Advances in personality psychology 9780203000953, 0203000951, 9781315812175, 9780415227681

Table of contents :
Content: Book Cover
Half-Title
Title
Copyright
Contents
List of Figures
List of Tables
List of Contributors
Series Preface
Preface to Volume 2
1. Cognitive Approaches to Trait Anxiety
2. The Functional Significance of Temperament Empirically tested: Data Based on Hypotheses Derived from the Regulative Theory of Temperament
3. Personality and Information Processing: Biological Foundations of Though and Action
4. Three Superfactors of Personality and Three Aspects of Attention
5. Personality, Trait Complexes and Adult Intelligence. 6. Phenotypes and Genotypes of Personality and Intelligence: Similarities and Differences7. Personality Structure Across Cultures: Indigenous and Cross-Cultural Perspectives
8. Situations That Matter to Personality
Index.

Citation preview

Advances in Personality Psychology Volume One Edited by

Sarah E. Hampson

Advances in Personality Psychology

In the first volume of this new series, Sarah E. Hampson brings together a unique collection of critical reviews of key areas of personality psychology and integrative accounts of important work by internationally recognised experts in the field. Advances in Personality Psychology includes chapters on cross-cultural evidence for the Big-Five framework for personality description, type and trait approaches to understanding childhood personality, developments in psychometrics, the relationship between hostility and cardiovascular disease, and the connections between personality and emotions. In further chapters the view that personality cannot change in adulthood is chal­ lenged and the importance of environmental factors is revealed by an observational study of tw ins. This state-of-the-art volume will provide students, teachers and researchers of contemporary personality psychology with a highly valu­ able resource on recent developments in this area.

Sarah E. Hampson is Professor of Psychology and Health at the University of Surrey.

Advances in Personality Psychology Series editors: Ivan Mervielde, University of Gent, Belgium and Sarah E. Hampson, University of Surrey, UK.

Advances in Personality Psychology is designed to bring together orig­ inal reviews and critical evaluations of important new developments in the expanding field of personality psychology. As the only serial publi­ cation in this area it will provide an invaluable resource for students, teachers and researchers alike. The first volume of this new series offers a collection of chapters based on papers drawn from the Ninth European Conference on Personality Psychology held at the University of Surrey, Guildford, U K in July 1998. This conference was supported in part by a European Commission Sponsorship grant awarded by General X II, Science Research and Development.

Advances in Personality Psychology

Volume One

Edited by Sarah E. Hampson

V p Psychology Press X

Taylor& Francis Croup HOVE AND NEW YORK

First published 2000 by Psychology Press 27 Church Road, Hove, East Sussex, B N 3 2FA http://wrww.psypress.com Simultaneously published in the U SA and Canada by Psychology Press 270 Madison Ave, New York N Y 100 16 Transferred to Digital Printing 2010 Routledge is an im print o f the Taylor & Francis Croup, an in forma business © 2000 Sarah E. Hampson A ll rights reserved. N o part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, o r other means, now known o r hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library. ISBN 978-0-415-22768-1 C over design by Keith Marten

P ublisher's Note The publisher has gone to great lengths to ensure the quality of this reprint but points out that some im perfections in the original may be apparent.

C o n te n ts

List o ffigures List o f tables List o f contributors Series preface Preface to volume 1 1 Cross-language studies of lexical personality factors

vii viii ix xi xiii 1

G E R A R D S A U C I E R , S A R A H E. H A M P S O N A N D L E W I S R. G O L D B E R G

2 Variable-centred and person-centred approaches to childhood personality

37

I V A N M E R V I E L D E A N D J E N S B. A S E N D O R P F

3 Personality development in adulthood: the broad picture and processes inone longitudinal sample

77

R A V E N N A H E L S O N A ND V I R G IN IA S.Y. K W A N

4 Behaviour genetics of personality: the case of observational studies

107

P E T E R B O R K E N A U . R A I N E R R I E M A N N , F R A N K M. S P I N A T H AND A LO IS A N G L E IT N E R

5 Linking reliability and factor analysis: recent developments in some classical psychometric problems

138

J O S M.F. T E N B E R G E

6 Personality and health: cardiovascular disease M A R T H A C. W H I T E M A N , I A N J. D E A R Y A N D F. G E R R Y R. F O W K E S

157

vi

C ontents

7

Personality and emotion: cognitive Science perspectives

199

G ER A LD M A T T H EW S, D O U G LA S D E R R Y B E R R Y AND G R E G J. S IE G L E

Index

238

Figures

2.1 Relations between the prototypicalities for the three Q-types identified at ages 4-6 years 2.2 Three Q-types as a function of ego-control and ego-resiliency 2.3 Three major personality prototypes characterized by their average Big Five patterns across four studies 3.1 Differences with age on Agreeableness in five countries 3.2 Progression of scores on Self-control in three longitudinal samples 3.3 Progression of scores on Dominance in three longitudinal samples 3.4 Path model depicting the main relations among identity consolidation, identity in marriage and ego-resilience in terms of an ongoing process from age 21 to age 27 4.1 Illustration of the hypothesis that genetic influences operate at a global level whereas environmental influences operate at various levels 4.2 The ACE version of the common pathway genetic models tested in Chapter 4 study 6.1 Study design and timing of data collection in the Edinburgh Artery Study 6.2 Putative pathways of risk for cardiovascular disease and the different models of biological mechanisms 7.1 Schematic outlines of alternative cognitive science explanations for personality-emotion associations 7.2 Levels of explanation in personality and emotion research 7.3 Siegle’s neural network model of emotional information processing 7.4 A latent-trait model for effects of neuroticism and cognitive process variables on stress outcomes 7.5 Levels of explanation for cognitive correlates of traits associated with negative affect

53 56 67 81 90 91

100

126 129 171 185 205 206 216 225 227

Tables

1.1 Big Five subcomponents found in English and German 1.2 Summary of the lexical studies compared in Chapter 1 2.1 Comparison of dimensions emerging from four temperament models 2.2 Proportions of free parental child descriptors allocated to the Big Five. 2.3 Principal component analysis of H/'PIC facets in two samples 2.4 Consistency of Q-factors across different studies 3.1 Correlations with age on CPI scales in five samples 3.2 Change on CPI scales in three longitudinal samples 4.1 Intraclass correlations between twins for behaviour observations at three levels of aggregation 4.2 Goodness of fit indices for the independent pathway genetic model and the common pathway genetic model 4.3 Loadings on the latent phenotypic trait and proportions of variance accounted for in three common pathway genetic models 5.1 M IN R ES loadings (after Varimax) for four factors 5.2 Varimax-rotated M RFA loading matrix for Lord’s tests 6.1 Baseline characteristics of the Edinburgh Artery Study 6.2 Means (s.d.) of Bedford-Foulds Personality Deviance Scales and Revised scales in men and women 6.3 Multivariate linear regression analysis of the PDS and personal characteristics on smoking in 777 men and 748 women 6.4 Multivariate linear regression analysis of the PDS and personal characteristics on alcohol consumption in 789 men and 759 women 7.1 Summary statistics for multiple regressions of three subjective state dimensions on appraisal, coping and personality measures in a study of 108 undergraduates

2 8 47 59 62 64 85 89 131 131

132 150 151 173 174

177

177

224

Contributors

Alois Angleitner is Professor of Psychology at the University of Bielefeld, Germany. Jens B. A sendorpf is Professor o f Psychology at Humboldt University o f Berlin, Germany. Peter Borkenau is Professor of Psychology at Martin Luther University, Germany. Ian J. Deary is Professor o f Differential Psychology, University of Edinburgh, UK. Douglas D erryberry is Associate Professor of Psychology, Oregon State University, USA. F. G erry R. Fowkes is Professor o f Epidemiology and Head o f Public Health Sciences at the University of Edinburgh, UK. Lewis R. G oldberg is Senior Scientist at Oregon Research Institute and Professor of Psychology at the University of Oregon, USA. S arah E. Hampson is Professor o f Psychology at the University of Surrey, Guildford, UK. Ravenna Helson is Research Psychologist at the Institute of Personality and Social Research (formerly the Institute o f Personality Assessment and Research), and Adjunct Professor o f Psychology at the University o f California, Berkeley, USA. Virginia S.Y. Kwan is a student affiliate at the Institute for Personality and Social Research and a graduate student in the Department of Psychology at the University of California, Berkeley, USA. G erald Matthews is Associate Professor of Psychology, University of Cincinnati, USA. Ivan M ervielde is Professor of Psychology, University of Gent, Belgium.

x

Contributors

Rainer Riemann is Professor of Psychology at the University of Bielefeld, Germany.

Gerard Saucier is Assistant Professor of Psychology at the University of Oregon, USA.

Greg J. Siegle is a post-doctoral fellow at the Western Psychiatric Institute and Clinic, University of Pittsburgh, USA.

Frank M. Spinath is an Assistant Professor of Psychology at the University of Bielefeld, Germany.

Jos M.F. ten Berge is Professor of Psychometrics at the University of Groningen, The Netherlands.

Martha C. Whiteman is Research Fellow in the Department of Psychology at the University of Edinburgh, UK.

Series preface

Over the past two decades, personality psychology has emerged, phoenix­ like, from the ashes of its demise in the 1970s. The increasing success of scientific meetings such as the European Conferences on Personality, the significant growth in the number of articles, books and handbooks on personality psychology, and the growing impact of the major scientific journals on personality show that personality psychology is enjoying a renaissance both in Europe and in North America. Compared with other psychological disciplines, personality psychology currently lacks a high-quality publication outlet for comprehensive review papers and integrative reports on programmatic research. Although occasional reviews are published in the Annual Review o f Psychology and Psychological Review, a dedicated serial publication that reports on current developments and advances in personality psychology should be valuable for researchers, teachers and students of personality psychology. Advances in Personality Psychology>aims to fill this gap. It is designed to bring together original reviews and critical evaluations of important new developments in personality psychology. This new series aims to be a scholarly source of information on new theoretical developments, emerging research paradigms, and promising applications of personality psychology. It provides a forum for a variety of contributions: integrative reviews of current developments, overviews of research programmes and international collaborative projects, methodological advances, collabora­ tive papers emerging from conference symposia, and updates on new developments in the application of personality psychology to fields such as psycho-diagnostics, clinical, industrial/organisational and educational psychology. We envisage at least one volume every two years, depending on the availability of manuscripts reaching the required standards. The series editors will appoint volume editors in consultation with the executive committee of the European Association of Personality Psychology and the

xii

Series preface

publisher. Publication in this series will be principally by invitation. A ll submissions will be peer reviewed. We hope you will enjoy this first volume in the series. Ivan Mervielde, Gent, Belgium and Sarah E. Hampson, Guildford, U K Series editors

Preface to volum e I

The chapters in this volume are by a selection of invited speakers and pre­ senters at the Ninth European Conference on Personality Psychology held at the University of Surrey, Guildford, UK in July 1998. These authors were invited to produce extended versions of their papers with additional collab­ orators if they wished. The result is a collection of chapters that take some of the most fundamental questions of our field and address them in depth. The first chapter, by Saucier, Hampson and Goldberg, reviews the evidence for the lexical Big Five in cross-language studies. This chapter addresses the fundamental question of whether the broad dimensions of personality description are the same across numerous different languages. The authors identify a series of methodological features on which cross­ language studies vary and which affect the assessment of similarity between the Big Five in English and the factor structures that emerge in other languages. The Big Five have become widely accepted as a unifying frame­ work in personality psychology and this framework is used by authors of several subsequent chapters. Mervielde and Asendorpf, in Chapter 2, relate the personality and temperament factors found in children to the Big Five in adults. In so doing, they evaluate the similarities among the various theories of temperament, and between theories of temperament and theories of personality in children. They review both variable-centred and person-centred approaches to the study of childhood personality and conclude that both make unique contributions to our understanding of chil­ dren’s personalities. The issue of personality stability versus change in adulthood is the focus of Chapter 3 by Helson and Kwan. This chapter reviews findings from several cross-sectional and longitudinal studies of adults, and then gives an overview of the many findings from the Mills study. The authors demonstrate that cross-sectional and longitudinal find­ ings are consistent in identifying certain aspects of personality that change over the course of adulthood. In their discussion of longitudinal studies, and of the Mills study in particular, they relate personality change in adulthood to the social influences that have affected different cohorts as they moved along their unique pathways through time.

xiv

Preface to volume I

In Chapter 4, Borkenau, Riemann, Spinath and Angleitner address the behaviour genetics of personality by focusing on observational studies. The genetics of adult personality is based exclusively on self-report measures of personality, which may have distorted the estimates of genetic, shared and non-shared contributions. They provide a comprehensive review of observational studies of young twins, and conclude that the shared envi­ ronment emerges as a more important source of variance in these studies than in self-report studies. They present findings from the German Observational Study of Adult Twins to support their hypothesis that genes affect behaviour at a more global level, whereas the environment is the only source of variance at more specific levels. The last three chapters address the interface between personality and other fields in psychology. In Chapter 5, ten Berge bridges the gap between psychometricians and personality psychologists by providing a discussion of some recent developments in factor analysis and the assessment of internal reliability. In Chapter 6, Whiteman, Deary and Fowkes consider the relationship between personality and health by examining in depth the association between Type A and hostility-related traits and cardiovascular diseases. Their review of past evidence indicates a small but consistent association that needs to be studied longitudinally so that causal relations and explanatory mechanisms can be identified. They present data from the Edinburgh Artery Study for both concurrent and prospective associa­ tions between various measures of hostility and disease variables, and point to possible underlying biological pathways. In Chapter 7, Matthews, Derryberry and Siegle examine the associations between personality and emotion by studying individual differences in emotional reactivity within a cognitive science framework. They focus on three cognitive science perspectives: cognitive neuroscience, connectionism, and transactional stress processes, each reflecting one of the authors’ research programmes. This chapter is an integrative and collaborative enterprise that has gener­ ated a novel approach to a classic issue in personality. All the contributions to this volume underwent the peer review process. I would like to take this opportunity to express my thanks to the colleagues whose advice I sought: John C. Barefoot, Paul Costa, Alexandra M. Freund, Lewis R. Goldberg, Robert Krueger, Robert R. McCrae, Roy P. Martin, Ivan Mervielde, Todd Q. Miller, Edward Necka, Mary K. Rothbart, Gerard Saucier, Warren W. Tryon and Gerdi Weidner. Finally I would like to thank the Oregon Research Institute for providing a congenial environ­ ment for my sabbatical during which time much of the editing of this book was completed. Sarah E. Hampson Guildford, U K

Chapter I

Cross-language studies of lexical personality factors Gerard Saucier, Sarah E. Hampson and Lewis R. Goldberg

The rationale for lexical studies rests on the assumption that the most meaningful personality attributes tend to become encoded in language as single-word descrip­ tors. Based on this rationale, a number of studies have been conducted examining the factor structure of adjectival descriptors extracted from dictionaries. Using as an initial working hypothesis an Anglo/German version of the Big Five factor struc­ ture, we review lexical studies in English and 12 other languages, and examine their fit with the Big Five. W e find that the inclusion of highly evaluative terms and phys­ ical descriptors may lead to factors beyond the Big Five. W ith more conventional variable selections emphasizing disposition descriptors, indigenous structures resem­ bling the Big Five seem to emerge most readily in northern European languages. The Big Five have been only partially reproduced in many other studies, although this might reflect the differences in their procedures for selecting variables. Factor structures with fewer factors (one, two or three) may be more generalizable crossculturally, perhaps because they are less affected by these methodological variations. At this point, more attention should be given to the influence of the major types of procedural differences on the resulting factor structures. Such variations in research design and analysis may account for much of what otherwise might be prematurely interpreted as cultural differences. Suggestions for improving the lexicalstudy paradigm are offered.

INTRODUCTION After many years of dispute, personality psychologists have recently shown some signs of agreement on a framework for classifying and organizing personality traits. A consensus among many in the field, which was evident by the early 1990s (e.g. McCrae and John, 1992), posits that five broad dimensions of personality capture the most important aspects of lexicalized personality traits. Table 1.1 indicates subcomponents associated with each of these ‘Big Five’ factors from a recent study of English and German personality descriptors (Saucier and Ostendorf, 1999).1 In an influential critique, Block (1995) cautioned against settling prema­ turely on the five-factor approach. However, even those most closely allied with the five-factor framework regard it as a working hypothesis, rather than a final solution (e.g. De Raad, 1998; Goldberg and Saucier, 1995).

2

Saucier, Hampson and Goldberg

Table l.l Big Five subcomponents found in English and German Example terms in English Domain

Subcomponent

Positive pole

Negative Pole

Extraversión (1)

Assertiveness Activity-adventu rousness Unrestraint Sociability

assertive daring talkative cheerful

weak unadventurous shy unsociable

Agreeableness (II)

Warmth-affection Generosity Gentleness Modesty-humility

affectionate generous agreeable modest

cold selfish harsh egotistical

Conscientiousness (III)

Orderliness Reliability Industriousness Decisiveness

organized responsible ambitious decisive

sloppy undependable lazy inconsistent

Emotional Stability (IV)

Insecurity3 Emotionality3 Irritability3

relaxed unemotional undemanding

insecure excitable irritable

Intellect (V)

Intellect Imagination Perceptiveness

intelligent creative perceptive

unreflective unimaginative shortsighted

Source: Saucier and Ostendorf (1999). Note a These subcomponents are labelled by their negative rather than positive pole.

Like any scientific model, its prime function is to stimulate research and either to be proved wrong or, more likely, to be refined in the light of new evidence. Additional dimensions could emerge in studies of different cultures or studies of different forms of personality language (e.g. typenouns), or different kinds of personality data. Moreover, findings from any of these sources could point to a better structure at a different hier­ archical level, one with more factors or fewer factors. Initial problems in developing a scientifically compelling descriptive classification include (a) defining the universe of personality attributes that will be classified and (b) determining the particular attributes from that universe that are important enough to be represented in the final classifi­ cation. The lexical approach to personality solves these problems by a major and far-reaching hypothesis: the most important individual differ­ ences eventually become encoded as single words in the natural (i.e. nonscientific) language (Allport and Odbert, 1936; Cattell, 1943; Goldberg, 1981; Saucier and Goldberg, 1996b). Therefore, at least for the major modem languages, their dictionaries contain a reasonably comprehensive

Studies of lexical personality factors

3

set of those individual differences that past speakers of those languages have needed to communicate. This ‘lexical hypothesis’ has two reason­ able corollaries: (a) frequency of use has a rough correspondence with importance, and (b) the number of words referring to a particular person­ ality attribute will be a rough guide to the importance of that attribute for the speakers of the language2 (Saucier and Goldberg, 1996b). These two conceptually independent criteria can both be satisfied if one examines the most frequently used subset of relevant terms in the language, and then searches for structure (i.e. factors, clusters) in that subset. Although the peculiarities of trait structure derived from any particular language could be of considerable interest, cross-language regularities have even more scientific import. In the search for such regularities, the most important personality dimensions will be those that replicate across samples of participants, targets of descriptions, variations in analytic procedures, and across languages (Saucier and Goldberg, 1996b). If a factor structure can be recovered in a wide variety of the world’s languages, then this would provide support for the possibility that those dimensions might represent cultural ‘universals’ of person description (Goldberg, 1981). This chapter reviews the evidence for the reproducibility of structural repre­ sentations - especially the Big Five - across those languages that have been studied to date. Some of the major studies in English are summar­ ized first to describe how the five-factor framework came to be established. Next, lexical studies in other languages are examined for the reproducibility of the five-factor framework. Because these studies differ in a number of their methodological details, we consider the potential impact of these differences on the findings obtained. Possible alternative structural frame­ works are also considered. Finally, we consider the broader significance of this body of research.

THE LEXICAL APPROACH IN ENGLISH The lexical assumption has been the guiding rationale for personality taxonomers dating back at least to the 19th-century British scientist, Francis Galton (1884), who used a dictionary to identify terms that described personality attributes. Particularly influential were Allport and Odbert (1936), who extracted from an unabridged English dictionary almost 18,000 words that could be used to describe individual differences. O f these, roughly 4500 were designated by Allport and Odbert as likely personality traits, the remainder having been classified as temporary moods or activities, social effects, physical or medical terms, predomin­ antly evaluations, and the like. This classification scheme proved highly influential, leading to the decision by Norman (1967) and Goldberg (1982) to exclude from their initial analyses terms describing one’s physical

4

Saucier, Hampson and Goldberg

appearance, the effects one has on others, sheer evaluations, and tempor­ ary activities, moods and states. The detailed history of the initial extraction and later modification of trait lists has been documented repeatedly (e.g. Digman, 1990; Goldberg, 1993, 1995; John, 1990; McCrae and John, 1992; Wiggins and Trapnell, 1997), and thus will not be repeated here. In this summary, we will focus primarily on the English-language studies that permit comparisons with those in other languages. At the dawn of the computer age, lexical studies in English found impetus from work by Norman (1963, 1967) and Goldberg (1976, 1982). Norman’s (1963) use of Roman numerals to identify the Big Five factors has become a hallmark of the lexically based taxonomic tradition; the five factors were ordered roughly by the number of reasonably familiar traitdescriptive adjectives in English available to describe that domain: I (Extraversión) and II (Agreeableness) with the most such terms, followed by III (Conscientiousness), and then IV (Emotional Stability) and V (first labelled Culture, here labelled Intellect) with the least adjectives. Norman (1967) retraced the footsteps of Allport and Odbert (1936) and extracted person-descriptive terms from a new edition of the same unabridged English dictionary that they had used. Norman’s four-person research team refined the resulting set of 18,125 terms by excluding those classified as pure evaluations and as physical and medical terms and cate­ gorizing the remainder as (a) stable traits, (b) temporary states and activities or (c) social roles, relationships and effects. Through this process, Norman identified roughly 2800 stable trait terms. Goldberg (1976, 1982) reduced Norman’s stable trait pool to 1710 by eliminating the least commonly used terms and those that seemed the least dispositional in nature. Using these same subjective criteria, Goldberg later developed a 540-term set that was employed in a number of studies in which university students described themselves and others (Goldberg, 1990). After additional samples of university students had provided familiarity ratings for these and other common English adjectives, Saucier and Goldberg (1996a) analysed the most familiar subset of 435 adjectives in a combined sample of nearly 900 (including 507 self-descriptions and 392 descriptions of other), and found a clear five-factor structure. Peabody (1987) developed 53 bipolar scales to represent a large pool of adjectives that included Goldberg’s 540 terms. Four college students made judgements of semantic similarity between all of the terms from each pole and each of the bipolar scales. Analyses of these judgements of 'internal structure’ revealed the Big Five factors, plus a small ‘Values’ factor; the first three factors (Assertiveness [Extraversión], Agreeableness, and Conscientiousness) were, however, much larger than the remaining ones. Further analyses of these bipolar scales in self and peer descriptions by college students consistently found the Big Five factors (Peabody and

Studies of lexical personality factors

5

Goldberg, 1989). But again, the first three factors (i.e. Extraversión [I], Agreeableness [II] and Conscientiousness [III]) were substantially larger than the other two (Emotional Stability [IV ] and Intellect [V]). The English-language studies of Goldberg and Peabody share at least one limitation: potentially relevant terms referring to states, social evalu­ ations, and physical and appearance characteristics were excluded from those item pools. A study by Tellegen and Waller (1987; summarized also by Waller, in press; Waller and Zavala, 1993) cast a wider net. Instead of relying on Allport’s or Norman’s compilations, an abridged dictionary was divided into roughly sixty 25-page sections, and seven or eight non­ contiguous pages from each section were randomly selected. On each selected page, the first adjective that could be fitted into the stems ‘tends to be ____ ’ and ‘is often ____ or otherwise appeared to be persondescriptive, was extracted.3 A set of 400 terms was selected, including terms describing social effects, pure evaluations and temporary states. Self-reports using these 400 terms were provided by almost 600 univer­ sity students. In order to encourage meaningful responses to these terms, many of which would be unfamiliar to most native speakers, participants were presented with portions of the dictionary definitions. Factor analyses (5 to 20 factors) of the 400 variables were conducted, and a seven-factor solution was judged to be the most compelling and meaningful (Waller, in press). The seven factors were judged to correspond to the Big Five, plus orthogonal dimensions labelled Positive Valence (e.g. Important, Outstanding) and Negative Valence (e.g. Evil, Vicious). The last two factors drew on the ‘pure evaluation’ descriptors excluded in all previous lexical-factor studies. Subsequent studies in Hebrew (Almagor, Tellegen and Waller, 1995) and Spanish (Benet-Martinez and Waller, 1997) used a similar methodology. These ‘Big Seven’ studies have several strengths. Solutions with various numbers of factors were examined, the range of variables was very broad, and samples of both variables and subjects were large. However, variable selection was based on dictionary pages, rather than on entries. This procedure approximates to representative variable selection only if the dis­ tribution of important descriptors (and their semantic content) has a rectan­ gular (flat) distribution across pages (or parcels of pages) in a dictionary. However, in fact some pages have many familiar person-descriptors (e.g. in English ‘in____ ’), whereas other pages have absolutely none (e.g. ‘x____ ’). As a result, page-sampled variable sets may often be unrepresentative of the lexicon, particulary if a language’s person-descriptors are disproportion­ ately prone to contain certain prefixes (e.g. good-, self-, well- in English). Additionally, such sampling procedures are likely to net many unfamiliar terms, violating the frequency-importance association that we believe is integral to the lexical approach. For example, in English, Rhadamanthine and Tenebrous are as likely to be sampled as Kind and Depressed.

6

Saucier, Hampson and Goldberg

Recently Saucier (1997) found a structure resembling the Big Five in a set of familiar English adjectives restricted to dispositions and states. When a wider range of terms, including social evaluations and physical attributes, was included, two additional factors emerged: Attractiveness and a factor resembling Big Seven Negative Valence. Whether one obtained the Big Five or the related seven-factor structure depended upon variable selection. In contrast, a three-factor solution was replicable across variable selections. The three factors, Extraversión, Agreeableness and Conscientiousness, were each somewhat broader than those from the Big Five; these are henceforth referred to as the Big Three. In further studies, Saucier (1998) combined eight American samples (total N = 3062) with representative lexical variable selections, and found that structures of one, two and three factors were all more replicable than five-factor structures, with their superior replicability especially evident in peer-rating samples. The factors in the three-factor structure closely resem­ bled the first three of the five-factor structure, suggesting that the most robust version of the Big Five may consist of the three large factors, along with two smaller factors whose replicability is weaker. In English, then, a Big Five structure has been replicated across different investigators, variable selections and participant samples. These studies have indicated that the form and replicability of this factor structure is affected by at least two major aspects of one’s methodology: (a) variable selection, and (b) the kind of targets (self, liked peers, diversely evalu­ ated peers, semantic concepts) that are rated (Peabody and Goldberg, 1989). The extensive lexical studies in English, all conducted in the United States, have pioneered many of the techniques subsequently applied to other languages. In the rest of this chapter, we will compare the findings from these cross-cultural lexical studies.

CROSS-LANGUAGE APPLICATIONS OF THE LEXICAL APPROACH Cross-language studies of the five-factor framework have used both etic (imported) and emic (indigenous) procedures (Berry, 1969). In the former, an imported (usually Western) framework is tried out in the new culture to see how well it fits (e.g. a Big Five measure is translated into another language), whereas emic approaches allow for the indigenous framework to emerge without imposing constraints (e.g. a representative sample of the language’s trait adjectives is analysed). If, for example, the Big Five is reproduced in etic studies, it does not constitute evidence for the crosscultural ubiquity of the personality factors, but only that people in other cultures are able to employ those dimensions. In most lexical studies, investigators use an emic approach to identify the indigenous factors of

Studies of lexical personality factors

7

personality description, and then use an etic measure to compare these dimensions with ones found in other languages. The present review is focused only on emic studies, and thus does not include studies primarily designed to develop reliable and valid measures of the English five-factor model in another language (e.g. Benet-Martinez and John, 1998; McCrae and Costa, 1997). Emic lexical research is summarized in Table 1.2. Not included are ongoing projects for which reports are not yet available; for example, lexical studies are presently being carried out in Portuguese, Quechua and Romanian (De Raad et al., 1998), as well as in Croatian, Greek, Norwegian and Slovak. In Table 1.2, the language under study is provided in the first column; the second column lists the names of the investigators; the third column indicates the source of the terms used in that study; and the fourth column provides the type of data that was obtained (most typi­ cally self or peer descriptions, or both). The method(s) used to evaluate factor equivalence are provided in the fifth column, and the final column provides a very rough summary of the findings. The particular Big Five factors that were seemingly identified in each study are indicated by their traditional Roman numerals (I to V). We begin our review with studies in three subgroups of Indo-European languages; Germanic (e.g. German, Dutch), Slavic (e.g. Czech, Polish, Russian), and Romance (e.g. Italian, Spanish).

GERMANIC LANGUAGES The German personality taxonomy Much of psychology has originated in Germany, and this is also true of lexical studies. Klages (1932) articulated the lexical rationale, and Baumgarten (1933) created the first list of (1629) German personality descriptors, both of which influenced the work of Allport and Odbert (1936). Lexical studies of the German language were taken up again in the 1980s when 5101 personality-relevant adjectives (e.g. cynical), 2212 type nouns (e.g. cynic) and 3607 attribute nouns (e.g. cynicism) were extracted from a comprehensive dictionary and supplementary lexical sources. Judges classified the adjectives into 5 broad and 13 subordinate categories, a refinement of the category systems used by Norman (1967); for greater detail, see Angleitner, Ostendorf and John (1990). Ostendorf (1990) studied those 430 familiar adjectives that a majority of judges had classified as either temperament and character traits or abil­ ities and talents, or their absence. Over 400 adult participants completed self-reports using the 430 adjectives and measures of imported Big Five and circumplex models, and 95 per cent of those individuals were described

Table 1.2 Summary of the lexical studies compared in Chapter I Language

Authors

Source of terms

English

Goldberg (1990)

Unabridged dictionary

Tellegen and W aller (1987)

Abridged dictionary

Saucier and Goldberg (1996a)

Familiar terms from Goldberg (1990)

Saucier (1997)

Familiar terms (broad set)

German

Ostendorf (1990)

Comprehensive dictionary

Dutch

Brokken (1978)

Unabridged dictionary

De Raad et al., (1992)

Brokken’s (1978) set of terms

Czech

Hrebícková et al. (1995)

Standard dictionary

Polish

Szarota and Ostendorf (1997)

Concise dictionary

Russian

Shmelyov and Pokhil’ko (1993)

Previous lists Dictionaries

Data collected

Method for determining factor equivalence

Replicated Big Five factors

Self-ratings Peer ratings

Inspection of factor content

1, II. III. IV, V

Self-ratings

Inspection of factor content

1, II. III. IV. V (plus NV, PV)

Self-ratings Peer ratings

Inspection of factor content

1. II. III. IV. V

Self-ratings Peer ratings

Inspection of factor content

1. II. III. IV. V (plus NV. Att)

Self-ratings Peer ratings Big-Five markers

Correlations among factor scores from different types of data

1. II. III. IV, V

Self-ratings Peer ratings

Inspection of factor content

1. Il, III, IV, but V less clearly Intellect

Self-ratings

Inspection of factor content

1, II, III, IV, but V less clearly Intellect

Self-ratings Peer ratings N EO

Correlations with N EO Inspection of factor content

1, II, III, IV, V, with IV less clear

Self-ratings Peer ratings N EO

Inspection of factor content Correlations with N EO

1, II, III, IV, V, with 1 and IV less clear

T rait-inference judgements

Congruence with English markers

1, II, III, IV, V, with IV less clear

Italian

Caprara and Perugini (1994)

Abridged dictionary

Self-ratings N EO BFQ

Inspection of factor content Correlations with N EO

1, III, and V, with II and IV rotational variants, and V similar to V in Dutch.

Di Bias and Forzi (1998)

Unabridged dictionary

Self-ratings Peer ratings

Inspection of factor content Correlations with Big Five markers

1, III, with II and IV each splitting into two factors

Spanish

Benet-Martinez and W aller (1997)

Abridged dictionary

Self-ratings Inspection of factor Big Seven measure Big Five measure Correlations between emic and etic factors

1, II, III (plus PV and NV) content

Hebrew

Almagor et al. (1995)

Abridged dictionary

Self-ratings

Inspection of factor content

II, III, V (plus NV), with 1 and IV splitting into two factors

Hungarian

Szirmàk and De Raad (1994)

Unabridged dictionary

Self-ratings

Inspection of factor content

1, II, III, IV, with V less clear

Turkish

Somer and Goldberg (1999)

3 abridged dictionaries

Self-ratings Peer ratings

Inspection of factor content

1, II, III, IV, V

Goldberg and Somer (1999)

Most familiar terms from Somer and Goldberg (1999)

Self-ratings

Correlations with Turkish Big Five markers

1, II, III, IV, V

Hahn, Lee and Ashton (1999)

2 dictionaries Terms from free descriptions

Self-ratings

Correlations with Korean Five Big markers

1 and II, with III, IV, and V less clear

Abridged dictionaries

Self-ratings N EO Big Five markers

Correlations with Big Five markers

1, II, III, V; with IV split into two factors

Korean

Filipino Church et al. (1997) (Tagalog)

Note I = Extraversión, II = Agreeableness, III = Conscientiousness, IV = Emotional stability, V = Intellect, PV = Positive Valence, NV = Negative Valence, Att = Attractiveness, NEO = NEO Personality Inventory, BFQ = Big Five Questionnaire.

10

Saucier, Hampson and Goldberg

by from one to three acquaintances. Analyses of the 430 adjectives gener­ ated five highly reliable factors in ipsatized ratings of either self or acquaintances. Correlations of these factors with corresponding factors from the imported measures of the Big Five averaged over 0.70 (see also Ostendorf and Angleitner, 1994, Table 1). The three-factor solutions were even more replicable; only at the threefactor level were all factor-comparability coefficients for both ipsatized and raw data above 0.90.4 Lexical studies in The Netherlands The first lexical studies in Dutch were carried out by Brokken (1978) and Hofstee (Hofstee, Brokken and Land, 1981). Further data and analyses were provided by De Raad, Hendriks and Hofstee (1992). In the original studies, 8690 person-descriptive adjectives were extracted from a large dictionary and then reduced to 6055 that were judged by at least one of four individuals to be stable traits. At a later stage of the project, those 1203 adjectives that fitted best into the sentence stems ‘He/she i s ____ by nature’ and ‘What kind of person is he/she?_____ ’ (determined by the average ratings of university students) were retained. These criteria seem to have served to retain temperament and character terms more abun­ dantly than those describing abilities and talents, with the result that factor V in the Dutch solution has a somewhat less intellectual character than the fifth factor in English and in German. Using the 1203 adjectives, 200 pairs of university students who knew each other well rated both themselves and their partners (Brokken, 1978). Later, De Raad (1992) collected 200 additional self-ratings from univer­ sity students on 551 of the 1203 adjectives selected by more stringent use of the same criteria used by Brokken. The ipsatized ratings of these 600 participants (400 self, 200 acquaintance) were pooled in a factor analysis of the 551 adjectives; four, five and six factors were rotated. In each of these solutions the first four factors included Extraversión (I), Agreeableness (II), Conscientiousness (III) and Emotional Stability (IV ) from the Big Five. The fifth factor in the five-factor solution contrasted intellectual autonomy and independence with conventionality (e.g. Critical, Rebellious versus Meek, Docile), and was interpreted as a variant of Big Five Intellect. In general, the Dutch studies supported the Big Five (though only weakly for the Intellect factor; see De Raad, 1994). De Raad and his colleagues have pioneered the application of the lexical approach to personality-type nouns (De Raad and Hoskens, 1990; De Raad and Ostendorf, 1996) and personality-relevant verbs (De Raad e ta i, 1988). These studies indicate that personality nouns tend to have stronger eval­ uative connotations than do adjectives. Some of the noun factors (labelled ‘Malignity’) are reminiscent of Negative Valence from the Big Seven. The

Studies of lexical personality factors

II

verb studies led to a two-factor representation, with one factor defined by verbs like Care for and Cooperate (versus Curse and Threaten), and the other factor defined by verbs like Decide and Lead (versus Flee and Brood); the factors have some resemblance to Big Five Agreeableness (II) and Extraversión (I), respectively. They probably also relate to factors in two-factor solutions from adjectives in some other languages (e.g. Saucier, 1998). Studies in these Germanic languages have been supportive of a strin­ gent emic Big Five model, according to which the same ‘Big Five’ factors emerge in five-factor solutions, when based on a representative set of single-word personality descriptors in a language.5 The stringent require­ ments of this model can be distinguished from a looser counterpart, in which the appearance of the Big Five factors in any solutions of five or more factors is interpreted as supportive evidence for the Big Five model.

SLAVIC LANGUAGES Polish In a study by Szarota (1996; Szarota and Ostendorf, 1997), two judges independently scanned a concise Polish dictionary for person-descriptive adjectives, and the 1811 terms selected by both judges were retained; to this list were added a further 28 terms taken from other lists of Polish personality descriptors. Ten judges then rated these 1839 terms for clarity of meaning and personality relevance, and classified them using the system developed by Angleitner, Ostendorf and John (1990). The 290 adjectives categorized as dispositions were used for self and peer ratings in a sample of 369 high-school students. In both the self and the peer ratings of ipsatized data, five factors similar to the Big Five factors could be iden­ tified. Agreeableness (II), Conscientiousness (III) and Intellect (V ) were virtually identical to the English and German counterparts. However, the Extraversión (I) factor had no sociability facet, and Emotional Stability (IV ) included content related to self-control. Czech Hrebícková, Ostendorf and Angleitner ( 1995) provided a preliminary report on a lexical analysis of the Czech language. All potentially personality­ relevant terms were extracted from a standard Czech dictionary. The 4145 such terms were categorized by six judges into 13 different types of descriptors, based on the classification system developed by Angleitner, Ostendorf and John (1990). Those 366 terms that the majority of judges had classified as dispositions were rated for self-descriptiveness by 397

12

Saucier, Hampson and Goldberg

research participants. Using ipsatized data, factor solutions with five and more factors were examined and interpreted both by inspection of the high-loading terms and by correlations with expert prototypicality ratings with respect to the Big Five. Four of the Big Five factors were clearly identifiable, with Emotional Stability being the only problematic factor (the Czech version of Emotional Stability was also related to Agreeableness and Intellect when correlated with the prototypicality indices for these factors). Generally, however, the Czech study provides substantial support for the Big Five structure. In a subsequent study, Hrebickova et al. (1999) studied the structure of Czech personality-relevant verbs. The two-factor solution was similar to the Dutch two-factor verb structure described above. The authors reported solutions with up to six factors; notably, in no solution did a factor emphasizing Intellect content appear. Russian Shmelyov, Pokhil’ko and Kozlovskaya-Telnova (1988, 1991; Shmelyov and Pokhil’ko, 1993) completed the earliest Russian lexical work without access to reports of the major earlier studies. Consulting two dictionaries and previous lists of trait-descriptive terms, they compiled a set of 2090 personality-relevant terms, including 666 nouns (most of which seemed to refer to undesirable traits) and 1424 adjectives. Studies of the internal structure (semantic similarity) of large subsets of this pool were reported by Shmelyov and Pokhil’ko (1993). The investigators rotated 15 factors, far more than in other studies, and these were compared with factors from translations of adjective clusters and cluster labels from the studies by Goldberg (1990). The four largest Russian factors (Altruism [II], Intellect [V], Energy-gaiety [I], and Conscientiousness [III]) and to a lesser degree the sixth (Emotional Stability [IV ]) were interpreted as highly congruent with the Big Five factors from Goldberg (1990). At present, the findings from these Russian studies are difficult to compare with those from other studies, due to their exclusive use of internal rather than external data, and the lack of evidence about the nature of factors at higher hierarchical levels. Nonetheless, most of the Big Five factors (Emotional Stability less strongly) appeared in these concept-rating analyses. Therefore, we can regard the Russian studies, like the other two in Slavic languages, as both (a) loosely supportive of the Big Five and (b) indicating possibly lower replicability for the Emotional Stability factor. An unpublished study of Croatian, a south Slavic language, also found a recognizable Big Five structure in ipsatized ratings, although there was some rotation in the plane defined by the Agreeableness and Emotional Stability factors (Mlacic, 1999). In contrast, the replication of the Big Five has been more problematic in analyses of Romance languages.

Studies of lexical personality factors

13

ROMANCE LANGUAGES Studies in Italy Two independent taxonomic projects have been conducted in Italy - a ‘Roman’ project conducted by Caprara and Perugini (1994), and a ‘Trieste’ project conducted by Di Bias and Forzi (1998, 1999). The procedures used in the Roman project were similar to those used in the Dutch studies, whereas the procedures used by the Trieste investigators were similar to those used in the German project (De Raad, Di Bias and Perugini, 1997). In the Roman project (Caprara and Perugini, 1994), an abridged diction­ ary was scanned for person-descriptive adjectives and 8532 were selected. Later, 1337 of these terms were retained by a criterion of utility for describing personality, based on the judgements of four experts. Further utility ratings by 22 lay judges were used to reduce the number of terms to 492, which were administered to 274 research participants for selfratings. The five-factor solution from analyses of these ipsatized data revealed Big Five Extraversión [I] and Conscientiousness [III] factors. A third factor was labelled Quietness [or Peacefulness] versus Irritability; a fourth factor, labelled Selfishness versus Altruism, included adjectives related to tough-mindedness and emotionality. The investigators concluded that the third and fourth factors were rotational variants of Agreeableness (II) and Emotional Stability (IV ). The fifth factor, labelled Conventionality, included descriptors related to conformity and traditionalism, and thus resembled the fifth factor in Dutch. In the Trieste project (Di Bias and Forzi, 1998), five judges searched through an unabridged Italian dictionary for person-descriptive terms. They extracted 11,010 terms, including 3780 adjectives, 1428 adjectives that can be used as type nouns, 2566 type nouns, and 3326 attribute nouns. The 1586 most personality-relevant, frequently used and clear terms (as determined by ratings from at least 10 secondary school students) of the adjectives and the adjectives that can be used as type nouns were later classified by university students into the 13 categories developed by Angleitner, Ostendorf and John (1990). The most prototypical disposi­ tional terms (as defined by high inter-judge agreement), which included 344 adjectives and 124 adjectives that could be used as type nouns, were then reduced to 314 and 118 of each kind by the elimination of synonyms and highly ambiguous terms. The set of 314 adjectives was administered to a heterogeneous sample of 427 individuals for self-ratings and another sample of 277 secondary school students for ratings of someone they knew well. The two data sets were ipsatized and analysed both separately and jointly, with three- to nine-factor solutions being examined. Self and other ratings produced comparable three-factor solutions that were easily identifiable as Big Five

14

Saucier, Hampson and Goldberg

factors I, II and III. In the five-factor solutions, Extraversión and Agreeableness each split into a pair of more specific factors (Assertiveness and Sociability; Quietness [or Peacefulness] and Tender-mindedness). Distinct aspects of factor IV appear to have been merged with the inter­ personal aspects of factors I and II to produce Assertiveness (versus Fearfulness) and Quietness (versus Irritability), with some other aspects associated with Tender-mindedness. A rather weak version of an Intellect factor only emerged when at least seven factors were rotated. Subsequently, Di Bias and Forzi (1999) attempted to replicate this struc­ ture in a study of ipsatized self-ratings from 369 participants using a set of 369 terms.6To construct this new variable set, the investigators excluded 63 terms with very skewed response distributions from their combined set of 314 adjectives and 118 adjectives that can be used as type nouns (described above). In these new analyses, four factors from the Italian five-factor solution, labelled Sociability (weakly related to Big Five factor I), Placidity/Quietness (a blend of Big Five factors II and IV ), Conscientiousness (III), and Self-assurance/Assertiveness (a blend of Big Five factors I, IV and V), included content from the first four of the Big Five, but there were no exact counterparts. Findings from the two Italian projects differ with regard to the nature of the Extraversión factor and the identification of a factor comparable to Big Five Intellect. The Roman set of descriptors included some non-dispositional terms, notably temporary states and social effects, and thus appears to be broader than the Trieste set. Generally, the Italian studies illustrate an important principle: a language containing terms describing Emotional Stability or Intellect may still not yield these Big Five lexical factors, because the terms may be highly associated with terms from other factors. Spanish Dictionary-sampling procedures were employed in a study of Spanish personality-descriptors (Benet-Martinez and Waller, 1997). Every fourth page of a 1666-page dictionary was inspected, using selection criteria like those of Tellegen and Waller (1987) and 299 adjectives were selected. As in the American study, the adjectives were accompanied by a synonym or short definition. Self-ratings were elicited from a sample of 894 university students. Prior to factor analysis, the effects of gender, age and primary language (Spanish versus Catalan) were statistically controlled, and the data ipsatized. Solutions of five to eight rotated factors were examined, and the seven-factor solution was reported to be ‘easily interpreted’, whereas the others were ‘structurally ambiguous’. The seven-factor solution corresponded in some respects to the seven factors reported by Tellegen and Waller (1987) and measured by the Inventory

Studies of lexical personality factors

15

of Personality Characteristics (IPC ; Tellegen, Grove and Waller, 1991). Correlations between the seven indigenous factors and those from the IPC were between 0.60 and 0.79 for ‘Pleasantness’ (which resembles both Positive emotions and Extraversión factors in other studies), Agreeableness, Conscientiousness (labelled ‘temperance’), and Positive Valence. The other three factors were substantially smaller in size. One of them correlated moderately (0.47) with Negative Valence, whereas the other two indige­ nous factors, Engagement and Openness (Introspectiveness might be an alternative label), had low correlations with the remaining IPC factors. The five-factor solution included Agreeableness, Conscientiousness, Conventionality and Negative Valence factors. There was also a large factor that counterposed Positive Valence and Positive Emotionality with Negative emotionality; this factor might resemble Assertiveness (from Italian) or the broader, Big Three factor of Extraversión. In summary, Benet-Martinez and W aller’s (1997) seven-factor solution might be seen as a replication of the three largest factors in Peabody and Goldberg (1989), as well as Positive Valence and (weakly) Negative Valence. As in the Italian studies, Emotional Stability subcomponents were spread across several factors, and Intellect was not clearly recovered. Given the use of page-sampling procedures, the potential for replication of either the Big Seven or Big Five models may be limited. Because the Spanish factors include versions of the first three Big Five factors, it would be useful to examine one-, two-, three- and four-factor solutions in these data. As our review has progressed from Germanic to Slavic and on to Romance languages, the emic replication of the English/German Big Five has waned steadily. Does this trend extend itself into non-Indo-European languages? On both linguistic and cultural grounds, if the Big Five are not culturally robust, then they should be even less evident in nonIndo-European languages.

NON-INDO-EUROPEAN LANGUAGES Hebrew Hebrew, like Arabic, is a member of the Semitic language group, and thus could provide an interesting contrast with other lexical studies. Almagor, Tellegen and Waller (1995) carried out such a study using a dictionary page-sampling technique similar to that used in the Spanish study just described. Two research assistants searched every fourth page of a 1600page Hebrew dictionary and extracted the first personality-descriptive adjective or noun they encountered. This process yielded 326 terms, which were reduced to 252 after removing synonyms. University students (N = 637) rated the 252 words for self-descriptiveness.

16

Saucier, Hampson and Goldberg

The investigators focused on a seven-factor rotation, which they compared with the Big Seven model (Tellegen and Waller, 1987) by observing the factor loadings of the Hebrew translations of Big Seven marker traits. The Hebrew factors labelled Agreeability (II), Dependability (III) and Negative Valence seem roughly equivalent to their Big Seven (and thus Big Five) counterparts. However, the terms with the highest loadings on the Positive Valence factor were all related to Intellect (e.g. Sophisticated, Sharp, Knowledgeable). Emotional Stability terms were found on factors labelled both Positive Emotionality/Agentic (e.g. Depressed, Sad) and Negative emotionality (e.g. Bad-tempered, Angry, Nervous), whereas the Positive Emotionality/Communal factor included terms that seem highly related to Sociability (e.g. Likeable, Enthusiastic, Friendly, Talkative). This Hebrew study was limited by its dictionary page-sampling method, which reduced the representativeness of the terms, as did the decision to reduce the item pool by almost 25 per cent by omitting all synonyms. None the less, some of its findings were similar to those found in a few other languages. At the seven-factor level: (a) Emotional Stability terms were associated with more than one factor; (b) Extraversión split into two factors (labelled Positive Emotionality/Agentic and Positive Emotionality/ Communal); (c) Intellect-related terms were mixed with some other highly desirable terms to form a factor labelled Positive Valence; and (d) a Negative Valence factor included all of the highly pejorative terms in the variable set, plus a few of the most desirable ones (e.g. Honest, Sincere, Dependable, Reliable). Thus the so-called Negative Valence factor was moderately bipolar, unlike its counterparts in English and Spanish Big Seven studies. Hungarian Szirmák and De Raad (1994) reported the findings from some recent lexical studies in Hungarian, one of the Finno-Ugric languages. Hungarian is of particular interest because it is a non-Indo-European language that is spoken by central Europeans who live in close proximity to numerous European languages and cultures. Over 8000 person descriptors were extracted from dictionaries. Five judges provided ratings of familiarity, personality-relevance, and stability (trait versus state) for nearly 4000 adjectives; the 624 trait-descriptive terms with the highest mean ratings on these three criteria were retained. Self-ratings on these 624 terms were provided by 400 university students, and were ipsatized prior to analysis. Three-, four-, five- and six-factor solutions were examined (De Raad and Szirmák, 1994). In the four-factor solution, the factors resembled Big Five Extraversión (I), Agreeableness (II), Conscientiousness (III) and

Studies of lexical personality factors

17

Emotional Stability (IV ), the first three of which made up the three-factor solution. In the five-factor solution, Agreeableness divided into two factors, one emphasizing calmness versus irritability, and the other (labelled Integrity) emphasizing egotism-related variables. The six-factor solution added an Intellect factor. Thus the Hungarian studies support the Big Three and Big Four (Big Five minus Intellect) models and also, by a loose rather than stringent criterion, the Big Five model. Turkish The Turkish language, a member of the Altaic group, was studied by Somer and Goldberg (1999) and Goldberg and Somer (in press). Five native speakers picked out 2200 person-descriptive adjectives from three modem abridged Turkish dictionaries. Most, but not all, terms describing physical attributes, sheer evaluations, and slang terms were removed, leaving 1300 terms that were judged for familiarity as personality descrip­ tors by 150 university students. Three separate studies were carried out with different subsets of the most familiar of these 1300 terms. The first two of these Turkish studies were reported by Somer and Goldberg (1999). In their Study 1, they used 474 familiar adjectives grouped into 358 synonym clusters and arranged as 179 pairs of antonym clusters. Ratings of self, liked-peers and disliked-peers were obtained for both poles of the antonym clusters from 232 university students. Two, three, four, five, six and seven factors were rotated. Analyses of the self and liked targets produced an easily interpretable five-factor solution reflecting the Big Five, whereas analyses of the liked plus disliked peers produced a five-factor solution in which the first four of the Big Five factors emerged quite clearly but the fifth factor appeared as a blend of Openness-Imagination with Attractiveness. The three-factor solution in both cases produced broad versions of factors I, II and III, as in most of the languages in which such solutions have been reported. In a second study, the 358 terms were reduced to a simpler and more user-friendly set of 235, and self and peer ratings were obtained from a much larger sample (945 university students); roughly half of the sample described themselves, and the other half described either liked, disliked or neutrally evaluated peers. Solutions for one to seven factors were obtained for the combined self and liked-peer targets, as well as for the combined liked, neutral and disliked-peer targets. The five-factor solution reproduced the Big Five quite cleanly, but with factor V closer to Openness than Intellect. The Big Five were also identifiable in the five-factor solu­ tion for the ratings of the liked, neutral, and disliked-peers. Again, for both sets of ratings, the three-factor solution produced broader versions of factors I, II and III.

18

Saucier, Hampson and Goldberg

To answer any possible charges that the synonym clusters used in the two studies by Somer and Goldberg (1999) might have been prestructured in some way, Goldberg and Somer (in press) administered the 498 most familiar single terms from the initial set of 1300 Turkish person-descriptive adjectives to over 600 university students for selfdescriptive ratings. In the total item pool, using ipsatized ratings, Big Five factors were identifiable, along with a broad Attractiveness factor and another factor composed of items of extremely low endorsement rates. When the item pool was restricted to terms that are less pejorative and more clearly related to personality traits, the Big Five factors were recog­ nizable. Again, as in Somer and Goldberg (in press), the fifth factor blended Intellect with Modemism/Progressivism. And Extraversión (I) and Emotional Stability (IV ) tended to be more highly associated with each other in the Turkish data than they are in the English/German version of the Big Five.7 The Turkish findings demonstrate that the Big Five can be reproduced in an emic study in a non-Indo-European language, at least as long as ipsatized ratings are employed, albeit with some possible cultural partic­ ularities for factor V. This factor may be the most susceptible of the five to effects of variable selection and to cultural differences (Somer and Goldberg, in press). Turkey has a predominantly Islamic cultural heritage while also being Westernized, and the traditionalism versus modernism theme in this factor may reflect that context. Korean In a study by Hahn, Lee and Ashton (1999), two Korean dictionaries and free descriptions of personality written by university students were searched for personality terms by a single investigator. Physical attributes, evaluations and temporary states were excluded, as well as some synonyms of included terms, resulting in about 1000 terms that were reduced to 785 when four judges assessed them for familiarity and adequacy for describing personality. These 785 were rated for frequency of use by 125 university students, and 406 of high familiarity were selected. Undergraduates (N= 435) from three Korean universities rated themselves on these 406 terms, and the data were ipsatized. A set of 46 Korean markers of the Big Five were identified and confirmed in factor analyses. Three to seven factor solutions were examined and correlated with the Korean Big Five marker factors to assist in their interpretation. In the three-factor solution, the familiar Big Three were found. In the four-factor solution, the Korean emic factors were labelled Extraversión (correlating 0.91 with Big Five Extraversión), Agreeableness (correlating 0.87 with Big Five Agreeableness), Conscientiousness/Precision (related

Studies of lexical personality factors

19

to both Big Five Conscientiousness and Intellect), and Masculinity/ Emotional Stability (correlating 0.79 with Big Five Emotional Stability). The Emotional Stability factor included a number of gender-related terms (e.g. Manly, Feminine), along with those reflecting self-efficacy or potency (e.g. Strong, Weak); these features may account for its lower replication coefficient. In the five-factor solution, the Extraversión, Agreeableness and Masculinity/Emotional Stability factors remained much the same, whereas terms reflecting Conscientiousness and Intel­ lect merged to form a Methodical intellect factor, with the other Conscientiousness factor now emphasizing Dependability. The six-factor solution produced a factor labelled Truthfulness that was judged to be similar to the Integrity factor identified in Hungarian by Szirmák and De Raad (1994). The investigators concluded that the Korean personality factors were quite similar to the Big Five. However, there were notable divergences from the Anglo-Germanic Big Five, at least with respect to the axis locations of several factors. Filipino Filipino (Tagalog), a member of the Austronesian language family, has borrowed some of its vocabulary from Spanish. It is the first and only native language of a Pacific Island or tropical region to be examined in a lexical personality study. Church, Katigbak and Reyes (1996) searched a com­ prehensive Filipino dictionary, extracting 6900 person-descriptive adjec­ tives, which were classified into the categories developed by Angleitner, Ostendorf and John (1990). The 682 most familiar and personality-relevant of these were used for self-ratings by both college and high school students (Church et al., 1997); these ratings were ipsatized. When five factors were rotated, four of them included content resembling that found in the Big Five, but rearranged into factors labelled Gregariousness, Socialization, Perceived competence, and Egotism; a fifth Negative Valence factor was composed entirely of pejorative terms. In the six-factor solution, a Temperamentalism factor split apart from Perceived competence. In a replication study (Church, Katigbak and Reyes, 1998), 740 Filipino college students provided self-ratings with a revised set of 502 terms, which also included some highly desirable and highly undesirable terms that might be expected to form Positive and Negative evaluation factors. The research participants also rated themselves on translations of Big Five marker scales. In analyses of ipsatized data, seven Filipino factors were labelled Gregariousness (I), Concern for others versus Egotism (II), Conscientiousness (III), Self-assurance (IV ), Intellect (V ), Negative Valence/Infrequency and Temperamentalness, another type of (negative)

20

Saucier, Hampson and Goldberg

Emotional Stability. Positive Valence markers (Big Seven) blended with terms related to Intellect to form factor V in this representation. In both of these Filipino studies, at least four of the seven Filipino factors seem to correspond to the English/German Big Five. Is this a replication of that model? John and Sri vastava (1999), applying a loose criterion for replication, interpreted it as supportive; however, by a strin­ gent criterion, the Filipino studies do not replicate the Big Five model. With respect to the Big Seven, a Negative Valence dimension replicated similar factors found earlier in English, Spanish and Hebrew. There was no unambiguous Positive Valence factor in the Filipino analyses, and the appearance of an Attractiveness factor (see Saucier, 1997) would be unlikely due to the exclusion of terms referring to physical characteristics and social effects. In several respects, the Filipino structure was the least like that found in English and other northern European (Germanic, Slavic) languages. The Conscientiousness factor was smaller than in previous studies, thus seeming to verify other reports indicating that this factor may be subject to differential emphasis according to the cultural setting (Slotboom etal.y 1998; Victor et al., 1998). As was the case in the Romance language and Hebrew studies, Emotional Stability content was distribu­ ted across at least two factors, which was also true of Extraversión content. Moreover, unlike the findings from every other language in which three-factor solutions have been examined, the Filipino threefactor solution did not include broad versions of Big Five factors I, II and III (the Big Three). Instead, Agreeableness and Conscientious­ ness content remained intertwined until a lower level in the hierarchical structure. Nonetheless, there are some similarities between the Filipino structures and those reported in Italian by Di Bias and Forzi (1998, 1999). In the seven-factor solutions from both languages, Extraversión content bifurcated into Gregariousness (or Sociability) and Self-assurance (or Assertiveness) factors, whereas Agreeableness content bifurcated into factors that might be labelled Peacefulness versus Irritability (or Temperamental ism) and Altruism versus Egotism. Moreover, at no hierarchical level in either language was there a single Emotional Stability factor; instead, this content was distributed across other factors. The structures that were derived from Hungarian, Hebrew and Spanish each showed at least some of the same bifurcations of Extraversión and Agreeabieness, suggesting that deviations from the Big Five structure are neither language-specific nor entirely random. Perhaps a useful alternative to the Big Five will spring from these initial findings.

Studies of lexical personality factors

21

PREVIOUS COMPARISONS AMONG LANGUAGES De Raad, Perugini and Szirmák (1997) used coefficients of congruence to make pair-wise comparisons of personality factors among five European languages, based on pairs of terms that seemed to be reasonably similar in each of the two languages. Congruence was consistently high (> 0.80) for the first three of the Big Five factors, somewhat lower for factor IV and even lower for factor V. In a study within one language, De Raad, Di Bias and Perugini (1997) demonstrated that structures found in two different Italian studies were similar, despite differences in their procedures for variable selection. Moreover, the two structures were more similar to one another than they were to the imported German factor structure. De Raad et al. (1998) compared seven languages (English, Dutch, German, Hungarian, Italian [Roman project], Czech and Polish). Within each language, they identified subsets of terms that had reasonably clear English equivalents, and they then computed congruence coefficients between (a) the factors based on all the terms that had been included in that variable selection and (b) the factors based on the subset of English translations. Across these seven languages, some evidence for the equivalence of the first four of the Big Five factors was consistently found, whereas factor V was not as consistently identified. Hofstee et al. (1997) compared Dutch, English and German Big Five structures. They identified a set of 126 triads of terms that were at least roughly equivalent in all three languages. They then compared the factor structure for these 126 terms across the three languages and found modest cross-language factor congruence. They concluded that the Big Five repli­ cated across these three quite similar languages, although the degree of equivalence was far from perfect. Each of these comparative studies focused solely on the five-factor struc­ tures. Future studies of this type might profitably examine a variety of hierarchical levels. As one example, Saucier and Ostendorf (1999) exam­ ined the replicability of specific subcomponents of the Big Five in parallel analyses conducted with English and German lexical data. They found 18 replicating subcomponents, indicating there are some cross-language regu­ larities at a hierarchical level more specific than that of the Big Five, at least in these two Germanic languages. Most of these cross-languagc comparison studies have focused on the question ‘Docs the Big Five replicatc in analyses of other languages?’ In part, the answer depends on whether one applies a loose or a stringent criterion for replication. Clearly, content from each of the Big Five factors can be found among person-descriptors in all of the languages studied so far. However, with the exception of Turkish, only studies of Germanic and Slavic languages have found a relatively dear Big

22

Saucier, Hampson and Goldberg

Five structure where it would most be expected - namely in five-factor solutions. Upon reflection, it becomes clear that a question like ‘Does the Big Five replicate in analyses of other languages?’ is more complex than it first appears.

WHEN IS A DIFFERENCE A REAL DIFFERENCE? How does one tell when a difference in factor representations between two languages reflects a true cultural difference, as compared with some methodological or other artefact? This problem is more complex than the related question of ‘factorial invariance' (e.g. Hofer, 1999; Horn, 1991; Meredith, 1993; Sorbom, 1974), where exactly the same variables are measured in different groups of persons. In cross-language studies, in addition to differences between the languages and the cultures, virtually everything else may differ as well, including: 1 the methods for selecting an initial set of terms (e.g. from unabridged versus abridged dictionaries, using one versus more than one diction­ ary, examining all entries versus sampling entries or pages); 2 the size and representativeness of this initial set of terms; 3 the procedures used to cull the initial set of terms into one or more smaller subsets (e.g. the number of judges involved, the nature of those judges, the types of judgements used in the selection process, the order of those judgements in the culling process); 4 the extent to which the final set of terms includes descriptors of indi­ vidual differences beyond those normally associated with personality traits (e.g. physical characteristics, temporary states, social effects, pure evaluations); 5 the size of the final set of terms; 6 the shape and extremity (skew) of the distribution of those terms on an evaluative dimension; 7 the nature of the final variables chosen for use (e.g. single terms, antonym pairs forming bipolar scales, synonym clusters, antonymic pairs of synonym clusters); 8 the nature of the judgements used to obtain indices of association among the variables (e.g. ‘internal’ versus ‘external’ data); 9 the size and nature of the subject sample used to provide those indices; 10 in the most common case, with external data, the nature of the targets being described (e.g. self versus others, and with others the extent of familiarity and liking of the subjects with the targets they describe); 11 the particular kinds of rating scales used to gather the final judge­ ments;

Studies of lexical personality factors

23

12 the kinds of procedures (if any) used to cull potentially off-task indi­ viduals from the participant sample (e.g. the use of indices of semantic consistency); 13 the handling of missing data; 14 the extent to which the sample has been culled of semantically incon­ sistent subjects or subjects with excessive missing data; 15 the methods (if any) used to handle individual differences in subjects’ use of the response scale (e.g. Z-scoring, acquiescence partialling; see Ten Berge, 1999);8 16 the methods of data analysis (e.g. multidimensional scaling, cluster analysis, exploratory or confirmatory factor analysis); 17 in the most common case, with component or factor analysis, the number of factors extracted and the procedures used for factor rota­ tion (e.g. orthogonal versus oblique rotations); and, finally, 18 the procedures used to compare the resulting factors across pairs of languages (e.g. expert or other judgements of factor similarity, corre­ lations between indigenous and imported dimensions, analyses of subsets of translation-similar terms in the two languages). Given these and other differences among studies, is it any wonder that investigators might disagree about the evidential basis for a particular structural representation? However, not all of these methodological differ­ ences are likely to be equally powerful as influences on the resulting structures. What is sorely needed at this juncture are some rigorous studies, within each of two or more single languages, examining the differential impact of these potential effects. Such purely methodological studies may not be of widespread interest, but their findings should have profound implications for our comprehension of the meaning of between-language factor differences. In the interim before such findings are available, it will be difficult to say anything definitive about cultural differences in factor representa­ tions. Certainly, the more replicable the factor structure, the less susceptible it will be to these method effects. And, other things being equal, the more terms that are associated with a particular factor, the more replic­ able should be that factor. This reasoning would suggest that the most hardy factor should be the first unrotated one, which typically differenti­ ates desirable from undesirable personality attributes. It is hard to imagine any language which would not include some descriptors of both types, and thus it ought to be the case that some sort of quasi-evaluative factor, contrasting good personality qualities with bad ones, should be recover­ able in any language. Indeed, Osgood, May and Miron (1975) found that a large evaluation factor was widely distributed cross-culturally in judgements about heterogeneous sets of concepts. Such an evaluative

24

Saucier, Hampson and Goldberg

factor could turn out to be the only truly ‘universal’ aspect of personality description. Clearly, then, there is value in looking for emic factors at a broader level in the personality-trait hierarchy than the Big Five. Two-factor solutions from several lexical studies suggest a consistent pattern: one factor tends to involve attributes associated with dynamism and indi­ vidual ascendancy, whereas the other tends to involve attributes associ­ ated with socialization, social propriety, solidarity, and community. As related to the Big Five, the first factor is most highly associated with Extraversión (I), Intellect (V ) and the fearfulness aspects of factor IV, whereas the second is most highly associated with Agreeableness (II), Conscientiousness (III) and the irritability aspects of factor IV. In addi­ tion to its appearance in lexical studies (e.g. Di Bias and Forzi, 1999; Hrebícková et al., 1999; Saucier, 1997, 1998), this pattern has appeared in other studies of personality structure (Digman, 1997; Paulhus and John, 1998; Shweder, 1972; White, 1980). It may represent projections of individualism and collectivism (Triandis et al., 1988) into within-group variation.9 The reproducibility of this two-factor solution in diverse languages needs more attention. A three-factor structure, which includes broad versions of Extraversión (1), Agreeableness (II), and Conscientiousness (111) can probably be found in many, if not most, major modern languages (e.g. De Raad and Szirmák, 1994; Goldberg and Somer, in press; Hahn, Lee, and Ashton, 1999; Ostendorf, 1990). However, it is important that readers understand that this ‘Big-Three’ factor structure is not equivalent to the dimensions of Psychoticism, Extraversión and Neuroticism proposed by Eysenck (1991) as ‘basic super-factors’. In the two models, only Extraversión is the same. Psychoticism has been shown to be a blend of the orthogonal Big Five factors II and III (Goldberg and Rosolack, 1994), whereas Neuroticism (the opposite pole of Big Five Emotional Stability) is not one of the ‘Big Three’ lexical factors. At the five-factor level, we can no longer expect that the exact English/ German representation of these broad domains will appear with any vari­ able set in all of the world’s languages. Tests of the Big Five structure remain useful, but alternative structures should be tested as well. Moreover, there may be factors beyond the Big Five (Saucier and Goldberg, 1998).

REFINING THE LEXICAL-STUDY PARADIGM Up to this point, we have been interpreting the findings from studies conducted within a standard framework, which we might label the ‘lexicalstudy paradigm’. This paradigm has evolved, as most do, from the

Studies of lexical personality factors

25

influential template of a few initial studies. As experience has accumu­ lated, the template has enabled researchers to learn a great deal, but some limitations of this template have also become evident. These limitations point to ways in which we might refine future lexical studies, and thereby learn even more. First, previous studies have been prone to rely on easily obtainable data from university students, who might in fact constitute a single cosmopolitan population across national and linguistic boundaries. It is possible that such samples represent the subgroup within each language community that is most Westernized and most likely to be have some facility in English, the language in which the Big Five was first discovered. Non­ student, rural or even non-literate populations would provide more diversity. Moreover, previous studies have probably leaned too heavily on self-descriptions. Extensive self-examination may be an unfamiliar and possibly bizarre task in some cultural settings, based as it is on contem­ porary Western norms of individualism, autonomy and introspection. More attention should be given to how personality descriptors actually are used in a wide range of cultural settings. Second, researchers should attend to the representativeness of the languages (and cultures) being sampled. Although nearly half of the world’s population inhabits tropical regions, only one of the lexical studies reviewed in this chapter (Filipino) examined a language that orig­ inated in the tropics. Indeed, European languages have been heavily sampled while other geographic regions and linguistic families have remained untouched. Better sampling of languages could be obtained by pursuing two distinct strategies. In strategy (a) studies should be under­ taken in non-European languages that have a very large number of speakers, such as Chinese, Bengali, Hindi and Japanese. Although useful studies of personality descriptors in Chinese (Yang and Bond, 1990) and Japanese (Isaka, 1990) have been conducted, they have not used a method compa­ rable to those in the studies reviewed here. In strategy (b) at least one language should be studied from each of the larger language families. For example, in Africa it would be useful to study one of the Niger-Congo languages and one of the Nilo-Saharan languages. In Asia, the larger unsampled language families include Sino-Tibetan (which includes Chinese), Austro-Asiatic, Daic and Dravidian. Studies of one or more languages native to Australia, New Guinea or the Americas would also contribute to a more representative sampling. Third, studies have largely tested a prototype Big Five structure devel­ oped originally from studies in English. Near-perfect reproduction of this structure has been characteristic mainly of emic structures from language families (i.e. Germanic, Slavic) with their origin in northern Europe. We ought to avoid ethnocentrism, with psychologists in effect imposing a

26

Saucier, Hampson and Goldberg

northern European folk psychology worldwide. It would be worthwhile to test other structures alongside the Big Five, such as those found in Italian or Filipino, and those based on fewer and broader factors. Fuller exami­ nation of unrotated factors might be useful; they may be as replicable as the rotated ones (Saucier, 1998). Fourth, there is some arbitrariness in the common practice of segre­ gating out a portion of the lexicon as referring to stable, dispositional (or ‘temperament’) traits, studying this portion, and ignoring other persondescriptors. Old assumptions, under which certain characteristics are heritable and others not, have been undermined by modem behaviour genetic findings (e.g. Bouchard et al., 1990). These findings indicate moderate heritability for a very wide range of psychological characteris­ tics, though not necessarily for very specific behaviours (see Chapter 4 for Borkenau et a l.'s review of studies). Therefore, so-called ‘traits’ are not so separate from other characteristics as Allport (1937) and others had assumed. Moreover, as Allport (1937, Chapter 2) made clear, there are many ways to define personality, and this is one reason there is little consensus about how to distinguish between personality and other terms. Past emphasis on a narrow band of disposition terms may have been based understandably on low computing power, a belief in the importance of comprehensive selection from the lexicon, and the need to begin the search for structure in some logical place. However, the understanding of personality characteristics can only benefit from being placed in context with other characteristics of persons (e.g. states, social effects, physical appearance). Fifth, restriction of variable selections to adjectives has been typical, but this procedure complicates extensions to other languages. As Dixon (1977) made clear, languages vary in their reliance on adjectives and some languages lack a substantial adjective class. Although person description can be called an ‘adjective function’ (Saucier and Goldberg, 1996b, p. 30), some of this function can be carried out with nouns. Moreover, in some languages, verbs may be used to denote person-related states and characteristics. Arguably, the goal of variable selection should be the development in each language community of a set of their most frequently used person-descriptors, regardless of whether these are found in adjec­ tive, noun or verb form. Such a set is more likely to be comparable across a wider range of languages than is a simple sampling of adjectives.

ASSUMPTIONS ABOUT PERSONALITY ATTRIBUTES Researchers have followed Allport (1937) in tending to assume that person­ ality descriptors are used by people mainly to denote individual differences

Studies of lexical personality factors

27

with respect to stable traits. But these descriptors may also function to comment on and regulate momentary behaviours, as in sentences like ‘Stop being such a grouch’ or ‘You should be more assertive’. Consistent with this view, the findings of Chaplin, John and Goldberg (1988) suggest that most personality descriptors can function to denote either stable traits or temporary states. Therefore, it is more accurate to label these descriptors as referring to, and yielding models of, ‘attributes’ or ‘characteristics’, rather than necessarily ‘traits’. Some, but not all, attributes ascribed to a person turn out to be stable traits; the trait nature of an attribute cannot be assumed a priori. One purpose of lexical personality research is to understand the semantic categories that members of a culture use to comprehend one another, and thereby be able to predict others’ subsequent behaviours. What is the origin of these categories? Some investigators adopt a biogenic assumption: in their view, cross-cultural regularities mean that we all share some genetically based personality hardware. An alternative is a sociogenic assumption: cross-cultural regularities indicate that various human cultures have found it useful to develop the same system for categorizing human behaviour, perhaps because of regularities in the human environment or in human social groups. Under the biogenic assumption, the personality factors derived from lexical studies are assumed to be indicators of a biologically based person­ ality structure. For example, McCrae and Costa’s (1998) five-factor model (FFM ), which was initially developed independently of the lexical Big Five (Costa and McCrae, 1980), later added two factors derived from lexical analyses (Costa and McCrae, 1985). At a purely descriptive level, the lexical Big Five and the FFM now look similar, and thus they are frequently confused. However, McCrae and Costa take a much more committed biogenic position than the typical lexical researcher, arguing for a genetic basis for their five dimensions, and citing cross-language studies as supporting this claim (McCrae and Costa, 1997). We remain sceptical of such a straightforward mapping of personality categories on to biological systems. It is quite possible that lexical personality factors represent, in part, a set of culturally derived categories into which biolog­ ically derived differences are fitted. Lexical factors cannot be assumed to reflect automatically the main lines of biological influence. Some critics of the lexical approach have argued against studying everyday language, even as a starting point for the development of scien­ tific theories (e.g. Block, 1995). They argue that the basic constructs in personality psychology should come from expert theories, rather than judgements by naive research participants. Our response is that such argu­ ments assume too extreme a discrepancy between lay and expert concepts. Scientists impose categories on what they observe, just as do lay people.

28

Saucier, Hampson and Goldberg

Lay conceptions affect scientists, and scientific conceptions that become widespread or popular tend to percolate into the language. Therefore, it is inevitable that the scientific language o f personality description will overlap substantially with the lay language. Accordingly, lexical studies can make an important contribution, helping to assure that personality models are parsimonious, yet comprehensive in content. Clearly, the study of different lexicons can lead to a useful and highly generalizable classification system for personality traits, but this classifi­ cation system should not be reified into an explanatory one. A model of descriptions does not provide a model o f causes, and the study of person­ ality lexicons should not be equated with the study of personality.

NOTES Work on this chapter was supported by Grant MH-49227 from the National Institute of Mental Health, US Public Health Service. The authors are grateful to Michael Ashton, Veronica Benet-Martinez, Jack Block, Roy D’Andrade, Boele de Raad, Mario Forzi, Janice C. Goldberg, Martina Hrebickova, John A. Johnson, Jeannette M. Mageo, Ivan Mervielde, Boris Mlacic, Dean Peabody, Marco Perugini, Leonard G. Rorer, Richard Shweder, Piotr Szarota, Auke Tellegen, Harry Triandis, Erika Westling and Jerry S. Wiggins for their thoughtful comments and suggestions. Correspondence may be addressed to any of the three authors: Gerard Saucier, Department of Psychology, University of Oregon, Eugene, OR 97403, USA (e-mail: [email protected]); Sarah E. Hampson, Department of Psychology, University of Surrey, Guildford GU2 5XH, UK (e-mail: s.hampson@ surrey.ac.uk); or Lewis R. Goldberg, Oregon Research Institute, 1715 Franklin Boulevard, Eugene, OR 97403-1983, USA (e-mail: [email protected]). 1 We use the phrase ‘Big Five’ to refer to the common elements in the fivefactor representations from lexical studies in English (e.g. Saucier and Goldberg, 1996a) and in German (Ostendorf, 1990). This practice should not be understood as an ethnocentric reification of the English/German structure, but only as a short-hand convention. 2 This corollary of the lexical hypothesis has important implications for the selection of variables in lexical studies. Because the most important attributes are likely to spawn a large number of synonyms (or quasi-synonyms), it defeats the goal of representative sampling to cull such redundancies from the final set of terms to be studied. Unfortunately, as we will indicate later in this chapter, the findings from a number of studies may have suffered as a result of such synonym deletion. On the other hand, just how many synonyms an attribute (e.g. Kind) might generate is somewhat indeterminate: Lexicons are open enough that new (but low-frequency) terms can be readily invented, especially using metaphors (e.g. Open-hearted) or nominalizations (e.g. a Mother Theresa) (see Henss, 1995). 3 The inclusion of the phrase ‘tends to be’ and the adverb ‘often’ may have served to bias the selection against attributes that are relatively non­ fluctuating, such as those referring to physical attractiveness. If this hypoth­ esis is correct, it might explain some of the differences between the ‘Positive

Studies of lexical personality factors

29

Valence’ factor found in this study and the ‘Attractiveness’ factor found by Saucier (1997). 4 In raw-data analyses in German, an Emotional Stability factor did not appear until at least six factors were extracted and rotated; in the raw-data five-factor solutions, Agreeableness was split into two factors, labelled SD (low social desirability) and ‘Warmherzigkeit’ (Warm-heartedness) (Ostendorf, 1990, p. 178). 5 A Norwegian lexical project has been carried out over the past few years, but reports of these findings have not yet been been published. Engvik (1999, personal communication) has reported that a clear Big Five factor structure was easily identifiable in analyses of 859 personality-trait adjectives. Indeed, it appears that even the Norwegian factor V was nearly identical to the English/German version of the Intellect factor. 6 Di Bias and Forzi (1999) analysed raw data as well as ipsatized data. The raw-data solutions are described, in a footnote, as having more unipolar factors but otherwise similar results. 7 When raw data instead of ipsatized ratings were analysed in this study of Turkish descriptors, a Big Five structure could not be identified, due to the fusion of aspects of Extraversión and Intellect into a single factor (probably related to Dynamism, Self-assurance, Assertiveness, and Positive Valence factors found in other lexical studies) beginning at the two-factor level. The raw-data five-factor solution included this Dynamism factor, plus factors that might be labelled Social propriety, Unpleasant affect. Conscientiousness, and Emotionality; the two-factor solution included the first two of these, and the three-factor solution the first three. Thus the raw-data analyses in the second study supported neither the Big Five nor the Big Three, but did find a twofactor structure resembling that in analyses of descriptors from other languages (e.g. Saucier, 1998). 8 There has been some controversy with respect to the advisability of ipsatizing. Ten Berge (1999) refuted some of the critiques made of that procedure, but noted that ipsatizing removes individual differences in mean item response, and 'to the extent that the mean reflects non-spurious differences between individuals' (p. 101), any removal of this component of responses leads to a loss of useful information. Ten Berge (1999) noted that when a personality scale is quite unbalanced with respect to positively and nega­ tively keyed items, ‘the sum of scores on the items represents a mixture of the trait under consideration and acquiescence, and ipsatizing the data will indeed remove the most important source of variation, an error of strategy' (pp. 93-94). In other words, ipsatization makes sense under the assumption that all indi­ vidual differences in response mean and standard deviation are spurious (i.e. due to individual differences in acquiescence and extremeness response tenden­ cies). But this assumption is most tenable when each of the attributes under consideration is represented by an equal number of positively and negatively worded items, a situation that cannot be expected to hold in most variable selections based on natural language descriptors. As Hofstee (1994) argued, removal of acquiescence variance might be accomplished with more preci­ sion by partialling the response mean computed across a balanced (sub)set of variables (e.g. a set of antonym pairs). In order to gain more clarity on the effect of these procedures, it would be useful if investigators reported their findings from both ipsatized and nonipsatized data (as in Goldberg, 1990; Ostendorf, 1990; Saucier, 1998), as well

30

9

Saucier, Hampson and Goldberg

as from residual data after removing the response means from a balanced subset of variables. As indicated in this chapter, most lexical analyses have been conducted with ipsatized data. Exceptions are (a) the Big Seven studies in English (Tellegen and Waller, 1987) and Hebrew (Almagor, Tellegen and Waller, 1995) in which only raw data were apparently used, and (b) some studies (Di Bias and Forzi, 1999; Goldberg and Somer, 1999; Ostendorf, 1990; Saucier, 1998) that have analysed both raw and ipsatized data. The few compar­ isons of the two kinds of data suggest that in analyses of raw data the Big Five is weaker, whereas the Big Three and higher-order solutions (one or two factors) are not weakened substantially. It is interesting that the third unrotated factor found in English, German and some other languages, which is orthogonal to these first two, has been labelled Tight-Loose (impulse control versus impulse expression). Triandis and his colleagues (1988) have discussed another cultural syndrome which they also labelled Tight-Loose.

REFERENCES Allport, G.W. (1937) Personality: A Psychological Interpretation. New York: Holt, Rinehart and Winston. Allport, G.W. and Odbert, H.S. (1936) ‘Trait names: A psycho-lexical study’, Psychological Monographs 47(1, Whole No. 211). Almagor, M., Tellegen, A. and Waller, N. (1995) ‘The Big Seven model: A crosscultural replication and further exploration of the basic dimensions of natural language of trait descriptions’, Journal o f Personality and Social Psychology’ 69: 300-307. Angleitner, A., Ostendorf, F. and John, O.P. (1990) ‘Towards a taxonomy of personality descriptors in German: A psycho-lexical study’, European Journal o f Personality 4: 89-118. Baumgarten, F. (1933) ‘Die Charaktereigenschaften’ [‘The character traits’], Beiträge zur Charakter- und Persönlichkeitsforschung (Whole No. 1). Bern: A. Francke. Benet-Martinez, V. and John, O.P. (1998) ‘Los Cinco Grandes across cultures and ethnic groups: Multitrait multimethod analyses of the Big Five in Spanish and English’, Journal o f Personality and Social Psychology 75: 729-750. Benet-Martinez, V. and Waller, N.G. (1997) ‘Further evidence for the crosscultural generality of the Big Seven factor model: Indigenous and imported Spanish personality constructs’, Journal o f Personality 65: 567-598. Berry, J.W . (1969) ‘On cross-cultural comparability’. International Journal o f Psychology 4: 119-128. Block, J. (1995) ‘A contrarian view of the Five-Factor approach to personality description’. Psychological Bulletin 117: 187-215. Bouchard, T.J., Lykken, D.T., McGue, M., Segal, N.L. and Tellegen, A. (1990) ‘Sources of human psychological differences: The Minnesota study of twins reared apart’, Science 250: 223-228. Brokken, F.B. (1978) The Language o f Personality. Meppel, The Netherlands: Krips.

Studies of lexical personality factors

31

Caprara. G.V. and Perugini, M. (1994) ‘Personality described by adjectives: The generalizability of the Big Five to the Italian lexical context’, European Journal o f Personality 8: 357-369. Cattell, R. B. (1943) ‘The description of personality. II. Basic traits resolved into clusters’. Journal o f Abnormal and Social Psychology 38: 476-507. Chaplin, W.F., John, O P. and Goldberg, L.R. (1988) ‘Conceptions of states and traits: Dimensional attributes with ideals as prototypes’, Journal o f Personality and Social Psychology 54: 541-557. Church, A.T., Katigbak, M.S. and Reyes, J.A.S. (1996) ‘Toward a taxonomy of trait adjectives in Filipino: Comparing personality lexicons across cultures’, European Journal o f Personality 10: 3-24. Church. A.T., Katigbak, M.S. and Reyes, J.A.S. (1998) ‘Further exploration of Filipino personality structure using the lexical approach: Do the Big Five or Big Seven dimensions emerge?’, European Journal o f Personality 12: 249-269. Church, A.T., Reyes, J.A.S., Katigbak, M.S. and Grimm, S.D. (¡997) ‘Filipino personality structure and the Big Five model: A lexical approach’, Journal of Personality>65: 477-528. Costa, P.T., Jr and McCrae, R.R. (1980) 'Still stable after all these years: Personality as a key to some issues in adulthood and old age’, in P. Baltes and O.G. Brim (eds) Life Span Development and Behavior, vol. 3, pp. 31-54. New York: Academic Press. Costa, P.T. Jr and McCrae, R.R. (1985) The NEO Personality Inventoiy Manual. Odessa, FL: Psychological Assessment Resources. De Raad, B. (1992) ‘The replicability of the Big Five personality dimensions in three word-classes of the Dutch language’, European Journal o f Personality 6: 15-29. De Raad, B. (1994) ‘An expedition in search of a fifth universal factor: Key issues in the lexical approach’, European Journal of Personality 8: 229-250. De Raad, B. (1998) ‘Five big Big Five issues: Rationale, content, structure, status and cross-cultural assessment’, European Psychologist 3: 113-124. De Raad, B „ Di Bias, L. and Perugini, M. (1997) ‘Two independent Italian trait taxonomies: Comparisons with Italian and between Italian and Germanic languages’, European Journal o f Personality 11: 167-185. De Raad, B., Hendriks, A.A.J. and Hofstee, W .K.B. (1992) ‘Towards a refined structure of personality traits’, European Journal o f Personality 6: 301-319. De Raad, B. and Hoskens, M. (1990) ‘Personality-descriptive nouns’, European Journal of Personality 4: 131 146. De Raad, B., Mulder, E., Kloosterman, K. and Hofstee, W .K.B. (1988) ‘Personality-descriptive verbs’, European Journal o f Personality 2: 81-96. De Raad, B. and Ostendorf, F. (1996) ‘Quantity and quality of trait-descriptivc type nouns’, European Journal o f Personality 10: 45-56. De Raad, B., Pcrugini, M. and Szirmak, Z. (1997) ‘In pursuit of a cross-lingual reference structure of personality traits: Comparisons among five languages’, European Journal o f Personality II: 167 185. De Raad, B „ Perugini, M., Hrebickova, M. and Szarota, P. (1998) ‘Lingua Franca of personality: Taxonomies and structures based on the psycholexical approach’, Journal o f Cross Cultural Psychology 29: 212-232.

32

Saucier, Hampson and Goldberg

De Raad, B. and Szirmák, Z. (1994) ‘The search for the “ Big Five” in a nonIndo-European language: The Hungarian trait structure and its relationship to the EPQ and the PTS’, European Review o f Applied Psychology 44: 17-26. Di Blas, L. and Forzi, M. (1998) ‘An alternative taxonomic study of personalitydescriptive adjectives in the Italian language’, European Journal o f Personality 12: 75-101. Di Blas, L. and Forzi, M. (1999) ‘Refining a descriptive structure of personality attributes in the Italian language: The abridged Big Three circumplex struc­ ture', Journal of Personality and Social Psychology 76: 451 —481. Digman, J.M . (1990) ‘Personality structure: Emergence of the five-factor model’, in M.R. Rosenzweig and L.W . Porter (eds) Annual Review o f Psychology, vol. 41, pp. 417 446. Palo Alto, CA: Annual Reviews. Digman, J.M . (1997) ‘Higher order factors of the Big Five’, Journal o f Personality and Social Psychology>73: 1246-1256. Dixon, R.M.W. (1977) ‘Where have all the adjectives gone?', Studies in Language 1: 19-80. Eysenck, H.J. (1991) ‘Dimensions of personality: 16, 5, or 3? Criteria for a taxo­ nomic paradigm’. Personality and Individual Differences 12: 773-790. Galton, F. (1884) ‘Measurement of character’, Fortnightly Review 36: 179-185. Goldberg, L.R. (1976) ‘Language and personality: Toward a taxonomy of traitdescriptive terms’, Istanbul Studies in Experimental Psychology 12: 1-23. Goldberg, L.R. (1981) ‘Language and individual differences: The search for universals in personality lexicons’, in L. Wheeler (ed.) Review o f Personality and Social Psychology, vol. 2, pp. 141-165. Beverly Hills, CA: Sage. Goldberg, L.R. (1982) ‘From Ace to Zombie: Some explorations in the language of personality’, in C.D. Speilberger and J.N. Butcher (eds) Advances in Personality Assessment, vol. I, pp. 203 234. Hillsdale, NJ.: Erlbaum. Goldberg, L.R. (1990) ‘An alternative “ Description of personality” : The Big-Five factor structure’. Journal o f Personality and Social Psychology 59: 1216-1229. Goldberg, L.R. (1993) ‘The structure of phenotypic personality traits’, American Psychologist 48: 26-34. Goldberg, L.R. (1995) ‘What the hell took so long? Donald Fiske and the BigFive factor structure’, in P.E. Shrout and S.T. Fiske (eds) Personality Research. Methods, and Theory: A Festschrift Honoring Donald fV. Fiske, pp. 29 43. Hillsdale, NJ: Erlbaum. Goldberg, L.R. and Rosolack, T.K. (1994) ‘The Big Five factor structure as an integrative framework: An empirical comparison with Eysenck’s P-E-N model’, in C.F. Halverson, Jr, G.A. Kohnstamm and R.P. Martin (eds) The Developing Structure o f Temperament and Personalityfrom Infancy to Adulthood, pp. 7-35. New York: Erlbaum. Goldberg, L.R. and Saucier, G. (1995) ‘So what do you propose we use instead? A reply to Block’, Psychological Bulletin 117: 221-225. Goldberg, L.R. and Somer, O. (in press) ‘The hierarchical structure of common Turkish person-descriptive adjectives’, European Journal o f Personality. Hahn D.-W., Lee, K. and Ashton, M.C. (1999) ‘A factor analysis of the most frequently used Korean personality trait adjectives’, European Journal o f Personality 13: 261-282.

Studies of lexical personality factors

33

Henss, R. (1995) ‘From Aal to Zyniker: Personality descriptive type nouns in the German language’, European Journal o f Personality 9: 135-145. Hofer, S.M. (1999) ‘Assessing personality structure using factorial invariance procedures’, in I. Mervielde, l.J. Deary, F. De Fruyt and F. Ostendorf (eds) Personality Psychology in Europe, vol. 7, pp. 35-49. Tilburg. Netherlands: Tilburg University Press. Hofstee, W .K.B. (1994) ‘Personality and factor analysis: Bind or bond’, Zeitschrift fü r Differentielle und Diagnostische Psychologie 15: 173-183. Hofstee, W .K.B., Brokken, F.B. and Land, H. (1981) ‘Constructie van een standaardpersoonlijkheids-eigenschappenlijst (S P E L )' [‘Construction of a standard list of personality descriptive adjectives’], Nederlands Tijdschrift voor de Psychologie 34: 443 452. Hofstee, W .K.B., Kiers, H.A.L., De Raad, B., Goldberg, L.R. and Ostendorf, F. (1997) ‘A comparison of Big-Five structures of personality traits in Dutch, English and German', European Journal o f Personality 11: 15-31. Horn, J.L. (1991) ‘Discussion of issues in factorial invariance’, in L.M. Collins and J.L. Horn (eds) Best Methods for the Analysis of Change, pp. 114-125. Washington. DC: American Psychological Association. Hrebickova, M „ Ostendorf, F. and Angleitner. A. (1995) ‘Basic dimensions of personality description in the Czech language’, paper presented at the 7th Meeting of the International Society for the Study of Individual Differences, Warsaw, Poland, July. Hrebickova, M., Ostendorf, F., Oseckä, L. and Cermäk, I. (1999) ‘Taxonomy and structure of Czech personality-relevant verbs’, in I. Mervielde, l.J. Deary, F. De Fruyt and F. Ostendorf (eds) Personality Psychology in Europe, vol. 7, pp. 51—65. Tilburg, The Netherlands: Tilburg University Press. Isaka, H. (1990) ‘Factor analysis of trait terms in everyday Japanese language’, Personality and Individual Differences 11: 115-124. John, O.P. (1990) ‘The “ Big Five” factor taxonomy: Dimensions of person­ ality in the natural language and in questionnaires’, in L.A. Pervin (ed.) Handbook o f Personality: Theory and Research, pp. 66 100. New York: Guilford. John. O.P. and Srivastava, S. (1999) ‘The Big Five trait taxonomy: History, measurement and theoretical perspectives’, in L.A. Pervin and O.P. John (eds) Handbook o f Personality: Theory and Research, 2nd edn, pp. 102-138. New York: Guilford. Klages, L. (1932) The Science of Character (translated 1932). London: George Allen and Unwin. (Original work published 1926). McCrae, R.R. and Costa, P.T., Jr (1997) ‘Personality trait structure as a human universal', American Psychologist 52: 509-516. McCrae, R.R. and Costa, P.T., Jr (1998) ‘Toward a new generation of person­ ality theories: Theoretical contexts for the Five-Factor Model’, in J.S. Wiggins (ed.) The Five-Factor Model o f Personality: Theoretical Perspectives, pp. 51-87. New York: Guilford. McCrae, R.R. and John, O.P. (1992) ‘An introduction to the five factor model and its applications’, Journal o f Personality 60: 175-215.

34

Saucier, Hampson and Goldberg

Meredith, W. (1993) ‘Measurement invariance, factor analysis and factorial invari­ ance’, Psychometrika 58: 525-543. Mlacic, B. (1999) ‘Hrvatska taksonomija deskriptora osobina liènosti’ [‘Croatian taxonomy of personality descriptive terms’], unpublished doctoral dissertation, University of Zagreb, Croatia. Norman, W.T. (1963) ‘Toward an adequate taxonomy of personality attributes: Replicated factor structure in peer nomination personality ratings’, Journal of Abnormal and Social Psychology 66: 574-583. Norman, W.T. (1967) 2800 Personality Trait Descriptors: Normative Operating Characteristicsfo ra University Population. University of Michigan; Ann Arbor, MI: Department of Psychology. Osgood, C.E., May, W. and Miron, M. (1975) Cross-Cultural Universals of Affective Meaning. Urbana: University of Illinois Press. Ostendorf, F. (1990) Sprache und Persönlichkeitsstruktur: 7mt Validität des FiinfFaktoren-Modells der Persönlichkeit [Language and Personality Structure: Toward the Validation of the Five-Factor Model of Personality], Regensberg, Germany: S. Roderer Verlag. Ostendorf, F. and Angleitner, A. (1994) Enthusiasts contra pessimists, Psychological Inquiry 5: 159-162. Paulhus, D.L. and John, O.P. (1998) ‘Egoistic and moralistic biases in self­ perception: The interplay of self-descriptive styles with basic traits and motives’, Journal of Personality 66: 1025-1060. Peabody, D. (1987) ‘Selecting representative trait adjectives’, Journal of Personality and Social Psychology 52: 59-71. Peabody, D. and Goldberg, L.R. (1989) ‘Some determinants of factor structures from personality-trait descriptors’, Journal ofPersonality and Social Psychology 57: 552-567. Saucier, G. (1997) ‘Effects of variable selection on the factor structure of person descriptors’, Journal of Personality and Social Psychology 73: 1296 1312. Saucier, G. (1998) ‘A hierarchy of personality factors in the natural language’, paper presented at the 9th European Conference on Personality’, Guildford, England, July. Saucier, G. and Goldberg, L.R. (1996a) ‘Evidence for the Big Five in analyses of familiar English personality adjectives’, European Journal of Personality 10: 61 77. Saucier, G. and Goldberg, L.R. (1996b) ‘The language of personality: Lexical perspectives on the five-factor model’, in J.S. Wiggins (ed.) The FiveFactor Model of Personality: Theoretical Perspectives, pp. 21-50. New York: Guilford. Saucier, G. and Goldberg, L.R. (1998) ‘What is beyond the Big Five?’, Journal of Personality 66: 495-524. Saucier, G. and Ostendorf, F. (1999) ‘Hierarchical components of the Big Five personality factors: A cross-language replication’, Journal of Personality and Social Psychology 76: 613 627. Shmelyov, A.G. and Pokhil’ko, V.I. (1993) ‘A taxonomy-oriented study of Russian personality-trait names’, European Journal of Personality 7: 1-17.

Studies of lexical personality factors

35

Shmelyov, A.G., Pokhil’ko, V.l. and Kozlovskaya-Telnova, A.Y. (1988) Praktikum po experimentaPnoi psichosemantike: Tezaurus lichnostnikh [Trait Project on Experimental Psychosemantics: Trait-names Thesaurus], Moscow: Moscow University Press. Shmelyov, A.G., Pokhil’ko, V.l. and Kozlovskaya-Telnova, A.Y. (1991) ‘Reprezentativnost’ lichnostnikh chert v soznanii nositelya russkogo yazika’ ['Representation of personality traits in the consciousness of a Russian-language speaker’], Psichologicheskii Zhurnal 2: 27-44. Shweder, R.A. (1972) 'Semantic structure and personality assessment’, unpub­ lished doctoral dissertation. Harvard University. Slotboom, A.-M., Havill, V.L., Pavlopoulos, V. and De Fruyt, F. (1998) ‘Developmental changes in personality descriptions of children: A cross-national comparison of parental descriptions of children’, in G.A. Kohnstamm, C.F. Halverson, I. Mervielde and V.L. Havill (eds) Parental Descriptions o f Child Personality, pp. 127-153. Mahwah, NJ: Lawrence Erlbaum. Somer, O. and Goldberg, L.R. (1999) ‘The structure of Turkish trait-descriptive adjectives’, Journal o f Personality and Social Psychology>76: 431-450. Sorbom, D. (1974) ‘A general method for studying differences in factor means and factor structures between groups’, British Journal of Mathematical and Statistical Psychology 28: 229-239. Szarota, P. (1996) ‘Taxonomy of the Polish personality-descriptive adjectives of the highest frequency of use’, Polish Psychological Bulletin 27: 342-351. Szarota, P. and Ostendorf, F. (1997) ‘Taxonomy and structure of Polish person­ al ity-relevant adjectives’, unpublished manuscript. Szirmák, Z. and De Raad, B. (1994) ‘Taxonomy and structure of Hungarian personality traits’, European Journal o f Personality 8: 95-118. Tellegen, A. and Waller, N.G. (1987) ‘Re-examining basic dimensions of natural language trait descriptors’, paper presented at the 95th annual convention of the American Psychological Association, New York, NY, August. Tellegen, A., Grove, W.M. and Waller, N.G. (1991) inventory of personality characteristics No. 7 (IPC7)’, unpublished manuscript. University of Minnesota, Department of Psychology, Minneapolis, MN. Ten Berge, J.M .F. (1999) ‘A legitimate case of component analysis of ipsative measures, and partialling the mean as an alternative to ipsatization’, Multivariate Behavioral Research 34: 89-102. Triandis, H.C., Bontempo, R., Villareal, M .J., Asai, M. and Lucca, N. (1988) ‘Individualism and collectivism: Cross-cultural perspectives on self-ingroup relationships’, Journal o f Personality and Social Psychology 54: 323-338. Victor, J.B., Dent, H E., Carter, B., Halverson, C.F. and Havill, V.L. (1998) ‘How African-American parents describe their children’, in G.A. Kohnstamm, C.F. Halverson, I. Mcrvielde and V.L. Havill (cds) Parental Descriptions o f Child Personality\ pp. 169-187. Mahwah, NJ: Erlbaum. Waller, N.G. (in press) ‘Evaluating the structure of personality’, in C.R. Cloninger (ed.) Personality and Psychopathology’. Washington, DC: American Psychiatric Press. Waller, N.G. and Zavala, J.D. (1993) ‘Evaluating the Big Five’, Psychological Inquiry 4: 131 134.

36

Saucier, Hampson and Goldberg

White, G.M. (1980) ‘Conceptual universals in interpersonal language’, American Anthropologist 82: 759-781. Wiggins, J.S. and Trapnell, P.D. (1997) ‘Personality structure: The return of the Big Five’, in R. Hogan, J. Johnson and S. Briggs (eds) Handbook o f Personality Psychology, pp. Til-764. San Diego, CA: Academic Press. Yang, K. and Bond, M.H. (1990) ‘Exploring implicit personality theories with indigenous or imported constructs: The Chinese case’, Journal o f Personality and Social Psychology 58: 1987-1095.

Chapter 2

Variable-centred and person centred approaches to childhood personality Ivan M ervie lde and Jens 8. A s e n d o r p f

The first part of the chapter reviews the variable- and the person-centred approach, as applied to the study of individual differences among children. Four temperament models and the five-factor personality model are discussed as exemplars of the variable-centred approach to child personality. A comparison of the higher-order temperament factors shows that emotionality, extraversión, activity, and to a lesser extent persistence, can be considered as common temperament factors that are related to the Big Five adult personality factors: emotional stability, extraversión and conscientiousness. Research on ego-control and ego-resiliency, consistency of Q-sort profiles and studies of personality types are reviewed as prototypical exam­ ples of person-centred studies that provide evidence for three person-types: resilient, overcontrolled and uncontrolled. The second part of the chapter presents current developments for both approaches. The analysis of natural language descriptions of children is presented as a viable strategy for the construction of a comprehensive and hierarchically organized system of variables representing individual differences among children. Current developments in the person-centred approach focus on the continuity and stability of person-types, their consistency across studies and clustering methods, and the replicability of the three major types in children and adults. It is concluded that variable-centred and person-centred approaches are not incompatible research strategies and that both approaches are necessary for a comprehensive account of individual differences in childhood.

IN TR O D U C TIO N In the 20th century, two major empirical approaches to the structure o f personality can be identified: variable-centred and person-centred approaches. Variable-centred approaches isolate psychologically mean­ ingful characteristics on which individuals reliably differ (traits) and study their correlational structure for a particular population. This structure is a property o f variables, not o f individual persons. In contrast, person-centred approaches describe the personality structure o f each member o f a popu­ lation by a configuration o f multiple variables within the person. Then the units o f analysis are persons, not variables. This person-centred view is well captured by Allport’s definition o f personality as ‘the dynamic organization within the individual o f those psychophysical systems that

38

Mervielde and Asendorpf

determine his unique adjustments to his environment’ (Allport, 1937, p. 48). The German psychologist William Stem (1911) nicely illustrated the complementary nature of the variable- and the person-centred approaches. Decades later Stem’s approach was rediscovered and gener­ alized to a three-dimensional covariation chart with persons, tests and occasions by Cattell (1946). However, Cattell (1946) and the subsequent factor analytic (Digman, 1990) and temperament approaches (Strelau and Angleitner, 1991) to personality differences preferred the variable-centred view. For a long time. Block (1961, 1971) and his wife (Block and Block, 1980) swam against the current of the variable-centred mainstream. Recently their approach has attracted increasing attention because it gives rise to personality types that are replicable, consistent across cultures, and continuous across age (Asendorpf and van Aken, 1999). The present chapter reviews how both approaches have been used to study individual differences in childhood. The renewed, long overdue interest from researchers in the person-centred approach is not due to the failure of the variable approach, but rather to the growing consensus about what types of variables are to be included in a person-centred approach in order to obtain replicable clusters of profiles based on measurement of individuals on a comprehensive and representative set of variables. Variable-centred and person-centred approaches do not compete with each other but rather complement each other and provide a more comprehensive picture of how individual differences can and should be addressed. The variable-centred approach delineates the features (e.g. colour, suit and values) of the cards that are used in the game, whereas the person-centred approach studies the hands that are received by the players. The first part of the chapter reviews variable- and person-centred approaches as they are applied to the study of individual differences among children. Four temperament models and the five-factor personality model are discussed and compared as exemplars of the variable-centred approach. Research on ego-control and ego-resiliency, consistency of Q-sort profiles and studies of personality types are reviewed as prototypical examples of person-centred studies. The second part of the chapter presents current developments for both approaches. The analysis of natural language descriptions of children is presented as a viable strategy for the construc­ tion of a comprehensive and hierarchically organized system of variables representing individual differences among children. The review of current developments in the person-centred approach covers recent studies on the continuity and stability of person-types, their consistency across studies and clustering methods, and the replicability of the three major types in children and adults.

Variable and person-centred approaches

39

VARIABLE-CENTRED APPROACHES Temperament models Researchers interested in individual differences among children and espe­ cially among young children have conceived these differences in terms of temperamental characteristics. Temperament is traditionally distinguished from personality because it refers to stable individual differences that appear from birth onwards and that presumably have a strong genetic or neurobiological basis. Theorists differ in the emphasis they put on the role of emotional processes, stylistic components, and attentional processes as the core of temperament. Goldsmith and Campos (1982), following Darwin, define temperamental categories as basic emotions while others go beyond emotions and include processes such as attentional self-regu­ lation (Rothbart and Derryberry, 1981; Rothbart and Bates, 1998) and activity (Buss and Plomin, 1984). The emphasis on the formal or stylistic characteristics of behaviour as the core of temperament is particularly well articulated in Strelau’s (1996, 1998) work on the regulative theory of temperament. Although a consensus on the nature of temperament has not yet been reached, Rothbart and Bates (1998) provide a useful qualifica­ tion of temperamental individual differences among children in terms of the ‘3As’: individual differences in the affective, activational and atten­ tional core of personality. Thom as and Chess m odel

The New York Longitudinal Study (N Y LS ) was a milestone for intro­ ducing the concept of individual differences in developmental psychology and paediatrics (Thomas and Chess, 1977; Chess and Thomas, 1996). Based on a content analysis of a small sample of interviews of parents with young children, Thomas and Chess developed a system of nine cate­ gories to classify behaviours that are relevant for child development and refer to basic psychological mechanisms of behavioural functioning: activity-level, rhythmicity, approach-withdrawal, adaptability, threshold of responsiveness, intensity of reaction, quality of mood, distractibility, atten­ tion span and persistence. This system was further operationalized in several questionnaires and rating forms to be used by parents or teachers of infants, pre-school and school-age children (Presley and Martin, 1994). Although standard psychometric criteria were used to construct these instruments, small samples and the limited number of items prohibited a thorough analysis of the alleged nine-dimensional structure proposed by Thomas and Chess. Presley and Martin (1994) compare the dimensions recovered from item-level factor analysis of several instruments. They conclude that six factors emerge from parental ratings of temperament: social inhibition

40

Mervielde and Asendorpf

(approach-withdrawal), task persistence, rhythmicity, adaptability, nega­ tive emotionality and activity. An extensive analysis of a sizeable combined sample of normal kindergarten children aged 3 to 7 years that were rated by their mother (N= 2,176) or teacher (A^ =514) on the Temperament Assessment Battery for Children (Martin, 1988), provided evidence for five factors: social inhibition, negative emotionality, adaptability, activity level and task persistence. The teacher ratings however were less differ­ entiated and could be represented by three factors: task persistence, inhibition and negative emotionality. In a similar review, using an updated and partially different list of studies, Martin, Wisenbaker and Huttunen (1994), suggest ‘that there is very little evidence for the Thomas and Chess nine-factor structure, although some of the original dimensions appear to be strongly represented. . .. It appears that there is evidence for seven factors, although the support for some factors is not as strong as for others’ (p. 161). Two of the seven factors, rhythmicity and threshold, are consid­ ered to be controversial because they tend to be more age-dependent. The activity, approach-withdrawal and task persistence factors emerging from the reviews of factor analytic studies are clearly similar to the ones proposed by Thomas and Chess. Two others, adaptability and emotion­ ality, only partially overlap with the similarly labelled categories of the Thomas and Chess system. Emotionality emerging from the factor analytic studies tends to be restricted to negative emotionality, whereas adaptability is a broader factor incorporating some elements of the agreeableness factor as conceived in the five-factor adult personality model. Additional factor analytic studies of single instruments are unlikely to challenge the present conclusions about the Thomas and Chess model. What is missing and might challenge the current view are joint factor analyses of several instruments, combining data from comprehensive longi­ tudinal studies involving several age groups and instruments and as such providing a broader sample of items, better safeguards against method bias and more stable factor solutions. The landmark N Y LS has undoubtedly sparked a renewed interest in temperament as the childhood precursor of adult personality. It generated new research and contributed important concepts and methods for the development of a more empirically based child psychology and paedi­ atrics. However, as a model for the basic dimensions of childhood temperament, it failed to pass a rigorous psychometric test of the proposed structure.

The EA5 m odel Buss and Plomin (1975) initially distinguished four temperamental dimen­ sions with their EA SI model: emotionality, activity, sociability and impulsivity. Emotionality is roughly equivalent to distress, it varies from

Variable and person-centred approaches

41

stoic lack of reaction to extreme emotional reactions such as crying and tantrums. It involves intense activation of the sympathetic nervous system and hence high emotional arousal. They expect that, by the second year of life, emotionality will comprise three dispositions: distress, fear and anger. Activity, the second temperamental dimension, has two major components: tempo and vigour. It is best measured by the rate and ampli­ tude of speech and movement, displacement of body movements and duration of energetic behaviour. In contrast to emotionality, activity involves behavioural arousal as opposed to physiological arousal. Sociability is described as the preference for being with others, the need to share activities and to receive rewarding attention as the result of social interaction. Typical measures for sociability are: the frequency of attempts to initiate social contacts, the number of affiliations, the amount of time spent with others and reactions to isolation. In later formulations of the model Buss and Plomin (1984) distinguish between ‘sociability’ and ‘shyness’ in accordance with the research by Kagan and his colleagues (Kagan, Reznick and Snidman, 1987, 1988) on ‘inhibition’. Shyness refers to feelings of tension and distress and a tendency to escape from social interactions with strangers. This distinction is supported by the differen­ tial relation of shyness and sociability to fear and anxiety and the development of anxiety disorders (Cheek and Buss, 1981; Kagan et al. 1987). The research of Asendorpf (1990, 1991, 1993) clearly supports ‘social inhibition with strangers’ as a robust dimension in early childhood that gradually differentiates from ‘shyness with known others’ and has different correlates (Asendorpf and van Aken, 1994). The original EA SI model also included impulsivity as a temperamental dimension referring to persistence, decision time, inhibitory control and sensation seeking. It was dropped in the 1984 revision because the evidence for the heritability of impulsivity was mixed. Various questionnaires were developed to measure the key constructs of the EA SI model with due concern for the factor-analytical structure of measures. Buss and Plomin (1975) provided evidence for the four EA SI factors in children aged 1 to 9 years, measured with the E A SI Temperament Survey. The 1984 revision produced the EA S Temperament Survey (Parental ratings) measuring: shyness, emotionality, sociability and activity. Boer and Westenberg (1994) tested the factor structure of the full EAS Temperament Survey with a sample of Dutch children aged between 4 and 12 years. Contrary to expectations, they discovered that a four-factor solution did not provide a better fit of the data than a three-factor solu­ tion. Moreover, in a four-factor solution the sociability items tended to cluster together with either shyness or activity, depending on the age group. For the youngest cohort (mean age = 6.4 years) sociability items merged with shyness, for the oldest group (mean age = 10.5 years) they clustered together with activity.

42

Mervielde and Asendorpf

A Norwegian translation of the same instrument was used by Mathiesen and Tambs (1999) to check the structure of the EA S Temperament Survey (Parental ratings), on a large population-based sample of Norwegian chil­ dren aged 18, 30 and 50 months. The factor structure obtained from a Varimax rotated principal component analysis of the 15 items measuring E, A and S (shyness), excluding sociability, was very similar to the one obtained for the youngest group of the Dutch sample. However, according to Mathiesen and Tambs, a four-factor solution of the full EA S, including the sociability items, provided the best representation of the data for each of the three Norwegian age groups. Combining the results of the Dutch and Norwegian studies, it is evident that it is more difficult to obtain sepa­ rate sociability and shyness factors than was expected on theoretical grounds or on the basis of previous research that did not analyse them jointly with other prominent temperament dimensions. Buss and Plomin explicitly refer to ‘inherited personality traits’ as defining characteristics of temperamental traits. The various questionnaires based on the EA S model have been extensively used in behaviour genetic studies with monozygotic (M Z) and dizygotic (D Z) twins aged from 1 to 9 years. The results of eight of these studies were combined in a meta­ analysis reported by Goldsmith, Buss and Lemery (1997). Weighted intraclass correlations (ICRs) for M Z twins were: 0.57 (emotionality), 0.64 (activity), 0.59 (sociability) and 0.66 (impulsivity). The DZ ICRs were: 0.11 (E), -0.08 (A), 0.10 (S) and 0.15 (I). The consistent and substan­ tial MZ ICRs provide evidence for moderate genetic effects on all four dimensions. A study by Spinath and Angleitner (1998) reports very similar results for a German version of the EA S Temperament Survey. However, because the DZ correlations are less than half the value of the M Z correlations and near zero or even negative they do not provide unambiguous evidence for additive genetic effects. Potential explanations for this pattern of DZ and M Z ICRs include gene-environment interaction, epistasis, and DZ contrast effects as a result of the mother’s or rater’s tendency to exag­ gerate differences among the DZ twins (see Borkenau et al., Chapter 4, this volume). A similar meta-analysis of M Z and DZ ICRs for studies based on measures of the N Y LS temperament dimensions, the Toddler Behaviour Assessment Questionnaire (TBAQ ; Goldsmith, 1996) and the Children’s Behaviour Questionnaire (CBQ; Rothbart and Ahadi, 1994; Rothbart, Ahadi and Hershey, 1994) casts some doubt on the ubiquitous role of contrast effects because DZ correlations for these measures were consistently different from zero. According to Goldsmith et al. (1997) this should be attributed to differences in item format (behavioural specificity) and the rating format (rating versus frequency judgement). If item format and rating format affect outcomes of behaviour genetic studies it might be worthwhile to study their effects on assessment of relations

Variable and person-centred approaches

43

among temperament measures or on assessment of temperament structure. Frequency judgements of specific behaviours would be expected to yield a higher dimensionality than ratings on, for example, trait adjectives. The R o th b art and D erryberry m odel

The key concepts in the temperament model developed by Mary K. Rothbart and Douglas Derryberry (Derryberry and Rothbart, 1984; Rothbart and Derryberry, 1981; Rothbart, Derryberry and Posner, 1994) are ‘reactivity’ and ‘self-regulation’. They assert that recent temperament models ‘are based on the assumption that personality differences arise in part from the reactivity of underlying neural systems’ (Derryberry and Rothbart, 1997, p. 633). Behaviourally, temperament can be observed across all ages as differences in patterns of emotionality, activity and atten­ tion. Motivational as well as attentional systems are considered to provide the link relating specific neural systems to the major dimensions of person­ ality. In a recent formulation of their model Derryberry and Rothbart (1997) discuss four motivational and three attentional systems (see also Matthews, Derryberrry and Siegle, Chapter 7, this volume). Each of the motivational systems is related to specific neural structures mainly located in the limbic system but also to emotional states and major personality dimensions. The appetitive system mobilizes approach behaviour to stimuli that predict rewards and is therefore related to Gray’s (1987) ‘Behavioral Activation System’. Signals predicting reward produce approach behaviour and the emotion of hope; whereas signals predicting non-punishment produce the emotion of relief. Individual differences in the reactivity of this appetitive system are related to extraversión or positive emotionality. The defensive or fearful motivational system is explicitly related to Gray’s ‘Behavioural Inhibition System’ and responds to novel stimuli, biologically prepared fear signals and signals predicting punishment and non-reward. Activation of this system inhibits ongoing motor behaviour, promotes passive avoidance, increases arousal and directs attention towards relevant environmental information. Novelty and expected punishment give rise to the emotional states of anxiety and fear, whereas anticipated non­ reward produces frustration. This system is related to neuroticism or negative emotionality. The frustrative and aggressive behaviour system is linked to Gray’s fight—flight system. Several different types of aggression (e.g. irritative, instrumental and defensive aggression) are distinguished and linked to emotions such as irritation, anger and rage. Given the different forms of aggression it is not surprising that this system is tied to several (adult) personality dimensions: agreeableness-hostility, extraversión (dominance and assertiveness) and neuroticism (irritability and anger).

44

Mervielde and Asendorpf

The system serving affliative nurturant needs and regulating social behaviours is called the affiliative or nurturant system. Although knowl­ edge of the related neural structures for such a system is limited, Derryberry and Rothbart (1997) consider Panksepp’s (1986) ‘separation distresspanic system’ that responds to the loss of social support by distress and MacDonald’s (1992) ‘affectional system’, a specialized social reward system that facilitates close family relations by promoting feelings of warmth, as promising candidates. Warmth and affection are the predominant emotional states and agreeableness (trust, warmth, tendermindedness) the related adult personality factor. Moreover, Cloninger’s (1987) reward dependence is also thought to be related to this affiativenurturant system. The first attentional system discussed by Derryberry and Rothbart (1997) is a vigilance system proposed by Posner (Posner and Rothbart, 1992) and thought to regulate tonic maintenance and phasic adjustments of alertness. The posterior attentional system involves components that allow a flex­ ible shift of attention from one location to another. ‘Effortful control’ of behaviour is supposed to be regulated by the anterior attentional system that is viewed as an executive system regulating the posterior attentional system as well as attention to semantic information (Posner and Rothbart, 1992). This last system is believed to be related to the adult personality factor conscientiousness (Ahadi and Rothbart, 1994). Ahadi and Rothbart (1994) describe the Children’s Behaviour Questionnaire (CBQ), which is a 15-scale instrument developed to assess some of the behavioural and emotional indicators that were linked to various neurological systems. Factor analysis of the 15 scales in two samples (U SA and China) of mothers judging their 6- and 7-year-old children, revealed three factors with some cross-cultural variations. The larger Chinese sample (N= 468) produced a first factor grouping: approach, high intensity pleasure, smiling and laughter, activity level, impulsivity and shyness. This factor can be identified as a positive emotionality/ extraversión factor. The second factor loaded discomfort, fear, anger, sadness and soothability/falling reactivity and is clearly related to neuroticism/negative emotionality. The third factor combined inhibitory control, attentional focusing, low intensity pleasure and perceptual sensi­ tivity and hence represents effortful control and the conscientiousness factor of the five-factor model (FFM ). The three-factor structure of the CBQ scales was confirmed for three subsamples reported by Goldsmith, Buss and Lemery (1997). Derryberry and Rothbart’s (1997) frustrative and aggressive behaviour system and their affiliative and nurturant system are apparently not emerging as factors from the analysis of the CBQ scales. Discrepancies between assumed neural systems and measured factors are not easily resolved. An item-level analysis instead of scale-level analysis could

Variable and person-centred approaches

45

perhaps provide more evidence for these systems but has not yet been reported. However, how behavioural researchers ultimately define the struc­ ture of temperament is critical for the mapping of biological structures and functions. The Goldsm ith and C am pos m odel

Goldsmith and Campos define temperament as individual differences in the probability of experiencing and expressing primary emotions and arousal (Goldsmith et a l., 1987). As inclusion criteria for temperament they specify that temperament is: emotional in nature, pertains to indi­ vidual differences, refers to behavioural tendencies, and is indexed by expressive acts of emotion. They exclude cognitive and perceptual factors as well as transitory states. Emotions are defined in terms of four char­ acteristics: (1) emotions regulate internal psychological processes; (2) emotions crucially regulate social and interpersonal behaviours; (3) basic emotions can be specified by unique patterns of facial, vocal or gestural expressions; (4) basic emotions use a non-codified communication process that has an innate basis. The basic emotions which form the content dimen­ sions of this model are: anger, sadness, fear, joy and pleasure, disgust, interest and surprise. Goldsmith (1996) constructed the Toddler Behaviour Assessment Questionnaire (TBAQ ), a care-giver report with acceptable psychometric properties, including reasonable convergent and discriminant validity (Goldsmith et al., 1997; Lemery et al., 1999). It consists of five fairly independent scales: activity level, pleasure, social fearfulness, anger prone­ ness, and interest/persistence. The TBA Q is clearly related to Rothbart’s CBQ. Thirteen CBQ scales correlate from 0.34 to 0.68 with one of the five TBAQ scales (Goldsmith et al., 1997). The activity level scales of both instruments correlate 0.68. TBAQ pleasure is related to CBQ smiling/laughter, approach, high pleasure and low pleasure. Social fear­ fulness is related to CBQ shyness and fear. Anger proneness as measured by the TBAQ is correlated with CBQ anger, impulsivity, sadness, activity level and approach. Interest/persistence is mainly related to attentional focusing as measured with the CBQ. Recent research on the Goldsmith and Campos model is mainly confined to behaviour genetic analyses. Goldsmith et al. (1997) present model fitting analyses for each of the TBAQ scales. Based on DeFries-Fulker regres­ sion analysis of a sample of 184 toddler twin pairs, aged 17 to 36 months, rated by their primary care-giver. Goldsmith et al. (1997) show that there is clear evidence for moderate genetic effects on activity level, social fear­ fulness, anger proneness and interest/persistence (h1 ranging from 0.26 to 0.78, depending on the fitted model). However, social fearfulness and anger proneness also exhibited moderate shared environmental effects

46

Mervielde and Asendorpf

(ic2 ranging from 0.28 to 0.63 depending on the model). Most interesting was the fact that the pleasure scale, referring to positive emotions showed the strongest evidence for only shared environmental effects (c2 of 0.50 or 0.66 depending on the model). Given that most temperament models do not include scales for positive emotions, this provides an important exception to the general finding that shared environmental effects are negli­ gible. A similar analysis for three component scores of the CBQ did not confirm this effect. Almost no evidence for shared environmental effects was detected from the model fitting analysis of CBQ component scales surgency, negative affectivity and effortful control, suggesting that char­ acteristics of the questionnaires can affect biometric inferences. Continuity across age levels is traditionally viewed as one of the defining characteristics of temperament. Lemery et at. (1999) conclude from a review of longitudinal studies of early temperament that there is evidence for stability, at least within a given research tradition based on the same or highly similar instruments. Using four composite measures (positive emotionality, fear, distress-anger and activity level), combining content from at least two questionnaires, they distinguish among different types of stability based on various statistical models for stability. A simple common factor model fits the stability data for a sample of toddlers (24 to 48 months), while an autoregressive simplex model, allowing progressive change, best fitted the data for infants (3-18 months), suggesting increasing stability from infancy to the toddler period. This review of temperament models shows that each model has a set of scales that partially overlap with those proposed by other theorists. Moreover, even within a single temperament model, it is often necessary to construct age-specific measures to cope with the expanding behavioural repertoire of the children. Because so many instruments have been devel­ oped it is not feasible to include most of them in a single study in order to determine the common factor structure. Part of the problem comparing the temperament models is that Thomas and Chess, and Rothbart and Derryberry, developed many scales that are moderately correlated, while Buss and Plomin used fewer but independent scales and Goldsmith prefers to integrate scales from different questionnaires into independent compos­ ites. Table 2.1 presents an attempt to compare the relatively independent dimensions emerging from the four temperament models. Emotionality is clearly present in each of the four models as a higher order dimension, and given the emphasis on distress and negative emotions, it might also be labelled as negative emotionality. The second consistent dimension refers to sociability versus social inhibition/shyness and may be best captured with the more general label extraversión. Activity is also promi­ nent as an independent dimension in three of the four models. Finally, persistence is present in two of the four models and hence might turn out to be a minor dimension at least for preschool children.

Variable and person-centred approaches

47

Table 2.1 Comparison of dimensions emerging from four temperament models Emotionality

Extroversion

Activity

Persistence

Thomas-Chess3

Negative emotionality

Inhibition

Activity level

Task persistence

EAS-model

Emotionality

Sociability Shyness

Activity

Rothbart-Derryberryb

Negative affectivity

Surgency

Surgency

Goldsmith-Camposc

Distress-Anger

Fear (social) Positive emotionality

Activity level

Effortful control

Notes a Based on Martin, Wisenbaker and Huccunen (1994). b Labels from Goldsmith, Buss and Lemery ( 1997). The surgency component includes activity level, c Composites from Lemery et al. (1999).

Given the greater comparability of dimensions emerging from the use of broader, more or less independent dimensions, future research might benefit from two strategies: ( 1) to determine the factor structure of scales used within a given tradition (e.g. N Y LS or Rothbart CBQ ) and to combine them into higher order factor scores or composites, and (2) to include higher order factor or composite scores from at least two temperament models in studies of the structure of temperament, stability and predictive validity. Five-factor personality model Although the terms temperament and personality are both used to refer to individual differences throughout the life-span, one cannot deny that temperament is the preferred term for designating differences among infants and children whereas personality is more commonly used to address differences among adults. In similar vein we use the term fivefactor or Big Five personality model to refer to research that takes adult personality dimensions as the target structure to model individual differences among children. Given the growing consensus about the Big Five as an appropriate model for charting individual differences among adults, we will confine this section to studies that are based on this model. Two lines of evidence have contributed to the consensus about the dimensions of adult personality: the lexical approach and the ques­ tionnaire approach. The lexical approach shows that five broad-band dimensions explain a significant and large proportion of the personality descriptive lexicon culled from standard dictionaries of different languages:

48

Mervielde and Asendorpf-

(I) Extraversión, (II) Agreeableness, (III) Conscientiousness, (IV ) Emotional stability and (V ) Intellect or Culture. The history of this type of research has been described by John (1990) and by Digman (1990, 1996), and for a recent review see Saucier, Hampson and Goldberg, Chapter 1 in this volume. The lexical approach alone did not convince many personality psychol­ ogist to adopt this five-dimensional model. It was argued that trait adjectives, the basic units of the lexical analysis, are too abstract or general compared with the typical behavioural items that figure as the basic units of most personality questionnaires. However, Goldberg’s taxonomic research was a major impetus for Costa and McCrae (1985) to update their NEO questionnaire (measuring Neuroticism, Extraversión and Openness) to include the two missing factors, conscientiousness and agree­ ableness. The revised version of this instrument, the NEO-PIR (Costa and McCrae, 1992) goes beyond the five domain scales by specifying six facets per factor, each measured by eight items. This model is commonly referred to as the five-factor model (FFM ), to contrast it with the Big Five emerging from the lexical research tradition (McCrae and Costa, 1996). However, four of the five factors included in both models are very similar both in terms of content as well as judging from empirical research. The fifth factor, however, has turned into a sort of ‘scientific embarrassment’ (Goldberg, 1994) with two major interpretations stemming from the lexical research (culture and intellect), Costa and McCrae’s emphasis on open­ ness, and divergent results of lexical studies of Hungarian, Dutch and Italian (Ostendorf and Angleitner, 1994). In an impressive series of studies, McCrae and Costa (1987) demon­ strated that most of the variance captured by traditional personality questionnaires could be accounted for by the five factors emerging from lexical research tradition. Recent research convincingly demonstrates the cross-cultural stability of this questionnaire measure of the FFM (McCrae and Costa, 1997; McCrae et al., 1999). Moreover, Costa and Widiger (1994) established the clinical relevance of the model by documenting the relationships between the FFM and personality disorders. Although the consensus about the validity and cross-cultural generalizability of the Big Five and the FFM is still growing, the models have also been criti­ cized for their reliance on the lexical approach, factor analysis and self­ presentation (Block, 1995; Eysenck, 1992; McAdams, 1992). T e a c h e r an d p a re n ta l ratings

Digman and his team were the first to show the usefulness of the Big Five model to represent individual differences among children. With a set of 20 (post hoc selected) scales from a group of 49 scales previously used to validate a ten-factor model, Digman and Takemoto-Chock (1981)

Variable and person-centred approaches

49

provided the first evidence for the validity of the Big Five as a model for individual differences among elementary school age children. Digman and Inouye (1986) recovered the five-factor structure from teachers’ nomina­ tions of sixth-grade children on 43 scales. Early in his career, when strongly influenced by Cattell, Digman was convinced that at least 10 common factors could be extracted from teacher ratings of children on Cattell’s scales (Digman, 1994). However, his later reanalysis of two of these early studies, using 35 common scales and comparability coefficients to deter­ mine the number of factors, produced a highly comparable five-factor structure for both studies. Victor (1994). using the same set of ‘Hawaii Scales’ as Digman and Inouye (1986), confirmed the Big Five factor struc­ ture for teacher ratings of fifth and sixth grade rural school children and documented relationships with measures of problem behaviours and school achievement. To document the validity of the Big Five as a model for individual differences among school-age children with a different set of rating scales, Mervielde, Buyst and De Fruyt (1995) asked teachers to rate 2240 chil­ dren, aged 4 to 12 years, on a set of 25 bipolar scales, derived from Goldberg’s (1992) markers of the Big Five. Principal component analysis of ratings of kindergarten children (aged 4 to 6 years) revealed four of the five factors, including a combined conscientiousness/intellect/openness factor. The complete five-factor structure emerged from the ratings of primary school children. To clarify the meaning of the fifth factor a sixfactor solution was obtained for the three primary school age levels. As a result, the fifth factor bifurcated in two components: one loading two openness scales and the other marking the intellect rating scales. The fifth and sixth factors correlated primarily with the fifth factor from the fivefactor solution, but the pattern of correlations changed with increasing age. For younger children the fifth factor was highly correlated with intel­ lect and to a lesser extent with openness. For the oldest children this pattern was reversed. The group aged 8 to 10 years showed intermediate correlations with both components. Moreover, openness and intellect had a differential pattern of correlations with some of the other factors as well. Openness was positively related to agreeableness, whereas intellect was negatively related to the same factor. Intellect, on the other hand, corre­ lated with conscientiousness, but openness tended to be inversely related to conscientiousness. Furthermore, multiple regression of grade point average on factor scores derived from the six-factor solution, indicated that the predictive validity of the Big Five steadily increased from 0.67 to 0.79 from grade one to grade six, with intellect and conscientiousness as the consistent predictors and openness growing in importance with increasing age. Finally, Mervielde (1998) showed that when mothers rate their own school-age children on the same set of bipolar scales, a similar five-factor structure emerges.

50

Mervielde and Asendorpf

Some studies recovered the Big Five from scales or item sets that were not constructed as measures of the Big Five and provide evidence that ‘prestructuring’ is not a prerequisite for the emergence of the fivefactor structure. Digman and Shmelyov (1996) collected teacher ratings of 8-10-year-old Russian children on 60 temperament and personality scales. An analysis of the 21 temperament scales showed four factors: sociability, anger, impulsiveness and fear. An analysis of a set of combined temperament and personality scales demonstrated the usefulness of the Big Five for representing personality characteristics of Russian children. There was a clear correspondence between the temperament dimensions and four of the five personality factors. The personality factors included an intellect/openness factor loading scales such as original, perceptive, knowledgeable, adaptable, curious, imaginative, dependent and rigid. Van Lieshout and Haselager (1994) used the Dutch California Child Q-set (CCQ) descriptors to examine the validity of the FFM as a model for the structure of parental and teacher ratings collected from a large sample of children and adolescents. An overall principal component analysis showed that seven factors provided the best fit for the data. Besides the adult Big Five dimensions, two additional factors referring to motor activity and dependency were identified. John et al. (1994) confirmed the findings of Van Lieshout and Haselager (1994) with Q-sorts of 350 ethnically diverse 12- to 13-year-old boys on the Caregiver CommonLanguage California Child Q-set. A seven-factor solution provided a good account of the data, showing two possible age-specific dimensions beyond the Big Five: irritability and positive activity. Moreover, John et al. (1994) showed significant relationships of the CCQ Big Five with theoretically and socially important criteria such as child psychopathology, juvenile delinquency, school performance and intelligence. Self-ratings and p eer nom inations

Although the Big Five can be successfully recovered from teacher and parental ratings of primary school children there is as yet little evidence that the model holds for preschool children or for self-ratings of school age children. Dreger (1995) was unable to recover the adult Big Five from self-ratings of 4- to 6-year-old children. Asendorpf and van Aken (1995) compared the factorial structure of self-, friend- and parental ratings of 155 German 12-year olds on 60 age-appropriate bipolar adjective markers of the Big Five. Whereas a clear five-factorial structure emerged for the parental ratings and to a lesser extent for the friend-ratings, only consci­ entiousness emerged as a clear factor in the self-ratings. Particularly problematic was a high (0.55) correlation between extraversión and emotional stability in the self-ratings. It seems that the Big Five are not sufficiently differentiated in self-ratings even in preadolescence.

Variable and person-centred approaches

51

Whereas self-ratings of young children are too unreliable or evaluative, aggregated peer nominations may provide an interesting and more reliable methodology for structural analysis of children’s personality perception. De Fruyt and Mervielde (1992) were unable to recover the Big Five from a principal component analysis of the 50 unipolar ends of 25 bipolar nomi­ nation scales, but showed that a separate analysis of the evaluatively positive nomination scales could be represented by four factors (one combining conscientiousness and intellect scales) and a three-factor solu­ tion for the negative unipolar scales. In contrast, Scholte, van Aken and Van Lieshout (1997) failed to find the Big Five in their analysis of peer ratings by adolescents. Mervielde and De Fruyt (in press) address the methodological problems of the analysis of peer nominations. They show that, given appropriate corrections for the skewness of the distrib­ ution of nominations, a replicable three-factor solution emerges, including a combined conscientiousness- intellect factor, a combined extraversionstability factor and a separate agreeableness factor. The basic dimensions emerging from temperament models and from the FFM are substantially related. Both approaches propose a sociability or extraversión dimension referring to the need for social contact or the lack or avoidance of social interaction. Negative emotionality or emotional stability versus neuroticism are recognized by all the reviewed tempera­ ment models and by both the lexical and the NEO-PIR model as essential and clinically relevant dimensions of individual differences in childhood. Two temperament models pay attention to task persistence or effortful control and as such provide a link to the conscientiousness factor from the Big Five. Activity level is a prominent independent dimension for three temperament models but is considered to be a facet of extraversión in the personality model. Agreeableness is not very prominent in tempera­ ment models but its significance for childhood has been cogently argued by Graziano and Eisenberg (1997). The fifth factor of the Big Five is not a prominent dimension in any of the temperament models because curiosity, creativity and exploratory behaviour are conceived as markers for cognitive development. Shiner (1998) integrates temperament and personality models in her proposal for a preliminary taxonomy of middle childhood traits. She proposes the following four broad dimensions: (1) positive emotionalityextraversion (sociability, social inhibition, dominance); (2) negative emotionality-neuroticism; (3) aggressiveness and prosocial tendencies, and (4) constraint-conscientiousness (persistence/attention, mastery motiva­ tion, inhibitory control and activity level). The relationship among the four broad dimensions proposed by Shiner and the first four factors emerging from the lexical approach and the FFM is once again clear and illustrates that bridges are being built between temperament and adult personality models. Future research is bound to cross these bridges more often.

52

Mervielde and Asendorpf

PERSON-CENTRED APPROACHES Most person-centred approaches to childhood personality are based on the Q-sort method. Each child is described by a knowledgeable informant who sorts trait descriptions according to how well they fit the child’s person­ ality. This method of personality assessment has sometimes been described as idiographic, but it should be noted that judges will base their judge­ ment of the saliency of a particular trait in a child on the child’s deviation in the trait from the age norm. Most studies have used the California Child Q-Set (CCQ; Block and Block, 1980) that consists of 100 trait descrip­ tions, or the German short version of the CCQ with 54 trait descriptions (Góttert and Asendorpf, 1989). Because the judges are instructed to assign the same number of traits to each level of saliency for the child’s person­ ality, the level and scatter of the Q-sort profiles are identical across different children, and the Pearson correlation between the profiles is an adequate measure of profile similarity. Three different kinds of Q-sort studies of children’s personality can be distinguished: studies on ego-control and ego-resiliency, studies on the consistency of the Q-sort profiles across ages and judges, and studies on personality types. The order of presentation reflects both their historical order and an increasing person-centredness. Early Q-sort studies reduced the information in the Q-sort profile to two dimensions of personality: ego-control and ego-resiliency. These dimensions were theoretically postulated in the dissertations by J. Block and J.H. Block in 1950/51 and were since then the focus of the Blocks’ empirical studies. Ego-control refers to the tendency to contain versus express emotional and motivational impulses (overcontrol versus under­ control). Ego-resiliency refers to the tendency to respond flexibly rather than rigidly to changing situational demands and particularly stressful situations. Block and Block (1980) operationalized ego-control and egoresiliency by the intraindividual correlations of a child’s Q-sort profile with the expert profiles of a prototypically ego-undercontrolled and a prototypically ego-resilient child. Thus person-centred assessments of personality were used to operationalize dimensions of personality that were subsequently applied in variable-centred analyses of the antecedents, corre­ lates and consequences of ego-control and ego-resiliency. Block and Block (1980) assumed that both extremely high and low ego-control are related to low ego-resiliency, and high ego-resiliency to intermediate ego-control. Surprisingly, this assumption remained untested until Asendorpf and van Aken (1999) confirmed the postulated inverted U-shaped relationship for both 4-6- and 10-year-old children (see Figure 2.1). Thus, the ego-control/resiliency model is a non-linear, twodimensional model of personality that can be viewed as an alternative to the FFM.

Prototypicality lor overcontrol

Prototypicality for underconlrol

Prototypicality for undercontrol Figure 2.1

Relations between the prototypicalities (Q-factor loadings) for the three Q-types identified at ages 4-6 years Source: Adapted from Asendorpf and van Aken (1999, Figure 8).

54

Mervielde and Asendorpf

Block and Block (1980) provided ample evidence for the convergent and discriminant validity of 130 preschool children’s ego-control and egoresiliency scores with regard to a large battery of experimental tests and observational procedures such as delay of gratification, actometer measures of motor activity and exploratory behaviour for ego-control, and level of aspiration, IQ and motor inhibition for ego-resiliency. If it is true, as the Blocks have always assumed, that the 100 CCQitems represent a broad, rather unselected sample of traits on which children can differ, it should be possible to recover the Big Five factors of person­ ality from the CCQ. Indeed, John et al. (1994) showed that it is possible to construct brief, internally consistent scales of CCQ-items for four of the Big Five. The fifth factor (openness/culture) could be scaled only with a questionable reliability (a = 0.53 for a seven-item scale). Thus, despite differences in theoretical approach and operationalization, the Q-sort approach is not incommensurable with the FFM. Huey and Weisz (1997) studied the relative advantage of CCQ-based indices of ego-control/resiliency and the Big Five for the prediction of externalizing problems such as aggression and school drop-out, and inter­ nalizing problems such as low self-esteem, loneliness and depression, in 116 clinically-referred children aged 7-17 years (median 11 years) (see Achenbach and Edelbrock, 1981 for the concept of externalizing and internalizing problems). Both approaches were similarly successful in their predictions, explaining approximately 40 per cent of the variance in exter­ nalizing problems, and 30 per cent of the variance in internalizing problems. When the two sets of predictors were combined, the predicted variance was 49 per cent for externalizing problems, and 36 per cent for internalizing problems, indicating modest unique contributions of each approach. A second, more person-centred application of the Q-sort approach is found in studies on the consistency of children’s personality across ages and judges. Whereas variable-centred approaches to consistency focus on the constancy of the rank-order of children in one trait, the Q-sort approach makes it possible to study the constancy of complete personality profiles in many traits for individual children. Ozer and Gjerde (1989) examined the consistency of the individual profiles of 84 children between 3-4 and 7 years of age. The resulting Pearson correlations showed a wide range of consistency, from r = -0.28 to r = 0.87. Thus, the children differed strongly in the long-term stability of their personality. Asendorpf and van Aken (1991) confirmed a similarly wide range for German and Dutch children for 2-year intervals (4-6 and 10-12 years). In addition, they found a surprisingly strong correlation between children’s ego-resiliency and their consistency (e.g. r = 0.57 between ego-resiliency at age 4 and the profile consistency over the following 2 years for 151 children). In a follow-up study of this sample, van Aken and Asendorpf (1999) found that the correlation between early resiliency and individual consistency

Variable and person-centred approaches

55

was even higher for the consistency between ages 4 and 10 years (r = 0.64) even though teachers served as judges at age 4 and parents at age 10. The reason for this robust correlation between resiliency and the stability of personality seems to be that the consistency of Q-sort profiles across different judges is also related to resiliency (van Aken and Van Lieshout, 1991). The conclusion from these studies is that resilient children are more consistent than overcontrolled or undercontrolled children. Asendorpf and van Aken (1991) discussed three different interpretations of this finding: resiliency promotes age-appropriate behaviour and therefore consistency across contexts and time; a consistent view of significant others promotes consistent social feedback and therefore resiliency; and a stable social envi­ ronment promotes both resiliency and consistency. The last application of the Q-sort approach is both the most recent and the most person-centred one. The Q-sort profiles of a sample of children are grouped into relatively homogeneous personality types. Although the idea that persons can be compared with regard to the similarity of their profiles in many traits can be traced back to the ‘Komparationsforschung’ by Stem (1911), it is not easy to implement this idea in a methodologi­ cally sound classification procedure that results in types that are robust with regard to variations in judges, ages and cultures. Robins et al. (1996) proposed such a method. The individual Q-sort profiles are factor analysed by Q-factor analysis (also called inverse factor analysis because the roles of persons and variables in ordinary factor analysis are reversed). The resulting Q-factors represent prototypic personality patterns. Only those factors that are replicable between random halves of the sample of persons are retained. Both Robins et al. (1996) and Hart et al. (1997) found that three, but not more, Q-factors were replicable in their studies of 13-yearold North American boys and 7-year-old Icelandic children. Subsequently, the individuals are classified into personality types according to the best-fitting Q-factor. In some cases, problems arise because a Q-sort loads equally high on two factors, or low on all factors. To solve this problem, Robins et al. (1996) introduced a more complex procedure. First, only persons who have a particularly high factor loading on one Q-factor are classified. Second, the remaining persons are assigned to these Q-types by discriminant analysis. Robins et al. (1996) related the three resulting types to their average ego-control and ego-resiliency scores. As Figure 2.2 indicates, the types can be characterized as resilients, overcontrollers and undercontrollers. Robins et al. (1996) related the three types also to their average scores in the Big Five CCQ scales. In all five scales, resilients had slightly more socially desirable scores (e.g. above-average scores in emotional stability). Overcontrollers had particularly low scores in extraversión and emotional stability, and undercontrollers had particularly low scores in conscien­ tiousness and agreeableness.

56

Mervielde and Asendorpf 60

1 Type 1

55-

gT o o g

50-

o c 5 gj5

ac 45 -

Type 2 □

o O)

UJ

40 -

Type 3 □

35 30

t------ 1------ 1--------------1------ 1------ r 35 40 45 50 55 60 65

70

Ego-control (high = undercontrol, T-score) Figure 2.2 Three Q-types as a function of ego-control and ego-resiliency in the study by Robins et al. (1996) Source: Adapted from Asendorpf and van Aken (1999, Figure I).

Robins et al. (1996) found lower IQ scores, lower teacher ratings of academic performance and school conduct, and a higher frequency of serious delinquency for undercontrolled boys as compared with both resilient and overcontrolled boys. Hart et al. (1997) identified the three Q-types at age 7, and studied their future development with regard to academic and social competence up to age 15. Resilients had a higher grade point average and fewer teacher-rated concentration problems than both undercontrollers and overcontrollers; overcontrollers had higher scores in teacher-rated social withdrawal and lower scores in self-esteem than the other two types; and undercontrollers had higher scores in teacher­ rated aggressiveness than their overcontrolled and resilient peers, but not below-average self-esteem. Overall, these findings can be summarized as externalizing tendencies in undercontrollers and internalizing tendencies in overcontrollers. Another person-centred approach of empirically deriving personality types is the cluster analysis of multiple trait scores (questionnaire scales, ratings or test scores). Each individual is described by the profile of scores

Variable and person-centred approaches

57

in these variables. These profiles are grouped by cluster analysis into rela­ tively homogeneous clusters. Each cluster represents a personality type, and the average profile of the cluster members describes a personality prototype. Profiles that differ with regard to overall level and scatter can be treated as different if an appropriate measure of profile similarity is used (e.g. the Euclidean distance of the profiles). Caspi and Silva (1995) obtained 22 observational ratings for a birth cohort of 1024 3-year-old New Zealand children from examiners who observed the children in various testing situations, reduced these ratings by factor analysis to three factors, and clustered the resulting profiles of factor scores. They found five clusters that could be replicated between two random thirds of the sample. Two clusters can be interpreted as under­ controlled and overcontrolled, respectively, and the remaining three clusters as subtypes of the resilient type (see Caspi, 1998). Thus, there was some consistency between these types and the Q-sort based types despite the different method of deriving the types. In a follow-up study of the children who were identified by Caspi and Silva (1995) at age 3 years as resilient, overcontrolled and undercontrolled, Caspi and Silva (1995) and Caspi et al. (1996) found that overcontrollers and undercontrollers continued to show internalizing versus externalizing tendencies through age 21years, including depression and alcohol depen­ dence for overcontrollers, and criminality and suicide attempts for undercontrollers. In terms of their social relationships, Newman et al. (1997) reported that overcontrollers at age 3 had lower levels of social support at age 21 but normative adjustment in romantic relationships and at work, and undercontrollers showed lower levels of adjustment and greater interpersonal conflict. The three resilient clusters were character­ ized in these outcome studies by a lack of both internalizing and externalizing problems. Together with the Q-sort based findings, these results provide a consistent picture of resilients being adjusted, overcon­ trollers showing internalizing tendencies, and undercontrollers showing externalizing tendencies, independent of their age.

CURRENT DEVELOPMENTS Variable-centred approaches Cross-cultural com parison o f p a re n ta l descriptions o f child personality

The dictionary is undoubtedly the most complete source of information for the analysis of the person-descriptive lexicon, but most of the persondescriptive adjectives listed in it are part of the passive person-descriptive

58

Mervielde and Asendorpf

language, that is understood by native speakers but not frequently used. An analysis of the natural (active) language as it is used in everyday discourse is an alternative approach to the study of the language of everyday, naturalistic discourse on personality (John, 1990; Church, Katigbak and Castenada, 1988). Natural language personality descriptions go beyond sampling of abstract trait adjectives and therefore may be of particular interest for the study of childhood personality that requires descriptions at the behavioural level in order to capture significant developmental trends. Moreover, the analysis of free descriptions takes into account both the range of terms and their frequency of use and hence assigns comparable weight to categories with few but frequently used terms as well as to categories with many different but infrequently used terms (Buyst, De Fruyt and Mervielde, 1994). The most obvious advantage of free descriptions is that they sample child descriptive information that is not framed in the language proposed by the researcher but rather in the language preferred by informants to encode the behaviour of the child. Finally, because free descriptions are predom­ inantly phrased in behavioural terms they tend to be more sensitive to cultural or even subcultural variations in child behaviour. The analysis of unconstrained, parental personality descriptions of chil­ dren was the cornerstone of a collaborative international research project conducted by research teams from seven countries: Belgium, China, Germany, Greece, Holland, Poland and the US. They collected oral parental personality descriptions of 2416 children aged 2 to 12 years, according to a common procedure described by Kohnstamm et al. (1995) and Kohnstamm et al. (1998). A literal transcription of each oral interview was segmented in small units that described individual differences. The more than 46,000 segmented units were assigned to 14 major categories. The system was inspired by the FFM and included three facets for each of the Big Five dimensions. Eight additional categories were added to accommodate almost all child characteristics reported by parents.1 Main categories and facets were further subdivided into an evaluatively posi­ tive, a neutral and an evaluatively negative subcategory. Across countries, 76 to 85 per cent of the descriptors could be assigned to the Big Five categories. However as reported in Table 2.2, the percentage allocated to each of the Big Five categories varied across countries. The percentage of descriptors coded as indicating openness/intellect varied from 11 per cent in Greece to more than 21 per cent in the US. The most frequently used category in each of the seven countries was extraversión, confirming the prominence of this factor in lexical research. Agreeableness was the second most frequently used category for five of the seven countries. Conscientiousness and emotional stability attracted on average less than 10 per cent with a significant exception for the Chinese sample. On average about one third of the descriptors provided by the parents were classified

Variable and person-centred approaches

59

Table 2.2 Proportions of free parental child descriptors allocated to the Big Five Belgium

China

Germany Greece

Nether­ lands

Poland

USA

No. of descriptors

9607

4458

9135

9744

6660

4567

2184

Extraversión Agreeableness Conscientiousness Emotional stability Openness/Intellect

27.7 19.6 7.9 9.2 12.6

27.2 17.4 19.4 7.1 14.1

29.9 21.3 8.3 8.0 17.6

25.0 25.5 10.7 8.7 1l.l

28.5 18.9 6.8 9.9 12.2

29.7 23.7 8.1 8.4 11.5

29.5 17.8 7.3 8.6 21.2

Total Big Five

77.0

85.2

85.1

81.0

76.3

81.4

84.4

Source: Based or> Table 1.3 in Kohnstamm et al. (1998).

as instances of the less desirable poles of the Big Five. European parents provided between 32 and 33 per cent negative descriptors, Chinese parents 42 per cent and US parents 21 per cent, indicating possible cross-cultural differences in the salience of the negative poles of the Big Five or in the willingness to report on less desirable aspects of the child’s behaviour. Age-related changes of parental descriptions of child characteristics were reported by Slotboom et al. (1998). The percentage of descriptors referring to extraversión dropped on average from 25 to 20 from age 3 to age 12 years. The percentage of descriptors referring to the low end of this dimension was much smaller and did not change across age. High agreeablencss descriptors were more frequently used to describe 9- and 12-year-olds than for 3- and 6-year-old children, whereas the frequency of descriptors categorized as ‘unagreeable’ dropped especially from age 3 to age 9 years. Overall these age trends suggest an increase in agree­ ableness of child descriptions. One of the strongest age-related shifts in parental descriptions was the significant increase in the use of descriptors referring to conscientiousness, in particular between age 3 and 9 years, coinciding with the demands of formal education. References by parents to instances of emotional stability were not age related and were equally salient for all age groups. The use of descriptors referring to openness and intellect dropped slightly from 13.4 per cent for 3-year-olds to 10.9 per cent for 12-year-olds, suggesting that at least in the minds of the parents, openness and intellect are salient aspects of the behaviour of their child. These findings should be contrasted with the absence of markers for openness and intellect in the traditional temperament literature, espe­ cially for younger children. Overall, the analysis of the distribution of free parental descriptors over age levels and countries suggests several inter­ esting age-related trends and is probably a more sensitive procedure to detect cross-scctional changes than comparison of ratings.

60

Mervielde and Asendorpf

V alid ity o f free descriptions

A major disadvantage of parental free descriptions is that they are guided by the perspective of the parent and therefore tend to cover only the char­ acteristics of the child that are salient at a given moment and for a particular parent. To check the content validity at the individual level of analysis, Mervielde (1998) counted how many of the Big Five categories were used at least once by the parents to describe the child. Combining the data for the seven countries he shows that 31 per cent of the parents use all five categories to describe their child, 37 per cent use four categories and 24 per cent sample three of the Big Five categories. Only 1 per cent of the parents sample one Big Five category to describe a child and 7 per cent refer to two out of five. The fact that almost 68 per cent of the parents refer to four or five of the Big Five when freely describing their child indicates the saliency of these categories not only at the group level but also when protocols of individual parents are analysed. Classification of free descriptions provides some indication for the rele­ vance of the five factors as dimensions for organizing natural language personality description, but it has many drawbacks from a methodolog­ ical point of view. It remains problematic because judges tend to fit the segmented descriptors to the available categories. One cannot exclude the possibility that relevant categories were not part of the system. Moreover, classification is a low level form of measurement that needs to be corrob­ orated by measurement procedures that go beyond the nominal scale level. Mervielde (1998) accomplished this by comparing the categorized free description data with parental ratings of the same children on indepen­ dently selected scales aimed at measuring the Big Five factors. The relationship between categorized free descriptions on the one hand and ratings on bipolar scales on the other hand was assessed by regressing the frequency of descriptors per category on factor scores extracted from the ratings. The multiple correlations varied from 0.60 for extraversión to 0.27 for openness, suggesting a moderate degree of correspondence between both methodologies. Moreover, a joint principal component analysis of 25 bipolar Big Five rating scales and the 10 proportions of free descriptions (one for each pole of the Big Five and for each described child) showed that the proportions of free descriptions consistently load on the Big Five component marked by the corresponding Big Five rating scales. Hence categorized free descriptions share common variance with ratings of the same children on scales selected to mark the Big Five. The co n structio n o f th e H ie r a rc h ic a l P e rso n a lity In v e n to ry fo r C hild ren

The correspondence between free descriptions and ratings on Big Five scales points to a common content tapped by both assessment tools and

Variable and person-centred approaches

61

is a necessary but not sufficient condition for a common structure. Structural analysis of the content of free descriptions requires ratings of children on a common and representative set of items covering the content captured by the aggregated free descriptions. Mervielde and De Fruyt (1999) describe how they sorted the extensive set of more than 9000 free parental descriptions provided by Flemish parents, into an organized set of more than 100 clusters covering three age groups: 5-7, 8-10 and 11-13 years. Adopting a set of formal rules, two to four items were produced for each of the clusters. A sample of 696 children was rated by one of the parents and one or two teachers on age-dependent item sets varying from 234 to 282 items. Principal component analyses at the item level indi­ cated that, for each age level, the first five principal components tended to group items according to the Big Five categories that were used to sort the free descriptors. The largest component was a combination of items from the agreeableness category together with dominance and irritability items. To mark the wider and different content of this factor it was labelled ‘benevolence’. Due to the broader content attracted by the benevolence factor, the emotional stability factor was reduced to anxiety items and items referring to self-confidence. The extraversión factor was primarily marked by shyness and positive emotionality items. Items indicating order­ liness, achievement motivation, attention and task persistence marked a classic conscientiousness factor. Finally, items referring to creativity, curiosity and intellect formed a combined openness/intellect component or imagination factor. To further structure the broad sample of items that loaded on each of the five extracted components, classic reliability analysis and principal component analysis were used to construct facets within each factor and for the three age groups (Mervielde and De Fruyt, 1999). Construction of facets for each of the age-specific instruments, resulted in three ques­ tionnaires with a rather similar structure and therefore it was decided to integrate the three age-related questionnaires into one Hierarchical Personality Inventory for Children (H/PIC) covering individual differences for children aged 5 to 13 years. The H/'PIC groups 144 items at the highest level into five domains: conscientiousness, benevolence, extraversión, imagination and emotional stability. Each of the five broad domains was further subdivided into facets measured by 8 items. The factor structure of the 18 facets was cross-validated on a sample of 719 twins and siblings rated by both parents. To illustrate the stability of the structure of the 18 facets, facet loadings derived from the principal component analysis of the construction and the cross-validation sample are reported in Table 2.3. The four conscientiousness facets are relatively pure markers of the first principal component. Egocentrism and irritability are the highest loading facets and the purest markers for the benevolence factor, whereas

62

Mervielde and Asendorpf

Table 2.3 Principal component analysis of H/PIC facets in two samples CON A Conscientiousness (CO N ) Achievement motiv. .87 Orderliness .86 Concentration .86 Perseverance .85 Benevolence (BEN) Egocentrism -.20 Irritability -.21 Compliance .50 Dominance .06 Altruism .14 Extraversión (EXT) Shyness -.09 Expressiveness .10 Optimism .03 Energy -.16 Imagination (IMA) Creativity .07 Curiosity .42 Intellect .45 Emotional stability (EMS) .08 Anxiety Self-confidence .22

BEN B

A

IMA

EXT B

A

B

A

EMS B

A

B

.83 .86 .85 .85

-.04 .10 -.19 -.19 -.24 -.14 -.28 -.23

.16 .14 -.06 -.07 -.03 -.14 .07 .00

.31 .33 -.01 -.05 .22 .25 .21 .11

.09 .10 -.20 -.06

.05 .08 -.16 -.10

-.13 -.19 .51 .11 .13

.91 .89 .85 .77 -.75 -.72 .74 .75 -.56 -.60

-.08 -.08 .07 .15 .00 .00 .44 .37 .46 .46

-.04 -.02 -.01 .00 -.01 -.04 .18 .16 .30 .21

.09 .22 .11 -.20 .40

.17 .39 .03 -.19 .28

.07 .00 .04 -.05

.02 .06 .12 .14 -.31 -.30 .32 .19

-.82 .79 .76 .66

-.73 .76 .67 .73

-.11 -.06 .24 .32 .19 .31 .04 .01

.30 .33 .03 .03 -.14 -.23 -.02 .01

.03 .25 .42

-.06 -.12 .05 .11 .06 .12

.22 .31 .11

.14 .27 .13

.08 .22

.10 .10

.14 .06

-.14 -.08 .48 .40

.84 .73 .69

.83 .78 .67

.02 .00 -.07 -.07 -.35 -.30

-.03 -.02 .36 .32

.93 .91 -.67 -.72

Notes Highest loadings are printed in bold. Sample A: N = 696, children aged 6-12 rated by a parent and one or two teachers. Sample B: N = 719, twins and siblings aged 6-12 rated by both parents.

compliance operates more like a blend of benevolence and conscien­ tiousness. Dominance and altruism combine a primary loading on benevolence with a substantial secondary loading on extraversión. Shyness is the highest loading facet of extraversión with a moderate loading on emotional stability. Two extraversión facets, optimism and expressiveness, broaden the scope of this factor and can be considered as indicators of positive emotionality. Although activity is a separate dimension in several temperament models, the related energy facet did not acquire the status of a separate factor in the item set derived from free parental descriptions. Creativity and curiosity are the defining facets of the imagination factor together with intellect. Finally, emotional stability turns out to be the smallest H/PIC component that is subdivided in an anxiety facet and one measuring self-confidence. A preliminary conceptual analysis of the content of the H/PIC scales (Mervielde and De Fruyt, 1999) shows that they cover a major part of

Variable and person-centred approaches

63

the variability captured by the adult FFM as specified by the NEO-P1R. A conceptual comparison with Achenbach’s Child Behaviour Checklist (C BC L) (Achenbach, 1991) suggests that the H/PIC facet scales cover most of the content specified in six of the eight clinical syndrome scales. Moreover, the H/PIC includes several facet scales that go beyond the indi­ vidual differences captured by the C BCL. Research is currently underway to provide an empirical basis for the conceptual content analyses, to establish the longitudinal stability of the structure and to document the convergent and discriminant validity of the domains and facets of this hierarchical model of individual differences in childhood. Although it would be premature to suggest that there is a consensus on the important variables and dimensions that are required to study indi­ vidual differences in childhood, it is fair to state that the search for communalities between temperament and personality models is a major step towards a more integrated conception of personality development across the life-span. The analysis of natural language descriptions of children’s personality as reviewed in this section of the chapter can be considered as a variant of the more traditional lexical approach as reviewed by Saucier, Hampson and Goldberg in Chapter 1 of this volume. Although the free description approach lacks the guarantees of the lexical approach for comprehen­ siveness, it may better represent the way individual differences are observed and conceived by the primary witnesses of children’s behaviour. Moreover, using this first hand knowledge as the basis for delineating the domain and the structure of variables revealed factors that are rather comparable to the traditional lexical factors and as such corroborate the convergent validity of the five-factor model. Person-centred approaches Current developments in the person-centred approach are focused on the replicability of the three person-types - resilients, overcontrollers and undercontrollers - across raters, methods for clustering profiles, cultures and languages and their longitudinal continuity and stability. Asendorpf and van Aken (1999) applied the Q-sort approach to person­ ality types to the large data set of the Munich Longitudinal Study on the Genesis of Individual Competencies (LO G IC ; Weinert and Schneider, 1999). They replicated Robins et al.'s (1996) and Hart et al.'s (1997) find­ ings that only three Q-factors were replicable for teacher-bascd Q-sorts at ages 4 6 years (aggregates across three yearly assessments) and parentbased Q-sorts at age 10 years. They also correlated the factor scores of the 54 items of the German CCQ between their Q-factors and those found by Robins et al. (1996), Hart et al. (1997), and as yet unpublished Dutch Q-factors. As Table 2.4 indicates, the consistency of these personality

64

Mervielde and Asendorpf

Table 2.4 Consistency of the Q-factors across different studies Germany Ages 4-6, teachers

Age 10, parents

Population

Judges

Resilient OverUndercontrolled ontrolled

UnderResilient Overcontrolled controlled

Dutch, ages 2-3 Dutch, ages 7-12 Iceland, age 7 US boys, age 10 US boys, age 10

Teachers Parents Experts Parents Parents

0.97

0.73

0.65

0.90

0.63

0.43

0.89

0.69

0.69

0.82 0.89a

0.66 0.65a

0.53 0.58*

Source: Adapted with permission from Asendorpf and van Aken (1999, Table I). Copyright (1999) by the American Psychological Association. Notes Reported are Pearson correlations between prototypic Q-sort profiles obtained for similar ages, a Consistency with German male factors.

prototypes was substantial across all studies despite their difference in culture, instrument (100-item versus 54-item Q-sort), judges, age and sex. Asendorpf and van Aken (1999) replicated Robins et al. ’s (1996) characterization of the three Q-types with regard to ego-control and egoresiliency (see Figure 2.2) and with regard to the Big Five. Contrary to Robins et al. (1996), they studied Big Five scores that were not based on the CCQ itself but on an independent assessment (parental ratings of 12 bipolar adjectives for each of the five factors). Moreover, they confirmed the quadratic relationship between ego-control and ego-resiliency for both ages 4-6 and 10 years (see Figure 2.1). Furthermore, they introduced a conceptual distinction between the continuity and stability of the types, and studied both properties between 4-6 and 10 years. They defined type continuity as the constancy of the Q-factor scores of the Q-sort items across age; thus continuity refers to the constancy of the personality prototypes. Continuity was high for all three factors (rs were 0.88, resiliency; 0.82, overcontrol; 0.78, undercon­ trol); these figures can be directly compared with the cross-study correlations reported in Table 2.4. In addition, the sizes of the types did not change significantly over age. In contrast, Asendorpf and van Aken (1999) understand type stability as the constancy of the membership of children for a particular type. Stability can be low even if continuity is high if many children switch between the types. In fact, stability was fairly low between 4-6 and 10 years of age, both for a categorical definition of type membership (yes/no; r = 0.30 for all three types) and for a continuous measure of type member­ ship (children’s Q-factor loading on a Q-factor; rs varied between 0.44

Variable and person-centred approaches

65

for the resilient type and 0.22 for the overcontrolled type). The lower stability relative to continuity suggests that there were changes in type membership that cannot be attributed to measurement error or a change in the nature of the type construct over time. Future studies on the reasons for such changes in type membership are needed. Asendorpf and van Aken (1999) studied long-term consequences of the three types that were identified at ages 4-6 years. With regard to cogni­ tive competence, they found a lower IQ and a lower than expected school grade for undercontrollers as compared with resilients up to age 12, repli­ cating Robins et al.'s (1996) findings. Interestingly, in line with Hart et al.’s (1997) results, undercontrollers did not report a lower cognitive self­ esteem than resilients. Furthermore, Asendorpf and van Aken (1999) found that overcontrollers’ IQ and cognitive self-esteem declined relatively to the other two types. The reasons for this decline are not clear, nor whether it may be specific to the LO GIC sample. With regard to social competence, Asendorpf and van Aken (1999 M.O.P.) found higher parental aggression ratings for undercontrollers as compared with resilients; higher parental inhibition ratings for overcon­ trollers; and both higher behavioural inhibition with strangers for overcontrollers and lower behavioural inhibition for undercontrollers. The last finding is noteworthy particularly because Kagan et al. (1987) oper­ ationalized behavioural inhibition in most of their studies by extremely high versus low inhibition, thus potentially confounding effects of high inhibition with effects of low inhibition. Finally, Asendorpf and van Aken (1999) tackled the thorny question of whether their Q-sort based types should be conceived as categories (as in lay theories of personality types) or rather as entities that are continuously distributed around a prototype. Note that in the case of three types, both possibilities can co-exist if the transitions from one type to the others are smooth, whereas the transitions between these two other types are discon­ tinuous. In fact, the distributions of children’s Q-factor loadings (i.e. their prototypicalities for the three types) showed such a pattern. When resilient children were disregarded, both the loadings on the overcontrolled factor and the loadings on the undercontrolled factor were bimodal, with clear peaks at both high and low scores, and the loadings for over- and under­ control were strongly negatively correlated. In contrast, the loadings for resiliency were fairly independent of the loadings for both over- and under­ control, and their distributions were not bimodal. Asendorpf and van Aken (1999) tended to interpret this finding as evidence for a mixed categorical/ continuous model where a continuous dimension of resiliency bifurcates at its lower part into two discrete types, overcontrol and undercontrol. However, they also acknowledged that this pattern may be due to a biased selection of the items in their 54-item Q-sort with regard to extremes in ego-control.

66

Mervielde and Asendorpf

In a more recent study, Asendorpf et al. (submitted) showed that the three personality types are also found for children when they are derived through replicated cluster analysis from parental Big Five ratings at age 12 years. Again, only three replicable types were found, not five as in the study by Caspi and Silva (1995). The reason for the discrepancy with the Caspi and Silva (1995) results may be that Asendorpf et al. (submitted) used a more strict method for evaluating cluster replicability. They found in various studies that it is not sufficient to rely on one random split of the sample because the replicability results vary greatly across different splits. Instead, stable results were only achieved if the median replicability for at least 10 different random splits was considered. Asendorpf et al. (submitted) also developed a new method of studying the consistency of types across different studies that are not exclusively based on the Q-factor approach. Basically, the children of one study are optimally assigned not only to the types that were derived from this study but also to the types that were derived from another study, and these two classifications of the same children are then compared for consistency through Cohen’s k. Asendorpf et al. (submitted) found for the LO GIC study that the Q-sort based types at age 10 showed a consistency of k = 0.52 with the Big Five based types that were derived through cluster analysis at age 12. For comparison, the replicability of the types was k = 0.60 for the Big Five based types. In addition, Asendorpf et al. (submitted) showed for two large samples of adult self-ratings of the Big Five that were based on different methods (NEO-FFI versus bipolar adjective ratings), that again only three types were replicable, and that these types were consistent with the childhood types (median k = 0.54 for the four comparisons between the two childhood studies and the two adulthood studies).Comparisons between the Big Five patterns of the types across the four studies did not reveal systematic differences due to age, type ofjudge or method of deriving types. Therefore, the average Big Five pattern of the three types across the four studies seems to be a useful overall description of the types (see Figure 2.3). Thus the person-centred approach to personality types results in three similar, replicable types both in childhood and adulthood, and indepen­ dent of the type of judge and method of deriving the types. Asendorpf et al. (submitted) could not replicate Asendorpf and van Aken’s (1999) mixed continuous/categorical model for the cluster-based types. This model seems to be specific to their short 54-item version of the CCQ, probably because of a biased item selection. Together, the studies by Asendorpf and van Aken (1999) and Asendorpf et al. (submitted) provide a solid starting point for a person-centred typology of children and adults. It should be noted, though, that this threetype model is not sufficiently differentiated. Person-centred assessments of types can only rival with variable-centred assessments if replicable

Variable and person-centred approaches

67

Undercontrolled

Resilient Overcontrolled

Figure 2.3 Three major personality prototypes characterized by their average Big Five patterns across four studies Source: Adapted from Asendorpf et al. (submitted. Figure I). Notes E = extraversión; N = neuroticism; O = openness; A = agreeableness; C = conscientiousness.

subtypes of the three types can be distinguished that show meaningful external social and cognitive correlates (see Robins, John and Caspi, 1998, for such an attempt that is based on the Q-sort approach). In that respect the three types of personality description resemble the Big Five factors: too broad for differentiated descriptions and specific pre­ dictions (see Paunonen, 1998), but with great potential for future differentiation.

CONCLUSION Current variable- and person-centred approaches to the study of child­ hood personality have their basic roots in the psychometric analysis of individual differences as pioneered by Cattell and commonly referred to

68

Mervielde and Asendorpf

as Q- and R-analysis. The variable-centred approach is concerned with the discovery of replicable broad categories of variables such as the Big Five or the major dimensions emerging from temperament research. The person-centred approaches share this concern for analysis at the highest level of the hierarchy but search for replicable broad clusters of profiles within persons. Both approaches adopt a top-down strategy, establishing a consensus on broad factors or types and decomposing them into smaller units. Representative sampling of a broad domain of variables has been the major concern for the lexical approach and the key factor for the growing consensus about the FFM . Representative sampling of variables is also an important step for establishing the generalizability of the resilient, over- and undercontrolled types as is documented in the review of the current developments in the person-centred approaches. The role of subject sampling is less well established but is expected to be more critical for a person-centred than for a variable-centred approach because the grouping of profiles clearly depends more on the composition of the sample than the structure of variables. The principal argument for adopting a person-centred approach has been formulated by Asendorpf et al. (submitted) as follows: A major task for personality psychology in the 21st century is, in our view, to take deviations from the traditional linear model seriously and exploit their inherent information about within-personality organ­ ization, its underlying genetic and environmental causes, and its personal and social consequences. The prototype approach to person­ ality differences makes this possible. Another advantage is that it facilitates communication with anybody interested in applications of personality description and prediction because differences between types are easier to communicate to clinicians and the general public than correlations with a trait. Therefore, researchers often use the type language to communicate correlational findings by presenting extreme group comparisons, however, with arbitrary cut-points along only one or two trait dimensions. The present prototype approach offers distinc­ tions that are less arbitrary because they are backed-up by the empirical analysis of many trait dimensions. The historical prominence and popularity of the variable-centred approach in research on childhood and adult personality is to some extent due to the lack of consensus about the domain of variables that should be studied. As long as there is little agreement on a basic taxonomy, new variables are invented and promoted as the fundamental dimensions because there is no agreed framework to check whether the new variables add specific variance or are merely clones with a different label. The

Variable and person-centred approaches

69

lexical approach to adult personality, the analysis of parental free descrip­ tions of children and the search for broad temperamental dimensions have contributed to a growing consensus about a common core of broad factors that can be studied across the life-span. The more difficult task of reaching a consensus on specific facets and their precise location in terms of multiple loadings is going to be the next test for the viability of the variable-centred approach. The broad dimensions emerging from the variable-centred approaches are the prime targets for the study of the genetic and physiological bases of individual differences. Without a solid consensus on how the domain of variables is organized, the search for reliable links between biological processes and behaviour is indeed premature. Recent developments in mapping the human genome provide new opportunities for linking specific genes with personality traits. New developments in behaviour genetic methods such as quantitative trait loci analysis are particularly relevant for the study of continuous variables such as personality traits. Moreover, according to Rutter et al. (1999), it has become apparent that genetic liability for many psychiatric disorders may lie in diagnostically non­ specific risk dimensions such as neuroticism. Variable- and person-centred approaches are both necessary for a complete account of individual differences. A consensus on the organi­ zation of the variables facilitates the search for replicable person-centred types. Decision makers are better served with a person-centred approach to individual differences because most of the real world decisions imply a choice among individuals. However, a scientific justification of the deci­ sion to hire a person or to advise a particular clinical treatment will always require a description of the person on the relevant variables. The review of the research on person-centred approaches illustrates that the basic types can be characterized in terms of the common dimensions emerging from the variable-centred approach. Therefore, future research may benefit from applying both variable-centred and person-centred approaches to the same data in order to compare the predictive validity of the two approaches and to assess their incremental validity.

NOTES I

The additional categories are: (V I) Independence; (V II) M ature for age; (VIM) Illness, Handicaps and H ealth; (IX ) Rhythm icity o f eating, sleeping, etc.; (X) G ender appropriate behaviour. Physical attractiveness; (X I) School perform ance, A ttitudes toward school; (X II) Contact com fort, Desire to be cuddlcd, Clinging; (X III) R elationships with siblings and parents; (X IV ) A m biguous phrases and descriptions that cannot be coded in other categories.

70

Mervielde and Asendorpf

REFERENCES Achenbach, T.A. (1991) Manualfo r the Child Behavior Checklist/4-18 and 1991 Profile. Burlington: University of Vermont, Department of Psychiatry. Achenbach, T.M. and Edelbrock, C.S. (1981) ‘Behavioral problems and compe­ tencies reported by parents of normal and disturbed children aged four through sixteen’, Monographs of the Society for Research in Child Development 46 (1, Serial no. 188). Ahadi, S.A. and Rothbart, M.K. (1994) ‘Temperament, development, and the Big Five’, in C.F. Halverson, G.A. Kohnstamm and R.P. Martin (eds) The Developing Structure of Temperament and Personality from Infancy to Adulthood, pp. 189-207. Hiilsdale, NJ: Erlbaum. Allport, G.W. (1937) Personality: A Psychological Interpretation. New York: Holt. Asendorpf, J.B. (1990) ‘Development of inhibition during childhood: Evidence for situational specificity and a two-factor model’, Developmental Psychology 26: 721-730. Asendorpf, J.B. (1991) ‘Development of inhibited children’s coping with un­ familiarity’, Child Development, 62: 1460-1474. Asendorpf, J.B. (1993) ‘Beyond temperament: A two-factorial coping model of the development of inhibition during childhood’, in K.H. Rubin and J.B. Asendorpf (eds) Social Withdrawal, Inhibition, and Shyness in Childhood, pp. 265-289. Hillsdale, NJ: Erlbaum. Asendorpf, J.B., Borkenau, P., Ostendorf, F. and van Aken, M A G. (submitted) ‘Carving personality description at its joints: Confirmation of three replicable personality prototypes for both children and adults.’ Asendorpf, J.B. and van Aken, M.A.G. (1991) ‘Correlates of the temporal consis­ tency of personality patterns in childhood’, Journal of Personality 59: 689-703. Asendorpf, J.B. and van Aken, M.A.G. (1994) ‘Traits and relationship status: Stranger versus peer group inhibition and test intelligence versus peer group competence as early predictors of later self-esteem’, Child Development 59: 688-703. Asendorpf, J.B. and van Aken, M.A.G. (1995) ‘Social competence’, in F. Weinert and W. Schneider (eds) The Munich Longitudinal Study on the Genesis of Individual Competencies (LO GIC). Report No. 12: Assessment Procedures and Results of Wave 9 (Tech. Rep.) Munich, Germany: Max Planck Institute for Psychological Research. Asendorpf, J.B. and van Aken, M.A.G. (1999) ‘Resilient, overcontrolled, and under­ controlled personality types in childhood: Replicability, predictive power, and the trait-type issue’, Journal ofPersonality and Social Psychology, 77: 815-832. Block, J. (1961) The Q-sort Method in Personality Assessment and Psychiatric Research. Springfield, IL: Charles C. Thomas. Block, J. (1971) Lives through Time. Berkeley, CA: Bancroft Books. Block, J. (1995) ‘A contrarian view of the five-factor approach to personality description’, Psychological Bulletin 117: 187-215. Block, J.H. and Block, J. (1980) ‘The role of ego-control and ego-resiliency in the organization of behavior’, in W.A. Collins (ed.) Minnesota Symposium on Child Psychology, vol. 13, pp. 39-101. Hillsdale, NJ: Erlbaum.

Variable and person-centred approaches

71

Boer, F. and Westenberg, P.M. (1994) ‘The factor structure of the Buss and Plomin EAS Temperament Survey (Parental ratings) in a Dutch sample of elementary school children’, Journal of Personality’ Assessment 62: 537-551. Buss, A.H. and Plomin, R. (1975) A Temperament Theory o f Personality Development. New York: Wiley. Buss, A.H. and Plomin, R. (1984) Temperament: Early Developing Personality Traits. Hillsdale, NJ: Erlbaum. Buyst, V., De Fruyt, F. and Mervielde, I. (1994) ‘Parental descriptions of chil­ dren’s personality: A five-factor model classification’, Psychologica Belgica 34: 231-255. Caspi, A. (1998) ‘Personality development across the life course’, in W. Damon (ed.) Handbook of Child Psychology, 5th edn (N. Eisenberg vol. ed.) Social, Emotional, and Personality Development, vol. 3, pp. 311-388. New York: Wiley. Caspi, A., Moffitt, T.E., Newman, D.L. and Silva, P.A. (1996) ‘Behavioral obser­ vations at age 3 years predict adult psychiatric disorders’, Archives of General Psychiatry 53: 1033-1039. Caspi, A. and Silva, P.A. (1995) ‘Temperamental qualities at age three predict personality traits in young adulthood: Longitudinal evidence from a birth cohort’, Child Development 66: 486-498. Cattell. R.B. (1946) The Description and Measurement o f Personality’. Yonkers, NY: World Book. Cheek, J.M. and Buss, A.H. (1981) ‘Shyness and sociability’, Journal of Personality and Social Psychology’, 41: 330-339. Chess, S. and Thomas, A. (1996) Temperament: Theory and Practice. New York: Brunner/Mazel. Church, A.T., Katigbak, M.S. and Castaneda, I. (1988) ‘The effects of language of data collection on derived conceptions of healthy-personality with Filipino bilinguals’, Journal of Cross-Cultural Psychology>19: 178-192. Cloninger, C.R. (1987) ‘A systematic method for clinical description and classi­ fication of personality traits’, Archives of General Psychiatry’ 44: 573-588. Costa, P.T. Jr and McCrae, R.R. (1985) The NEO Personality Inventory Manual. Odessa, FL: Psychological Assessment Resources. Costa, P.T. Jr and McCrae, R.R. (1992) Revised NEO Personality Inventory (NEO PI-R) and NEO Five-Factor Inventory (NEO -FFI). Odessa, FL: Psychological Assessment Resources. Costa, P.T. Jr and Widiger, T.A. (eds) (1994) Personality Disorders and the FiveFactor Model of Personality. Washington, DC: American Psychological Association. De Fruyt, F. and Mervielde, 1. (1992) ‘The Big Five in children’s sociometric judgements’, International Journal of Psychology 27: 353-354. Derryberry, D. and Rothbart, M.K. (1984) ‘Emotion, attention, and temperament’, in C.E. Izard, J. Kagan and R.B. Zajonc (eds) Emotions, Cognition and Behavior, pp. 132-166. Cambridge: Cambridge University Press. Derryberry, D. and Rothbart, M.K. (1997) ‘Reactive and effortful processes in the organization of temperament’. Developmental Psychopathology 9: 633-652. Digman, J.M. (1990) ‘Personality structure: Emergence of the five-factor model’, Annual Review of Psychology’ 41: 417-440.

72

Mervielde and Asendorpf

Digman, J.M. (1994) ‘Child personality and temperament: Does the five-factor model embrace both domains?’, in C.F. Halverson, Jr, G.A. Kohnstamm, and R.P. Martin (eds) The Developing Structure of Temperament and Personality from Infancy to Adulthood, pp. 323-338. Hillsdale, NJ: Erlbaum. Digman, J.M. (1996) ‘The curious history of the five-factor model’, in J.S. Wiggins (ed.) The Five-Factor Model of Personality, pp. 1-20. New York: Guilford Press. Digman, J.M. and Inouye, J. (1986) ‘Further specification of the five robust factors of personality’, Journal of Personality and Social Psychology 50: 116-123. Digman, J.M. and Shmelyov, A.G. (1996) ‘The structure of temperament and personality in Russian children’, Journal of Personality and Social Psychology 71:341-351. Digman, J.M. and Takemoto-Chock, K. (1981) ‘Factors in the natural language of personality: Re-analysis, comparison, and interpretation of six major studies’, Multivariate Behavioral Research 16: 149-170. Dreger, R.M. (1995) ‘Testing the five-factor model of personality in preschool children’, Journal of Social Behavior and Personality 10: 51-73. Eysenck, H.J. (1992) ‘Four ways five factors are not basic’, Personality and Individual Differences 13: 667-673. Goldberg, L.R. (1992) ‘The development of markers of the Big-Five factor struc­ ture’, Psychological Assessment 4: 26-42. Goldberg, L.R. (1994) ‘How not to whip a straw dog’, Psychological Inquiry 5: 128-130. Goldsmith, H.H. (1996) ‘Studying temperament via construction of the Toddler Behavior Assessment Questionnaire’, Child Development 67: 218-235. Goldsmith, H.H., Buss, K.A. and Lemery, K.S. (1997) ‘Toddler and childhood temperament: Expanded content, stronger genetic evidence, new evidence for the importance of environment’, Developmental Psychology 33: 891-905. Goldsmith, H.H., Buss, A.H., Plomin, R., Rothbart, M.K., Thomas, A., Chess, S., Hinde, R.A. and McCall, R.B. (1987) ‘Roundtable: What is temperament? Four approaches’. Child Development 58: 505-529. Goldsmith, H.H. and Campos, J.J. (1982) ‘Toward a theory of infant tempera­ ment’, in R.N. Emde and R.J. Harmon (eds) The Development of Attachment and Affiliative Systems, pp. 161-193. New York: Plenum. Göttert, R. and Asendorpf, J. (1989) ‘Eine deutsche Version des Califomia-ChildQ-Sort, Kurzform’ [‘A German short version of the California Child Q-Set’], Zeitschrift fur Entwicklungspsychologie und Pädagogische Psychologie 21: 70-82. Gray, J.A. (1987) The Psychology of Fear and Stress, 2nd edn. New York: McGraw-Hill. Graziano, W.G. and Eisenberg, N.H. (1997) ‘Agreeableness: A dimension of personality’, in R. Hogan, J. Johnson and S. Briggs (eds) Handbook of Personality Psychology, pp. 795-824. New York: Academic Press. Hart, D., Hofmann, V., Edelstein, W. and Keller, M. (1997) ‘The relation of childhood personality types to adolescent behavior and development: A longitudinal study of Icelandic children’, Developmental Psychology 33: 195-205.

Variable and person-centred approaches

73

Huey, S.J., Jr. and Weisz, J.R. (1997) 'Ego control, ego resiliency, and the fivefactor model as predictors of behavioral and emotional problems in clinic-referred children and adolescents’, Journal o fAbnormal Psychology 106: 404-415. John, O.P. (1990) ‘The “ Big Five” factor taxonomy: Dimensions of personality in the natural language and in questionnaires’, in L.A. Pervin (ed.) Handbook o f Personality: Theory>and Research, pp. 66-100. New York: Guilford Press. John, O.P., Caspi, A., Robins, R.W., Moffit, T.E. and Stouthamer-Loeber, M. (1994) ‘The “ Little Five” : exploring the nomological network of the five-factor model of personality in adolescent boys’. Child Development 65: 160-178. Kagan, J., Reznick, S. and Snidman, N. (1987) ‘The physiology and psychology of behavioral inhibition in children’, Child Development 58: 1459-1473. Kagan, J., Reznick, S. and Snidman, N. (1988) ‘Biological bases of childhood shyness’, Science 240: 167-173. Kohnstamm, G.A., Halverson, C.F., Jr, Mervielde, I. and Havill, V. (eds) (1998) Parental Descriptions o f Child Personality: Developmental Antecedents of the Big Five? Mahwah, N J: Erlbaum. Kohnstamm, G.A., Mervielde, I., Besevegis, E. and Halverson C.F. (1995) ‘Tracing the Big Five in parents’ free descriptions of their children’, European Journal o f Personality 9: 283-304. Lemery, K.S., Goldsmith, H.H., Klinnert, M.D. and Mrazek, D.A. (1999) ‘Developmental models of infant and childhood temperament’, Developmental Psychology' 35: 189-204. McAdams. D.P. (1992) ‘The five-factor model in personality: A critical appraisal’, Journal of Personality 60: 329-361. McCrae, R.R. and Costa, P.T., Jr (1987) ‘Validation of the five-factor model of personality across instruments and observers’, Journal o f Personality and Social Psychology 52: 81-90. McCrae, R.R. and Costa, P.T., Jr (1996) ‘Toward a new generation of personality theories: Theoretical contexts for the five-factor model’, in J.S. Wiggins (ed.) The Five-Factor Model o f Personality, pp. 51-87. New York: Guilford Press. McCrae, R.R. and Costa, P.T., Jr (1997) ‘Personality trait structure as a human universal’, American Psychologist 52: 509-516. McCrae, R.R., Costa, P.T., Jr, de Lima, M.P., Simoes, A., Ostendorf, F., Angleitner, A., Marusic, I., Bratko, D., Caprara, G.V., Barbaranelli, C., Chae, J.H. and Piedmont, R.L. (1999) ‘Age differences in personality across the adult life span: Parallels in five cultures’, Developmental Psychology 35: 466-477. MacDonald, K. (1992) ‘Warmth as a developmental construct: An evolutionary analysis’, Child Development 63: 753-773. Martin, R.P. (1988) The Temperament Assessment Battery’fo r Children, Brandon, VT: Clinical Psychology Publishing. Martin, R.P., Wisenbaker, J. and Huttunen, M. (1994) ‘Review of factor analytic studies of temperament measures based on the Thomas-Chess structural model: Implications for the Big Five’, in C.F. Halverson, Jr, G.A. Kohnstamm and R.P. Martin (eds) The Developing Structure o f Temperament and Personality from Infancy to Adulthood, pp. 157 172. Hillsdale, NJ: Erlbaum. Mathiesen, K.S. and Tambs, K. (1999) ‘The EA S temperament questionnaire Factor structure, age trends, reliability and stability in a Norwegian sample’, Journal o f Child Psychology and Psychiatry 40: 431439.

74

Mervielde and Asendorpf

Mervielde, I. (1998) ‘Validity of results obtained by analyzing free personality descriptions’, in G.A. Kohnstamm, C.F. Halverson, I. Mervielde and V. Havill (eds) Parental Descriptions of Child Personality: Developmental Antecedents of the Big Five?, pp. 189-203. Mahwah, NJ: Erlbaum. Mervielde, I., Buyst, V. and De Fruyt, F. (1995) ‘The validity of the Big-Five as a model for teachers’ ratings of individual differences among children aged 4-12 years’, Personality and Individual Differences 18: 1827-1836. Mervielde, I. and De Fruyt, F. (1999) ‘Construction of the Hierarchical Personality Inventory for Children (H/PIC)’, in I. Mervielde, I. Deary, F. De Fruyt and F. Ostendorf (eds) Personality Psychology in Europe, vol. 7, pp. 107-127. Tilburg, The Netherlands: Tilburg University Press. Mervielde, I. and De Fruyt, F. (in press) ‘The Big Five personality factors as a model for the structure of children’s peer nominations’, European Journal of Personality. Newman, D.L., Caspi, A., Moffitt, T.E. and Silva, P.A. (1997) ‘Antecedents of adult interpersonal functioning: Effects of individual differences in age 3 temperament’, Developmental Psychology 33: 206-217. Ostendorf, F. and Angleitner, A. (1994) ‘The Five-Factor Taxonomy: Robust dimensions of personality description’, Psychologica Belgica 34: 175-194. Ozer, D.J. and Gjerde, P.F. (1989) ‘Patterns of personality consistency and change from childhood through adolescence’, Journal of Personality 57: 483-507. Panksepp, J. (1986) ‘The psychobiology of prosocial behaviors: Separation distress, play, and altruism’, in C. Zahn-Waxler, E.M. Cummings, and R. Iannotti (eds) Altruism and Aggression: Biological and Social Origins, pp. 19-57. Cambridge: Cambridge University Press. Paunonen, S.V. (1998) ‘Hierarchical organization of personality and prediction of behavior’. Journal of Personality and Social Psychology 74: 538-556. Posner, M.l. and Rothbart, M.K. (1992) ‘Attentional mechanism and conscious experience’, in D. Milner and M. Rugg (eds) The Neuropsychology of Consciousness, pp. 91-111. San Diego, CA: Academic Press. Presley, R. and Martin, R.P. (1994) ‘Toward a structure of preschool tempera­ ment: Factor structure of the Temperament Assessment Battery for Children’, Journal of Personality 62: 415-448. Robins, R.W., John, O.P. and Caspi, A. (1998) ‘The typological approach to studying personality’, in R.B. Cairns, L.R. Bergman and J. Kagan (eds) Methods and Modelsfo r Studying The Individual, pp. 135-160. Thousand Oaks, CA: Sage. Robins, R.W., John, O.P., Caspi, A., Moffitt, T.E. and Stouthamer-Loeber, M. (1996) ‘Resilient, overcontrolled and undercontrolled boys: Three replic­ able personality types’, Journal o f Personality and Social Psychology 70: 157-171. Rothbart, M.K. and Ahadi, S.A. (1994) ‘Temperament and the development of personality’. Journal of Abnormal Psychology 103: 55-66. Rothbart, M.K., Ahadi, S.A. and Hershey, K.L. (1994) ‘Temperament and social behavior in childhood’, Merril-Palmer Quarterly 40: 21-39. Rothbart, M.K. and Bates, J.E. (1998) ‘Temperament’, in W. Damon and N. Eisenbcrg (cds) Handbook o f Child Psychology: Vol. 3. Social Emotional and Personality Development, 5th edn, pp. 105-176. New York: Wiley.

Variable and person-centred approaches

75

Rothbart, M.K. and Derryberry, D. (1981) ‘D evelopm ent o f individual differences in tem peram ent’, in M.E. Lam b, A.L. Brow n (eds) Advances in Developmental Psychology, vol. 1, pp. 37-86. H illsdale, N J:Erlbaum . Rothbart, M .K., Derryberry, D. and Posner, M .l. (1994) ‘A psychobiological approach to the developm ent o f tem peram ent’, in J.E. Bates and T.D. W achs (eds) Temperament: Individual Differences at the Interface óf Biology and Behavior, pp. 83-116. W ashington, DC: Am erican Psychological Association. Rutter, M., Silberg, J., O ’Conner, T. and Sim onoff, E. (1999) ‘G enetics and child psychiatry: I A dvances in quantitative and m olecular genetics’, Journal of Child Psychology and Psychiatry 40: 3 -18. Schölte, R.H.J., van Aken, M.A.G. and Van Lieshout, C.F.M . (1997) ‘A dolescent personality factors in self-ratings and peer nom inations and their prediction o f peer acceptance and peer rejection’. Journal ofPersonality Assessment 69: 534-554. Shiner, R.L. (1998) ‘How' shall we speak o f children’s personality in middle child­ hood? A prelim inary taxonom y’, Psychological Bulletin 124: 308-332. Slotboom , A.M ., Havill, V.L., Pavlopoulos, V. and De Fruyt, F. (1998) ‘D evelop­ mental changes in personality descriptions o f children: A cross-national com parison o f parental descriptions o f children’, in G.A. K ohnstam m , C.F. Halverson, I. M ervielde and V. Havill (eds) Parental Descriptions of Child Personality: Developmental Antecedents of the Big Five?, pp. 127-153. M ahwah, NJ: Erlbaum. Spinath, F.M. and Angleitner, A. (1998) ‘Contrast effects in Buss and P lom in’s EAS questionnaire. A behavioral-genetic study on early developing personality traits assessed through parental ratings’, Personality and Individual Differences 25: 947-963. Stem , W. (1911) Die Differentielle Psychologie in ihren methodischen Grundlagen [.Methodological foundations of differential psychology], Leipzig, Germ any: Johann A m brosius Barth. Strelau, J. (1996) ‘The regulative theory o f tem peram ent: C urrent status’, Personality and Individual Differences 20: 131-142. Strelau, J. (1998) Temperament: A Psychological Perspective. N ew York: Plenum. Strelau, J. and A ngleitner, A. (eds) (1991) Explorations in Temperament. New' York: Plenum Press. Thom as, A. and Chess, S. (1977) Temperament and Development. New York: Brunner/M azel. van Aken, M .A.G. and Asendorpf, J.B. (1999) ‘A person-centered approach to developm ent: The tem poral consistency o f personality and self-concept’, in F.E. W einert and W. Schneider (eds) Individual Developmentfrom 3 to 12: Findings from a Longitudinal Study, pp. 301-319. Cam bridge: C am bridge University Press. van Aken, M .A.G. and Van Lieshout, C.F.M . (1991) ‘C hildren’s com petence and the agreem ent and stability o f self- and child descriptions’. International Journal of Behavioral Development 14: 83-99. Van Lieshout, C.F.M . and H aselager, G.J.T. (1994) ‘The Big Five personality factors in Q-sort descriptions o f children and adolescents’, in C.F., Halverson, Jr, G.A. K ohnstam m and R.P. M artin (eds) The Developing Structure of Temperament and Personality from Infancy to Adulthood, pp. 293-318. Hillsdale, NJ: Lawrence Erlbaum Associates.

76

Mervielde and Asendorpf

Victor, J.B. (1994) ‘The five-factor model applied to individual differences in school behavior’, in C.F. Halverson, Jr, G.A. Kohnstamm and R.P. Martin (eds) The Developing Structure o f Temperament and Personality from Infancy to Adulthood, pp. 339-354. Hillsdale, N J: Erlbaum. Weinert, F.E. and Schneider, W. (eds) (1999) Individual Development from 3 to 12: Findings from a Longitudinal Study. Cambridge: Cambridge University Press.

Chapter 3

Personality development in adulthood: the broad picture and processes in one longitudinal sample Ravenna H elson and Virginia S.Y. Kw an

The authors suggest that personality is a relatively enduring organization of motives and resources that helps the individual adapt to changes over the life-span and that undergoes modifications itself in the process. Supporting this position are the find­ ings of recent studies of rank-order consistency and of normative change on personality inventories in both cross-sectional and longitudinal samples. Rank-order consistency does not reach a peak until late middle age and is never complete. Cross-sectional studies using a variety of inventories in a variety of cultural settings suggest that people increase in norm-orientation (control) with age and decrease in intensity of social involvement. Results show no consistent relation to age for social assurance and complexity; some relationships are curvilinear. Three longitu­ dinal samples representing different birth cohorts show findings similar to those of the cross-sectional studies. However, the longitudinal studies show that different samples change on different aspects of norm-orientation and social involvement, in different amounts, and at different times, consistent with the idea that each sample faced a distinctive course of adaptational challenges. Findings from the Mills Longitudinal Study test hypotheses about the relation of gender, life style, and social influences such as the women’s movement to personality change. They demonstrate generality across samples in some studies and explore cohort differences in others. Two kinds of personality change involved in adult development are described, and studies of how personality changes in terms of self and identity processes are illustrated.

IN TR O D U C TIO N Personality trait psychologists have characteristically used techniques and strategies to show the consistency and predictive power o f personality rather than how it changes. They have tended to assume that biological factors or early childhood experience underlie a continuity that is an essen­ tial aspect o f the construct o f personality. Particularly when confronted with a challenge to the very idea that personality predicts behaviour (e.g. Mischel, 1968), they have regarded evidence o f change in personality as indication o f error or failure. O f course, there have been exceptions. For example, Eysenck (1967) studied the biological basis o f personality, but also personality differences associated with age (Eysenck and Eysenck,

78

Helson and Kwan

1969). Block (1971) distinguished several kinds of personality change and demonstrated individual differences in patterns of change from childhood to early middle age, but he and Jeanne Block (Block and Block, 1980) also conceptualized and measured the master traits of ego-control and ego-resilience. Conley (1985a) demonstrated the consistency of traits but also (1985b) took evidence of change seriously and tried to clarify what aspects of personality changed most and least. We believe that the stability model of adult personality is giving way to a conception in which personality is relatively enduring but is also expected to change. One factor promoting this trend is the cognitive revo­ lution with its attention to cognitive and motivational middle-level variables (e.g. Cantor and Zirkel, 1990; Pervin, 1989) and a view of the individual as engaged in efforts to succeed in self-chosen goals. Baltes (1987, 1997) has applied such a cognitive and motivational perspective to a general conceptualization of adaptation over the life-span. According to Baltes (1997), adaptation is required to maximize benefits and minimize losses. Young people have the advantage of benefits gained from evolutionary selection, but those past the childbearing age do not. A major loss for all ageing adults, for example, is a decline in fluid intelligence. In the interests of adaptation there is an increasing need or demand for culture over the life-span, though losses bring decreasing efficacy in the use of culture. To offer our own example of what Baltes has in mind, meeting new people brings information and opportunity to young adults, but older people, though more dependent than the young on the medical care or technology for temperature control that their culture provides, may find new contacts more strenuous and less informative, and any opportunities harder to seize. We would add that older people may thus become less extraverted. Baltes is not a personality psychologist, but if personality is regarded as playing an adaptive and proactive role in the individual’s development, he and his colleagues (e.g. Baltes and Baltes, 1990; Staudinger, Marsiske and Baltes, 1995) offer a very broad perspective in which to expect both normative and individual personality change with age. Personality can be regarded as a relatively enduring organization of motives and resources that helps the individual to adapt to life changes and itself changes in the process (Helson, 1993; Helson, Mitchell and Moane, 1984). A second factor inclining personality psychologists toward acceptance of the idea of personality change in adulthood is the increasing amount of empirical evidence that it takes place. The recent interest in cultural psychology has led to the demonstration of similarities across cultures, including the relation of personality inventory data to age. To date these studies have been cross-sectional. However, there is also an increasing amount of personality inventory data from longitudinal studies. We value other kinds of measures of personality, but in this

Personality development in adulthood

79

chapter we accept inventories as useful tools and focus on what these recent data show. We begin with a selective review of the recent literature on normative personality consistency and change in adulthood, as assessed with inven­ tories. This literature is largely empirical. Then, still emphasizing personality inventories as tools, we turn to the question of why change takes place. We use work from the Mills Longitudinal Study to illus­ trate how social and psychological factors affect personality change and development. Researchers have become increasingly aware that change has many meanings and needs to be measured in multiple ways. In particular, the consistency of personality tells nothing about mean-level change. We begin with the rank-order stability of personality over time, then turn to meanlevel change in cross-sectional and longitudinal studies.

RANK-ORDER CONSISTENCY OF PERSONALITY OVER THE LIFE-SPAN The Jesuits and early psychoanalysts thought basic beliefs or personality structures were in place by age 5. Others (e.g. Bloom, 1964) have put the age at the end of adolescence, after hormonal and intellectual changes are integrated and identity begins to take shape; still others put it at age 30, after a person settles into the grooves of adulthood (James, 1890; McCrae and Costa, 1990). One way of studying the question is to ask when the rank-ordering of individuals on personality characteristics becomes highly stable. Bloom (1964) and Schuerger, Zarrella and Hotz (1989) addressed this question, but Roberts and Friend-DelVecchio (in press) include find­ ings from the last decade and cover most of the life-span. They identified 152 relevant studies based on 124 samples. The average study covered an interval of 6.7 years. Their meta-analysis showed that rank-order consis­ tency increased from about 0.31 in childhood to 0.54 during the college years to 0.64 at age 30, and then reached a plateau around 0.74 between the ages of 50 and 70. The type of trait measured (e.g. extraversión, open­ ness, etc.) had little effect on consistency. Thus, personality does not become highly stable until middle age or later. Why should personality become more consistent with age? Some have suggested that young adults face challenging new experiences which affect their developing self-concept and goals, but that there are fewer such experiences and a firmer self-concept as personality matures (Glenn, 1980). According to Baltes (1997), the ability to utilize ‘culture’ decreases in the later years, so that psychological and social factors have less influence, and resources are spent less in growth than in maintenance or management of loss.

80

Helson and Kwan

MEAN-LEVEL CHANGE WITH AGE In the language of personality inventories and trait psychology, is there mean-level change with age? If so, what is the nature of change and when does it occur? Even whether it occurs is a difficult question, due not only to problems in statistical assumptions in the measurement of change but also to the confounds of age, cohort, and time of testing pointed out by Schaie (1965). Though a common conclusion has been that personality was primarily stable, there were suggestions that social extraversión declined in older people (Neugarten, 1977) and that along with this trend impulse control increased, with mixed evidence for neuroticism (Conley, 1985b).

CROSS-SECTIONAL STUDIES OF PERSONALITY DIFFERENCES WITH AGE Ideas and conflicting evidence Among trait psychologists, Costa and McCrae (1980, 1997; McCrae and Costa, 1990) have published the most about personality in adulthood and articulated the most influential position. They have emphasized the stability of adult personality, based on findings in both cross-sectional and longi­ tudinal work. Here we treat their ideas in relation to some of their cross-sectional studies. A particularly impressive study (Costa and McCrae, 1986) showed very little relation between age and scores on short forms of scales to assess extraversión, openness, and neuroticism in a sample of 10,000 adults in the US between the ages of 35 and 85. The authors graphed lines that were dramatically horizontal for white males, white females, black males, and black females. However, Costa and McCrae had other data showing that college students differed in personality from adults. In terms of their own five-factor model of personality, adults had higher scores on conscientiousness and agreeableness, lower scores on extraversión, openness and neuroticism. On the basis of these and other data Costa and McCrae adopted the view that personality may change during the 20s. After age 30, they said, any change in adult personality is negligible. What may appear to be change is usually attributable to cohort differences, time of testing or unreliability (Costa and McCrae, 1997; McCrae and Costa, 1990). However, Johnson et al. (1983) reported relationships between age and Adjective Check List factor scores in a Hawaiian sample including 3402 Caucasians, Chinese and Japanese, each group with different cohort expe­ rience. From adolescence to ‘over 51’ both males and females showed more socially desirable characteristics (kind, warm, pleasant), increased

Personality development in adulthood

81

in ego-organization (industrious, thorough, efficient), became less abra­ sive and neurotic, decreased in perceived social attractiveness, and became somewhat more introverted and intraceptive. Note that the themes of change were generally consistent with those Costa and McCrae had artic­ ulated, but change was not confined to the teens and 20s. The investigators obtained similar patterns of correlations when they conducted analyses separately for the parents and offspring who composed the sample. In recent work, McCrae and collaborators (1999) themselves reported differences with age on the Revised NEO Personality Inventory (NEOPI-R; Costa and McCrae, 1992) in five countries. Their samples were casually constituted or intended for other purposes, but where findings agree, this informality of procedure can be overlooked. The hypotheses about the nature of age differences were generally supported across samples. However, with respect to the point at which differences between age groups occurred, the results suggest that change extends throughout the period of life represented. For example, in the case of Agreeableness (Figure 3.1), the prediction that scores would increase with age was supported across samples, but except in the Croatian sample (which consisted of high school students and their parents) there were substan­ tial mean changes in the later years.

AG REEABLEN ESS

■ 18-21 I *22-29 □ 30-49 E350+

60b = 0.87cd

i> = 1.45d

b= 1.81b

b = 2.66abb = 1.70

co

+

German

Italian

Portuguese

Croatian

Korean

Figure 3.1 Differences with age on Agreeableness in five countries Source: McCrae et Association.

ol.

(1999. p. 473). Copyright by the American Psychological

82

Helson and Kwan

Studies using the California Psychological Inventory W e have described several cross-sectional studies that illustrate discrepant outcomes in large, multiethnic samples. Now we turn to three additional multiethnic studies that all used the California Psychological Inventory (C PI; Gough and Bradley, 1996). Our reasons for this focus are that these studies are recent; findings are clearer when comparisons are made in terms of the same instrument; and findings from the cross-sectional studies will provide a basis for hypotheses addressing findings from three longi­ tudinal studies that also used the CPI, to be presented in the next section. Furthermore, the conceptualization of CPI scales in terms of dispositions of social importance makes them useful for studying personality change in a social context, which is a major focus of the M ills Longitudinal Study. Findings from the M ills Study, some of which were based on the CPI, will be presented in the last section of the chapter. C P I scales and groupings

C PI scales do not measure traits abstracted through factor analyses but dispositions selected for their importance in social life (Gough and Bradley, 1996). There are three main categories of scales. One consists of measures of regulation or control of behaviour, described as norm-orientation. The most basic is the Socialization scale, which measures attitudes of trust, role-taking ability, respect for rules, and considerateness of others. The Self-control scale measures the tendency to control one’s temper and sexual impulses. Others in this category include Responsibility, Good impres­ sion, Well-being, Communality, and Achievement via conformance. For the purposes of this chapter the Flexibility scale is also included, because high-scorers may be flighty and low-scorers show a rigid constriction of behaviour. In Big Five terms, these scales are correlated with agreeable­ ness, conscientiousness and (low) neuroticism. We expect them to be positively related to age. A second set of scales assesses various patterns of interest and skill in social interaction. These scales can be considered as measures of extra­ versión. In this chapter we will subdivide these into measures of assurance and measures of vitality. We will expect change (decrease with age) only on the latter. Similar distinctions within the domain of extraversión are found on the Hogan Personality Inventory (Hogan and Hogan, 1995) and the NEO-R-PI. On the latter, the facets of assertiveness and activity are not as much related to age as excitement-seeking and positive emotion (McCrae et al., 1999). C PI measures of assurance include scales that emphasize social competence. For example, the Dominance scale assesses initiative, confidence, and leadership, or task-orientation. Other scales in

Personality development in adulthood

83

this set are Independence and Capacity for status. Measures of vitality are those that emphasize interest and affective involvement in social interac­ tion. The Social presence scale measures spontaneity, wit and cleverness; the Empathy scale measures interest and resourcefulness in understanding others; the Self-acceptance scale assesses liking for oneself, including one’s physical characteristics; and the Sociability scale taps interest and energy in social life, including social organizations. A third set of scales assess cognitive breadth, interests and skills. Correlations between these scales and NEO-R-PI Openness are modest (Gough and Bradley, 1996), and we will refer to them as measures of complexity. The high-scorer on Tolerance has respect for the fact that people do not see the world in the same way, the high-scorer on Psychological mindedness likes to understand why people behave in the ways they do, and the high-scorers on Achievement via independence like to pursue complex or long-range goals of their own choosing in their own way. The idea that at least some people become more differentiated and better integrated over time is important in many conceptions of adult devel­ opment (e.g. Loevinger, 1976; Vaillant, 1977). However, people in general may not increase in these respects, and the relation to age may be curvi­ linear (Labouvie-Vief et al., 1995). Almost all of the CPI scales fit at least fairly well into one of these three categories (see Table 3.1). Table 3.1 omits only Femininity/ Masculinity, Intellectual efficiency, and two of the scales measuring normorientation from this overly large set (Well-being and Communality). Sam p les

In brief, one of the studies to be examined utilizes a very large British sample and two studies each use one American and one Chinese sample. Further detail is as follows. First, a sample representative of the residents of Great Britain was selected by a professional polling agency for studies being conducted by Robert McHenry and his colleagues at the Oxford Psychologists Press. Data from 855 men and 1139 women between the ages of 16 and 87 were gathered in 1997. Second, in a study by Yang, McCrae and Costa (1998), the US sample consisted of participants in the Baltimore Longitudinal Study of Aging or adults recruited as peer raters. There were 153 men and 195 women ranging in age from 19 (women) or 27 (men) to 92. The Chinese sample was recruited in 14 cities of the People’s Republic of China (PRC ) by word of mouth, with an effort to obtain a representative sample of men and women between ages 18 and 67. Data were collected from 1159 men and 934 women between 1989 and 1991. Third, in a study by Labouvie-Vief et al. (in press), the US sample consisted of Detroit participants in an ongoing longitudinal study, selected to be representative of community-based adults in three suburban

84

Helson and Kwan

communities between ages 20-70 and over. There were 134 men and 151 women. Data were collected in 1994. Their Chinese sample was recruited from nine communities in the Beijing area, 219 men and 231 women between the ages of 20 and 85. Findings across studies

On the left in Table 3.1 are CPI measures classified into four categories: Norm-orientation, Social vitality, Social assurance and Complexity. The first two columns show for each C PI scale the average correlation with age for men and for women across the studies. Subsequent columns show findings for men and women residents of Great Britain, then for the Baltimore and PRC men and women of the Yang et al. study, and then the Detroit and Beijing men and women from the Labouvie-Vief et al. study. Because the British and PRC samples are large, the significance level for them is 0.001, and p < 0.05 for the three smaller samples. The average correlations shown in the first two columns are moderate to low, as would be expected if personality shows considerable consis­ tency over time. Note that correlations for men and women are generally quite similar. For the measures of Norm-orientation, 8 of the 10 groups showed signif­ icant increases with age on all of the scales. Baltimore women and Detroit men show generally lower correlations, but the relation to age is signifi­ cant for them on four of the six scales. Overall, this is a high level of consistency. Older people in these studies present themselves as more selfdisciplined, cooperative and conventional, and less open to change. On measures of Social vitality, Detroit men show no significant corre­ lations with age. However, the 9 other groups show significant negative correlations on at least three of the four scales. Older people are less sparkling in social life, less involved in the emotional understanding of others, and find themselves less attractive or interesting. For the measures of Social assurance, eight correlations are significant and all these are negative. However, they are predominantly low and the other 22 of the 30 correlations (10 groups, three scales) are insignificant. Thus, social assurance does not change with age across these groups in any consistent linear way. Finally, for measures of Complexity, 20 of 30 correlations are non­ significant, and of the remaining 10, 5 are positive and 5 negative. Most correlations are again low. Note that all five significant positive correla­ tions come from the two Chinese samples, especially the males. Though Complexity may show a relation to age in some samples, it is not consis­ tently related to age across samples. So far only linear change has been discussed. Some authors reported that they found little evidence of anything else. However, an ANOVA for

Table 3.1 Correlations with age on C PI scales in five samples

Measures of Norm-orientation Responsibility Socialization Self-control Good impression Achievement by conformity (Low ) Flexibility3

Baltimore Great Britain Women Men Women Men N = 855 N= 1139 N - 153 N - 195

Detroit Men Women N = ¡34 N = 151

Beijing Men Women N = 219 N = 23l

0.26 0.34 0.45 0.40 0.36 0.32

0.17 0.28 0.45 0.40 0.33 0.30

-0.1 l NS -0.17 -0.36 -0.29 -0.31 -0.17 -0.20 -0.28

-0.11 -0.31 -0.11 -0.25

-0.15 -0.32 -0.20 -0.30

0.04NS -0.04NS _0O2NS ~0.06NS 0.04NS -0.03NS

-0.22 -0.14NS -0.04NS -0.17 -0.16 -0.23

0.08NS 0.03NS -0.17 -0.14 0.03NS -0.02NS

0.05NS 0.06NS -0.04NS 0.04NS 0.02NS

0.1 l NS 0.15 -0.16 -0.10NS 0.03NS

0.03NS -0.14NS -0.10NS -0.21 0.03NS -0.22

0.13 0.05NS 0.02NS 0.0 l NS 0.17 0.12

0.09NS 0.05NS 0.04NS -0.10NS -0.06NS -0.05NS

0.31 0.1 l NS 0.06NS -0.03NS 0.17 0.08NS

0.25 0.12 0.35 0.28 0.15 0.39

0.24 0.14 0.36 0.25 0.18 0.37

-0.11 -0.13 -0.29 -0.32 -0.10NS -0.12 -0.28 -0.28

Measures of Social assurance Dominance Capacity for status Independence

0.01 -0.07 -0.01

0.00 -0.17 -0.03

Measures of Complexity Tolerance Achievement by independence Psychological mindedness

0.13 -0.03 0.03

0.03 -0.09 -0.02

-0.16 -0.14 -0.08NS -0.10

0.03NS -0.08 -0.02NS -0.32 -0.02NS -0.07NS -0.09NS -0.24

0.44 0.39 0.49 0.51 0.36 0.49

0.26 0.27 0.39 0.39 0.25 0.43

-0.03NS -0.06NS -0.33 -0.25 -0.33 -0.22 -0.28 -0.24

tS) z

-0.12 -0.30 -0.16 -0.27

0.42 0.19 0.40 0.37 0.26 0.25

s

-0.15 -0.26 -0.17 -0.22

0.20 0.19 0.15NS 0.1 l NS 0.20 0.18

o

Measures of Social vitality Sociability Social presence Self-acceptance Empathy

0.23 0.33 0.26 0.21 0.21 0.22

oo z o o o

0.25 0.18 0.35 0.31 0.21 0.32

2 CO o o

0.28 0.27 0.34 0.30 0.26 0.32

0.17 0.02NS 0.15 0.14 0.02NS 0.23

PRC Men Women N = II59N -9 34

o o 00 z on

CPI Scale

Average r Men Women

Notes a Rekeyed so that high scores are in direction of norm-orientation. Except as labelled NS (non-significant), all correlations are significant at p