that – several decades after the emergence of the configurational perspective in organization studies – the theory of configurations still requires further development, even as empirical

Configurational Theory and Methods in Organizational Research Research in the Sociology of Organizations, Volume 38, 311–319 Copyright r 2013 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0733-558X/doi:10.1108/S0733-558X(2013)0000038018

311

312

BART CAMBRE´ ET AL.

research on configurations is finally beginning to deliver on its promise. In this concluding chapter, our aim is to attempt a look forward, both theoretically and methodologically. Our focus here lies on what combines the contributions of this volume, how the approach taken here can enrich organizational research, and what future directions appear perhaps most promising. We furthermore point to some issues that remain unsolved and need to be addressed in future research to further establish the configurational approach in the field of organizational studies.

THEORETICAL CONSIDERATIONS Configurational thinking is of course not restricted to the perspective outlined here. In fact, configurational arguments arguably pervade most organizational theories since they allow researchers to study characteristics that can be considered core to the (theoretical) notion of organizing. Perhaps most prominently, configurational thinking forces us to move toward understanding how distinct characteristics jointly cause an outcome. For instance, Mintzberg (1979) developed a configurational form of structural contingency theory by considering the main coordination mechanisms as core elements of organizing that are found in different combinations in different organizational forms. Two other properties of configurations that are core to contemporary thinking about organizations are the notions of nonlinearity and equifinality (e.g., Meyer, Tsui, & Hinings, 1993). Nonlinearity refers to the fact that ‘‘variables found to be positively related in one configuration may be unrelated or even inversely related in another’’ (Meyer et al., 1993, p. 1178). As such, the concept of nonlinearity is not new (see for instance all U-shaped relations in standard regression analysis), but it allows organizational scholars to think further than a two- or three-way interaction and to move beyond the positive or negative net-effects of variables. The notion of equifinality refers to the fact that ‘‘a system can reach the same final state from different initial conditions and by a variety of different paths’’ (Katz & Kahn, 1978, p. 30). The configurational building blocks outlined here are thus present in one form or another in most organization theories. While one might argue that the configurational perspective at this time presents more of an analytic approach than a substantive theory in itself, the development outlined in the current volume would suggest that the boundaries are beginning to blur. This opens up the potential to the further development of existing (and new) organization theories and broadens the field of applications, allowing

Conclusion

313

scholars in multiple domains of organizational and management research to adopt the configurational approach in their field of study. At the same time, this development calls for a renaissance of configurational thinking about organizations, to reorient current theoretical conceptualization toward the domain of configurations. Indeed, Grandori and Furnari (Chapter 4) suggest that a configurational approach may be helpful to constructing significant ‘‘missing pieces’’ in organization theory. The chapters of this volume provide ample examples of new forms of theorizing that combines configurational thinking with substantive theory about their respective phenomena. For instance, starting with an essential issue of configurational thinking, Jackson and Ni (Chapter 6) explore the field of complementarities and draw attention to how organizational structures, practices, and institutions have interdependent effects that call for a configurational approach. To study their phenomenon of interest, they point out, researchers need to go beyond the traditional bivariate relationships associated with organizational elements and study the complex interplay of structures and practices from an equifinal and configurational point of view. Similarly, to evaluate the effectiveness of corporate governance practices, a complex multilevel approach is often recommended, but few studies analyze practices in conjunction with each other and on multiple levels. The configurational approach can in fact ‘‘contextualize’’ practices, allowing researchers to examine them jointly instead of independently. In this regards, Bell, Aguilera, and Filatotchev (Chapter 7) reveal how firm-level governance practices interact with each other as well as with macro institutions. Traditional multivariate econometric techniques cannot fully analyze this complex interaction, but new theory about corporate governance practices will also be required. Another example of how a configurational approach may be used to identify the combined influence of effects at multiple levels of analysis is the contribution by Crilly (Chapter 8). He investigates corporate social responsibility and more specifically how individual psychology and social context simultaneously affect the managers response to pressures for social responsibility, showing how effects at any level can depend on effects at all other levels. A similar insight is offered by Park and El Sawy (Chapter 9), who demonstrate the value of a configurational approach for inquiring the holistic nature of digital ecodynamics, a field marked by mutual causality, synergetic effects, and nonlinear changes. Their fuzzy set analysis of digital ecodynamics combines a more textured understanding of the causal complexity with a holistic perspective on their phenomenon of interest. Indeed, the notion of digital ecodynamics might be perhaps best understood

314

BART CAMBRE´ ET AL.

as a powerful example of novel substantive theory that is truly configurational in nature and incorporates many of the elements we have discussed above. As shown by Raab, Lemaire, and Provan (Chapter 10), a configurational approach may furthermore offer novel ways to study ‘‘whole networks’’ of organizations. The complexity of their nested constellations – organizations within networks within broader environmental contexts that jointly contribute to the outcome – is hard to grasp with more traditional techniques such as hierarchical clustering and multilevel analysis, but it is also difficult to conceptualize theoretically without the conceptual tools of the settheoretic approach (Lacey & Fiss, 2009). A similar pattern of causal complexity is the study of Pajunen and Airo (Chapter 11), who investigate the causal complexity of institutional and country characteristics on strategic decision making from a configurational perspective. The effect of such country characteristics and institutional conditions are likely to depend on the presence or absence of the effects of other conditions. Focusing mainly on the net effects may produce an incomplete understanding of whether or not a particular location is appropriate for a firm to deploy activities. While these examples have shown the effect of configurations beyond the firm boundaries, Whittington, McKee, Goodwin, and Bell (Chapter 12) focus on intra-organizational processes in their analysis of so-called ‘‘bundles’’ or configurations of leadership, task, and motivational factors that impact employee attitude and even performance outcomes. While the level of analysis is a different one, the insight offered is again powerful in that the authors demonstrate how the presence, and sometimes absence, of leadership still enable firms to achieve reasonable performance outcomes by employees.

METHODOLOGICAL CONSIDERATIONS The chapters of this volume use both crisp and fuzzy set QCA. While there appears to be a preference for crisp set QCA among small-N researchers and a preference for fuzzy set QCA among large-N researchers, we do not expect one of them to become the dominating approach. Both have their distinctive strengths, and there are vigorous attempts to optimize both methods and further strengthen the quality of the tools and measures. For instance, Marx, Cambre´, and Rihoux (Chapter 2) discuss how to address two of the most important critiques toward csQCA. First, building on Marx (2010), they focus on the assumption of naturally occurring contradictions and argue that researchers should balance conditions and cases according to

Conclusion

315

established benchmarks (Marx & Dusa, 2011). Adjusting the proportion of conditions over cases according to these benchmarks overcomes the issue that QCA is not able to distinct real from random data. With regard to the second critique, the sensitivity to individual cases, these authors discuss several scenarios and show that only in specific circumstances the sensitivity to individual cases is problematic. Another remaining challenge is the analysis of temporally ordered configurations. As Hak, Jaspers, and Dul (Chapter 5) point out, many organizational theories are inherently temporal. Yet, this temporal nature is frequently not taken into account in empirical analyses. Even the regular configurational approach shows a ‘‘temporality’’ problem (Rihoux & Ragin, 2009) with its current difficulties in tracking shifting configurations over time and explaining the ‘‘how’’ of causal configurations (Park and El Sawy, Chapter 9). While temporal qualitative comparative analysis (TQCA) aims to address these issues, it faces its own considerable technical limitations. In TQCA, co-occurrences (‘‘ties’’) are difficult to code and a code cannot be assigned to a pair of which a condition is missing. In response, Hak, Jaspers, and Dul promote the use of Necessary Condition Analysis (NCA or TNCA for the temporal approach) to study the temporal sequences of conditions between cases. Based on a truth-table, the analysis resembles the QCA-approach, but the focus is on searching for necessary instead of sufficient configurations, making it a somewhat different approach to be applied in time-related and longitudinal configurational research. Clearly, important challenges remain in developing a truly dynamic configurational approach. An important recent methodological development is the application of QCA to large-N situations. In line with Gerring (2001), Greckhamer, Misangyi, and Fiss (Chapter 3) hold that large-N QCA can also be used for hypothesis testing and theoretical deduction, thus going beyond the traditional approach of using small-N QCA for theory building. From this perspective, large-N QCA can be considered as an alternative to the widespread general linear approaches to studying organizational phenomena. This allows the researcher to study configurational ‘‘recipes’’ instead of the common focus on net-effects and individual causes. However, as far as external validity (generalization) of the findings is concerned, challenges remain, as suggested by Park and El Sawy (Chapter 9), and perhaps only ‘‘modest’’ generalizations can currently be achieved. However, QCA might be applied both as a substitute and as a complement to linear approaches in large-N organizational research. If we consider a configurational analysis as a stand-alone alternative (substitute) to regression techniques, it allows

316

BART CAMBRE´ ET AL.

researchers to make the leap from net-effect thinking to configurational thinking, which is more in line with some theoretical considerations of organizational research. On the other hand, many authors in this volume (Greckhamer, Misangyi, and Fiss, Chapter 3; Jackson and Ni, Chapter 6; Pajunen and Airo, Chapter 11; Whittington, McKee, Goodwin, and Bell, Chapter 12) consider the configurational approach using QCA not as a substitute but as an important alternative to more traditional linear techniques. Of course, QCA and correlational methods are very different in their goals and assumptions (Ragin, 2008). For instance, the goal of QCA is not to isolate the net independent effect of each condition on an outcome (Ragin, 2006). For much empirical research in management and organization studies, calculating the net effect and variances remains one of the key concerns. Considered as a complement of conventional regression analysis, QCA allows mixed-method research that employs the strength of each approach. In this view (see also Greckhamer, Misangyi, and Fiss, Chapter 3), a statistical analysis can be complemented with a QCA-analysis (Fiss, Sharapov, & Cronqvist, 2013). Similarly, the researcher might consider the two-step approach suggested by Jackson and Ni (Chapter 6), combining statistical analysis of net effects with set-theoretical analysis. Here, a preliminary statistical analysis is used to narrow down the number of conditions to be entered into a set-theoretical analysis. Finally, the analysis of multilevel issues in organizational research can benefit from a configurational approach (Greckhamer, Misangyi, Elms, & Lacey, 2008; Lacey & Fiss, 2009). For instance, QCA can bridge levels of analysis by combining behavioral variables at micro level with characteristics at meso (e.g., organization) and macro level (e.g., institutional, countries). Hence, instead of controlling for effects at other levels to measure the net effects at a given level, as in a multilevel regression analysis, a configurational approach is explicitly interested in the combined effects. In addition, a QCA approach does not require units at lower levels to be fully nested within higher level units, allowing for the application of QCA to a wider range of multilevel, dynamic phenomena where membership in different levels may be partial and fleeting.

THE PATH FORWARD The configurational approach presented in this volume holds the promise of understanding organizations in a more complex, systematic way more in line

Conclusion

317

with the notion that organizations ‘‘are best understood as clusters of interconnected structures and practices, rather than as modular or loosely coupled entities whose components can be understood in isolation’’ (Fiss, 2007, p. 1180). It is evident that the emergence of a set-theoretic approach to configurations presents a considerable leap forward and many innovations are currently being developed (Rihoux & Marx, 2013; Schneider & Wagemann, 2012; Thiem & Dusa, 2012). However, considerable challenges still remain, both with respect to configurational theorizing and to configurational methodology. Configurational theorizing became less common with the rise of the correlation studies and the focus on net-effects and variances rather than on configurations and causal complexity – it is time to take the notion of configurations seriously again. One important path for doing so continues the tradition of typology research in organization studies, an approach that has led to some of the most influential and widely validated theories (e.g., Miles & Snow, 1978). However, beyond typologies, there is also a need for a more general configurational theory of organizations. Our hope is that the seeds have been planted. With respect to configurational analysis, QCA can be considered the most widespread analysis technique to analyze configurational systems. Some of the critiques on QCA have already been addressed; others remain. Yet, there are also challenges that stem from the very success of QCA as a research approach. Specifically, we see a growing need for homogeneity in how the analysis is conducted and how the results are presented. For instance, different notational systems persist, one employing capitalized and small letters to represent the presence or absence of a cause, the other using Boolean expressions. Furthermore, while configuration tables (Ragin & Fiss, 2008) that use full and crossed out circles appear to have become a common form of representation, other forms of presentation exist and possess different advantages. While a more homogeneous and clear approach will make it easy to report and interpret results, convincing more scholars and reviewers to adopt this method, it is also important not to inhibit experimentation and innovation. This issue also relates to standards for calibration, coverage, consistency, selecting causal conditions, choosing cutoffs, and causal inference, where the challenge is finding a balance between rigor and standards on the one hand and allowing theoretical interpretations on the other hand. A potential danger here is that the use of QCA becomes too mechanical, with a strong focus on cut-offs, scores, and numbers, thus shifting toward a more statistical and standardized analysis technique that no longer captures the

BART CAMBRE´ ET AL.

318

complexity of organizational configurations that may require a more flexible approach and dialogue with the data (Ragin, 1987). In line with such concerns, the relationship between QCA and other techniques has yet to be further revealed. Since QCA and more standard correlational methods such as OLS regression are based on different starting assumptions, we should perhaps not expect their results to be comparable. While QCA has so far been used as a stand-alone technique to analyze causal complexity in organizations, it may also hold potential as a complement of variable-based approaches (in large-N) or case-based approaches (in small-N). Considerable work remains to be done to explore the intersection and the potential complementarities between QCA and standard regression analysis (Fiss et al., 2013; Greckhamer, Misangyi, and Fiss, Chapter 3; Vis, 2012). In the current volume, we have aimed to both take stock of the state of configurational theory and methods using a set-theoretic approach and to outline an agenda going forward. It is evident that significant challenges remain. Yet, the momentum of the configurational approach appears to be stronger than in a long time.

REFERENCES Fiss, P. (2007). A set-theoretic approach to organizational configurations. Academy of Management Review, 32, 1180–1198. Fiss, P., Sharapov, D., & Cronqvist, L. (2013). Opposites attract? Opportunities and challenges for integrating large-N QCA and econometric analysis. In B. Rihoux & A. Marx (Eds.), QCA, 25 years after ‘The Comparative Method’: Mapping, challenges and innovations – Mini Symposium. Political Research Quarterly, 66(1), 191–198. Gerring, J. (2001). Social science methodology. A criterial framework. Cambridge: Cambridge University Press. Greckhamer, T., Misangyi, V. F., Elms, H., & Lacey, R. (2008). Using qualitative comparative analysis in strategic management research: An examination of combinations of industry, corporate, and business-unit effects. Organizational Research Methods, 11, 695–726. Katz, D., & Kahn, R. L. (1978). The social psychology of organizations (2nd edn.). New York, NY: Wiley. Lacey, R., & Fiss, P. (2009). Comparative organizational analysis across multiple levels: A set-theoretic approach. In B. G. King, T. Felin, & D. A. Whetten (Eds.), Studying differences between organizations: Comparative approaches to organizational research (Vol. 26, pp. 91–116). Research in the Sociology of Organizations. Bingley, UK: Emerald Group Publishing Limited. Marx, A. (2010). Crisp-set qualitative comparative analysis (csQCA) and model specification: Benchmarks for future csQCA applications. International Journal of Multiple Research Approaches, 4(2), 138–158.

Conclusion

319

Marx, A., & Dusa, A. (2011). Crisp-set qualitative comparative analysis (csQCA), contradictions and consistency benchmarks for model specification. Methodological Innovations Online, 6(2), 103–148. Meyer, A. D., Tsui, A. S., & Hinings, C. R. (1993). Configurational approaches to organizational analysis. Academy of Management Journal, 36, 1175–1195. Miles, R. E., & Snow, C. C. (1978). Organizational strategy, structure and process. New York, NY: McGraw-Hill. Mintzberg, H. (1979). The structuring of organizations: A synthesis of the research. Englewood Cliffs, NJ: Prentice Hall. Ragin, C. C. (1987). The comparative method: Moving beyond qualitative and quantitative strategies. Berkeley, CA: University of California Press. Ragin, C. C. (2006). Set relations in social research: Evaluating their consistency and coverage. Political Analysis, 14(3), 291–310. Ragin, C. C. (2008). Redesigning social inquiry: Fuzzy sets and beyond. Chicago, IL: University of Chicago Press. Ragin, C., & Fiss, P. (2008) Net effects versus configurations: An empirical demonstration. In C. Ragin (Ed.), Redesigning social inquiry: Fuzzy sets and beyond. University of Chicago Press. (pp. 190–212). Rihoux, B., & Marx, A. (Eds.). (2013). QCA, 25 years after ‘The Comparative Method’: Mapping, challenges and innovations – Mini Symposium. Political Research Quarterly, 66(1), 167–235. Rihoux, B., & Ragin, C. (Eds.). (2009). Configurational comparative methods. Qualitative comparative analysis (QCA) and related techniques. Thousand Oaks, CA: Sage. Schneider, C. Q., & Wagemann, C. (2012). Set-theoretic methods in the social sciences. A guide to qualitative comparative analysis. Cambridge: Cambridge University Press. Thiem, A., & Dusa, A. (2012). Qualitative comparative analysis with R: A user’s guide. New York, NY: Springer. Vis, B. (2012). The comparative advantages of fsQCA and regression analysis for moderately large-N analyses. Sociological Methods & Research, 41(1), 168–198.