European Review of Social Psychology
European Review of Social Psychology Editorial Board Dominic Abrams Herbert Bless Nyla Branscombe Marilynn Brewer Nanne De Vries Michael Diehl Alice Eagly Naomi Ellemers Klaus Fiedler Susan Fiske Michael Hogg Klaus Jonas Charles Judd Arie Kruglanski John Levine Jacques-Philippe Leyens Anne Maass Anthony Manstead David Messick Gerold Mikula Amélie Mummendey Brian Parkinson Kate Reynolds Bernard Rimé Charles Stangor Fritz Strack Paul van Lange Alberto Voci Vincent Yzerbyt
University of Kent, UK University of Mannheim, Germany University of Kansas, USA Ohio State University, USA University of Maastricht, The Netherlands University of Tübingen, Germany Northwestern University, USA Leiden University, The Netherlands University of Heidelberg, Germany Princeton University, USA University of Queensland, Australia Technical University of Chemnitz, Germany University of Colorado, USA University of Maryland, USA University of Pittsburgh, USA University of Louvain, Belgium University of Padua, Italy University of Cambridge, UK Northwestern University, USA University of Graz, Austria Friedrich-Schiller-University of Jena, Germany University of Oxford, UK Australian National University, Australia University of Louvain, Belgium University of Maryland, USA University of Würzburg, Germany Free University of Amsterdam, The Netherlands University of Padua, Italy University of Louvain, Belgium
Editors Wolfgang Stroebe Miles Hewstone
Utrecht University, The Netherlands Oxford University, UK
iii
About the editors Wolfgang Stroebe has published widely on the topics of attitudes, group processes and health psychology. A former president of the European Association of Experimental Social Psychology and a fellow of the British Psychological Society, the Society for the Psychological Study of Social Issues, and the Society of Personality and Social Psychology, he has recently received an honorary doctorate from the University of Louvain. Having held academic positions in England, USA and Germany he is currently Professor of Social, Organizational and Health Psychology at the University of Utrecht (The Netherlands). Miles Hewstone has published widely on the topics of social cognition and intergroup relations. He was awarded the British Psychological Society’s Spearman Medal in 1987 and its Presidents’ Award for Distinguished Contributions to Research in 2001. He was a Fellow at the Center for Advanced Study in the Behavioral Sciences, Stanford, California from 1987–1988 and 1999–2000, and is an Academician of the Academy of Learned Societies in the Social Sciences, and a Fellow of the British Academy. He is Professor of Social Psychology and Fellow of New College, University of Oxford.
European Review of Social Psychology Volume 13 Edited by
Wolfgang Stroebe Utrecht University, The Netherlands and Miles Hewstone Oxford University, UK
Published in 2002 by Psychology Press Ltd 27 Church Road, Hove, East Sussex, BN3 2FA, UK www.psypress.co.uk Simultaneously published in the USA and Canada by Taylor & Francis Inc. 29 West 35th Street, New York, NY, 10001, USA This edition published in the Taylor & Francis e-Library, 2005. “To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.” Psychology Press is part of the Taylor & Francis Group © 2002 by European Association of Experimental Social Psychology All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or 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 0-203-49553-5 Master e-book ISBN
ISBN 0-203-59547-5 (Adobe eReader Format) ISBN 1-84169-940-3 (hbk) ISSN 1046–3283 Cover design by Jim Wilkie
Contents
List of Contributors Acknowledgments Preface “Me and us” or “us and them”? The self as a heuristic for defining minimal ingroups Sabine Otten A dual-route model of crossed categorisation effects Richard J.Crisp, Nurcan Ensari, Miles Hewstone and Norman Miller Stereotype accuracy Carey S.Ryan The psychology of system justification and the palliative function of ideology John T.Jost and Orsolya Hunyady Leader endorsement in social dilemmas: Comparing the instrumental and relational perspectives Mark van Vugt and David De Cremer Gender-related inequalities in the division of family work in close relationships: A social psychological perspective Esther S.Kluwer and Gerold Mikula Does scientific thinking lead to success and sanity? An integration of attribution and attributional models Friedrich Försterling Morality and political orientations: An analysis of their relationship Nicholas Emler Testing attitude-behaviour theories using non-experimental data: An examination of some hidden assumptions Stephen Sutton Author Index Subject lndex
vii ix x 1
35
75 110
154
185
219
262
296
327 339
Contributors
R.J.Crisp, School of Psychology, University of Birmingham, Edgbaston,Birmingham, B15 2TT, UK. David De Cremer, Department of Experimental Psychology, University ofMaastricht, P.O.Box 616 6200 MD Maastricht, The Netherlands Nicholas Emler, Department of Psychology, University of Surrey, GU2 7XH, UK. Nurcan Ensari, 10868 Walnut Street, Los Alamitos CA 90720, USA. Friedrich Försterling, Institute für Psychologie, Ludwig-MaximiliansUniversität,Leopoldstrasse 13, 80803 München, Germany. Miles Hewstone, Department of Experimental Psychology, University of OxfordSouth Parks Road, Oxford, OXl 3UD, UK. Orsolya Hunyady, University of Debrecen, Hungary. John T.Jost, Stanford University, Graduate School of Business, 518 MemorialWay, Stanford, CA 94305–5015, USA. Esther S.Kluwer, Utrecht University, Department of Social and OrganisationalPsychology, PO Box 80140, 3508 TC Utrecht, The Netherlands. Norman Miller, Dept of Psychology, SGM 501, University of Southern CaliforniaLos Angeles, CA 90089–1061, USA. Gerold Mikula, Institut für Psychologie, Universität Graz, Universitätsplatz 2,A-8010 Graz, Austria Sabine Otten, University of Groningen, Department of Social and OrganizationalPsychology, Grote Kruisstraat 2/1, NL-9712, TS Groningen, The Netherlands. Carey S.Ryan, Department of Psychology, University of Nebraska at Omaha,Omaha, NE 68182–0274, USA.
viii
Stephen Sutton, University of Cambridge, Institute of Public Health, RobinsonWay, Cambridge CB2 2SR, UK. Mark Van Vugt, Department of Psychology, University of Southampton, Highfield, Southampton, SO17 1BJ, UK.
Acknowledgements
We would like to thank the following reviewers who helped us and the authors to shape these chapters into their final versions: Georgios Abakoumkin, Dominic Abrams, Richard Crisp, Faye Crosby, Alice Eagly, Klaus Fiedler, Ken Gergen, Denis Hilton, Michael Hogg, Klaus Jonas, Charles Judd, Norbert Kerr, Anne Maass, Greg Maio, Anthony Manstead, Gillian Marks, David Messick, Gerold Mikula, Diana Mutz, Mark Rubin, Geoff Thomas, Frank van Overvalle, Ulrich von Hecker, Norbert Vanbeslaere, Thomas Wills.
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY, 2002, 13, viii–ix
Preface
The European Review of Social Psychology was conceived when one of us (Wolfgang Stroebe) was President of the European Association of Experimental Social Psychology. The idea was to create a series that would reflect the dynamism of social psychology in Europe and the attention paid to European ideas and research. Even though the European Review has always been intended as a publication of the European Association of Social Psychology, it took more than a decade for the Association to agree to this. We are delighted that from this volume onwards (Volume 13), the European Review of Social Psychology is an official publication of the European Association of Experimental Social Psychology and we are grateful to Jacques-Philippe Leyens and Naomi Ellemers who, as Presidents of the Association, paved the way for the official adoption. As significant as the official adoption by the European Association is the fact that we have also changed publishers. The European Review is now being published by Psychology Press, a European publishing house that is part of the international Taylor & Francis Group. We are delighted to be working with a dynamic publishing house and an enthusiastic team of editors with novel ideas on how to further the image (and increase the distribution) of the European Reviewof Social Psychology. From Volume 14 onwards, the European Review of SocialPsychology will be published as an electronic journal. With each paper being published as soon as the editorial process has been completed, this will not only considerably reduce the publication lag, making the papers available immediately, rather than awaiting publication at the end of the year, but it will also allow libraries to subscribe to the Review, rather than having to order volume by volume. However, readers will not have to do without the familiar blue volumes, because at the end of the year, the set of chapters for that year will be published as a printed volume.
xi
In the past decade, the European Review of Social Psychology has been widely accepted as one of the major international series in social psychology and we are confident that these changes will only further this image. Social Psychology is an international endeavour and this fact underpinned our decision to make the European Review of Social Psychology an international review publishing outstanding work of authors from all nations rather than restricting it to Europeans. With the help of an editorial board consisting of senior scholars from various European countries, Australasia, and North America, the editors invite outstanding researchers to contribute to these volumes. Invitations are based either on suggestions from editorial board members or made in response to proposals submitted to the editors. The emphasis of these contributions is on critical assessment of major areas of research and of substantial individual programmes of research as well as on topics and initiatives of contemporary interest and originality. Volumes contain three types of contributions: (1) Reviews of the field in some specific area of social psychology, typically one in which European researchers have made some special contribution; (2) Reports of extended research programmes which contribute to knowledge of a particular phenomenon or process; (3) Contributions to a contemporary theoretical issue or debate. All manuscripts are externally reviewed and typically extensively revised. The final decision on whether to publish a given manuscript is subject to a positive outcome of the review and editorial process. Thanks to the quality of the authors as well as of the editorial process (assisted by our editorial board as well as outside reviewers), the European Review has become internationally renowned. Wolfgang Stroebe Miles Hewstone
© 2002 European Association of Experimental Social Psychology
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY, 2002, 13, 1–33
“Me and us” or “us and them”? The self as a heuristicfor defining minimal ingroups Sabine Otten University of Jena, Germany
The question of why even a minimal ingroup is typically evaluated more positively than the respective outgroup has stimulated extensive theoretical and empirical work in social psychology. Integrating findings from various domains of research, this chapter summarises a comprehensive research programme that focuses on cognitive rather than motivational factors that contribute to positive ingroup distinctiveness. More specifically, evidence is presented showing that (a) there is a positive ingroup default, such that novel ingroups are immediately associated with positive affect; (b) people make inferences from the self in order to define their novel groups; and (c) this process of using the self as a means of cognitive structuring is based on heuristic rather than systematic information processing. Implications for our understanding of the role of self in intergroup evaluations and of factors determining ingroup favouritism in both minimal and real groups are discussed.
In 1971, Tajfel and collaborators (Tajfel, Billig, Bundy & Flament, 1971) published their experiments using the Minimal Group Paradigm (MGP), which showed that after categorisation into novel, arbitrary social groups, anonymous ingroup members were treated more positively than outgroup members. In 1979, Tajfel and Turner explained this finding in the
© 2002 European Association of Experimental Social Psychology http://www.tandf.co.uk/journals/pp/02699931.html DOI: 10.1080/10463280240000028
2
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
framework of social identity theory (Tajfel & Turner, 1979, 1986). However,in 1989, Messick and Mackie wrote that “…nearly 20 years after the discovery that mere categorization produced intergroup bias, an adequate theory of the phenomenon has yet to be developed” (p. 62). Today, we can state that the scientific debate over the origins of ingroup favouritism between novel, minimal groups continues (see e.g., Cadinu & Rothbart, 1996; L.Gaertner & Insko, 2000). The present chapter contributes to this debate by outlining a model of the self as heuristic for defining minimal groups, which integrates theory and data from enon has yet to be developed” (p. 62). Today, we can state that the scientific debate over the origins of ingroup favouritism between novel, minimal groups continues (see e.g., Cadinu & Rothbart, 1996; L.Gaertner & Insko, 2000). The present chapter contributes to this debate by outlining a model of the self as heuristic for defining minimal groups, which integrates theory and data from research on intergroup behaviour and social information processing. Consistent with social identity theory, the present approach assumes that once a categorised individual accepts the novel group membership as a meaningful part of the self (Oakes, 2001), ingroup favouritism can result from social categorisation, irrespective of realistic conflict of interests (Sherif, 1966). However, this approach considerably deviates from social identity theory with regard to the basic mechanisms that may account for these effects. The MGP (Rabbie & Horwitz, 1969; Tajfel et al., 1971) was originally designed to identify the necessary and sufficient conditions to elicit ingroup favouritism and outgroup derogation. Key features of the MGP are (a) anonymous assignment to novel, arbitrary social categories; (b) neither intra- nor intergroup interaction; (c) no functional relation between categorisation dimension, on the one hand, and the intergroup comparison dimension, on the other hand; and (d) exclusion of direct self-interest (there is no opportunity to favour oneself). Unexpectedly, studies using the MGP revealed that categorising people into arbitrary, completely novel groups was already sufficient to elicit ingroup favouritism in intergroup allocation decisions made by novel group members. Subsequent studies replicated this finding for both intergroup allocation of resources and intergroup evaluations on trait dimensions (for reviews, see Brewer, 1979; Brewer & Brown, 1998; Messick & Mackie, 1989; Tajfel, 1982). In 1979, Tajfel and Turner provided an interpretation for this “mere categorisation effect” in the more general framework of social identity theory. A crucial assumption in this theory is that the self-concept
Address correspondence to: Sabine Otten, University of Groningen, Dept. of Social and Organizational Psychology, Grote Kruisstraat 2/1, NL-9712 TS Groningen, The Netherlands, Email:
[email protected]
FAVOURITISM TOWARDS MINIMAL INGROUPS 3
comprises both a personal identity, defined by idiosyncratic characteristics that distinguish the individual from others, and a social identity, defined by shared characteristics associated with subjectively relevant and valued social groups to which the individual belongs. When a social category becomes meaningful, group members identify with their ingroup, and compare it to relevant outgroups. Based on the premise of a need for positive self-evaluation, it is hypothesised that this comparison will be guided by a search for positive ingroup distinctiveness. In short, an ingroup’s positive value is defined in relation to the outgroup; its superiority in intergroup allocations or evaluations—even in a minimal group setting— can enhance or consolidate positive social identity and, accordingly, support group members’ positive self-concept. While social identity theory has contributed substantially to research on intergroup behaviour, in general (see Brown, 2000), its interpretation of the findings in the MGP has been criticised (e.g., Diehl, 1990; Hinkle & Brown, 1990; Messick & Mackie, 1989; Mummendey, 1995; Rabbie, Schot & Visser; 1989). One critical shortcoming is the lack of unequivocal empirical support for the self-enhancing function of ingroup favouritism. Abrams and Hogg (1988; see also Hogg & Abrams, 1990) derived two corollaries from social identity theory with regard to the links between selfesteem and ingroup favouritism: (a) intergroup discrimination should elevate self-esteem, and (b) low self-esteem should increase intergroup discrimination.1 The second derivation in particular received only weak empirical support, and both corollaries need further qualification, most of all with regard to the self-esteem measures involved (see Aberson, Healey, & Romero, 2000; Long, Spears, & Manstead, 1994; Rubin & Hewstone, 1998). Summing up the evidence at present, Brown (2000) concluded that “Whatever the merits of these methodological arguments, it seems that the original self-esteem plank of the SIT model is now much less surely established than it once was” (p. 756). Another challenge for the explanation of ingroup favouritism in terms of a motivated striving for positive ingroup distinctiveness stems from a research programme conducted by Mummendey and collaborators on the positive-negative asymmetry in social discrimination (for reviews, see Mummendey & Otten, 1998; Otten & Mummendey, 2000). A series of studies showed that the probability of ingroup favouritism is affected by the valence of the intergroup comparison dimension. Whereas Mummendey and her co-workers replicated ingroup favouritism in a minimal intergroup setting when group members allocated goods between ingroup and outgroup (e.g., monetary rewards), or evaluated the groups on positive trait dimensions, the findings differed for intergroup treatment on negative comparison dimensions. When allocating negative resources (e.g., the number of unpleasant tasks to be solved) or when evaluating in- and outgroup on negative trait dimensions, parity was the dominant outcome
4
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
(Blanz, Mummendey, & Otten, 1995; Gardham & Brown, 2001; Otten, Mummendey & Blanz, 1996).2 However, in principle positive ingroup distinctiveness (getting more or losing less than the outgroup) can be established or secured in both valence conditions. Why should positive social identity be more contingent upon having greater positive than fewer negative group characteristics, relative to the outgroup? On the one hand, this question elicited a body of research demonstrating mechanisms that can account for a decreased probability of ingroup favouritism in the negative domain (e.g., Mummendey, Otten, Berger & Kessler, 2000; Otten & Mummendey, 1999; Otten, Mummendey & Buhl, 1998). On the other hand, and more importantly for the present chapter, the positive-negative asymmetry in social discrimination casts doubts upon the traditional view that a striving for positive ingroup distinctiveness can sufficiently account for ingroup favouritism in the positive domain (see Mummendey, 1995). The present chapter is concerned with a question that was Tajfel’s concern when first implementing the MGP (Tajfel et al., 1971), namely to examine the basic conditions and processes that can account for intergroup biases. Specifically, the focus is on the emergence of favourable ingroup images—representations of the ingroup that are not yet shaped by actual experiences or well-established social stereotypes. A central feature in the present analysis is the assumption that ingroups need not necessarily be defined positively in relation to outgroups. Mummendey (1995) outlined that a striving for positive social identity and intergroup differentiation, though linked in the framework of social identity theory, need not necessarily operate in conjunction. In the same vein, Allport (1954, p. 41) had asked “Can there be an in-group without an out-group?”, a question recently reconsidered by Brewer (1999). In line with Allport, she argued that in a logical sense the very act of perceiving and identifying with an ingroup (“us”) necessitates the existence of an outgroup (“others”). However, once group membership became salient, the treatment and evaluation of ingroup and outgroup might function quite independently. Consequently, attachment to own groups and rejection of
1 See, however, Turner (1999; Turner & Reynolds, 2001) for a critique of the corollaries derived by Abrams and Hogg (1988). 2 It is important to note that the positive-negative asymmetry does especially apply to minimal intergroup settings, where groups are both novel and ill-defined. Mummendey and coworkers showed that intergroup differentiation occurs on positive as well as negative dimensions once the group context is further specified (e.g., by introducing status differentials, see Blanz et al., 1995; Otten et al., 1996) or, as Reynolds, Turner and Haslam (2000) demonstrated, when there are prototypical representations of the groups with regard to the trait dimensions that are evaluated.
FAVOURITISM TOWARDS MINIMAL INGROUPS 5
or hostility towards other groups need not correlate (see also Hinkle & Brown, 1990). Whereas Brewer (1999) discusses the conditions under which positive attitudes towards ingroups and outgroups may coincide, the main question of the present chapter has a somewhat different focus. If favourable ingroup representations are not based on positive differentiation from an outgroup, where does positive social identity originally stem from? In this context, Allport (1954) referred to the higher familiarity (e.g., amount of contact) that individuals tend to have with their ingroups as compared to outgroups. However, this argument cannot apply to arbitrary, laboratory ingroups that are judged and treated favourably (for reviews, see Brewer, 1979; Messick & Mackie, 1989; Mullen, Brown, & Smith, 1992; Tajfel, 1982). Little controversy is elicited by the claim that—by and large—self categories tend to be positive (e.g., Taylor & Brown, 1988; Turner, Hogg, Oakes, Reicher, & Wetherell 1987, pp. 58–59). However, we know very little about how and why novel ingroups, with minimal ones just being an extreme example, initially acquire their positive value. We pursued this question in a series of studies, which (a) mostly took advantage of the MGP as an experimental setting that allows us to analyse impression formation concerning completely novel groups, and which (b) built upon a rich set of findings from various domains within social psychological research on intergroup behaviour. In the following, this chapter introduces an integrative research programme that mainly focuses on the factors that contribute to the positive evaluation of novel, minimal ingroups in absolute terms (why is the ingroup seen as good?) rather than relative terms (why is the ingroup seen as better than the outgroup?). In this vein, in the following the term ingroup favouritism will be used in order to refer to a positive, above-average evaluation of ingroups. The term positive ingroup distinctiveness is relational or comparative, and is meant to imply a significant differencebetween ingroup and outgroup evaluation. Although this distinction is subtle and both phenomena are clearly interrelated, it is assumed to be worthwhile for a deeper understanding of the processes operating in minimal intergroup settings. The research programme focused on two distinct, but interrelated assumptions. First, there is an automatically generated positive ingroupdefault which, on an implicit level, gives the ingroup an evaluative and affective advantage over the outgroup. Second, as already inherent in the very act of category assignment, self and ingroup are associated with each other. As the self tends to be evaluated positively, this association is assumed to contribute to favourable ingroup evaluations. The link between self and positive ingroup evaluation is analysed threefold: (a) Selfand ingroup are assumed to be interrelated in connectionist networks of memory; supporting this claim, response-time evidence for overlapping
6
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
mental representations of self and ingroup will be reported. (b) We assume that lacking any other information, there will be a tendency to use the self as an anchor for ingroup judgements. Like the positive ingroup default, this process can be understood as fairly effortless pathway towards a (positive) ingroup definition. (c) Finally, empirical evidence will be presented suggesting that self-anchoring in ingroup judgements varies asa function of information processing, more specifically, as a function of both the motivation and the ability for cognitive structuring. In sum, the findings from these three domains converge on a model of the self as heuristic for the definition of minimal ingroups. This model integrates several fields of research, and it can contribute not only to an explanation of the striking findings in the MGP, but also to our more general understanding of ingroup favouritism and social discrimination. Importantly, each assumption refers to the intragroup rather than intergroup level of categorisation. An intergroup context is needed to introduce social categorisation. However, once group assignment has taken place, the link between self and ingroup might be sufficient to account for ingroup favouritism. POSITIVE INGROUP DEFAULT At least in laboratory intergroup situations, and in settings that are not characterised by conflicts and/or firm social stereotypes, most of the variance in positive ingroup distinctiveness can be traced back to variations in the evaluation of the ingroup (e.g., Brewer, 1979; Brewer & Brown, 1998; L.Gaertner & Schopler, 1998). Thus, an analysis of the factors initially determining individuals’ ideas about their ingroups seems worthwhile and necessary. Is the positive ingroup definition generated by social comparison with and positive differentiation from the corresponding outgroups? As discussed above, several authors suggest that this is not necessarily the case (Allport, 1954; Brewer, 1999; Hinkle & Brown, 1990; Mummendey, 1995). Alternatively, an ingroup’s positive value may also be understood as a default and as a premise rather than as a consequence of social comparison. In this vein, Maass and Schaller (1991) assume an “initial categorization-based ingroup bias”: Group members assume that their ingroup is better than the outgroup, as long as the given evidence does not obviously conflict with this assumption. In other words, “the cognitive process of forming group impressions is guided and biased by ingroup favoritism” (p.204; see also Schaller & Maass, 1989). Social cognition research supports the assumption of an automatically generated implicit ingroup bias (e.g., Bargh, 1997; see also Howard & Rothbart, 1980). There is strong empirical evidence suggesting that selfinclusive social categories can be associated with positive affect irrespective of an explicit social comparison with a specific outgroup. Perdue, Dovidio,
FAVOURITISM TOWARDS MINIMAL INGROUPS 7
Gurtman, and Tyler (1990) used unspecific ingroup designators (“us” or “we”) and outgroup designators (“they” or “them”) as primes in a lexical decision task. Their findings revealed that ingroup designators facilitate positive responses, whereas outgroup labels are rather neutral with regard to their affective association. The authors conclude that “in-group and outgroup associated words may function as linguistic vectors that establish an evaluative predisposition towards targets previously uninfected by prejudice” (Perdue et al., 1990, p. 482). However, primes like “we” might elicit associations with social groups like family and peers, whose positive value is rooted in a high degree of familiarity (as discussed by Allport, 1954). Hence, in order to understand when and why an ingroup is initially evaluated and treated positively, further research was needed. The question for our research was: Are completely novel ingroups already associated with positive affect? A series of experiments by Otten and collaborators (Otten & Moskowitz, 2000; Otten & Wentura, 1999) provided a strong affirmative answer to this question. In order to test implicit bias towards novel ingroups, Otten and Wentura (1999) combined a typical minimal group categorisation procedure with an affective priming paradigm (e.g., Fazio, Sanbonmatsu, Powell, & Kardes, 1986), in which participants were subliminally primed with the labels of a novel ingroup and outgroup in a lexical decision task (similar to Perdue et al., 1990, experiment 2). Ostensibly, the categorisation criterion was the temporal sequence of participants’ performance in previously administered concentration tasks (defined as “convex” and “concave”). Participants received anonymous feedback on their group membership via a personal computer. Following the feedback, group labels were subliminally administered as primes in a seemingly unrelated task, in which participants classified a series of traits as either positive or negative. Comparing the effects of a priori positive or negative primes (like “good” and “bad”) with the effects of group labels (ingroup or outgroup) revealed a comparable relative facilitation effect of congruent responses, especially in the case of positive congruency (i.e., a priori positive or ingroup prime with positive trait word). In their second experiment, which was a modified replication of the first one, Otten and Wentura (1999) did not measure response latencies, but error rates in the lexical decision task. The probability of errors was increased by introducing the response window technique used by Draine and Greenwald (1998). With this procedure, participants are forced to respond to the target stimuli within a time span that is too short to perform highly accurately. The results confirmed the findings from the first experiment: Lowest error rates for the combination of a positive target stimulus with an ingroup label as preceding prime indicated ingroup favouritism on an implicit level. The respective error rate was considerably lower than for the outgroup or the neutral (i.e., a priori positive or negative) prime condition. However, type of priming had no effect on
8
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 1. Mean error rates (in percentages) as a function of target valence (positive versus negative) and priming (ingroup label versus outgroup label versus neutral) (Experiment 2, block 2) (adapted from Otten & Wentura, 1999, with permission).
responses to negative stimuli (see Figure 1). Importantly, in this experiment there was also a positive correlation between the facilitation effect for the ingroup label and evaluations of the ingroup on explicit measures: The stronger the congruency effect, the stronger the preferential judgement of the ingroup. For outgroup labels, there were no comparable effects. Otten and Moskowitz (2000) corroborated these findings using a spontaneous trait inference paradigm (Uleman, Hon, Roman, & Moskowitz, 1996). Again, participants underwent a minimal categorisation procedure, this time purportedly based upon “perceptual styles in the organisation of complex pictorial material”. In the subsequent task, which allegedly investigated the processing of verbal material, participants were led to believe that the sentences presented to them had been written either by ingroup or by outgroup members, who, in a previous study, had described their everyday activities (“Before leaving the house, she did all the dishes”). Sentences were presented on a computer screen, followed by a single word, which was either implied by the previous sentence (e.g., “neat”) or not implied (e.g., “intelligent”). The participants’ task was to decide whether the word appeared in the previous sentence. In line with the assumption that ingroups (and their members) are positively associated by default, the rejection of positive trait words implied by the preceding sentence was significantly delayed when the sentence presumably described behaviour by an ingroup rather than an outgroup member. Again, no comparable effects emerged for negative traits (see Figure 2). In this respect, the present findings on implicit bias towards minimal ingroups are fully consistent with empirical evidence about explicit biases in minimal or quasi-minimal group contexts: Positive ingroup distinctiveness is based
FAVOURITISM TOWARDS MINIMAL INGROUPS 9
Figure 2. Spontaneous trait inferences as a function of trait valence and group affiliation (adapted from Otten & Moskowitz, 2000).
upon a relative upgrading ofthe ingroup rather than on a relative downgrading of the outgroup (Brewer, 1979, 1993). Due to their reference to minimal group labels rather than to generalised ingroup designators (like “we”), the Otten et al. findings (Otten & Moskowitz, 2000; Otten & Wentura, 1999) concerning implicit intergroup bias can hardly be explained by evaluative conditioning or by differences in the degree of familiarity. Similarly, an automatic activation of ingroup and outgroup stereotypes (e.g., Bargh, 1997; Fazio et al., 1986; Wittenbrink, Judd, & Park, 1997) does not convincingly apply to novel, minimal groups. Hence, the notion of a positive ingroup default (or an initial categorisationbased ingroup bias, as labelled by Maass & Schaller, 1991) seems adequate. However, demonstrating that there is unintentional implicit favouritism towards minimal ingroups does not contribute to our understanding of what is the basis for an ingroup’s positively distinct evaluation. In the case of minimal groups, neither personal nor socially transmitted experiences with the ingroup can convincingly account for a positive priming effect. However, even a minimal ingroup comprises at least one element with which the novel group member is well acquainted: the self. In fact, Perdue and collaborators (1990) already refer to this aspect as an alternative (even though not their preferred) interpretation of their findings. They propose that using self-inclusive entities as primes (like “us” or “we”, but also like the novel ingroup labels), implies that information processing about these entities might be affected by self-schemata. More specifically, Perdue et al. assume that due to the typically favourable content of the self-concept (e.g., Baumeister, 1998; Diener & Diener, 1996; Taylor & Brown, 1988) priming with first-person pronouns (singular or collective) is sufficient to facilitate responses to positive attributes.3 In similar vein, Gaertner and collaborators (S.L.Gaertner, Dovidio, Anastasio,
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
10
Bachman, & Rust, 1993, p. 3) postulate a “pro-ingroup bias” that is based upon the ingroup’s association with the self. In the following, we will focus in more detail on the nature of such an associative link between self and ingroup, and on its relevance to ingroup favouritism. SELF AND POSITIVE INGROUP EVALUATION Not only social stimuli can profit from their association with the self. For example, the “name letter” effect (Nuttin, 1987; see also Hoorens & Nuttin, 1993) shows that people tend to prefer the initials of their own name when choosing their favourites from a list of randomly selected letters. More generally, Nuttin named this phenomenon as “mere ownership”, the preference for any self-object above similar objects belonging to other persons, because of the bare belongingness to the self (see Feys, 1995, p. 560). While the name letter effect can be explained not only in terms of an affective self-bias, but also—in line with Allport’s assumptions (1954; see below)—in terms of familiarity and previous experience, a similar evaluative preference can be shown for novel,unfamiliar stimuli (patterns, symbols) that were introduced as belonging to the self. This was demonstrated in several studies by Feys (1991, 1995), who measured the attractiveness of novel stimuli (in the 1991 study: abstract symbols; in the 1995 study: Chinese characters) that were previously introduced as belonging to the self (e.g., by claiming that the Chinese characters depicted the participant’s own name) as compared to stimuli that were not assigned to the self. In both studies, liking for selfrelated stimuli was highest, which, according to Feys (1995) supports the view that “…the mere ownership effect reveals a purely affective self-bias, which is more primary than any better known cognitive self-serving bias or self-bias…” (p. 560). Turning to the domain of interpersonal relations, Aron and collaborators (Aron, Aron, & Smollan, 1992; Aron, Aron, Tudor, & Nelson, 1991; Aron & Fraley, 1999), found that the degree of perceived overlap between self and other in close relationships predicted the positive evaluation of the other and both the stability and well-being in the relationship. Finally, analysing impression formation in groups, Park and Judd (1990) reported that after a minimal categorisation procedure
3
In a logical sense, this implies that implicit ingroup favouritism relies on (and should positively correlate with) self-esteem. Otherwise, the affective congruency effect for the combination of positive targets with ingroup designators would be hard to explain. However, while there are many empirical tests on the link between self-esteem and explicit measures of ingroup favouritism (see above), evidence about such link in the implicit domain—to our knowledge—still needs to be collected.
FAVOURITISM TOWARDS MINIMAL INGROUPS 11
participants in a think-aloud paradigm referred to the self when uttering ideas about the novel ingroup, while no self-reference was made when impressions were formed about the outgroup. Applying this evidence to the present investigation of why ingroups are evaluated positively, one can conclude that the very act of social categorisation implies both a certain kind of “ownership” with regard to the category label and a certain degree of similarity or overlap between self and novel ingroup. Thus, the social category becomes associated with the self, and it can profit from the fact that the self tends to be evaluated positively (e.g., Baumeister, 1998). However, while this reasoning sounds convincing, it still does not tell us enough about the process underlying the link between self and ingroup and accounting for ingroup favouritism towards novel groups. Self and ingroup in connectionist networks Basic assumptions and findings. Smith and colleagues (Coats, Smith, Claypool, & Banner, 2000; Smith, Coats, & Walling, 1999; Smith & Henry, 1996) analysed the relationship between self and ingroup in the theoretical context of a connectionist model of human memory. In short, this model implies that mental representations of both self and ingroup (and other social and non-social targets) are flexible and context-dependent constructions. According to Smith et al. (1999) "Connectionist networks do consist of nodes or units connected by links over which activation can be sent" (p. 878). The underlying network structure does not provide fixed combinations of entities (e.g., the self) and specific information (e.g., trait characteristics). Rather, it "can affect judgments and overt responses only insofar as it affects the spread of activation [between nodes or units] and therefore the nature of the representation ... that is constructed" (Smith et al., 1999, p. 879). In the case of self and ingroup, these mental representations are assumed to overlap to a certain extent (see Smith et al., 1999); in other words, there is a common pool of underlying features (e.g., traits, attitudes). In conclusion, when making self-descriptions, ingroup representations will be activated simultaneously, and vice versa (Smith, 2002). Smith and colleagues tested these assumptions, adopting a paradigm from Aron and collaborators (Aron et al., 1991) to the domain of intergroup relations. First, in a paper-and-pencil task the self, the ingroup, and the outgroup are rated on trait lists (90 traits for each entity) using 7point scales (l=does not apply, 7 = does fully apply). Later on, ostensibly to analyse effects of measurement type, one target is evaluated another time at the computer with a dichotomous response format ("yes", does apply, or "no", does not apply). Response latencies in this last trial are the decisive dependent variable (see Figure 3 for a schematic representation of the
12
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 3. Simplified representation (neglecting outgroup ratings) of the design by Smith and collaborators (1999) and Epstude and Otten (2000). Response latencies are measured for ingroup ratings; an X indicates that representations of self and ingroup match.
design). Smith and collaborators (e.g., Smith et al., 1999; Smith & Henry, 1996) demonstrated that reports on attributes of both self and ingroup were facilitated for trait dimensions on which self and ingroup matched (i.e., trait applied or did not apply to both targets), and inhibited when trait judgements on self and ingroup mismatched. In addition, the size of the match/mismatch effect as manifested in response latencies correlated significantly with measures of group identification (Coats et al., 2000). Interestingly, matches and mismatches with the outgroup had no crucial impact on response latencies. Referring to self-categorisation theory (Turner et al., 1987), Smith and colleagues interpret their findings in terms of a merging between self and other implied by group membership; “the self extends out beyond the individual” (Coats et al., 2000, p. 314). Thus, judgements about the self are affected by knowledge about the ingroup and vice versa; their representations are constructed from and affected by a common pool of underlying knowledge (Smith, 2002). Within the connectionist model, the social context affects whether self-representations determined by activation patterns are dominated by individuating characteristics or by traits and attributes defining the respective ingroup. Hence, this model is fully compatible with the distinction between interpersonal and intergroup levels of self-categorisation as distinguished in self-categorisation theory and social identity theory (see Smith et al., 1999). While there is the notion of context-specific patterns of activation, on the one hand, there is the idea of the similarly context-specific salience of either personal or social identity, on the other hand. The case of judgemental ambiguity The paradigm used by Smith and collaborators necessitates that the self, the ingroup, and the outgroup are rated on a large number of trait dimensions. Hence, this procedure can hardly be applied to the MGP, where participants’ judgeability ratings of ingroup and outgroup (see
FAVOURITISM TOWARDS MINIMAL INGROUPS 13
Yzerbyt, Leyens, & Schadron, 1997) can be expected to be quite low. Nevertheless, is there any possibility to analyse the creation rather than the mere activation of links between self and ingroup with this method? In an attempt to do so, Epstude and Otten (2000) replicated the study by Smith et al. (1999) with a realistic categorisation (gender). Moreover, they added an analysis, which focused on trait dimensions on which participants in the paper-pencil-task had indicated judgemental ambiguity with regard to the ingroup. In the original studies, Smith and collaborators excluded all traits that received an undecided rating (i.e., “4”, the midpoint of the 7-point bipolar scale) in the paper-and-pencil part of their studies, so that they could define clear instances of matches and mismatches between judgements. However, in order to analyse whether the self is used as a reference point to form impressions about the ingroup, precisely these cases are most interesting. Fully in line with the data pattern reported by Smith and collaborators (1999) for well-defined ingroup trait dimensions, Epstude and Otten (2000) found for trait dimensions on which the ingroup received “neutral”, undecided judgements on the paper-pencil-task that subsequent judgements in a dichotomous response format (“trait applies”; “trait does not apply”) were facilitated when they matched self-ratings. In other words, a “yes [applies to the ingroup]” response to an ambiguous trait dimension was faster when this trait additionally applied to the self than when it did not apply to the self (and the reverse for “no” responses). Thus, reliance on self-ratings seems to provide a means to generate quick and straightforward decisions about characteristics of the ingroup. One could argue that focusing on these ambiguous traits allows us to build a bridge between the present paradigm and studies by Hogg and collaborators, who identified feelings of uncertainty as a crucial variable motivating ingroup identification and ingroup favouritism in the MGP (see Hogg, 2001; Hogg & Abrams, 1993; Hogg & Mullin, 1999). The close link between self and ingroup, as revealed in the original studies by Smith and collaborators can be read as evidence for group identification (as indicated by the correlation with the match-mismatch effect; Smith et al., 1999, see also Tropp & Wright, 2001). Hence, the evidence indicating that group members are able to construe such links when uncertain about the ingroup is in line with research on uncertainty reduction. At the same time, findings are in accordance with the connectionist network model: If we assume that there is a direct link between self and ingroup, then a spread of activation from previously solely self-related traits to the ingroup reveals the assumed dynamic character of these mental representations (see Smith, 2002; Smith et al., 1999).
14
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Self-categorisation as an ingroup member Although it is rooted in different theoretical and empirical traditions, the logic of self and ingroup as linked in connectionist networks in memory is quite compatible with self-categorisation theory, where the relation between self and ingroup is one of the core issues (e.g., Onorato & Turner, 2001; Turner, 1999; Turner et al., 1987; Turner, Oakes, Haslam, & McGarty, 1994; Turner & Onorato, 1999). Similar to the idea of situationspecific activation patterns in the framework of a connectionist model (Smith, 1996), the self is seen as a dynamic, malleable entity, formed in the process of social judgement and determined by the social context (Turner et al., 1994). Whereas the self on the personal level encompasses individual, idiosyncratic characteristics that distinguish “me” from others, the self on the intergroup level is defined in terms of characteristics that are shared with other group members. The self as “me” on the interpersonal level can vary considerably from the self as “us” on the intergroup level of categorisation. Nonetheless, both views are similarly valid and elaborated self-representations in a given situation; neither level is, in principle, more simple or more correct, but characterised by changes in the nature and content of the self-concept (see Turner et al., 1987, p. 51). Moreover, selfcategorisation theory postulates that when the intergroup level becomes salient, the self becomes depersonalised: It is defined according to features that are prototypical for the group. This process of depersonalisation is seen as crucial for intergroup behaviour, such as ingroup favouritism and outgroup derogation. However, the idea of depersonalisation necessitates the existence of an ingroup prototype—a concept about traits that are typical and relevant for the ingroup (see Turner et al., 1987). How can such a process operate in the case of minimal groups? How can the self be defined in terms of a group prototype that has not yet been established? In such a setting, we assume self and ingroup will merge, such that the personal self (or certain aspects of it) provides the prototype to define the novel ingroup. Hence, the ingroup definition will follow certain aspects of the (personal) selfdefinition (see Reynolds at al., 2000, for a similar point). Lacking information about both ingroup and outgroup prototypes, rather than altering the content of the self-concept, in the first place, the salient, but ambiguous intergroup context—as given in the MGP—will imply a transfer of this content from self to the new ingroup (as tentatively revealed by Epstude & Otten’s, 2000, findings). Both self-categorisation theory and the connectionist network model (e.g., Smith, 2002) imply that when group membership is salient, the concepts of self and ingroup become closely linked. However, while this link applies to the content of concepts about self and ingroup, it does not necessarily concern the quality or valence of these concepts (i.e., a
FAVOURITISM TOWARDS MINIMAL INGROUPS 15
superiority of the ingroup over the outgroup in intergroup evaluations). In fact, Smith and colleagues (Coats et al., 2000; Smith et al., 1999; Smith & Henry, 1996;) and Epstude and Otten (2000) found that congruency of self and ingroup ratings facilitates ingroup judgements on both positive and negative trait dimensions. However, the paradigm used in these studies conflates valence and the response type. As there is a general tendency to evaluate the self positively (see e.g., Baumeister, 1998), a match between self and ingroup on positive dimensions usually implies that a certain trait is assigned to both targets (as the self tends to be evaluated positively), whereas a match on negative dimension typically implies that a trait is rejected in both cases. Thus, the findings are still supportive of the existence of a positive ingroup default in the MGP (Otten & Moskowitz, 2000; Otten & Wentura, 1999). Nonetheless, a more specific analysis is needed to explore how close links between self and ingroup can translate into positive ingroup ratings, and whether and how this process can account for positive ingroup distinctiveness and ingroup favouritism. The self as an anchor for ingroup judgements Self-anchoring. Along these lines, Cadinu and Rothbart (1996) proposed a model linking the cognitive association between self and ingroup and the “mere categorisation effect” in the MGP. In contrast to social identity theorists (Tajfel & Turner, 1979), who cite motivational factors as mainly responsible for positive ingroup distinctiveness in the MGP, Cadinu and Rothbart point to cognitive processes. They postulate that novel social categories are defined by application of two heuristics, one referring to similarity, one referring to oppositeness. First, based upon the premise that people generally possess favourable beliefs about themselves, they argue that ingroup characteristics are inferred from characteristics of the self, resulting in an ingroup that is “a less extreme copy of the self image” (Cadinu & Rothbart, 1996, p. 663). Second, as self and ingroup are regarded favourably, by a principle of differentiation the outgroup will be regarded as less favourable (see Cadinu & Rothbart, 1996, p. 662). Cadinu and Rothbart (1996) provided evidence for their model through a series of experiments. First, participants were assigned to novel, minimal categories, after which they were provided with specific trait information about either the ingroup or the outgroup. Finally, they rated the respective complementary group (outgroup or ingroup) on the very same trait dimensions. As a further variation, either before or after the group evaluation task, participants made self-ratings on the respective trait dimensions. Results revealed that self was rated as more similar to the ingroup than to the outgroup, and that ingroup judgements were independent of outgroup information. In contrast, outgroup judgements relied on ingroup information, such that the more a certain trait had been
16
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
ascribed to the ingroup, the less it was ascribed to the outgroup. Finally, self-anchoring was stronger when the judgemental sequence started, rather than concluded, with the self-evaluation task. Although these studies imply that self-anchoring supports positive ingroup judgements, and that the outgroup is located relative to (as different from) the ingroup, the paradigm used by Cadinu and Rothbart (1996) does not fully match the minimal intergroup setting that it claims to explain. First, participants were provided with explicit and differentiated information about one of the two groups involved in the novel intergroup setting. Second, self-ratings were obtained after the social categorisation was introduced. Thus, it is possible that self-ratings were already affected by social categorisation; they were ratings on the self as a group member. To address this concern, Otten (1999) conducted a modified replication of the third experiment by Cadinu and Rothbart (1996), which resembled more closely the typical minimal intergroup setting. Participants lacked the critical ingroup and outgroup trait information, and care was taken that self-evaluations referred to the individual self rather than to the self as a group member. In addition, ingroup and outgroup judgements were obtained as repeated measures, thus also allowing for the analysis of relative group judgements. Intergroup judgements were varied such that either ingroup or outgroup evaluations were obtained first. Replicating the findings by Cadinu and Rothbart, ingroup ratings were most positive when they were given immediately after self-evaluations, that is, when the self was highly accessible. At the same time, outgroup judgements were least positive when they preceded both ingroup judgements and self-evaluations. Correspondingly, there were two cells with significant ingroup distinctiveness, the one driven by favourable ingroup evaluations, the other by unfavourable outgroup ratings. In addition, the correlation between selfratings and ingroup ratings, controlling for outgroup ratings, was significant, whereas there was no substantial correlation between self- and outgroup ratings, when controlling for ingroup ratings. Taken together, these findings imply that self-anchoring and differentiation might contribute independently to positive ingroup distinctiveness in the MGP, and that these two processes are elicited by different contextual conditions. While self-anchoring is supported by increasing the situational salience of the self (see also Vorauer & Ross, 1999), differentiation increases with the salience of the intergroup situation (which tends to be higher when group members focus on the outgroup rather than on the ingroup; see Haslam, Oakes, Turner, & McGarty, 1995). Egocentric social projection. The more general principle behind the selfanchoring process as described by Cadinu and Rothbart (1996) has been labelled egocentric social projection. The concept of the self as a judgemental anchor or as a crucial source of information in order to generate judgements about others (consensus estimates on attitudes,
FAVOURITISM TOWARDS MINIMAL INGROUPS 17
behavioural intentions, traits) is central in the work by Krueger and collaborators (e.g., Clement & Krueger, 2000; Krueger, 1998; Krueger & Clement, 1996). Krueger (1998) assumes that characteristics of own groups can be inferred by projection from the self. As social categorisation provides group members with the information that they are more similar to ingroup members than to outgroup members, they “have reasons to believe that ingroup members share their own responses more than outgroup members do”(pp. 221–222). Hence, there is more egocentric projection to ingroups than to outgroups (Krueger & Clement, 1996; Krueger & Zeiger, 1993; Mullen, Dovidio, Johnson, & Copper, 1992). The projection process supports the salience of social categorisation (Turner et al., 1987), as it simultaneously increases intragroup similarities, on the one hand, and intergroup differences, on the other hand. Krueger (1998) explicitly applies his model to the case of minimal groups. He claims that the ingroup-outgroup asymmetry in projection helps to create intergroup differences “where none exist” (p. 228). As people typically describe themselves positively, projection fosters positive ingroup descriptions, while outgroup descriptions—lacking the “benefits of projection” (p. 228)–remain neutral. Clement and Krueger (1997, cited by Krueger, 1998) reported evidence for asymmetric egocentric projection as a source for positive ingroup distinctiveness; they found that, when controlling for self-ratings, relative ingroup favouritism decreased considerably. It is important to note that Krueger (1998), unlike Cadinu and Rothbart, assumes only one process, an egocentric projection from the positive self to the ingroup, to account for ingroup favouritism and positive ingroup distinctiveness in the MGP. This viewpoint is fully compatible with Brewer’s (1979) observation that significant differences in the treatment of ingroup and outgroup variations in typical minimal group experiments typically rely on ingroup favouritism rather than outgroup derogation. However, the egocentric projection model differs from social identity theory’s perspective on the processes operating in the MGP: According to Krueger, a comparison between ingroup and outgroup is no longer a necessary condition for ingroup favouritism to emerge. It is merely the ingroup’s access to the “benefits” of egocentric social projection, thus, an intragroup process, that is assumed to be the mechanism underlying positive ingroup distinctiveness. Self-ratings and intergroup ratings: Evidence from intra-individual regression analyses. Otten and Wentura (2001) provided further evidence for the relevance of self-evaluations in intergroup judgements. They measured not only how strongly self-ratings predict intergroup ratings, but also compared the impact of a self-ingroup similarity heuristic and a mere valence-based heuristic on ingroup judgements. Embedded in a MGP, after learning about their novel category membership participants rated the
18
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
degree to which positive and negative traits (10 traits, each) applied to the ingroup versus the outgroup (on bipolar scales ranging from “fully applies to group A” to “fully applies to group B”). Ingroup favouritism could be realised by assigning positive traits predominantly to the ingroup and negative traits predominantly to the outgroup. The same trait dimensions that were used for the intergroup evaluations had been embedded in a selfevaluation questionnaire that participants completed at the very beginning of the experiment, before the social categorisation was introduced. With these data, it was possible to compare the similarity of profiles for self- and intergroup ratings (see Figure 4 for two hypothetical examples). Analyses of intra-individual regressions revealed that self-ratings predicted intergroup differentiation in favour of the ingroup more powerfully than a mere valence-based heuristic. In addition, there was a significant valence by self-rating interaction. Self-ratings on positive traits predicted relative ingroup ratings more strongly than self-ratings on negative traits. However, in both valence conditions the self was a significant predictor and there was significant ingroup favouritism (i.e., positive traits were assigned more strongly to the ingroup, whereas negative traits were assigned more strongly to the outgroup). Hence, the present data suggest that egocentric projection, though differing in strength, operates not only in group judgements on positive, but also on negative comparison dimensions.4 A similar point was recently made by Gramzow, Gaertner, and Sedikides (2001) who studied the impact of the self-ingroup link on intergroup memory. Fully in line with the Otten and Wentura findings, they present evidence supporting their view that information processing about novel ingroups is not simply based on the valence of the information available, but also on its self-congruency. In sum, research on egocentric projection in social categorisation and on self-anchoring in intergroup contexts strongly supports the assumption that the self is an influential, and perhaps even a pivotal, source for favourable ingroup judgements. At the same time, the existence of an ingroupoutgroup asymmetry in social projection, as well as the findings by Otten (1999), imply that, at least under certain conditions, assimilating self and ingroup can be sufficient to establish positive ingroup distinctiveness. An explicit process of social comparison and differentiation between ingroup and outgroup can be seen as an alternative rather than an additive means to this end. Krueger (1998) emphasised that when lacking other information, relying on self information in order to make estimates about others can successfully increase accuracy in social judgement. From this viewpoint, the role of self in the process of social categorisation and ingroup favouritism (both absolute and relative) is quite different from the self-esteem hypotheses as derived from social identity theory (see Hogg & Abrams, 1990; Rubin & Hewstone, 1998): Rather than motivating or profiting from
FAVOURITISM TOWARDS MINIMAL INGROUPS 19
Figure 4. Ingroup evaluation and self-ratings for two fictious participants, (a) and (b) (adapted from Otten & Wentura, 2001, with permission).
a positive ingroup description, the self provides a source ofinformation that can be used to give meaning to the novel ingroup. As indicated by the findings of Otten and Wentura (2001), the application of the self as a heuristic for ingroup evaluation does not function unconditionally, but can differ according to the self-aspects involved (see Simon & Hastedt, 1999). The finding that the self more strongly predicted positive than negative intergroup trait evaluations can imply both differences in the degree of
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
20
“fit” of the self as prototypical exemplar of the novel group (as argued by Reynolds et al., 2000) and differences in the certainty with which positive and negative attributes are assigned to the self. Self-anchoring in ingroup judgements as a functionof information processing As already mentioned above, Hogg and collaborators (e.g., Hogg, 2001; Hogg & Abrams, 1993) posit that ingroup favouritism and positive ingroup distinctiveness in the MGP are (at least partly) motivated by the goal of reducing judgemental uncertainty (endeavouring to give meaning to a novel, ill-defined social category). A series of experiments support this model: Situational manipulations of uncertainty affected ingroup identification and relative ingroup favouritism in the minimal intergroup setting (see Grieve & Hogg, 1999; Hogg, 2001; Hogg & Mullin, 1999). Consistent with this reasoning, Shah, Kruglanski, and Thompson (1998) demonstrated that ingroup identification and group evaluation varied as a function of individual’s need for cognitive closure (defined as people’s motivation to end up with quick and firm judgements as opposed to judgemental ambiguity; see Kruglanski, 1996; Kruglanski & Webster, 1996). However, irrespective of this body of empirical evidence, we are still left with a crucial question. How can identification with a novel ingroup function as a means to reduce uncertainty, as long as these ingroups lack definition and meaning? Hence, the assumption that ingroup favouritism in the MGP is guided by a search for meaning and cognitive structure, returns us to the above argument about the self as an informational source for ingroup definition. Clement and Krueger (2000; see also Krueger, 1998) point out that the process of egocentric projection is applicable with little effort and requires little cognitive mediation. Therefore, integrating the idea of egocentric social projection with cognitive efficiency explanations, we can expect that generalising from self to novel ingroup is a process that varies according to information-processing modes. In this context, a study by Forgas and Fiedler (1996, experiment 3) is noteworthy. They manipulated mood to elicit either heuristic information processing (positive mood) or systematic information processing (negative mood). Interestingly, they found that in a typical minimal group situation, where participants were given no 4
Thus, our data are not fully in line with the reasoning suggested by Reynolds et al. (2000). As an account for the positive-negative asymmetry in social discrimination, they suggested that negative traits typically do not “fit” self-ratings. Consequently, intergroup comparisons on these dimensions would be rather meaningless and should, therefore, not be subject to intergroup differentiation.
FAVOURITISM TOWARDS MINIMAL INGROUPS 21
information as to whether the categorisation criterion related systematically to any other characteristics, positive mood resulted in both strongest evaluative ingroup bias and strongest similarity between self and ingroup ratings. Bar-Tal and collaborators argue that efficient cognitive structuring requires not only the activation of a need for cognitive structure,5 but also the subjective ability to achieve cognitive structure (Bar-Tal, 1994; Bar-Tal, Kishon-Rabin, & Tabak, 1997). In support of this claim, a series of experiments revealed that reliance on heuristics, like schemata or stereotypes, was strongest when both need and ability for cognitive structure were high (e.g., Bar-Tal & Guinote, 2002; Bar-Tal et al., 1997). Consequently, Otten and Bar-Tal (2002) tested whether both factors also affected the use of the self as a heuristic for ingroup judgements in a minimal group situation. As expected, the regression weight for self-ratings as a predictor of ingroup ratings was strongest in the condition maximising the probability of heuristic processing, namely when participants were both motivated and able to arrive quickly at concrete judgements. Again, no similar effects were found for outgroup evaluations. In sum, the empirical evidence gives rise to the assumption that epistemic motivations and other situational and individual determinants of information processing affect how strongly people rely on the self as a means to give meaning to a novel ingroup. Returning to the beginning of this section, one could speculate that uncertainty reduction in the MGP is— at least in part— mediated by self-anchoring. However, further empirical tests are required to test this hypothesis directly. IMPLICATIONS AND LIMITATIONS At the beginning of this chapter, I questioned the assumption that ingroup favouritism and positive ingroup distinctiveness in the MGP necessarily require a comparison with and explicit differentiation from the corresponding outgroup. The research programme presented here, which integrated various theoretical and empirical approaches to intergroup behaviour and social cognition, supports this hypothesis: There is evidence that positive affect towards a new ingroup can arise as an immediate consequence of social categorisation, irrespective of explicit intergroup comparisons. Moreover, the given evidence, not only from own studies, but also from other research in this field, indicates that the self is used as referent to define a novel, minimal ingroup. In the following, I will discuss in more detail both implications and limitations of the above findings for our understanding of the processes that contribute to ingroup favouritism and positive ingroup distinctiveness. In addition, some perspectives for future research will be outlined.
22
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Ingroup versus outgroup evaluation: Autonomoussources of positive ingroup distinctiveness As already stated by Allport (1954) and as recently reaffirmed by Brewer (1999), a positive attitude towards a novel ingroup can emerge independently of a negative attitude towards the corresponding outgroup. As will be outlined in more detail below, this might be due to different processes underlying ingroup and outgroup treatment. In the scope of the present chapter, quite consistently, and over various domains of research, there was hardly any empirical evidence for complementary processes affecting ingroup and outgroup evaluations in the MGP. First, while there is a positive default towards the ingroup, affect towards the novel outgroup is neutral rather than negative (Otten & Moskowitz, 2000; Otten & Wentura, 1999; Perdue et al., 1990). Second, while the self is strongly correlated with ingroup judgements, it is basically unrelated to outgroup judgements; correspondingly, only the ingroup judgement profits from a previous focus on self-evaluation (Cadinu & Rothbart, 1996; Krueger, 1998; Otten, 1999). Third, only ingroup evaluations were affected by situational manipulations of heuristic information processing (Otten & Bar-Tal, 2002). When judging ingroups, group members can rely on two sources: intragroup comparison (me versus the others), or intergroup comparison (us versus them). Features of the situational context define which of the levels of categorisation will prevail (Turner et al., 1987), and sometimes the two levels interact (see also L.Gaertner, Sedikides, & Graez, 1999; Simon, 1993; Simon & Hastedt, 1999; Simon, Pantaleo, & Mummendey, 1995). In the framework of this chapter we can state that both self-anchoring and social comparisons between ingroup and outgroup may underlie positive ingroup distinctiveness. This thought is expressed schematically in Figure 5 (the dashed rather than solid arrow from intergroup differentiation to ingroup favouritism simply expresses that this path was not tested in the research programme summarised in this chapter; however, it is well documented in the intergroup literature; see e.g., Brewer & Brown, 1998; Brown, 2000; Hewstone, Rubin, & Willis, 2002). Different from the ingroup, the outgroup evaluation is determined by the salience of the intergroup distinction and by further contextual and motivational determinants of intergroup differentiation, exclusively. Both processes can contribute to the phenomenon of positive ingroup distinctiveness, either separately or in conjunction.
5 Note, that in the context of this article the terms “need for cognitive closure” and “need for cognitive structure” are used as equivalents. Nonetheless, we are aware of the fact that the two terms stem from slightly different theoretical models (see Bar-Tal et al., 1997, p. 1158).
FAVOURITISM TOWARDS MINIMAL INGROUPS 23
Figure 5. A dual process model of positive ingroup distinctiveness.
However, for the case of minimal groups I propose to make an even stronger point, as favourable ingroup judgements rather than unfavourable outgroup judgements typically account for significant ingroup-outgroup differences (e.g., Brewer, 1979; Brewer & Brown, 1998). We do need to implement social categorisation, thus, there needs to be the notion of an “us” as opposed to a “them”, for the sequence to be launched. Then, however, projection from self to ingroup can be the main if not only source of favourable ingroup ratings. Therefore, when the overall salience of the intergroup distinction is not strong, positive ingroup distinctiveness, i.e. deviation from the outgroup, might be a statistical rather than a psychological phenomenon: it is only the ingroup that matters. But what about the oppositeness heuristic as postulated by Cadinu and Rothbart (1996)? In their studies they found a linear trend, such that the higher the ingroup ratings that were given by the experimenters as comparative information, the lower the participants’ subsequent outgroup judgements. However, when no fixed information about the ingroup is provided (as in the MGP), there is little if any evidence for such negative interdependence between ingroup and outgroup evaluations (see Brewer, 1999). Rather, when both ingroup and outgroup were defined only on the categorisation dimension, a positive correlation emerged (Otten, 1999). Accepting and identifying with a social categorisation implies that—at least on the categorisation dimension—ingroup and outgroup are defined as different from each other, but they are not necessarily contrasted as opposites. Implications regarding social identity theory’sexplanation for the “mere categorisation effect” This chapter has summarised mechanisms supporting favourable ingroup evaluations in the MGP that neither necessitate social comparisons between
24
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
ingroup and outgroup nor a motivation to establish positive ingroup distinctiveness. However, as Figure 5 implies, this does not mean that there are no instances where social comparisons (guided by striving for positive social identity) are crucial constituents for significant and ingroupfavouring differentiation between groups, as originally assumed by social identity theory (Tajfel & Turner, 1979). Whether self-anchoring or rather intergroup comparisons will be major sources of positive ingroup distinctiveness in the MGP is assumed to be affected by the situational salience of either the self within the group or rather the ingroup in relation to the outgroup. Intergroup allocations versus evaluations. The studies summarised in this chapter focused on intergroup evaluations, exclusively. However, the original studies on the mere categorisation effect referred to intergroup allocations (typically of positive resources). Thus, the question arises of whether and how the self-anchoring process is similarly applicable to the favourable and distinct treatment of ingroups in intergroup allocations.6 Two approaches seem plausible and could be tested in future research: First, one might argue that a similarity between self and ingroup might apply to relative ingroup favouritism in both intergroup evaluations and intergroup allocations. Second, one could link the idea of self-anchoring to recent work on intergroup allocations by L.Gaertner and Insko (2000), who showed that an expectation of reciprocity is an important determinant of ingroup favouritism in the MGP (see also Locksley, Ortiz, & Hepburn, 1980). Lacking any further information, the expectation that fellow ingroup members will reciprocate own allocation behaviour might also rely on the construal of similarity between self and the novel ingroup. In conclusion, an interesting empirical endeavour would be to test whether perceived self-ingroup similarity (as indicator of self-anchoring) affects expectations of reciprocity, and, as a result, ingroup-favouring allocations. Maximising intergroup differences. However, the above reasoning still leaves one phenomenon unexplained: The strongest argument for a positive ingroup distinctiveness striving stems from the observation that group members maximised the difference between ingroup and outgroup at the expense of maximising their ingroup’s profit (see e.g., Brewer, 1979; Brown, 2000). Can these findings be reconciled with the self-anchoring approach? Considering the dual process model of positive ingroup distinctiveness as outlined in Figure 5, we can give an affirmative answer. The model assumes that when the salience of the ingroup-outgroup distinction increases, differentiating ingroup and outgroup will become a more relevant source of positive ingroup distinctiveness. Exactly this was found by Otten (1999, see above), who showed that there were two pathways to positive ingroup distinctiveness, either an especially favourable ingroup judgement supported by previous self-ratings, or a below-average outgroup judgement, when the judgemental focus was initially on the
FAVOURITISM TOWARDS MINIMAL INGROUPS 25
outgroup. Obviously, the allocation matrices used in the original MGP studies, which force participants to decide simultaneously about ingroup and outgroup treatment, increase awareness about the outgroup. Thus, we can assume a rather high salience of the intergroup level of selfcategorisation (see also Diehl, 1990) and, accordingly, a strong impact of the lower path in the dual process model depicted in Figure 5. In conclusion, an important endeavour for further research in this domain is to systematically differentiate and vary minimal group experiments with regards to the salience of the intra- or intergroup level of self-categorisation. Here, as already argued by Deschamps and Devos (1998), one should consider an orthogonal variation of these factors rather than treating them as opposing poles on a continuum. Self-anchoring and the content of group evaluation We have already discussed in detail that inferences from the self to (minimal) social groups are target-specific: They apply to the ingroup, but not to the outgroup. But even for novel ingroups, self-anchoring is not assumed to be a universal process, but will be moderated by the content of generalisation, that is, by the quality of dimensions on which social judgements are made. The study by Otten and Wentura (2001) revealed that valence had an impact on self-anchoring; although significant in both cases, the regression weights for self-ratings on intergroup ratings were stronger for positive rather than for negative traits (see also Davis, Conklin, Smith, & Luce, 1996). On the one hand, one could read this finding as tentative evidence that self-anchoring is affected by the self-enhancement motive: only extending positive characteristics of the individual self to the collective self “pays” in this respect. On the other hand, due to the positivity bias in selfperception (e.g., Baumeister, 1998), there should be a higher probability that positive traits will be both accessible and perceived as appropriate for generalisations from the self to the ingroup as a whole (see also Reynolds et al., 2000). In addition to valence, trait dimensions vary with regard to their typicality, that is, with regard to their applicability not only to the self, but
6 Note, however, that according to a social identity perspective on the findings in the MGP no distinction needs to be made between intergroup allocations and evaluations. Rather, according to Turner (1978) allocation decisions have an impact on social identity only due to their evaluative implications: the differential allocation of resources is seen as a means to establish differences on positively valued dimensions. However, Otten, Mummendey, and Blanz (1995) report very low correlations between subsequently measured intergroup allocations and evaluations.
26
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
to a larger proportion of people (see Gramzow et al., 2001). In selfdescriptions both positive and typical traits are predominant. Correspondingly, Gramzow and collaborators (2001) provided evidence that especially those characteristics that are both positive and typical are integrated into impressions about novel ingroups. Finally, Simon and Hastedt (1999) proposed that the subjective importance of trait dimensions can determine the links between personal self and representations of the ingroup. In their self-aspect model they posit that positive and important aspects of the individual self form the basis for the construction of the collective self. Minimal versus real group settings The main aim of this chapter was to offer an alternative account for favouritism towards minimal groups. Correspondingly, the majority of studies reported above referred to laboratory rather than real groups. What about the relevance of these findings with regard to real groups? In fact, one might argue that the specific conditions of the MGP maximise the probability that the self is used as a heuristic to define the novel ingroup (as has been suggested by Reynolds et al., 2000). As long as the ingroup is only defined with regard to the categorisation criterion, its prototype will most probably be derived from the self, which is not only the sole exemplar of the group available, but also more generally a preferred standard in social judgement (e.g., Dunning & Hayes, 1996; Krueger, 1998). Thus, in the MGP the process of “depersonalisation” (Turner et al., 1987), if applicable at all, takes a very specific form: Personal characteristics of the self are extended to the ingroup as a whole, thereby giving them a social rather than a unique meaning. Certainly more research is needed to clarify the role of self-anchoring in ingroup and intergroup judgement for natural groups. Recently we have begun to analyse self, ingroup, and outgroup judgements at different measurement points during the formation of new groups (Eisenbeiss & Otten, 2002). In addition to other factors determining group socialisation (e.g., Levine & Moreland, 1994) self-anchoring and depersonalisation will be examined as determinants of ingroup identification, ingroup favouritism, and intergroup differentiation in a longitudinal design. It is assumed that in early stages of group formation the self will be a predictor for both ingroup and intergroup judgement. However, in the long run, after a stable ingroup stereotype has emerged, the impact of similarities between personal self and ingroup on relative ingroup preference and intergroup differentiation should decrease.
FAVOURITISM TOWARDS MINIMAL INGROUPS 27
SUMMARY AND CONCLUSION Social categorisation substantially contributes to the meaning with which we imbue our social reality (Tajfel, 1978), and people are motivated to define the novel social categories to which they belong (Hogg & Abrams, 1993; Turner et al., 1987). Moreover, social identity theory (Tajfel & Turner, 1979) assumes that this meaning is acquired by locating the ingroup in relation to the outgroup, a process that is affected by a striving for positive social identity. Thus, classical accounts tackling the question of why even a minimal ingroup can become not only positive, but also positively distinct from the corresponding outgroup, focus on intergroup comparison. However, the present research programme demonstrates that a positive attitude towards novel ingroups can emerge not only from intergroup contrasts, but also from the intragroup assimilation between ingroup and individual self. An integrative summary of the present research programme reads as follows: Forming a positive attitude towards a novel ingroup is a spontaneous and relatively automatic process (Otten & Moskowitz, 2000; Otten & Wentura, 1999). Moreover, there is evidence that initial positive ingroup attitudes are based on a close association between (positive) self and ingroup. Evidence for such link has been provided both on implicit measures (Epstude & Otten, 2000; Smith et al., 1999; Smith & Henry, 1996) and on explicit measures (Otten, 1999; Otten & Wentura, 2001). Finally, data indicate that application of the self as a heuristic for ingroup evaluation is contingent upon both situational factors (relevance of categorisation; availability of information about the group, etc.) and processing conditions (such as mood, and both need for and ability to achieve cognitive structure; see Otten & Bar-Tal, 2002). In conclusion, the individual self can be seen as an important factor contributing to the perception of ingroups as positively distinct; as Krueger (1998) put it: “…, ingroup favoritism is not only ethnocentric, but also egocentric in nature” (p. 228). Correspondingly, the findings can be read as further support for a primacy of the individual self in social judgement (e.g., Dunning & Hayes, 1996; L.Gaertner et al., 1999; Sedikides & Skowronski, 1993; Simon, 1993; Simon et al., 1995). However, considering the dynamic nature of the social self (see Turner & Onorato, 1999), rather than talking about “primacy”, one might better refer to the rich interplay between representations of individual and collective self (see Sedikides & Brewer, 2001). Individuals who are categorised and who form impressions about their novel ingroup can use both group-level information (as far as it is available in the given social context) and their knowledge about relevant personal characteristics to create their—positive—collective identity.
28
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
REFERENCES Aberson, C.L., Healy, M., & Romero, V. (2000). Ingroup bias and self-esteem: A metaanalysis. Personality and Social Psychology Review, 3,157–173. Abrams, D., & Hogg, M.A. (1988). Comments on the motivational status of selfesteem in social identity and intergroup discrimination. British Journal of Social Psychohgy, 18,317– 334. Allport, G.W. (1954). The nature of prejudice.Reading, MA: Addison-Wesley. Aron, A., Aron, E.N., & Smollan, D. (1992). Inclusion of other in the self scale and the structure of interpersonal closeness. Journal of Personality and Social Psychology, 63,241– 253. Aron, A., Aron, E.N., Tudor, M., & Nelson, G. (1991). Close relationships as including other in the self. Journal of Personality and Social Psychology, 60, 241–253. Aron, A., & Fraley, B. (1999). Relationship closeness as including other in the self: Cognitive underpinnings and measures. Social Cognition, 17,140–160. Bargh, J.A. (1997). The automaticity of everdyday life. In R.S.Wyer (Ed.), Advances in socialcognition(Vol. 10, pp. 1–61). Mahwah, NJ: Lawrence Erlbaum Associates Inc. Bar-Tal, Y. (1994). The effect of need and ability to achieve cognitive structure on mundane decision-making. European Journal of Personality, 8,45–58. Bar-Tal, Y., & Guinote, A. (2002). Who exhibits more stereotypical thinking? The effect of need and ability to achieve cognitive structure on stereotyping. European Journal of Psychology,16,313–331. Bar-Tal, Y., Kishon-Rabin, L., & Tabak, N. (1997). The effect of need and ability to achieve cognitive structuring on cognitive structuring. Journal of Personality and Social Psychology,73,1158–1176. Baumeister, R.F. (1998). The self. In D.T.Gilbert, S.T.Fiske, & G.Lindzey (Eds.), Thehandbook of social psychology(Vol. 1, pp. 680–740). New York: McGrawHill. Blanz, M., Mummendey, A., & Otten, S. (1995). Positive-negative asymmetry in social discrimination: The impact of stimulus-valence, size- and statusdifferentials in intergroup evaluations. British Journal of Social Psychology, 34, 409–419. Brewer, M.B. (1979). In-group bias in the minimal intergroup situation: A cognitivemotivational analysis. Psychological Bulletin, 86,307–324. Brewer, M.B. (1993). The role of distinctiveness in social identity and group behaviour. InM. A.Hogg & D.Abrams (Eds.), Group motivation: Social psychological perspectives (pp. 1– 16). London: Harvester Wheatsheaf. Brewer, M.B. (1999). The psychology of prejudice: Ingroup love or outgroup hate? Journal ofSocial Issues,55, 429–444. Brewer, M.B., & Brown, R.J. (1998). Intergroup relations. In D.T.Gilbert, S.T.Fiske & G. Lindzey (Eds.), The handbook of social psychology(Vol. 2, pp. 554–594). New York: McGraw-Hill. Brown, R. (2000). AGENDA 2000-Social Identity Theory: Past achievements, current problems and future challenges. European Journal of Social Psychology, 30,745–778.
FAVOURITISM TOWARDS MINIMAL INGROUPS 29
Cadinu, M.R., & Rothbart, M. (1996). Self-anchoring and differentiation processes in the minimal group setting. Journal of Personality and Social Psychology, 70, 661–677. Clement, R.W., & Krueger, J. (1997). Estimates of social consensus by majorities and minorities: The case for social projection. Personality and Social Psychology Review, 1,299– 313. Clement, R.W., & Krueger, J. (2000). The primacy of self-referent information in perception of social consensus. British Journal of Social Psychology, 39,279– 299. Coats, S., Smith, E., Claypool, H.M., & Banner, M.J. (2000). Overlapping mental representations of self and in-group: Reaction time evidence and its relationship with explicit measures of group identification. Journal of Experimental Social Psychology, 36,304–315. Davis, M.H., Conklin, L., Smith, A., & Luce, C. (1996). Effect of perspectivetaking on the cognitive representations of persons: A merging of self and other. Journal of Personality andSocial Psychology, 70,713–726. Deschamps, J.C., & Devos, T. (1998). Regarding the relationship between social and personal identity. In S.Worchel, J.F.MoralesD.Páez, & J.-C.Deschamps. Social identity.International perspectives(pp. 1–12). London: Sage. Diehl, M. (1990). The minimal group paradigm: Theoretical explanations and empirical findings. In W.Stroebe & M.Hewstone (Eds.), European review of social psychology(Vol. 1, pp. 263–292). Chichester, UK: John Wiley & Sons. Diener, E., & Diener, C. (1996). Most people are happy. Psychological Science, 7, 181–185. Draine, S.C, & Greenwald, A.G. (1998). Replicable unconscious semantic priming. Journal ofExperimental Psychology: General, 127,286–303. Dunning, D., & Hayes, A.F. (1996). Evidence for egocentric comparison in social judgment. Journal of Personality and Social Psychology,71, 213–229. Eisenbeiss, K., & Otten, S. (2002). The role of self-anchoring in group formation: A longitudinal study. Manuscript in preparation. Epstude, K., & Otten, S. (2000). The ingroup as part of the self or the self as part of the ingroup?Response time evidence for self-anchoring and self-stereotyping in realistic intergroup contexts.Paper presented at the 3rd Jena Workshop on Intergroup Processes, Jena, June 28-July 1. Fazio, R.H., Sanbonmatsu, D.M., Powell, M.C., & Kardes, F.R. (1986). Attitudes and social cognition: On the automatic activation of attitudes. Journal of Personality and SocialPsychology, 50,229–238. Feys, J. (1991). Briefly induced belongingness to self and preference. European Journal of SocialPsychology,21, 547–552. Feys, J. (1995). Mere ownership: Affective self-bias or evaluative conditioning? EuropeanJournal of Social Psychology, 25,559–575. Forgas, J.P., & Fiedler, K. (1996). Us and them: Mood effects on intergroup discrimination. Journal of Personality and Social Psychology, 70,28–40. Gaertner, L., & Insko, C.A. (2000). Intergroup discrimination in the minimal group paradigm: categorization, reciprocation, or fear?Journal of Personality and Social Psychology, 7,77– 94.
30
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Gaertner, L., & Schopler, J. (1998). Perceived ingroup entitativity and intergroup bias: An interconnection of self and others. European Journal of Social Psychology, 28,963–980. Gaertner, L., Sedikides, C., & Graetz, K. (1999). In search of self-definition: Motivational primacy of the individual self, motivational primacy of the collective self, or contextual primacy?Journal of Personality and Social Psychology, 76,5–18. Gaertner, S.L., Dovidio, J.F., Anastasio, P.A., Bachman, B.A., & Rust, M.C. (1993). The common ingroup identity model: Recategorization and the reduction of intergroup bias. In W.troebe & M.Hewstone (Eds.), European review of social psychology(Vol. 4, pp. 1–26). Chichester, UK: John Wiley & Sons Ltd. Gardham, K., & Brown, R. (2001). Two forms of intergroup discrimination with positive and negative outcomes: Explaining the positive-negative asymmetry effect. British Journal ofSocial Psychology, 40,23–34. Gramzow, R.M., Gaertner, L., & Sedikides, C. (2001). Memory for in-group and out-group information in a minimal group context: The self as informational base. Journal ofPersonality and Social Psychology, 80,188–205. Grieve, P.G., & Hogg, M.A. (1999). Subjective uncertainty and intergroup discrimination in the minimal group situation. Personality and Social Psychology Bulletin, 25,926–940. Haslam, S.A., Oakes, P.J., Turner, J.C, & McGarty, C. (1995). Social categorization and group homogenity: Changes in the perceived applicability of stereotype content as a function of comparative context and trait favourableness. British Journal of SocialPsychology, 34,139–160. Hewstone, M., Rubin, M., & Willis, H. (2002). Intergroup bias. Annual Review of Psychology,53,575–604. Hinkle, S., & Brown, R.J. (1990). Intergroup comparisons and social identity: Some links and lacunae. In D.Abrams & M.Hogg (Eds.), Social identity theory: Constructive and criticaladvances(pp. 48–70). Hemel Hempstead, UK: Wheatsheaf. Hogg, M.A. (2001). Subjective uncertainty reduction through self-categorization: A motivational theory of social identity processes. In W.Stroebe & M.Hewstone (Eds.), European review of social psychology(pp. 223–255). Chichester, UK: Wiley. Hogg, M.A., & Abrams, D. (1990). Social motivation, self-esteem and social identity. In D. Abrams & M.A.Hogg (Eds.), Social identity theory. Constructive and critical advances(pp. 28–47). New York: Harvester Wheatsheaf. Hogg, M.A., & Abrams, D. (1993). Towards a single-process uncertainty reduction model of social motivation in groups. In M.A.Hogg & D.Abrams (Eds.), Group motivation. Socialpsychology perspectives.Hemel Hempstead, UK: Harvester Wheatsheaf. Hogg, M.A., & Mullin, B.A. (1999). Joining groups to reduce uncertainty: Subjective uncertainty reduction and group identification. In D.A.Abrams & M.A.Hogg (Eds.), Social identity and social cognition(pp. 249–279). Oxford: Blackwell. Hoorens, V., & Nuttin, J.M. (1993). Overvaluation of own attributes: Mere ownership or subjective frequency? Social Cognition, 11,177–200.
FAVOURITISM TOWARDS MINIMAL INGROUPS 31
Howard, J.W., & Rothbart, M. (1980). Social categorization and memory for ingroup and out-group behavior. Journal of Personality and Social Psychology, 38,301–310. Krueger, J. (1998). Enhancement bias in the description of self and others. Personality andSocial Psychology Bulletin, 24,505–516. Krueger, J., & Clement, R.W. (1996). Inferring category characteristics from sample characteristics: Inductive reasoning and social projection. Journal of ExperimentalPsychology: General, 128,52–68. Krueger, J., & Zeiger, J.S. (1993). Social categorization and the truly false consensus effect. Journal of Personality and Social Psychology, 65,670–680. Kruglanski, A.W. (1996). Motivated social cognition: Principles of the interface. In E.Higgins & A.W.Kruglanski (Eds.), Social psychology: Handbook of basic principles(pp. 493–520). New York: Guilford Press. Kruglanski, A.W., & Webster, D.M. (1996). Motivated closing of the mind: “Seizing” and “freezing”. Psychological Review, 103,263–283. Levine, J.M., & Moreland, R.L. (1994). Group socialization: Theory and research. In W. Stroebe & M.Hewstone (Eds.), European review of social psychology (Vol. 5, pp. 305–336). London: Wiley. Locksley, A., Ortiz, V., & Hepburn, C. (1980). Social categorization and discriminatory behavior: Extinguishing the minimal intergroup discrimination effect. Journal of Personalityand Social Psychology, 39,773–783. Long, K.M., Spears, R., & Manstead, A.S.R. (1994). The influence of personal and collective self-esteem on strategies of social differentiation. British Journal of Social Psychology, 33,313–329. Maass, A., & Schaller, M. (1991). Intergroup biases and the cognitive dynamics of stereotype formation. In W.Stroebe, & M.Hewstone (Eds.), European review of social psychology(Vol. 2, pp. 189–209). Chichester, UK: John Wiley & Sons Ltd. Messick, D.M., & Mackie, D.M. (1989). Intergroup relations. Annual Review of Psychology,40, 45–81. Mullen, B., Brown, R., & Smith, C. (1992). Ingroup bias as a function of salience, relevance and status: An integration. European Journal of Social Psychology, 22, 422–404. Mullen, B., Dovidio, J.F., Johnson, C., & Copper, C. (1992). In-group and outgroup differences in social projection. Journal of Experimental Social Psychology, 28,103–122. Mummendey, A. (1995). Positive distinctiveness and intergroup discrimination: An old couple living in divorce. European Journal of Social Psychology, 25,657– 670. Mummendey, A., & Otten, S. (1998). Positive-negative asymmetry in social discrimination. In W.Stroebe & M.Hewstone (Eds.), European review of social psychology(Vol. 9, pp. 107– 143). New York: John Wiley & Sons Ltd. Mummendey, A., Otten, S., Berger, U., & Kessler, T. (2000). Positive-negative asymmetry in social discrimination: Valence of evaluation and salience of categorization. Personality andSocial Psychology Bulletin, 26,1258–1270. Nuttin, J.M. (1987). Affective consequences of mere ownership: The name letter effect in twelve European languages. European Journal of Social Psychology, 17,381–402.
32
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Oakes, P. (2001). The root of all evil in intergroup relations? Unearthing the categorization process. In R.Brown & S.Gaertner (Eds.), Blackwell handbook of social psychology:Intergroup processes(pp. 3–21). Malden, MA: Blackwell. Onorato, R.S., & Turner, J.C. (2001). The “I”, the “me” and the “us”: The psychological group and self-concept maintenance and change. In C.Sedikides & M.B.Brewer (Eds.), Individual self, relational self, collective self (pp. 147– 170). Philadelphia: Psychology Press. Otten, S. (1999). [Unpublished data.] University of Massachusetts. Otten, S., & Bar-Tal, Y. (2002). Self-anchoring in the minimal group paradigm: The impact of need and ability to achieve cognitive structure. Group Processes and Intergroup Relations, 5,267–284. Otten, S., & Moskowitz, G.B. (2000). Evidence for implicit evaluative in-group bias: Affect-biased spontaneous trait inference in a Minimal Group Paradigm. Journal of ExperimentalSocial Psychology, 36,77–89. Otten, S., & Mummendey, A. (1999). To our benefit or at your expense? Justice considerations in intergroup allocations of positive and negative resources. Social Justice Research, 12,19– 38. Otten, S., & Mummendey, A. (2000). Valence-dependent probability of ingroup favoritism between minimal groups: An integrative view on the positivenegative asymmetry in social discrimination. In D.Capozza & R.Brown (Eds.), Social identity processes(pp. 33–48). London: Sage. Otten, S., Mummendey, A., & Blanz, M. (1995). Different measures of social discrimination in laboratory settings: Doubts in validity are advisable. Revue Internationale de PsychologieSociale, 2,7–21. Otten, S., Mummendey, A., & Blanz, M. (1996). Intergroup discrimination in positive and negative outcome allocations: The impact of stimulus valence, relative group status and relative group size. Personality and Social Psychology Bulletin, 22,568–581. Otten, S., Mummendey, A., & Buhl, T. (1998). Accuracy in information processing and the positive-negative asymmetry in social discrimination. Revue Internationale de PsychologieSociale, 11,69–96. Otten, S., & Wentura, D. (1999). About the impact of automaticity in the Minimal Group Paradigm: Evidence from affective priming tasks. European Journal of Social Psychology, 29,1049–1071. Otten, S., & Wentura, D. (2001). Self-anchoring and ingroup favoritism: An individual-profiles analysis. Journal of Experimental Social Psychology, 37, 525–532. Park, B., & Judd, C.M. (1990). Measures and models of perceived group variablility. Journal ofPersonality and Social Psychology, 59,173–191. Perdue, C.W., Dovidio, J.F., Gurtman, M.B., & Tyler, R.B. (1990). Us and them: Social categorization and the process of intergroup bias. Journal of Personality and SocialPsychology, 59,475–486. Rabbie, J.M., & Horwitz, M. (1969). The arousal of ingroup-outgroup bias by a chance win or loss. Journal of Personality and Social Psychology, 69,223–228. Rabbie, J.M., Schot, J.C, & Visser, L. (1989). Social identity theory: A conceptual and empirical critique from the perspective of a behavioural interaction model. European Journalof Social Psychology, 19,171–202.
FAVOURITISM TOWARDS MINIMAL INGROUPS 33
Reynolds, K.J., Turner, J.C, & Haslam, S.A. (2000). When are we better than them and they are worse than us? A closer look at social discrimination in positive and negative domains. Journal of Personality and Social Psychology, 78,64–80. Rubin, M., & Hewstone, M. (1998). Social identity theory’s self-esteem hypothesis: A review and some suggestions for clarification. Personality and Social Psychology Review, 2,40–62. Schaller, M., & Maass, A. (1989). Illusory correlation and social categorization. Toward an integration of motivational and cognitive factors in stereotype formation. Journal ofPersonality and Social Psychology, 56,701–721. Sedikides, C., & Brewer, M.B. (Eds.) (2001). Individual self, relational self, collective self.Philadelphia, PA: Psychology Press. Sedikides, C, & Skowronski, J.J. (1993). The self in impression formation: Trait centrality and social perception. Journal of Experimental Social Psychology, 29,347–357. Shah, J., Kruglanski, A.W., & Thompson, E.P. (1998). Membership has its (epistemic) rewards: Need for closure effects on in-group bias. Journal of Personality and SocialPsychology, 75,383–393. Sherif, M. (1966). Group conflict and co-operation: Their social psychology.London: Routledge , & Kegan Paul. Simon, B. (1993). On the asymmetry in the cognitive construal of ingroup and outgroup: A model of egocentric social categorization. European Journal of Social Psychology, 23,131– 147. Simon, B., & Hastedt, C. (1999). Self-aspects as social categories: The role of personal importance and valence. European Journal of Social Psychology, 29, 479–487. Simon, B., Pantaleo, G., & Mummendey, A. (1995). Unique individual or interchangeable group member? The accentuation of intragroup differences versus similarities as an indicator of the individual self versus the collective self. Journal of Personality and Social Psychology,69,106–119. Smith, E.R. (1996). What do connectionism and social psychology offer each other?Journal ofPersonality and Social Psychology, 70,893–912. Smith, E.R. (2002). Overlapping mental represantations of self and froup: Evidence and implications. In J.P.Forgas & K.Williams (Eds.), The social self: Cognitive, Interpersonaland intergroup perspectives(pp. 21–35). Philadelphia: Psychology Press. Smith, E.R., Coats, S., & Walling, D. (1999). Overlapping mental representaions of self, in-group, and partner: Further response time evidence and a connectionist model. Personalityand Social Psychology Bulletin, 25,873–882. Smith, E.R., & Henry, S. (1996). An in-group becomes part of the self: Response time evidence. Personality and Social Psychology Bulletin, 20,635–642. Tajfel, H. (Ed.) (1978). Differentiation between social groups: Studies in the social psychology ofintergroup relations. London: Academic Press. Tajfel, H. (1982). Social psychology of intergroup relations. Annual Review of Psychology, 33,1–39. Tajfel, H., Billig, M.G., Bundy, R.P., & Flament, C. (1971). Social categorization and intergroup behaviour. European Journal of Social Psychology, 1,149–178.
34
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Tajfel, H., & Turner, J.C. (1979). An integrative theory of intergroup conflict. In W.G.Austin, & S.Worchel (Eds.), The social psychology of intergroup relations (pp. 33–47). Monterey, CA: Brooks/Cole Publishers. Tajfel, H., & Turner, J.C. (1986). The social identity theory of intergroup behavior. In S. Worchel & W.G.Austin (Eds.), Psychology of intergroup relations(pp. 7–24). Chicago: Nelson-Hall Publishers. Taylor, S.E., & Brown, J.D. (1988). Illusion and well-being: A social psychological perspective on mental health. Psychological Bulletin, 103,193–210. Tropp, L.R., & Wright, S.C. (2001). Ingroup identification as the inclusion of ingroup in the self. Personality and Social Psychology Bulletin, 27,585–600. Turner, J.C. (1978). Social categorization and social discrimination in the minimal group paradigm. In H.Tajfel (Ed.), Differentiation between social groups(pp. 101–140). London: Academic Press. Turner, J.C. (1999). Some current issues in research on social identity and selfcategorization theories. In N.Ellemers, R.Spears & B.Doosje (Eds.), Social identity: Context, commitment,content (pp. 6–34). Oxford: Blackwell. Turner, J.C., Hogg, M.A., Oakes, P.J., Reicher, S.D., & Wetherell, M.S. (1987). Rediscovering the social group. A self-categorization theory.Oxford: Basil Blackwell. Turner, J.C, Oakes, P.J., Haslam, S.A., & McGarty, C. (1994). Self and collective: Cognition and social context. Personality and Social Psychology Bulletin, 20, 454–463. Turner, J.C, & Onorato, R.S. (1999). Social identity, personality, and the selfconcept: A self-categorization perspective. In T.R.Tyler, R.M.Kramer & O.P.John (Eds.), Thepsychology of the social self (pp. 11–46). Mahwah, NJ: Lawrence Erlbaum Associates Inc. Turner, J.C., & Reynolds, K.J. (2001). The social identity perspective in intergroup relations: Theories, themes, and controversies. In R.Brown & S.Gaertner (Eds.), Blackwell handbookof social psychology: Intergroup processes(pp. 133– 152). Malden, MA: Blackwell. Uleman, J.S., Hon, A., Roman, R.J., & Moskowitz, G.B. (1996). On-line evidence for spontaneous trait inferences at encoding. Personality and Social Psychology Bulletin, 22,377–394. Vorauer, J.D., & Ross, M. (1999). Self-awareness and feeling transparent. Failing to suppress one’s self. Journal of Experimental Psychology, 35,415–440. Wittenbrink, B., Judd, C.M., & Park, B. (1997). Evidence for racial prejudice at the implicit level and its relationship with questionnaire measures. Journal of Personality and SocialPsychology, 72,262–274. Yzerbyt, V.-Y, Leyens, J.-P, & Schadron, G. (1997). Social judgeability and the dilution of stereotypes: The impact of the nature and sequence of information. Personality and SocialPsychology Bulletin, 23,1312–1322.
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY, 2002, 13, 35–73
A dual-route model of crossed categorisation effects Richard J.Crisp University of Birmingham, UK Nurcan Ensari University of Southern California, USA Miles Hewstone University of Oxford, UK Norman Miller University of Southern California, USA Categorisation is a pervasive psychological process that structures our social world and provides the basis for in-group favouritism. Crossed categorisation is the logical extension to theorising and research in this domain in its description and explanation of the processes and outcomes that occur when two or more dimensions of group membership become simultaneously salient or accessible. Building on previous theoretical, empirical, and meta-analytical work, we propose an integrative model of crossed categorisation effects. Adopting a “dual-route” perspective, we suggest that moderators of crossed category evaluations will produce specifiable outcome patterns via two mediational routes: affective and cognitive. We review recent empirical work within the framework of this model, finding general support for the proposed processes. Finally, we suggest refinements and outline possibilities and implications for the model and future work into crossed categorisation, intergroup relations, and social discrimination.
In recent years there has been an increasing interest in research on multiple social categorisation in intergroup relations (i.e., intergroup contexts where more than one basis for social categorisation is salient, accessible, or important at the same time). Such multiple systems of social classification
© 2002 European Association of Experimental Social Psychology http://www.tandf.co.uk/journals/pp/02699931.html DOI: 10.1080/10463280240000091
36
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
are important to social psychologists because they qualify and refine existing theoretical approaches to the study of social categorisation and intergroup relations, and they are arguably more realistic than conceptualisations of prejudice and discrimination based on simple ingroup versus out-group comparisons. In this chapter we will consider recent research that has helped us understand how social categories are combined, represented, and used in social perception. We will then consider the conclusions from recent reviews of the crossed categorisation literature, and on the bases of these reviews and recent work in the “simple” categorisation domain, we suggest an integrative model with which to interpret crossed categorisation effects. We will go on to examine empirical evidence for the model’s proposed antecedents and processes, before finally suggesting refinements and possible extensions to the model in future work on this developing and important field within social categorisation and intergroup relations. MULTIPLE CATEGORY USE IN SOCIALPERCEPTION Do people really use multiple social categories in person perception? Put another way, does crossed categorisation have any validity as a realistic reflection of intergroup relations? Although the collected literature on crossed categorisation has implicitly assumed that people can and do activate and use complex category structures (Crisp & Hewstone, 1999a; Migdal, Hewstone, & Mullen, 1998; Urban & Miller, 1998), this assumption has yet to be considered in detail. There is some evidence that multiple alternative social categories are, in fact, mutually exclusive, and that only single bases for classification are activated and utilised in social perception at any one time. Studies by Macrae, Bodenhausen, and Milne (1995) and Van Twuyver and van Knippenberg (1999) have suggested, in contrast to crossed categorisation work, that social classification is inherently competitive and mutually exclusive. Macrae et al. found that the activation of one social category (e.g., Chinese) led to facilitated activation of the associated stereotype, but inhibition of the stereotype of a relevant, but not activated, second category (e.g., Women). As well as this inhibition of category content (stereotypical attributes), Van Twuyver and van Knippenberg found Address correspondence to: RJ Crisp, School of Psychology, University of Birmingham, Edgbaston, Birmingham, B15 2TT, UK, Email:
[email protected] Preparation of this review was supported by an Economic and Social Research Council doctoral studentship (R00429534071) and British Academy (SG-31041) grant to R.J.Crisp, an Economic and Social Research Council grant (R000239382) to R.J.Crisp and M.Hewstone, and a National Science Foundation Grant SBR 931 9752 to N.Miller.
CROSSED CATEGORISATION 37
evidence supportive of “negative interdependence” (van Knippenberg & Dijksterhuis, 2000) in category use in a study employing the category confusion paradigm (Taylor, Fiske, Etcoff, & Ruderman, 1978). Van Twuyver and van Knippenberg found the use of a gender categorisation was negatively correlated with the use of a student/teacher categorisation. Such competitive activation of categories and stereotypes may be functional in many situations (i.e., enabling attentional focus to increase the efficiency of social perception; Dijksterhuis & van Knippenberg, 1998). However, although these findings by Macrae et al. (1995) and Van Twuyver and van Knippenberg (1999) suggest that one category will usually dominate over another in social perception, other work, using similar paradigms, has demonstrated that sometimes categories can be used simultaneously in both social and non-social perception. Deschamps (1977), and later Crisp and Hewstone (1999b), found that perceivers could, in fact, use multiple dimensions of categorisation in nonsocial perception. Crisp and Hewstone presented participants with eight squares varying in size. There were four between-participants conditions across which two dimensions of classification were varied. In the “simple” condition all the squares were the same colour but the smallest were labelled “A”and the largest were labelled “B”. In the “superimposed” condition colour co-varied perfectly with the letter labelling (i.e., all the small squares were green and all the large squares were blue). In the “crossed” condition, however, the colour was perfectly orthogonal to the letter labelling (i.e., half the “A”s were green and half were blue). Finally, in the “random” condition, the colour categorisation was unsystematically crossed with the letter labelling. In a comparison of the perceived interclass differences (the difference between the largest A and the smallest B), Crisp and Hewstone replicated Deschamps’ findings. In support of processes outlined by the category differentiation model (Doise, 1978), the difference between the two classes was greater in the superimposed than in the simple condition, and the difference between the two classes was smaller in the crossed condition than in the simple condition. In addition to Deschamps’ original findings of reduced interclass differentiation in the crossed condition, Crisp and Hewstone also found evidence for increased intraclass differentiation. The category differentiation model does indeed predict such an intraclass effect (the inverse of the interclass effect), so these findings add more support to the notion that in terms of basic-level cognitive operations, perceivers can process and use at least two dimensions of categorisation at the same time. Multiple category dimensions can also be used in social perception. Arcuri (1982) used the category confusion paradigm (Taylor et al., 1978) in an experiment similar to that of Van Twuyver and van Knippenberg (1999). Student participants were presented with a recording of a conversation between eight people, with the speaker’s face being presented
38
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
on a screen every time s/he spoke. Participants were required to match the individual sentences uttered with each of the eight speakers in a subsequent cued-recall test. Two categorical dimensions (the same ones as used by Van Twuyver & van Knippenberg; gender and academic status [student/ teacher]) were used to create three different experimental conditions: “simple” (four male and four female students); “superimposed” (four male students and four female teachers); and “crossed” (four students, two male and two female; and four teachers, two male and two female). By examining the inter- and intracategory errors on the matching task, Arcuri (1982) found a pattern consistent with multiple category use and Doise’s (1978) category differentiation model. Relative to the baseline level of simple categorisation, where intracategory errors were greater than intercategory errors, the difference between intra- and intercategory errors was greatest in the superimposed condition (where two categorisations coincided) and lowest in the crossed-category condition. In a recent set of studies also employing a memory-based paradigm, Crisp, Hewstone, and Cairns (2001a) and Crisp and Hewstone (2001; see also Crisp, 1998) have also found evidence that in real social contexts perceivers can and do use multiple bases for classification in social information processing. In these experiments Crisp and colleagues adapted a paradigm used by Park and Rothbart (1982) to allow subtle manipulation of multiple category memberships. Prior to a lecture, undergraduate participants were presented with a short paragraph ostensibly taken from a local newspaper and were told that the experimenter was interested in how people read such stories. The eight stories used concerned different positive (e.g., citizens’ awards) and negative (e.g., drunk driving) events and were counterbalanced over participants. After the participant had read the story and answered several filler questions, the experimenter left the room as if the “survey” were over. Forty-five minutes later (at the end of the lecture) the experimenter reappeared and gave the participants a surprise cued recall test in which they were required to recall some specific information regarding the character in the story (e.g., age, occupation). In line with Park and Rothbart (1982) this memory measure assessed recall of target person attributes. Crisp et al. (2001a) used this paradigm in a study in Northern Ireland and found that participants processed information, about females and males, and Catholics and Protestants, as an interactive function of the two category dimensions. Gender was specified by the use of relevant pronouns and name, and religion was also cued by name (e.g., Elizabeth Wriht [Protestant] versus Bridget O’Kane [Catholic]) —in this context names are a clear indicator of social category membership. We found that recall was influenced by particular combinations of category memberships rather than just an additive function of both independently. More specifically, a significant two-way interaction between gender and
CROSSED CATEGORISATION 39
religion indicated that both gender and religion group memberships were used to guide the processing of information about the main story character. Recall was greater for religion in-group characters than religion out-group characters only when the character was also out-group on the gender dimension of categorisation. This provided evidence that people can actually use more than one dimension of group membership at a time (even in an intergroup context like Northern Ireland where one dimension, religion, is commonly thought to dominate) and that complex combinations of category dimensions can be used to guide group-relevant information processing. Crisp and Hewstone (2001) found similar effects in a study that employed the same paradigm, but in a different cultural context (Singapore), using an alternative category dimension (ethnicity: Chinese versus Malay). An interaction between gender and ethnicity in recall of information from positive contexts again indicated the use of multiple bases of social categorisation in intergroup perception. Other work by Stangor, Lynch, Duan, and Glass (1992), van Knippenberg, Van Twuyver, and Pepels (1994), and van Twuyver and van Knippenberg (1995) has supported the notion that multiple dimensions of categorisation can be used in social perception (all using the category confusion paradigm). In particular Stangor et al. found interactions between category dimensions, which strongly supports the notion that combined category “subtypes” can be created “on line” in social perception. Whilst in many cases chronically accessible dimensions of categorisation (such as gender, race, and possibly age; Higgins & King, 1981) may dominate, significant contributions of simultaneously salient multiple dimensions seem to be made to the structuring of social encoding and/or retrieval of information. How can work demonstrating mutual inhibition of social categories be reconciled with that showing that multiple categories can be simultaneously activated and used in information processing, representation, and explicit judgement? A crucial distinction between studies that have found inhibition of additional dimensions and those that have shown a simultaneous use of multiple bases for classification is the use of priming of one of the two available categorisations in the former set. Studies showing mutual inhibition in category activation have manipulated the salience1 of one of the categories over the other, enhancing category (and stereotype) accessibility for the primed category in relation to the nonprimed category (and a baseline level of category/stereotype activation). In many of the studies demonstrating multiple category use no priming is employed to enhance the situational applicability of one categorisation over the other. If one category dimension were primed compared to a second possible basis for classification, we would indeed expect the primed (or otherwise highly accessible/situationally relevant) categories to be used more than less relevant or less accessible categories. This is a functional use
40
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
of multiple categorisations. However, when neither social categorisation is situationally more relevant than the other, as with the work reviewed above, the simultaneous use of multiple social categorisations seems to become likely. Overall then, there are instances where additional categories are inhibited in favour of a dominant, highly accessible category, including priming (differential accessibility; van Knippenberg et al., 1994) and differential situational relevance (“normative fit”; Oakes, 1987). There is also clear evidence, however, that people can and do process, represent, and use multiple dimensions of social categorisation in a variety of real intergroup settings. In the next section we examine specifically the implications of affiliation in multiple categorisation contexts for social perception, and examine how multiple intergroup comparisons can be used to guide evaluative judgements. MODELS OF CROSSED INTERGROUP AFFILIATION Intergroup relations between perceivers and target groups can be specified with reference to a pattern of in-group (self-including) and out-group (selfexcluding) differences. Thus, if the perceiver is Black, other Black individuals are in-group members whilst White individuals are out-group members. When two dimensions of social categorisation are simultaneously available to guide judgements, this can result in a “crossed categorisation” situation. We can define the four composite groups created by, for instance, crossing race and gender group memberships as follows. For the Black female perceiver, other Black females are double in-group members (sharing categorisations on both dimensions of categorisation), Black males and White females are mixed group members (sharing a categorisation with the perceiver on only one of the two dimensions of categorisation)2 whilst White males are double out-group members. Thus, within the typical paradigm for studying crossed categorisation, two dimensions of group membership are made salient simultaneously and participants are asked to evaluate (or make some other social judgement about) the different crossed category composite groups relative to each other. A number of different patterns of evaluations across the composite groups have been observed using the crossed categorisation paradigm (Brewer, Ho, Lee, & Miller, 1987; Hewstone, Islam, & Judd, 1993). The six main patterns can be described using contrast weights, or by notation, as shown in Table 1. The additive pattern (pattern 1) documents the predicted outcome across the four crossed categorisation composite groups
1“Salience” in this context refers to the situational “importance” that is distinguished from pervasive importance of established categories.
CROSSED CATEGORISATION 41
TABLE 1 A priori predictions of the six main outcome patterns of evaluation resulting from crossed categorisation
Differences in sign (+/–) specify where differences are to be expected across the four crossed category subgroups. i=in-group membership, o=out-group membership. Dominant (relatively important, salient, or accessible categories) are denoted in upper case (i.e., I vs. i, O vs. o). An additional pattern can be termed “equivalence” which specifies no differences across the four crossed category subgroups.
when categories are of equal salience and are combined in a purely additive fashion. Thus, the double in-group is evaluated most positively (being in-group on both dimensions), the double out-group least positively (being out-group on both dimensions), and crossed targets in between these two extremes (being in-group on one of the two dimensions, but out-group on the other dimension). The category dominance pattern (pattern 2) describes the case where a single categorisation dominates, and classification based on the second category distinction is ignored. This situation could occur as the result of contextual factors (e.g., a single minority female entering a room filled with an exclusively male majority) or be the result of historical factors (e.g.,
2
These mixed membership groups have also been called “partial” groups in the crossed categorisation literature, but here we use the term “mixed” to avoid any misconception that these crossed category groups are incomplete in any way.
42
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
ethnicity in the Balkans). There are two versions of this model based on which dimension is dominant. Social inclusion (pattern 3; sometimes referred to as the category conjunction [similarity] pattern) describes a situation in which all groups are evaluated equally positively as long as they are in-group on at least one dimension. Double out-groups are therefore evaluated less positively than all other groups. The social exclusion pattern (pattern 4; sometimes referred to as the category conjunction [dissimilarity] pattern) is the opposite of social inclusion. Here, the mixed membership targets are evaluated as negatively as double out-group targets because they are outgroup on at least one dimension. Hierarchical acceptance (pattern 5; sometimes referred to as hierarchical ordering) specifies that the effects of one category distinction are dependent on prior categorisation on the other dimension, i.e., in-group classification on the first dimension will lead to greater in-group/out-group differentiation on a second dimension than if the first dimension had been an out-group membership. This pattern is essentially a combination of category dominance and social exclusion patterns. There is category dominance because one of the categorisation dimensions is more important than the other, and there is social exclusion because there is something unique about the double in-group condition (i.e., it is especially positively evaluated). Like the category dominance pattern, there can be two versions of this pattern depending on which dimension is dominant. Hierarchical rejection (pattern 6; sometimes referred to as hierarchical derogation) is a variant of this hierarchical ordering pattern in that out-group membership on the more important dimension determines differentiation. Out-group classification on the important dimension will increase differentiation on a second dimension, e.g., White supporters only derogating the Black players on an opposing football team. Finally, an additional pattern may also be evident in some situations. The equivalence pattern (Miller, Urban, & Vanman, 1998) cannot be represented by contrast weights because it specifies equality in the evaluation of the four crossed category groups. Clearly, this is problematic to some extent because its confirmation rests on the absence of a reliable effect. However, when statistical power to detect differences is high, it is possible to test for this pattern and it does represent an empirically obtainable pattern of evaluation relative to others in the presence of moderating factors. These patterns have been observed across a number of studies, and reviews of the literature have confirmed the most common amongst them in varied antecedent contexts (Crisp & Hewstone, 1999a; Migdal et al., 1998; Urban & Miller, 1998). Here we consider the findings from previous reviews, and then go beyond these findings to propose a theoretical model
CROSSED CATEGORISATION 43
with which to interpret how crossed categorisation evaluations vary as a function of context, and the processes that may mediate these changes. EXTANT THEORETICAL AND EMPIRICAL WORK The evidence for each of the patterns outlined in Table 1 has been rigorously investigated in meta-analytic (Migdal et al., 1998; Urban & Miller, 1998;) and narrative (Crisp & Hewstone, 1999a) reviews. These reviews found that the baseline pattern, which seemed not only most common but also the most likely pattern to be observed in neutral antecedent contexts (i.e., devoid of differential cognitive, affective, or motivational influences) was additive (see pattern 1, Table 1). To illustrate, Urban and Miller compared target evaluations meta-analytically by computing effect sizes that represented the relevant comparisons: ii versus mixed (io or oi), ii versus oo, mixed versus oo, and io versus oi (where i=ingroup component membership, o=out-group component membership). The mean effect size estimates supported an additive pattern in that the mean ii versus oo effect size exceeded that of the ii versus the mixed, and the mixed versus oo (viz., ii>io=oi>oo). Put another way, when the two dimensions of categorisation are equally important and there are no other moderating factors, this additive model is the fundamental crossed categorisation effect. Below we discuss theorising subsequent to these prior reviews that has qualified and extended our understanding of the processes that give rise to the baseline additive pattern of evaluation. THE ADDITIVE PATTERN Previous work in the crossed categorisation literature has suggested that the additive pattern was the de facto pattern to be expected when all other potential moderators were controlled. Furthermore, both Crisp and Hewstone (1999a) and Migdal et al. (1998) concluded that category differentiation (Doise, 1978, see also Campbell, 1956; Tajfel, 1959; Tajfel & Wilkes, 1963) was a key process in defining how perceivers represent and use multiple social categories (see Arcuri, 1982; Crisp, Hewstone, & Rubin, 2001b; Deschamps, 1977; Deschamps & Doise, 1978; MarcusNewhall, Miller, Holtz, & Brewer, 1993; Rehm, Lilli, & Van Eimeren, 1988; Vanbeselaere, 1991, 1996). Theoretically, however, a purely differentiation-based explanation of evaluative differentiation in crossed category contexts is as problematic as in simple categorisation situations. Put simply, a basic cognitive mechanism invoked to simplify the perceptual and cognitive context cannot explain the in-group favouring direction of bias (it can only explain the pervasive tendency towards differentiation). Crisp and Hewstone also concluded that there was little support for motivational (self-esteem) processes in explaining crossed categorisation
44
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
effects (reflecting difficulties with self-esteem as an explanation for simple category discrimination; Rubin & Hewstone, 1998; but see also Crisp & Hewstone, 2000b). If we accept the well-supported category differentiation element of crossed categorisation processes, but not the motivational explanation, a second putative process is needed to explain evaluative differentiation. We propose here that the second process relevant to explaining crossed categorisation effects is affective in nature. Below we outline what we believe are the two key processes that operate in baseline crossed categorisation situations, and which are essential to understanding how the additive pattern changes as a function of context. Category differentiation As mentioned above, there is a great deal support for the notion that the same category differentiation processes that operate in simple contexts are also affected in predictable ways in crossed categorization contexts. In “simple” category contexts a general principle of exaggeration of difference between categories and minimisation of difference within categories seems a fundamental process in concept formation (Campbell, 1956; Doise, 1978; Tajfel, 1959). Deschamps and Doise (1978) proposed that when two orthogonal dichotomous categorical dimensions are crossed then the accentuation of differences between, and similarities within, each of the categorical dimensions work against, or in combination with, each other depending on the in-group or out-group composition of the crossed category group. Crisp and Hewstone (1999b) and Deschamps (1977) both found evidence in line with category differentiation processes with object classification, as did Arcuri (1982) with (unaffiliated) social categorisation. In affiliated contexts, the category differentiation model implies that the double in-group will be differentiated from the double out-group to a greater extent than from the mixed membership groups because in the former case there is accentuation of differences between both categories, whilst in the latter case this occurs on one dimension only. MarcusNewhall et al. (1993, Study 1) found mediational evidence that this does indeed occur. When perceived similarity was controlled, a crossed vs convergent factor was no longer related to reward allocations, suggesting a mediating role for category differentiation in explaining the effects of crossed categorisation on bias reduction. Category differentiation, then, does have a role in explaining how the additive pattern arises. However, it cannot account for in-group favouritism, and is more a cognitive prerequisite for bias than a complete explanation.
CROSSED CATEGORISATION 45
Affective priming In-group and out-group membership seems to carry an acquired affective valence (Dovidio, Evans, & Tyler, 1986; Fazio, Jackson, Dunton, & Williams, 1995; Perdue, Dovidio, Gurtman, & Tyler, 1990). In-group categories appear inherently positive in connotation, and out-group categories inherently negative. Whether this association develops via a process of associative learning with real groups over time (Bargh, 1997; Perdue et al., 1990; see also Cacioppo, Marshall-Goodell, Tassinary, & Petty, 1992; Das & Nanda, 1963; Staats & Staats, 1958) or automatic generalisation of self positivity with novel groups (Cadinu & Rothbart, 1996; Otten & Wentura, 1999),3 there seems at least to be a strong positive “in-group default” (Maass & Schaller, 1991), that can account for in-group favouritism. Activation of positive and negative affect with respect to the in-group and out-group respectively can in many cases be expected to influence judgements of in-groups and out-groups. A tendency to feel more positively towards in-group memberships (and more negatively towards out-group memberships) will lead to the baseline additive pattern (where evaluation occurs as a function of in-group membership) in crossed category contexts.4 Together, the cognitive category differentiation process and the affective priming process can explain when crossed category groups are evaluated as a function of distinct in-group and out-group memberships in “neutral” antecedent contexts. The category differentiation process provides the mental differentiation between “us” and “them” (in crossed category contexts, the distinctions between different degrees of “us” and different degrees of “them”). Affective priming can explain how these differential affiliations provoke different degrees of evaluative bias (given a prerequisite mental definition of the different degrees of in-group and out-group). On the basis of these two key processes, we have developed a model that will predict how different moderators might change evaluative patterns of crossed categorisation. A DUAL-ROUTE MODEL OF CROSSEDCATEGORISATION EFFECTS Previous reviews of the literature did not offer an integrated theoretical model of moderation and mediation of crossed categorisation effects. In this section we propose such a model on the basis of past reviews and our integrative theorising. As discussed above, the additive pattern is the 3
Self-anchoring can putatively provide the affective connotation for in-groups (but not out-groups) with novel or minimal categorisations (Otten & Wentura, 1999).
46
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
baseline from which all deviations should occur in the presence of moderating factors. In two meta-analyses of crossed categorisation patterns, significant residual variation remained after removal of the additivity effect (Migdal et al., 1998; Urban & Miller, 1998). This suggests that the variation among the obtained patterns is not random sampling error around the fundamental additivity effect but, instead, reflects the effects of moderating variables. Urban and Miller identified four factors that should produce deviation from the additive pattern. The affective valence induced by the experimental setting; cognitive overload; the discrepant importance of the two social category dimensions used to define in-group/out-group status; and access to individuating information about the four types of target persons. Some recent work by Crisp and Hewstone (2000b; see also Crisp, Hewstone, Richards, & Paolini, in press) also advocates contextual inclusiveness as a potential moderator. We propose that the moderating factors exert their influence on the pattern of evaluations of crossed category groups via either, or both, of the processing routes putatively responsible for the baseline additive pattern of evaluations (see Figure 1). By specifying the nature of particular moderators, using this dual-route model one can predict not only the outcome pattern of evaluations but also the mediational process that will give rise to it. The specification of distinct (but not mutually exclusive) affective and cognitive routes to evaluative outcome in this way allows a clear explanatory link to be made between the additive outcome and deviations from this baseline. We consider each of the moderators proposed above with respect to the two key underlying processes involved in crossed categorisation via this dual-route model. Affective valence can clearly be linked to the processes that might account for the additive pattern in neutral affective contexts. If in-/outgroup status has an acquired baseline affective valence (Otten & Wentura, 1999; Perdue et al., 1990), then positive or negative mood should lead to changes in the salience of the inherent affective connotation of in-group and out-group membership. Urada and Miller (2000) suggest that positive mood will effectively cue positive material in memory (Isen, Shalker, Clark,
4 As with stereotypes (Wyer, Sherman, & Stroessner, 2000), when perceivers are aware that they are being asked to make an explicit evaluative judgement about a socially sensitive category they are likely to suppress all negative evaluations. However, when the target groups are less socially sensitive (according to egalitarian social norms) or when an evaluative judgement is not explicitly requested, implicit affective reactions would be expected to correlate highly with measures of evaluation. In the crossed categorisation work reviewed in this chapter, it is appropriate to invoke the affective priming explanation, since the target groups are invariably minimal groups (low social sensitivity) or the measures do not obviously ask participants to evaluate the out-group (e.g., preference for discussion partner).
CROSSED CATEGORISATION 47
Figure 1. A dual-route model of moderating and mediating effects in crossed categorisation contexts.
& Karp, 1978; Laird, Wagener, Halal, & Szegda, 1982; Teasdale & Fogarty, 1979; Teasdale & Russell, 1983), and have a priming effect whereby attention is directed primarily to other positive aspects in the environment (Higgins & King, 1981). As such, a baseline additive pattern (ii>io=oi>oo; where positive affective connotation of the categories is a function of in-group membership) will be modified in the presence of an affective context that enhances the positive (in-group) characteristics of combined groups. In-group components of crossed category composites will thus achieve dominance, leading to a social inclusion pattern (ii=io=oi>oo).5 By the same token, negative mood should prime out-group membership, leading to a relative dominance of the negatively valenced out-group components of the composite groups, and the social exclusion pattern; ii>oi=io=oo). This leads to specifiable predictions regarding the effects of positive and negative mood on crossed categorisation evaluations that are clearly linked to the underlying processes that we claim drive
48
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
baseline additive bias. Since mood primes only the affective (and not cognitive structural) components of crossed category groups, we would expect no effect of affective priming (via mood) on non-evaluative measures of category structure and differentiation. Mood should affect crossed category evaluations only via the affective priming route. Consequently we expect a dissociation between evaluative judgement and cognitive representation in the different affective versus neutral contexts (see also Vescio, Hewstone, Crisp, & Rubin, 1999). Meta-analytic evidence showed that greater cognitive load was associated with reduced evaluative differentiation among the target persons of the crossed categorisation paradigm (Urban & Miller, 1998). Task difficulty and annoying distractions (load) produce stressful arousal in participants (Marco & Suls, 1993; Repetti, 1993). Consequently, similar to the anticipated effects of negative mood, such negatively stressful cognitive overload is likely to have a priming effect in which attention is oriented towards other negative stimuli. Because out-group category membership often is a negative stimulus, this will in turn produce a social exclusion pattern. As such, in this context, cognitive load can be expected to modify the additive pattern of crossed categorisation via the affective priming route, but not the cognitive structure route, in the same way as negative mood. Although distinct in Urban and Miller’s (1998) meta-analysis, cognitive overload is essentially expected to change composite group evaluation via changes in affective valence. In other words, it is basically an additional (although perhaps more ecologically valid) way of manipulating differential affect. As such, we do not treat it as a separate moderator in our analysis below. Category importance can be influenced directly by category salience. Category salience depends on an interaction between the relative accessibility of that categorisation for the perceiver and the fit between input and stored category specifications (Bruner, 1957; Oakes, Haslam, & Turner, 1994; Turner, Hogg, Oakes, Reicher, & Wetherell, 1987). When a category becomes highly accessible to a person, or when a stimulus to be judged readily fits the category, that category dimension is likely to be more important and dominant as compared to others (Miller et al., 1998). In a crossed categorisation context, when one dimension is more important than the other it induces more favourability towards the mixed target that
5 It could be argued that the double in-group, comprising of two primed ingroup memberships, would remain more positive than the mixed groups. However, in relative terms, all groups containing at least one in-group membership will be evaluated much more positively than in baseline conditions, to the exclusion of the double out-group (containing no in-group memberships). This accentuated differentiation between all in-group including groups and the double out-group would lead to a pattern most closely resembling the social inclusion pattern.
CROSSED CATEGORISATION 49
has an important in-group identity (Io) than one that has an important outgroup identity (Oi: i.e., Io>Oi). Put simply, category importance will lead to category dominance. Because category importance is assumed to increase the accessibility, and salience, of particular category dimensions, we would expect it to have effects on evaluations via both affective valence (more accessible categories will be more available for affective “tagging”) and category structure (more available categories will enhance differentiation). Meta-analytic results show that category importance moderates evaluative bias, and that it is associated with greater structural differentiation between the two mixed targets (Urban & Miller, 1998) Access to individuating information may also moderate the baseline pattern via processes clearly specified in simple category contexts. Personalisation and decategorisation refer to types of intergroup interactions in which in-group and out-group members respectively have more individuated information about each other, by comparison with situations in which their interaction more predominantly reflects their reactions to their respective category labels (Brewer & Miller, 1984). Exchanging individuating information and presenting significant aspects of self to another, are viewed as important factors in the development and maintenance of interpersonal relationships (e.g., Cozby, 1972; Goodstein & Reinecker, 1974; Jourard, 1971). These factors are also important in reducing the negativity bias towards the out-group in intergroup relations (e.g., Bettencourt, Brewer, Croak, & Miller, 1992; Brewer & Miller, 1984, 1988; Marcus-Newhall et al., 1993). As a result of personalised contact, the negative stereotype of members of out-groups can be disconfirmed (Wilder, 1978), and in-group/out-group distinctions diminished, perceived variability among out-group members is increased, and thereby, a negative view of the out-group can be reduced (Brewer & Miller, 1984, 1988; Cook, 1978; Harrington, 1988; Wilder, 1978). Importantly, access to individuated information alone is not sufficient to promote a decategorisation effect, but in addition the contact must be personalised (i.e., include self-other comparisons and self-disclosure, Ensari & Miller, 2001). Thus, when individuated information is available, and social perception/interaction is personalised, it may lead to a decategorisation effect, whereby categories are abandoned as functional guides to social judgement in favour of focusing on unique individuals’ characteristics (Fiske & Neuberg, 1990). In a crossed categorisation situation, like any other, we would therefore predict an abandonment of all bases for classification, and a shift from the additive to the equivalence pattern. Thus, in decategorised contexts, with no categories to affectively “tag”, no differential affective valence can be assigned to categories. As such, we would expect access to individuating information to lead to the equivalence pattern due to this lack of any mental definition of who the in-group and out-group are. This moderator should not then produce effects on crossed category evaluation
50
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
via the affective route, but instead via cognitive structure. Intercategory differences should disappear on non-evaluative indices of differentiation. Contextual inclusiveness is a novel potential moderator not suggested by previous reviews. Category inclusiveness can be defined as the extent to which categorisation is at a more superordinate level, subsuming other social categories in the immediate intergroup context; e.g., “European” is a more inclusive social category than “British” or “French”). In the same way that superordinate categorisation helps to alleviate intergroup bias on a single subordinate category dimension (Gaertner & Dovidio, 2000; Gaertner, Dovidio, Anastasio, Bachman, & Rust, 1993; Gaertner, Mann, Dovidio, Murrell, & Pomare, 1990), making a superordinate categorisation salient in a crossed categorisation situation will promote a reduction in bias, via recategorisation of (mixed) or double out-group members into a common in-group. Put another way, when the four composite groups in a crossed categorisation context are subsumed by a salient superordinate categorisation, we suspect that it will lead to the formation of a more inclusive pattern. Manipulating the inclusiveness of category contexts will directly influence the underlying category representation of the crossed groups, leading to a moderating effect on evaluations via cognitive changes in category structure. We can thus formulate a model that makes clear predictions of the effects of antecedent moderators on both theoretically meaningful processes and evaluative outcome patterns in crossed categorisation contexts (see Figure 1). Distinguishing between affective and cognitive processes, which both contribute to the baseline additive pattern of crossed category evaluation, we can specify circumstances under which we expect a number of the different patterns previously observed in the literature. Compared to the baseline additive pattern, we expect the following deviations to occur: (1) Positive and negative affect (mood) should lead to social inclusion and exclusion patterns of evaluation respectively because the affective context should prime the congruent components (in-group vs outgroup) of crossed category composites. Since these different mood contexts would be expected only to prime affective, and not structural, components of crossed category composites, we expect measures of category structure and differentiation to be unaffected by mood. (2) Category importance is expected to increase the accessibility and salience of particular dimensions of categorisation over others. As such, it will lead to greater affective priming, cognitive differentiation, and in-group evaluation on such dimensions. (3) Access to individuating information will decategorise the crossed category context. In other words, the moderating effect of individuated interaction will occur via the cognitive rather than affective route. As
CROSSED CATEGORISATION 51
such, in these situations the baseline additive pattern of evaluations and category differentiation will be degraded, leaving an equivalence pattern on evaluative (outcome) and cognitive structure (mediational) measures, but there will be no effect on measures of affect. (4) Contextual inclusiveness will have a re-categorising effect, leading to the equivalence pattern in terms of both evaluations and structural differentiation, but not on measures of affective mediation. Interestingly, the proposed recategorisation effect caused by priming superordinate identity is predicted to have the same process and outcome implications as the decategorisation effect proposed following access to individuating information [(2) above]. EMPIRICAL SUPPORT FOR THE MODEL In this section we assess whether empirical studies, which are subsequent to previous reviews, and which directly test moderation of crossed category patterns, provide support for the dual-route processes advocated in our theoretical model. We consider recent studies under three broad headings: Affective state and category importance (we consider these two moderators together since no new work has examined category importance independent from affective state), access to individuating information, and contextual inclusiveness. Affective state is predicted to moderate evaluations via the affective route, whilst access to individuating information and contextual inclusiveness are predicted to take the cognitive route towards moderation. Affective state and category importance The dual-route model makes broad predictions regarding the routes and outcomes expected in different antecedent contexts. In the case of affect, the prediction is that the route taken will be affective priming (not cognitive structuring) and that generally inclusive patterns will emerge. In the studies of affective moderation we discuss below, a distinction can be made between “incidental affect”, which arises from sources unrelated to any intergroup aspect of the setting, and “integral affect”, which is affect produced directly by out-group persons in the contact situation (Bodenhausen, 1993). We consider both types of affect in relation to the prediction of generalised outcome inclusiveness and affective priming mediation. Incidental affect. Several studies have explored the potential moderating effect of mood on the emergent pattern of evaluation in the crossed categorisation paradigm. Crisp and Hewstone (2000a) studied the effects of positive incidental affect by manipulating performance feedback in a crossed categorisation version of the minimal group paradigm (Tajfel, Flament, Billig, & Bundy, 1971). The participants received either positive
52
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
feedback (i.e., “You have scored 84 out of 100”) or neutral feedback (i.e., “You have scored 51 out of 100”) on a computerised visual memory task. The dependent measures included liking, similarity to self, and trait differentiation. The anticipated additive pattern of target evaluations was found in the no feedback (neutral affect) condition (ii>io=oi>oo). As discussed above, this additive pattern serves as the baseline against which to compare evaluations in other induced affect conditions. In the positive affect condition, evaluations of the two mixed membership groups (io and oi) moved closer to evaluation of the double in-group (ii). Importantly, measures of intergroup similarity and trait differentiation did not deviate from the baseline additive pattern when participants were in a positive mood. This is in line with the notion that mood moderates crossed category patterns via an affective process, but has no effect on the cognitive structure of the intergroup situation. Crisp and Hewstone’s (2000a) study thus offers some support to the notion that positive affect can promote inclusion in crossed categorisation contexts. In another set of experiments Urada and Miller (2000) tested whether participants in a positive mood would see mixed targets (io and oi) primarily in terms of factors that cue their positive (i.e., in-group) features, leading towards a social inclusion pattern (ii=io=oi>oo; in line with the proposed dual-route model). However, Urada and Miller went beyond the postulated priming effect of mood and suggested a role for category importance in conjunction with affect. In one study, Urada and Miller used mixed targets that had group memberships of equal but simultaneously of high importance (IO, OI; where dominance/importance is indicated by upper-case). In this situation, they postulated that the two mixed targets, containing both an important in-group and an important out-group, would preclude any priming effect of positive affect because of an overriding importance of both dimensions (since positive affect is, in relative terms, a weak contextual cue; Taylor, 1991). Put another way, any affective priming would enhance the influence of in-group membership on both dimensions to the same extent, precluding a social inclusion pattern, and instead leading to a persistent additive pattern (II>IO=OI>OO). To induce positive affect Urada and Miller elicited participants’ detailed recall of one of their happiest life experiences. In neutral mood conditions participants recalled an affectively neutral everyday event. As expected, positive affect had no impact on evaluations of mixed targets when the categories of those targets were of equal importance (i.e., IO and OI). Put another way, in the absence of a dominant in-group category membership, incidental positive affect did not increase the attractiveness of mixed targets and the additive pattern (II> IO=OI>OO) was observed. Urada and Miller (2000) also examined incidental positive affect in a context wherein one dimension of social categorisation was more important than a second category dimension. Here, the authors argued that the
CROSSED CATEGORISATION 53
priming effect of positive incidental affect, with respect to the in-group/ out-group difference between the mixed targets on the dominant dimension, would increase the positivity of the Io target relative to that of the Oi target, because here importance is differentially distributed across dimensions. As a result, evaluations of the Io target will be assimilated towards the double in-group target, whereas the Oi target will be evaluated less favourably than the Io, leading to the following pattern, Ii=Io>Oi>Oo. Confirming this expectation, across four studies, incidental positive affect consistently increased preferences for the mixed target who shared a dominant in-group membership (Io). By contrast, it did not alter evaluations of mixed targets who had a dominant out-group membership (Oi). In summary then, the work of Urada and Miller broadly supports the dual-route model via a combination of the proposed effects of mood and category importance. Integral affect. Ensari and Miller (1998) examined the effects of positive and negative integral affect on preferences for four potential discussion partners (with memberships comprising all four combinations of group membership: ii, io, oi, and oo). Ensari and Miller hypothesised that integral affect (affect specifically linked to the intergroup context) would not only prime the affectively congruent in-group or out-group, but also that the specific out-group which would be the source of the affect will give that category an acquired dominance. The out-group social category membership of the mixed category target oi was made dominant by experimentally inducing affect from that particular out-group category (dominance hereafter indicated by upper-case; i.e., Oi). All participants were college-age (early twenties) students at the University of Southern California. Thus if the Oi target was an older adult (out-group, “o”) and a University of Southern California student (in-group, “i”), then older adults were made the source of affective arousal (“o” became “O”). At the same time, the other three targets (ii, io, and oo) did not possess a group membership on the dominant category dimension (i.e., no information about age was given for any of these other targets). Instead, alternative social category dimensions were used to provide the category cues that identified the in-group and out-group memberships of the ii, io, and oo targets and the particular dimensions that were used differed for each target.6 Preference for the Oi target was then compared with that for the other target-person combinations (ii, io, oo) when either positive or negative affect had or had not been induced by members of the out-group category of the oi mixed group. In their first experiment, Ensari and Miller (1998) induced positive integral affect via compliments from an out-group category. The participants in the compliment condition were given a specially prepared page allegedly photocopied from the university newspaper that consisted of complimentary interviews with members of one of the out-group categories
54
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
of the participant. For instance, some of the ideas contained in the alleged interviews with older adults were the following: “young people today adopt a more adult attitude that warrants respect from old people”, “young adults have a purpose in life”, and “young adults make an effort to create a significant impact on society’s view about important social issues”. After the participant read the article, he or she was asked to indicate a relative preference among the four crossed categorisation targets as partner for the discussion task. According to the above theorising, after having read compliments to one’s in-group from members of the Oi group, the positivity of the Oi target relative to that of the other mixed target should be elevated (i.e., Oi>io). Thus, Ensari and Miller expected compliments from an out-group category to cause assimilation of that out-group (the Oi target) into a broader in-group category with the ii targets (i.e., ii=Oi). Compared to a no compliment condition, where the predicted additive pattern was obtained, the difference between the mixed and double in-group target was ignored when the out-group component of the mixed target had previously been a source of compliments. This led to a hierarchical derogation (or rejection) pattern of evaluation (ii=Oi>io>oo; see Table 1, pattern 6). Ensari and Miller (1998) also explored the effects of negative integral affect on the pattern of evaluation. Similar to the processes outlined above, when attention is drawn to out-group members of a social category dimension by their insults to one’s in-group that too should increase the importance of that category dimension. That is, attention should be directed towards the category that was made a source of the insult. By making the out-group component of the mixed target dominant, in-group status on the second dimension (Oi) should become less important. Paralleling the procedure of Experiment 1, insulting articles were created (again allegedly photocopied from the university newspaper). Examples of the information contained in the insulting interviews with older adults include the following: “young adults never make a sufficient effort to create a significant impact on society’s views about important social 6 It is important to note that the structure of the crossed categorisation situation in these studies was not of the traditional form. This was not a situation in which two dimensions were orthogonally crossed, rather, the actual in-group or out-group composites of the crossed category subgroups were made up from different categorical dimensions, i.e., one of the mixed groups shared membership with the double in-group on one dimension but was out-group on a dimension that was not a composite of the double in-group. An example would be Welsh female engineers comparing themselves with Welsh females (the double in-group ignoring occupation) with Welsh accountants (the mixed group ignoring gender). However, there is no theoretical reason why this should change which processing route is taken in the dual-route model, or change the predicted pattern of evaluations as a function of in-group or out-group membership.
CROSSED CATEGORISATION 55
issues”, “young people need to adopt a more adult attitude that would enable them to gain respect from old people”, and “young adults’ activities have a negative effect on the image that other countries have of our own society”. After the participants read these interviews, they were asked to indicate how much they wished to be a partner with each of the four crossed categorisation targets for participation in a discussion task. Confirming expectations, relative to the baseline additive pattern in a no insult condition, preference for the mixed target who shared an out-group status with the source of insult (the Oi target) was about as negative as that for an oo target, yielding a hierarchical acceptance pattern (ii>io>Oi=oo). Finally, in Experiment 3, Ensari and Miller expanded the generality of the effects found in Experiments 1 and 2 by replicating them in another culture, Turkey. Within the context of a single experiment, positive and negative integral affect from an out-group category again respectively yielded contrasting hierarchical patterns of evaluation (relative to the baseline additive pattern). Mediational analyses (Baron & Kenny, 1986), based on manipulation check data that had confirmed the effectiveness of the mood inductions, supported the view that mood had, in fact, mediated these obtained effects. In support of the proposed integrative model, differential affect again moderated the pattern of evaluations across the crossed category groups (positive affect promoting some degree of inclusiveness, Oi=ii, and negative affect some degree of exclusiveness, Oi=oo). Furthermore, the mediational process here was supported (although the predicted absence of effect on structural measures of cognitive representation was not tested in these studies). As in Urada and Miller (2000) category importance seemed to be linked with affect to predict the outcome pattern of evaluation: Affect linked to out-group members on a specific component of mixed targets led to an affective priming effect only for those specific (dominant) out-groups. To summarise, Ensari and Miller (1998) offer strong support for the dual-route model. In their studies positive and negative affect mediated an increase in social inclusivity towards mixed category groups to whom the affect was relevant. This suggests that affective priming of the in-group and out-group components of mixed category groups produced the selective increased or decreased evaluation in response to integral (group specific) affect. Affect in this case has an effect on evaluations that is limited to specific groups in accordance with nature of integral affect. Put another way, integral (vis-à-vis incidental) affect seems not only to prime affective components of crossed category groups, but also increase category importance for the specific out-group responsible for the negative feedback. Thus, whilst consistent with the key processing distinction in the dual-route model, Ensari and Miller’s findings also refine the predicted outcome of affective moderation.
56
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Access to individuating information All the studies we consider here were published, or carried out, subsequent to the previous reviews of the literature. Here, however, we make one exception, because Marcus-Newhall et al. (1993) can be interpreted now in line with our new theoretical model. Marcus-Newhall et al. divided four participants into artificial social categories based on an alleged dotestimation judgement task. Then they were reassigned to four-person teams, each consisting of two “overestimators” and two “underestimators”. Next, the teams performed a co-operative task that required team members to list the most important traits for the selection of NASA astronauts. Prior to engaging in the co-operation task, two team members were assigned the role of “cognitive-traits expert” for the upcoming task, whereas the other two shared the role of “emotional-traits expert”. Each team member was given a set of preliminary materials to read that established their expertise regarding their respective roles for the NASA task. In the “convergent role assignment” condition, both overestimators were assigned to the cognitivetraits experts, whilst both underestimators were assigned to the emotional-traits experts (i.e., targets were either double in-group or double out-group). In the “cross-cutting” condition, work roles cross-cut initial group membership, creating four possible crossed-category groups (i.e., the four targets were overestimator/ cognitive-traits expert, overestimator/emotional-traits expert, underestimator/cognitive-traits expert, underestimator/emotional-traits expert). In the first study participants had the opportunity for personalised interaction, and negative task expectations (task anxiety) were minimised by creating the perception of individual control over own outcomes. This reduced intrateam competition and undermined the occurrence of any negative associations that might have been produced by the knowledge that participants had to work with an out-group member (see Vanman, Paul, Ito, & Miller, 1997). Analyses revealed that, compared to convergent role assignment, cross-cutting role assignment decreased perceptions of similarity within categories and increased perceptions of similarity across categories, leading to a reduction of bias (equivalence pattern). These findings are thus consistent with the proposed moderating role of access to individuating information (in conjunction with personalised contact and increased self-other comparisons/self-disclosure) and support the mediating route of the effect (i.e., cognitive structural processes, here measured by perceived similarity). Ensari and Miller (2001) examined the generalised effects of heightened category salience (emphasis on category memberships) versus decategorisation (emphasis on individual attributes) within the context of a minimal group version of the crossed categorisation paradigm. Specifically,
CROSSED CATEGORISATION 57
Ensari and Miller manipulated the rationale for co-operative group formation (i.e., category-salient vs individuated) and presented this to participants as the basis for constructing a heterogeneous work team from participants exemplifying the four combinations of the two artificial category dimensions. The experimental variations of this rule functioned to focus participants’ attention on either both category dimensions (doublecategory condition), one dimension (single-category condition) or, instead, on team member unique personal attributes (decategorisation condition). Personalisation was made a constant feature of all experimental conditions by employing an experimental task that simultaneously required both the disclosure of unique personal attributes and invited self-other comparisons with respect to them. Ratings of the members of the other team in the experimental session, a measure of generalised bias, showed that the personalised interaction that was imposed as a constant feature in all conditions had a considerably reduced benefit when the rule for assigning persons to work teams emphasised their category memberships. Even in the presence of personalised interaction, an emphasis on one or both categories, as produced by manipulation of the rule for assigning persons to teams, yielded differential bias among the evaluations of members of the other team. Specifically, it produced category dominance and additive patterns respectively (as would be expected). When, instead, the personalised interaction occurred under the condition in which the rule for assignment to work teams emphasised group members’ unique personal attributes (i.e., the decategorisation condition), a generalised equivalence pattern was obtained (see Table 2). These findings are thus in line with the proposed moderating effects on crossed category evaluations as a function of category-based versus individuated perception. As expected, on the basis of the cognitive route to changing evaluations, individuated perception led to a decategorisation effect, where category distinctions were abandoned as useful guides to evaluative judgement. These findings support the dual-route proposition—that different moderators will have effects on the pattern of evaluations via different processing routes. In particular, access to individuating information has effects on evaluations mediated by changes in the representation of cognitive structure (Marcus-Newhall et al., 1993)—with “knock-on” effects for evaluations (the cognitive prerequisite for bias having been removed). Mood, in contrast, seems to modify evaluations via affective priming (Ensari & Miller, 1998), effectively by-passing the cognitive structure of the intergroup representation. Contextual inclusiveness Crisp and Hewstone (2000b) examined the effects of priming with a generic superordinate membership on implicit intergroup bias. Perdue et al.
Means with different subscripts within a row of each condition are reliably (p<.05) different from one another, ii=in-group/ingroup; io=in-group/ out-group; oi=out-group/in-group; oo=out-group/out-group. Data reported from Ensari and Miller (2001).
TABLE 2 Evaluations of the other team members as a function of rule for assignment to teams
58
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
CROSSED CATEGORISATION 59
(1990) found that responses to positive traits in a lexical decision task were facilitated when subliminally paired with the pronoun “we”, “us” or “our”. Such pronouns can be conceptualised as being indicative of a superordinate common in-group identity (see Gaertner & Dovidio, 2000; Gaertner at al., 1993). Extrapolating the notion that it is possible to prime inclusiveness via such pronouns to a crossed categorisation situation, we suggested that the priming of a superordinate in-group identity (such as “we”) could promote a more inclusive use of categorisation. In turn, this will lead to a more socially inclusive pattern of evaluation. Using names that clearly indicated in-group or out-group membership according to gender and nationality, Crisp and Hewstone (2000b) found some support for these predictions using reaction times to positive and negative traits in a lexical decision task as a measure of evaluation. The general notion that in-group and out-group membership is evaluatively positive and negative in connotation (and has an affective valence in the same way as other positive and negative words; Fazio, Sanbonmatsu, Powell, & Kardes, 1986) is well supported. In-group label primes typically facilitate reaction times to positive words (vis-à-vis out-group primes and negative words, e.g., with real groups, Dovidio et al., 1986; Purdue et al., 1990; Perdue & Gurtman, 1990, with ad hoc [novel] groups, Otten & Moskowitz, 2000; Otten & Wentura, 1999). In crossed category situations, we might expect automatic evaluation of in-groups and outgroups to follow a specified pattern, given certain moderating conditions such as the contextual inclusiveness manipulated here. No differences were found across negative traits, which is in line with the general finding that intergroup discrimination is driven by in-group enhancement rather than out-group derogation (Brewer, 1979; Hewstone, Rubin, & Willis, 2002; Mullen, Migdal, & Hewstone, 2001). Although these findings should be considered tentative, since no additive pattern was obtained in the no prime condition, they do broadly support a moderating role for inclusive in-group identities. For positive traits a social inclusion pattern was observed across the mixed category groups in the “we” primed condition. The reaction time for identifying a positive trait as positive was slower when the target name being remembered was double out-group (i.e., Male/Welsh) than if the target name was indicative of any other crossed categorisation (i.e., double in-group or mixed group status; Female/English, Female/Welsh, or Male/English). Crisp et al. (in press, b) recently expanded on this finding by using an alternative priming procedure, explicit ratings, and the minimal group paradigm. In this study inclusive or exclusive priming was operationalised using a “proof-reading” task in which participants were required to circle the relevant pronoun each time they encountered it in a short story (cf. Brewer & Gardner, 1996). Over- and underestimators and being better at large versus small shape estimations were the minimal groups. On explicit
60
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
evaluations, compared to the baseline additive pattern, participants primed with the “they” pronoun were more exclusive in their judgements, leading to the social exclusion pattern. Although inclusive primes had no significant effect on evaluations, this work supports the general notion of inclusiveness priming having a moderating effect on judgements. Overall, the studies reviewed above support the central tenets of the dual-route model. Affective moderators seem to modify the additive pattern via affective priming, but not via changes in cognitive structure. More category-based moderators seem to exert their influence via changes in the differentiation between categories, in contrast to affective priming. Table 3 contains a summary of the moderating effects that have been observed in these recent studies using the crossed categorisation paradigm. REFINEMENTS TO THE MODEL Broadly, the evidence outlined above is consistent with the main theoretical premise of the model, that moderators can be distinguished meaningfully in the route they take towards modification of the basic additive pattern of evaluation. On the one hand, affectively based moderators appear to change evaluations via an affective priming process, and there is mediational evidence that perceived mood drives changes in affective reactions to groups as a function of in-group (positive) or out-group (negative) membership (e.g., Ensari & Miller, 1998; Urada & Miller, 2000). Furthermore, we posit that such affectively based moderators will change evaluations withoutchanging patterns of cognitive structure. There is also some support for this premise in studies that assessed both cognitive and affective/evaluative measures (Crisp & Hewstone, 2000a). On the other hand, moderators that are cognitively based are expected to alter evaluative patterns of crossed categorisation via the structural route. This contention was also broadly supported. Manipulations designed to alter category differentiation distinctions led to predicted “decategorised” patterns (Ensari & Miller, 2001; or more socially inclusive patterns, Crisp & Hewstone, 2000b) and there is mediational evidence that changes in category structure are responsible for such changes in evaluation (MarcusNewhall et al., 1993). Despite the broad support for the theoretically specified process and outcomes predicted in the proposed model, there are some deviations from predictions that will require further investigation and possible refinement of the model. Below we consider these deviations and suggest how future research can clarify their implications.
CROSSED CATEGORISATION 61
TABLE 3 Observed moderators of crossed categorisation evaluations
Contextual inclusiveness should mediate evaluations via the cognitive route when a superordinate identity is made salient without affective implication, but via affective route when the moderator carries affective significance (e.g. grouprelevant pronouns as used in Crisp & Hewstone 2000b; Crisp et al., in press).
Affect and importance interaction Different types of mood, namely incidental and integral, had different effects on the evaluative pattern (Ensari & Miller, 1998; Urada & Miller, 2000). Both, however, are broadly consistent with the affective mediational route suggested by the model. The hierarchical patterns observed with both incidental and integral manipulations can be regarded
62
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
as more inclusive than the additive pattern (for the particular mixed membership group moving psychologically closer to the double in-group). The deviations from the predicted social inclusion pattern can be explained within the framework of the model, and an interaction between category importance and mood. The work by Urada and Miller (2000) suggests that sometimes affect will influence evaluations only in conjunction with certain levels of importance. In their studies, Urada and Miller demonstrated a moderating effect for positive mood only when one dimension was more important than the other. The combination of proposed processes in the model— affective priming and category salience—would be expected to lead to the social inclusion and category dominance patterns respectively. Importantly, the resulting pattern found by Urada and Miller is entirely consistent with a combination of these two patterns (i.e., social inclusion+category dominance=hierarchical rejection; see Hewstone et al., 1993). As such, whilst Urada and Miller’s findings qualify the model, they are consistent with a combination of factors that are proposed to have moderating effects. These results suggest that an interaction between affectively and cognitively oriented moderators is an important consideration when predicting outcome pattern and mediational route (s). However, we might reasonably expect differential importance to be influential in the affective moderation effect only for positive moods (as in Urada & Miller, 2000). Since negative stimuli exert a more powerful effect on social information processing (Taylor, 1991), negative moods might exert a moderating influence (i.e., social exclusion pattern) in the absence of the differential importance needed for (inherently less salient) positive mood to have an effect. As such, the postulated negative mood inducing effects of cognitive overload (yet to be tested empirically) may also operate without conjunctive differential importance of category dimensions. Such issues are interesting possibilities for future work. Category importance may also be a key concept for explaining why there is a discrepancy between the results obtained by Urada and Miller (2000) and Crisp and Hewstone (2000a), who both examined the effects of incidental positive mood. Crisp and Hewstone included no manipulation of category importance, and found that positive mood shifted evaluations from an additive to an equivalence pattern. In contrast, in Urada and Miller’s (2000) baseline conditions (no dominance) positive mood did not influence perceivers’ evaluations of the four targets, leaving the additive pattern intact. Only when there was differential category importance (dominance) did a moderating effect occur. Whilst both studies broadly supported the notion that positive affective context primes positive affective components of crossed category groups (leading to more socially inclusive reactions to crossed category composites), Crisp and Hewstone demonstrated equivalence whilst Urada and Miller demonstrated
CROSSED CATEGORISATION 63
hierarchical rejection. One possible explanation for this divergence (and qualification to the dual-route model) is the overall level of importance of the groups used in these studies. Crisp and Hewstone used minimal groups of arguably lower importance to their participants than the real groups used by Urada and Miller. It is possible that when category distinctions are of lower importance, positive affect can increase the tendency to form a superordinate categorisation (Dovidio, Gaertner, Isen, & Lowrance, 1995; Dovidio, Gaertner, Isen, Rust, & Guerra, 1998). As a result, differential bias among the four mixed category targets disappears, leading towards an equivalence in the evaluation of the targets, consistent with Crisp and Hewstone’s findings. With real groups, however, the presence of a particularly important negative social category (O) might prevent the formation of a superordinate categorisation (and perhaps simultaneously provide an organising structure among the four different types of target). If that is the case, then although the two categories used to form the mixed targets (I and O) were equal in importance in Urada and Miller’s study, the participants nevertheless differentiated the IO target from the II target because of the overall more meaningful and important negative out-group identity of the IO target. Similarly, in this case, the two negative out-group identities of the OO target preclude its assimilation towards the evaluation of the mixed targets. Supporting this idea, Urban and Miller’s (1998) metaanalysis found a larger effect size for the comparison of mixed versus double out-group target with real (and so more important) groups (effect size of .36) than with artificial (and so less important) groups (effect size of . 18). This indicates more differentiation between the mixed and double outgroup targets in real than in minimal group situations. As such, category importance (via real/minimal or other distinctions) may be an important additional moderator when considering the effects of affect on the evaluation pattern in crossed categorisation situations. Experimental manipulations of these variables within a single study will be required to provide clearer resolution of this issue. Process specificity Despite the broad support for the distinction between processing routes, there are some findings that require close inspection. Crisp and Hewstone’s (2000a) finding of the equivalence pattern under conditions of positive affect does initially pose a problem for the dual-route model. Rather than attributing the obtained equivalence pattern following positive mood induction to an affective priming route, it may reflect a change in structural representation. Some work suggests that positive affect can augment perceivers’ tendency to include more diverse people in their ingroup category (see Isen & Daubman, 1984; Isen, Niedenthal, & Cantor, 1992; Mackie & Worth, 1991; Worth & Mackie, 1987). According to this
64
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
argument, positive incidental affect in the crossed categorisation paradigm (when categories are of relatively low importance) may facilitate the recategorisation of composite group members into a common superordinate group, and thereby reduce intergroup bias via the structural route (Dovidio et al., 1995, 1998). As suggested above, it might be that mood leads to recategorisation at a superordinate level when categories are of relatively low importance. This, in turn, suggests that mood might in some circumstances take the cognitive structural route rather than the affective priming route. On the one hand, Crisp and Hewstone’s study supports this notion—the equivalence pattern was obtained and mood mediated an increase in double out-group liking (which is difficult to explain via affective priming, since no affectively congruent in-group membership comprised the double out-group for positive mood to prime). On the other hand, however, moderation via the cognitive route predicts mediation via measures of cognitive structure, which were unaffected by positive mood in Crisp and Hewstone’s study. One potential solution to this problem is that there is no necessity that affective and cognitive mediators be mutually exclusive (for instance, the hierarchical patterns of evaluation predicted from the simultaneous presence of category importance and mood moderators). In Crisp and Hewstone’s study postulation of a complex interaction between a pull towards structural inclusiveness and concurrent pull towards affective priming can explain the findings. If, as suggested above, the effects of affective moderation are themselves moderated by category salience, then categories of relatively lower importance than others (such as minimal groups) will be more amenable to the structural change following positive mood induction. Put another way, whilst in many cases mood would be expected to follow an affective priming mediational route, in some specifiable cases (e.g., categories of low versus high importance), cognitive processes may also be invoked and thereby alter category structure. To investigate this possibility, further tests of the model are needed that systematically manipulate category importance and mood, whilst measuring perceived inclusiveness, category differentiation, affective priming, and evaluative outcome. Whilst positive mood may sometimes have an inclusiveness effect in crossed categorisation contexts (Dovidio et al., 1995, 1998), we would expect the effect of priming of inclusiveness in the absence of a mood manipulation to differ from that obtained with mood manipulation. Mood may sometimes be sufficient for an enhanced perception of inclusiveness, but it is possible to manipulate inclusiveness without mood, which goes some way to validate the distinction between moderating conditions that lead to changes in evaluation via the affective versus cognitive route. One potential criticism of using pronouns such as “we” or “they” to promote inclusivity or exclusivity, respectively (Crisp & Hewstone, 2000b; Crisp et
CROSSED CATEGORISATION 65
al., in press), however, is that such primes may also prime affective components of composite groups. Put another way, priming with a pronoun such as “we” does not only potentially affect inclusiveness but it could also act as an in-group (positive) prime for in-group (positive) target attributes. Indeed, the finding that “we” primes lead to a social inclusion pattern (Crisp & Hewstone), and “they” primes a social exclusion pattern (and no moderation of cognitive structure; Crisp et al.), rather than an equivalence pattern (which would indicate de- or recategorisation) suggests that such pronouns were acting as affective primes rather than structural moderators in these studies. The evaluative outcome and ineffectiveness of priming on category differentiation suggests that it is the affective priming route that is taken following presentation of these pronouns. This does not, of course, invalidate the model. On the contrary, interpreting the findings of the above research in the context of the model allows us to refine future empirical tests of the theoretical basis of crossed categorisation effects. Regarding the effects of pronoun priming, the analysis above suggests that in future work alternative ways of manipulating inclusiveness, which would not be expected to exert influence via the affective priming route, should be examined. We predict that such moderators would have a recategorising effect, and promote an equivalence pattern, mediated by changes in cognitive structure. It may be that the current evidence for inclusiveness moderation is actually taking the affective route and promoting social inclusion this way instead of changing category structure. Future work should thus use manipulations of category inclusiveness that are independent of any possible affective influence (i.e., not pronouns and not mood) to perform a refined test of whether superordinate identity has cognitive effects on crossed category representation and evaluation independent of mood. In summary, the slight deviations from the model’s predicted pattern and process that are observed in the literature can be accounted for by simultaneous activation of the two routes. Future work should specify exactly how this dual access might operate, and whether the general predictions of the model hold up in different situations. In future work then, we advocate that tests of the model that include refined moderators that would be expected to have effects on only one of the two possible mediational routes (i.e., instead of moderators that may have overlapping effects such as pronoun primes). In addition to refined moderators, we also recommend measurement of both category structure and affective processes for all potential moderators. The combination of clearly defined moderators (predicted to have effects via only one of the two routes, excluding the other) and measures of both processing routes can reveal a “double dissociation” or mutual exclusivity between affective and cognitive processes that will help to support or refute the model.
66
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Specific emotion An additional consideration for future work and possible refinements to the model is specific emotion, which is broadly expected to take the affective route but may have more specific effects depending on which emotion is primed. Reactions to crossed categorisation targets who share an out-group category membership with those who are responsible for a negative emotion will elicit generalized tit-for-tat behaviour that is negative (Borden, Bowen, & Taylor, 1971; Ohbuci & Kambara, 1985). From this perspective and in accord with the Classical Neo-Associative Model of aggression, the shared negativity among specific states such as anger, sadness, and disgust may all yield similar effects in the evaluations of the targets (Berkowitz, 1993). Nevertheless, specific emotions have distinct attentional (Blaney, 1986) and information feedback functions (Miller et al., 1998). Sadness may produce a broader categorisation (Bodenhausen, 1993; possibly via the cognitive route), less evaluative extremity (Gross, Holtz, & Miller, 1995) and, in turn, less differentiation across the four crossed categorisation targets by comparison with other equally strong, but different negative states such as, for instance, anger. The priming function of anger may more strongly direct attention towards the negative valence of out-group stimuli. This is likely to augment the negativity of evaluation of targets who have an out-group identity, thereby more strongly inducing a social exclusion pattern. The notion that differential reactions may arise as a consequence of the particular type of negative emotion that is elicited does not contradict Berkowitz’s (1993) theorising. Instead, it argues for additional moderating effects as a consequence of the specific negative emotion that is elicited. Anger implies movement against the out-group (Miller et al., 1998, p.404). By contrast, disgust or fear have different cognitive and behavioural implications. They imply movement away from the out-group. As shown by Mackie, Devos, and Smith (2000) the experience of fear or contempt prompts non-offensive behaviours that reflect the desire to avoid the outgroup, whereas the experience of anger or irritation is linked to offensive behaviours that reflect a desire to move against or harm the out-group. Under sadness, if the situation readily allows withdrawal to a place where one can be alone, actors are less likely to exhibit either hostile approach or strong avoidance tendencies towards mixed or double out-group targets. In response to induced sadness, attributions of a lack of sympathy and understanding will more readily be made to persons with out-group identities. Neuropsychological evidence for a differential hemisphere activation for approach (aggression) versus avoidance behaviours (HarmonJones & Sigelman, 2001) adds some credence to the possibility of dissociating specific emotional effects on crossed categorisation contexts.
CROSSED CATEGORISATION 67
Whilst specific emotion may be expected to fit broadly into the dualroute model, the specification of different emotions may lead to qualitatively and quantitatively different patterns of evaluative outcome, or even different routes being taken. Investigating the extent to which the broad distinction between affective and cognitive processing routes can account for specific contextual antecedents such as emotions, and how it might help us understand the complexities of such moderators in relation to multiple categorisation, is an intriguing goal for future research. CONCLUSIONS In this chapter we have presented a new theoretical framework within which to understand the effects of different antecedent contexts on observed patterns of evaluation across crossed category composite groups. By making a significant theoretical leap from previous work, we have specified a general model within which all work examining the moderating influence on evaluations can be assessed. Importantly, this model makes clear predictions as to which processes and outcomes are expected in different antecedent contexts, and makes a significant theoretical development by suggesting a dual-route model for predicting the processing and evaluative implications of context and categorisation in these situations. Our review of the most recent literature within the framework of this model has generally supported this dual-route view of process. Importantly this analysis has also demonstrated that whilst interactive and additive combinations of contextual factors can lead to (sometimes complex) deviations from the general predictions of the model, examination of the precise affective and cognitive antecedents means that the outcomes of such interactions can be explained by the general principles outlined in the model. We have also suggested how further research can help to support, refine (or refute) this model. In particular, investigation and examination of the precise interactive and additive effects of moderators and which route is taken is an important line of work for future research. Crossed categorisation represents the logical extension and detailed specification of basic social categorisation theory and research in social psychology. The recent advances discussed in this chapter demonstrate that whilst crossed categorisation phenomena represent a varied and multifarious area of intergroup relations, basic affective and cognitive processing routes seem to lie at the heart of observed effects. In a world that is increasingly characterised by multiple, cross-cutting group affiliations the study of the psychological processes and implications of such intergroup relations is becoming increasingly pertinent. Experimental work into crossed categorisation is still in relative infancy, but the theoretical and empirical advances outlined above suggest that the future
68
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
clarification and refinement of the extant phenomena associated with such complex intergroup affiliation will greatly aid our attempts to understand the nature of social classification, prejudice, and intergroup conflict. REFERENCES Arcuri, L. (1982). Three patterns of social categorization in attribution memory. EuropeanJournal of Social Psychology, 12,271–282. Bargh, J.A. (1997). The automaticity of everyday life. In R.Wyer Jr. (Ed.), The automaticity ofeveryday life: Advances in social cognition Vol. 10 (pp. 1–61). Mahwah, NJ: Lawrence Erlbaum Associates Inc. Baron, R.M., & Kenny, D.A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal ofPersonality and Social Psychology, 51,1173–1182. Berkowitz, L. (1993). Aggression: Its causes, consequences, and control.New York: McGraw-Hill. Bettencourt, B.A., Brewer, M.B., Croak, M., & Miller, N. (1992). Cooperation and the reduction of intergroup bias: The role of reward structure and social orientation. Journal ofExperimental Social Psychology, 28,301–319. Blaney, P.H. (1986). Affect and memory: A review. Psychological Bulletin, 99,229– 246. Bodenhausen, G.V. (1993). Emotions, arousal, and stereotypic judgments: A heuristic model of affect and stereotyping. In D.M.Mackie & D.L.Hamilton (Eds.), Affect, cognition, andstereotyping: Interactive processes in group perception(pp. 13–37). San Diego, CA: Academic Press. Borden, R.J., Bowen, R., & Taylor, S.P. (1971). Shock-setting behavior as a function of physical attack and extrinsic reward. Perceptual & Motor Skills, 33,563–568. Brewer, M.B. (1979). In-group bias in the minimal intergroup situation: A cognitive-motivational analysis. Psychological Bulletin, 86,307–324. Brewer, M.B., & Gardner, W. (1996). Who is this “We”? Levels of collective identity and self representations. Journal of Personality and Social Psychology, 71,83–93. Brewer, M.B., Ho, H.K., Lee, J.Y., & Miller, N. (1987). Social identity and social distance among Hong Kong schoolchildren. Personality and Social Psychology Bulletin, 13,156–165. Brewer, M.B., & Miller, N. (1984). Beyond the contact hypothesis: Theoretical perspectives on desegregation. In N.Miller & M.B.Brewer (Eds.), Groups in contact: The psychology ofdesegregation(pp. 281–302). Orlando, FL: Academic Press. Brewer, M.B., & Miller, N. (1988). Contact and cooperation: When do we work? In P.Katz & D.Taylor (Eds.), Eliminating racism: Profiles in controversy(pp. 315–326). New York: Plenum. Bruner, J.S. (1957). On perceptual readiness. Psychological Review, 64,123–152. Cacioppo, J.T., Marshall-Goodell, B.S., Tassinary, L.G., & Petty, R.E. (1992). Rudimentary determinants of attitudes: Classical conditioning is more effective
CROSSED CATEGORISATION 69
when prior knowledge about the attitude stimulus is low than high. Journal of Experimental Social Psychology, 28,207–233. Cadinu, M.R., & Rothbart, M. (1996). Self-anchoring and differentiation processes in the minimal group setting. Journal of Personality and Social Psychology, 70, 661–677. Campbell, D.T. (1956). Enhancement of contrast as a composite habit. Journal of Abnormaland Social Psychology, 53,350–355. Cook, S.W. (1978). Interpersonal and attitudinal outcomes in cooperating interracial groups. Journal of Research & Development in Education, 12,97– 113. Cozby, P.C. (1972). Self-disclosure, reciprocity and liking. Sociometry, 35,151–160. Crisp, R.J. (1998). Crossed categorization and intergroup bias: Context, process, and socialconsequences.Unpublished doctoral dissertation, Cardiff University. Crisp, R.J., & Hewstone, M. (1998). Dynamic crossing of minimal categories.Unpublished dataset. Cardiff University. Crisp, R.J., & Hewstone, M. (1999a). Differential evaluation of crossed category groups: Patterns, processes, and reducing intergroup bias. Group Processes and Intergroup Relations,2,303–333. Crisp, R.J., & Hewstone, M. (1999b). Subcategorization of physical stimuli: Category differentiation and decategorization processes. European Journal of Social Psychology, 29,665–671. Crisp, R.J., & Hewstone, M. (2000a). Crossed categorization and intergroup bias: The moderating roles of intergroup and affective context. Journal of Experimental SocialPsychology, 36,357–383. Crisp, R.J., & Hewstone, M. (2000b). Multiple categorization and social identity. In D. Capozza & R.Brown (Eds.), Social identity theory: Trends in theory and research(pp. 149– 166). Beverly Hills, CA: Sage. Crisp, R.J., & Hewstone, M. (2001). Multiple categorization and implicit intergroup bias: Differential category dominance and the positive-negative asymmetry effect. EuropeanJournal of Social Psychology, 31,45–62. Crisp, R.J., Hewstone, M., & Cairns, E. (2001a). Multiple identities in Northern Ireland: Hierarchical ordering in the representation of group membership. British Journal of SocialPsychology, 40,501–514. Crisp, R.J., Hewstone, M., Richards, Z., & Paolini, S. (in press). Inclusiveness and crossed categorization: Effects on co-joined category evaluations of in-group and out-group primes. British Journal of Social Psychology,. Crisp, R.J., Hewstone, M., & Rubin, M. (2001b). Does multiple categorization reduce intergroup bias? Personality and Social Psychology Bulletin, 27,76–89. Das, J.P., & Nanda, P.C. (1963). Mediated transfer of attitudes. Journal of Abnormal andSocial Psychology, 66,12–16. Deschamps, J.-C (1977). Effects of crossing category membership on quantitative judgment. European Journal of Social Psychology, 7,517–521. Deschamps, J.-C, & Doise, W. (1978). Crossed category memberships in intergroup relations. In H.Tajfel (Ed.), Differentiation between social groups(pp. 141– 158). Cambridge: Cambridge University Press. Dijksterhuis, A., & van Knippenberg, A. (1998). Inhibition, aberdeen, and other cloudy subjects. In R.S.Wyer Jr. (Ed.), Advances in Social Cognition, 11,83–96.
70
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Doise, W. (1978). Groups and individuals: Explanations in social psychology.Cambridge: Cambridge University Press. Dovidio, J.F., Evans, N., & Tyler, R.B. (1986). Racial stereotypes: The contents of their cognitive representations. Journal of Experimental Social Psychology, 22, 22–37. Dovidio, J.F., Gaertner, S.L., Isen, A.M., & Lowrance, R. (1995). Group representations and intergroup bias: Positive affect, similarity, and group size. Personality and Social PsychologyBulletin, 21,856–865. Dovidio, J.F., Gaertner, S.L., Isen, A.M., Rust, M., & Guerra, P. (1998). Positive affect, cognition, and the reduction of intergroup bias. In C.Sedikides, J.Schopler, & C.A.Insko (Eds.), Intergroup cognition and intergroup behavior (pp. 337–366). Mahwah, NJ: Lawrence Erlbaum Associates Inc. Ensari, N., & Miller, N. (1998). Effect of affective reactions by the out-group on preferences for crossed categorization discussion partners. Journal of Personality and Social Psychology, 75,1503–1527. Ensari, N., & Miller, N. (2001). Decategorization and the reduction of bias in the crossed categorization paradigm. European Journal of Social Psychology, 31, 193–216. Fazio, R.H., Jackson, J.R., Dunton, B.C., & Williams, C.J. (1995). Variability in automatic activation as an unobtrusive measure of racial attitudes: A bona fide pipeline?Journal ofPersonality and Social Psychology, 69,1013–1027. Fazio, R.H., Sanbonmatsu, D.M., Powell, M.C., & Kardes, F.R. (1986). On the automatic activation of attitudes. Journal of Personality and Social Psychology, 50,229–238. Fiske, S.T., & Neuberg, S.L. (1990). A continuum of impression formation, from category-based to individuating processes: Influences of information and motivation on attention and interpretation. In M.P.Zanna (Ed.), Advances in experimental social psychology(Vol. 23, pp. 1–74). New York: Random House. Gaertner, S.L., & Dovidio, J.F. (2000). Reducing intergroup bias: The Common Ingroup IdentityModel.Philadelphia, PA: Psychology Press. Gaertner, S.L., Dovidio, J.F., Anastasio, P.A., Bachman, B.A., & Rust, M.C. (1993). The common in-group identity model: Recategorization and the reduction of intergroup bias. In W. Stroebe and M.Hewstone (Eds.), European review of social psychology(Vol. 2, pp. 247– 278). Chichester, UK: John Wiley. Gaertner, S.L., Mann, J.A., Dovidio, J.F., Murrell, A.J., & Pomare, M. (1990). How does co-operation reduce intergroup bias?Journal of Personality and Social Psychology, 59,692–704. Goodstein, L.D., & Reinecker, V.M. (1974). Factors affecting self-disclosure: A review of the literature. In B.A.Maher (Ed.), Progress in experimental personality research(pp. 49–77). San Diego, CA: Academic Press. Gross, S.R., Holtz, R., & Miller, N. (1995). Attitude certainty. In R.E.Petty & J.A.Krosnick (Eds.), Attitude strength: Antecedents and consequences(pp. 215– 245). Mahwah, NJ: Lawrence Erlbaum Associates Inc. Harmon-Jones, E., & Sigelman, J. (2001). State anger and prefrontal brain activity: Evidence that insult-related relative left-prefrontal activation is associated with experienced anger and aggression. Journal of Personality and Social Psychology, 80,797–803.
CROSSED CATEGORISATION 71
Harrington, H.J. (1988). The effects of personal contact on intergroup relations.Unpublished dissertation , University of Southern California, Los Angeles. Hewstone, M., Islam, M.R., & Judd, C.M. (1993). Models of crossed categorization and intergroup relations. Journal of Personality and Social Psychology, 64,779–793. Hewstone, M., Rubin, M., & Willis, H. (2002). Intergroup bias. Annual Review of Psychology,53,575–604. Higgins, E.T., & King, G.A. (1981). Accessibility of social constructs: Information processing consequences of individual and contextual variability. In N.Cantor & J.F.Kihlstrom (Eds.), Personality, cognition, and social interaction(pp. 69– 122). Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Isen, A.M., & Daubman, K.A. (1984). The influence of affect on categorization. Journal ofPersonality and Social Psychology, 47,1206–1217. Isen, A.M., Niedenthal, P.M., & Cantor, N. (1992). An influence of positive affect on social categorization. Motivation and Emotion, 16,65–78. Isen, A.M., Shalker, T., Clark, M.S., & Karp, L. (1978). Affect, accessibility of material and behavior: A cognitive loop? Journal of Personality and Social Psychology, 36,1–2. Jourard, S.M. (1971). The transparent self (Rev. ed.). New York: Van Nostrand Reinhold. Laird, J.D., Wagener, J.J., Halal, M., & Szegda, M. (1982). Remembering what you feel: The effects of emotion on memory. Journal of Personality and Social Psychology, 42,646–657. Maass, A., & Schaller, M. (1991). Intergroup biases and the cognitive dynamics of stereotype formation. In W.Stroebe, & M. Hewstone (Eds.), European review of social psychology(Vol. 2, pp. 189–209). Chichester, UK: John Wiley & Sons Ltd. Mackie, D.M., Devos, T., & Smith, E.R. (2000). Intergroup emotions: Explaining offensive action tendencies in an intergroup context. Journal of Personality and Social Psychology, 79,602–616. Mackie, D.M., & Worth, L.T. (1991). “Feeling good but not thinking straight”: Positive mood and persuasion. In J.P.Forgas (Ed.), Emotion and social judgments(pp. 201–220). Oxford: Pergamon. Macrae, C.N., Bodenhausen, G.V., & Milne, A.B. (1995). The dissection of selection in person perception: Inhibitory processes in social stereotyping. Journal of Personality and SocialPsychology, 69,397–407. Marco, C.A., & Suls, J. (1993). Daily stress and the trajectory of mood: Spillover, response assimilation, contrast, and chronic negativity. Journal of Personality and Social Psychology,64,1053–1063. Marcus-Newhall, A., Miller, N., Holtz, R., & Brewer, M.B. (1993). Cross-cutting category membership with role-assignment: A means of reducing intergroup bias. British Journal ofSocial Psychology, 32,125–146. Migdal, M., Hewstone, M., & Mullen, B. (1998). The effects of crossed categorization in intergroup evaluations: A meta-analysis. British Journal of Social Psychology, 37,303–324. Miller, N., Urban, L.M., & Vanman, E.J. (1998). A theoretical analysis of crossed social categorization effects. In C.Sedikides, J.Schopler, & C.Insko (Eds.),
72
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Intergroup cognitionand intergroup behavior.Hillsdale NJ: Lawrence Erlbaum Associates Inc. Mullen, B., Migdal, M.J., & Hewstone, M. (2001). Crossed categorization vs simple categorization and intergroup evaluations: A meta-analysis. European Journal of SocialPsychology, 31,721–736. Oakes, P.J. (1987). Salience and categorization. In J.C.Turner, M.A.Hogg, P.J.Oakes, S.D. Reicher & M.S.Wetherell (Eds.), Rediscovering the social group: A self-categorizationtheory.Oxford: Blackwell. Oakes, P.J., Haslam, S.A., & Turner, J.C. (1994). Stereotyping and social reality.Oxford: Blackwell. Ohbuchi, K., & Kambara, T. (1985). Attacker’s intent and awareness of outcome, impression management, and retaliation. Journal of Experimental Social Psychology, 21,321–330. Otten, S., & Moskowitz, G.B. (2000). Evidence for implicit evaluative in-group bias: Affect-based spontaneous trait inference in a minimal group paradigm. Journal of ExperimentalSocial Psychology, 36,77–89. Otten, S., & Wentura, D. (1999). About the impact of automaticity in the minimal group paradigm: Evidence from affective priming tasks. European Journal of Social Psychology, 29,1049–1071. Park, B., & Rothbart, M. (1982). Perceptions of out-group homogeneity and levels of social categorization: Memory for the subordinate attributes of in-group and out-group members. Journal of Personality and Social Psychology, 42, 1051–1068. Perdue, C.W., Dovidio, J.F., Gurtman, M.B., & Tyler, R.B. (1990). “Us” and “Them”: Social categorization and the process of intergroup bias. Journal of Personality and SocialPsychology, 59,475–486. Perdue, C.W., & Gurtman, M.B. (1990). Evidence for the automaticity of ageism. Journal ofExperimental Social Psychology, 26,199–216. Rehm, J., Lilli, W., & Van Eimeren, B. (1988). Reduced intergroup differentiation as a result of self-categorization in overlapping categories. A quasi-experiment. European Journal ofSocial Psychology, 18,375–379. Repetti, R.L. (1993). Short-term effects of occupational stressors on daily mood and health complaints. Health Psychology, 12,125–131. Rubin, J.M., & Hewstone, M. (1998). Social identity theory’s self-esteem hypothesis: A review and some suggestions for clarification. Personality and Social Psychology Review, 2,40–62. Staats, A.W., & Staats, C.K. (1958). Attitudes established by classical conditioning. Journal ofAbnormal and Social Psychology, 57,37–40. Stangor, C., Lynch, L., Duan, C., & Glass, B. (1992). Categorization of individuals on the basis of multiple social features. Journal of Personality and Social Psychology, 62,207–218. Tajfel, H. (1959). Quantitative judgment in social perception. British Journal of Psychology, 50,16–29. Tajfel, H., Flament, C., Billig, M., & Bundy, R.F. (1971). Social categorization and intergroup behaviour. European Journal of Social Psychology, 1,149–178. Tajfel, H., & Wilkes, A.L. (1963). Classification and quantitative judgment. British Journal ofPsychology, 54,101–113.
CROSSED CATEGORISATION 73
Taylor, S.E. (1991). Asymmetrical effects of positive and negative events: The mobilization/ minimization hypothesis. Psychological Bulletin, 110,67–85. Taylor, S.E., Fiske, S.T., Etkoff, N.L., & Ruderman, A.J. (1978). Categorical and contextual bases of person memory and stereotyping. Journal of Personality and Social Psychology, 36,778–793. Teasdale, J.D., & Fogarty, S.J. (1979). Differential effects of induced mood on retrieval of pleasant and unpleasant events from episodic memory. Journal of Abnormal Psychology, 88,248–257. Teasdale, J.D., & Russell, M.L. (1983). Differential aspects of induced mood on the recall of positive, negative, and neutral words. British Journal of Clinical Psychology, 22,163–171. Turner, J.C, Hogg, M.A., Oakes, P.J., Reicher, S.D., & Wetherell, M.S. (1987). Rediscovering the social group: A self-categorization theory.Oxford: Basil Blackwell. Urada, D., & Miller, N. (2000). The impact of positive mood and category importance on crossed categorization effects. Journal of Personality and Social Psychology, 78,417–433. Urban, L.M., & Miller, N.M. (1998). A theoretical analysis of crossed categorization effects: A meta-analysis. Journal of Personality and Social Psychology, 74,894–908. Vanbeselaere, N. (1991). The different effects of simple and crossed categorizations: A result of the category differentiation process or of differential category salience? In W.Stroebe & M. Hewstone (Eds.), European review of social psychology(Vol. 2, pp. 247–278). Chichester, UK: John Wiley. Vanbeselaere, N. (1996). The impact of differentially valued overlapping categorizations upon the differentiation between positively, negatively and neutrally evaluated social groups. European Journal of Social Psychology, 26, 75–96. van Knippenberg, A., & Dijksterhuis, A. (2000). Social categorization and stereotyping: A functional perspective. In W.Stroebe and M.Hewstone (Eds.), European review of socialpsychology(Vol. 11, pp. 105–144). Chichester, UK: John Wiley. van Knippenberg, A., van Twuvyer, M., & Pepels, J. (1994). Factors affecting social categorization processesin memory. British Journal of Social Psychology, 33, 419–431. Vanman, E.J., Paul, B.Y., Ito, T., & Miller, N. (1997). The modern face of prejudice and structural features that moderate the effect of cooperation on affect. Journal of Personality &Social Psychology, 73,941–959. Van Twuyver, M., & van Knippenberg, A. (1995). Social categorization as a function of priming. European Journal of Social Psychology, 25,695–702. Van Twuyver, M., & van Knippenberg, A. (1999). Social categorization as a function of relative group size. British Journal of Social Psychology, 38,135– 156. Vescio, T.K., Hewstone, M., Crisp, R.J., & Rubin, M. (1999). Perceiving and responding to multiply categorizable individuals: Cognitive processes, identity motives and discriminatory consequences. In D.Abrams & M.A.Hogg (Eds.), Social identity and social cognition(pp. 111–140). Oxford: Blackwell.
74
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Wilder, D.A. (1978). Reduction of intergroup discrimination through individuation of the out-group. Journal of Personality and Social Psychology, 36,1361– 1374. Worth, L.T., & Mackie, D.M. (1987). Cognitive mediation of positive mood in persuasion. Social Cognition, 5,76–94. Wyer, N.A., Sherman, J.W., & Stroessner, S.J. (2000). The roles of motivation and ability in controlling the consequences of stereotype suppression. Personality and Social PsychologyBulletin, 26,13–25.
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY, 2002, 13, 75–109
Stereotype accuracy Carey S.Ryan University of Nebraska at Omaha, USA
Methodological issues and empirical findings from stereotype accuracy research are reviewed. Methodological issues include the limitations of accuracy criteria and three methods of comparing perceived to actual group characteristics—signed discrepancies, absolute discrepancies, and within-subject correlations. Empirical findings concern the roles of cognitive processes, status and power, and social ideology in intergroup perceptions; stereotype development; individual differences in stereotyping; and stereotype use. It is argued that stereotype accuracy research is neither easy to do nor politically popular, but that it may challenge existing theory and stimulate new ideas about the nature of stereotyping processes. Scientists and lay persons alike have long assumed that stereotypes are inaccurate and that their use necessarily results in inaccurate judgements. The reasons are not hard to understand. The stereotypes that women are passive or that Blacks are athletic, for example, are no doubt erroneous if they are meant to imply that all women or all Blacks are so. And on what basis would one determine the actual passivity of women or the actual athleticism of Blacks anyway? Further, the notion that stereotypes can be accurate seems to imply that group attributes should be applied to individual group members so long as those attributes (or stereotypes) are accurate. This implication seems highly offensive in a society that values
© 2002 European Association of Experimental Social Psychology http://www.tandf.co.uk/journals/pp/02699931.html DOI: 10.1080/10463280240000037
76
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
the individual and his or her unique merits. Finally, the evidence of discrimination against certain groups is clear and disturbing; people should be concerned about stereotypes being used to judge individual group members—whether the stereotype is generally accurate at the group level or not. Stereotypes, however, are no longer conceptualised as beliefs about the characteristics of all group members. And although this has actually been true for some time (Allport, 1954), only relatively recently have researchers devised ways of empirically assessing the extent to which individuals perceive group members to conform to the cultural stereotype or to vary on stereotype-relevant attribute dimensions (Linville, Fischer, & Salovey, 1989; Park, Judd, & Ryan, 1991). In recent work, stereotypes have been conceptualised as including at least two components: beliefs about the modal attributes or central tendency of a group (i.e., stereotypicality) and perceptions of the variability of individual group members with respect to stereotype-relevant attributes. These two components of stereotypes have been shown to be both methodologically and conceptually distinct from each other and from prejudice. In other words, even people who perceive a group to be highly stereotypic (i.e., to have a higher mean rating on stereotypic than on counterstereotypic attribute dimensions) may perceive a great deal of variability among individual group members. And people who perceive a group to be both highly stereotypic and homogeneous do not necessarily feel negatively towards the group. In short, stereotypes have come to be conceptualised as beliefs about the central tendency and variability of group members. And although many stereotypes may be unjustly negative, beliefs about a group are no longer required to be either negative or inaccurate in order to be defined as stereotypes (Ashmore & Del Boca, 1981). The implicit belief that actual differences between groups either do not exist or are so minimal as to be unworthy of investigation may also be changing (Lee, Jussim, & McCauley, 1995). Although such a colour-blind perspective is undoubtedly well intentioned and has certainly been useful in highlighting the many similarities between groups and the serious problems that often result from stereotyping (e.g., Ryan, Bogart, & Vender, 2000;
Address correspondence to: Carey S Ryan, Department of Psychology, University of Nebraska at Omaha, Omaha, NE, 68182–0274, USA, Email:
[email protected] Preparation of this manuscript was supported by National Institute of Mental Health Grant R01 MH62061. I am grateful to Alice Eagly, Miles Hewstone, Charles Judd, Wolfgang Stroebe, Molly Wernli, and two anonymous reviewers for their helpful comments on earlier versions of this chapter.
STEREOTYPE ACCURACY 77
Ryan, Judd, & Park, 1996a; Sagar & Schofield, 1980), it has also restricted the range of issues that researchers consider. Moreover, the failure to recognise and appreciate some group differences may itself be considered offensive, inasmuch as it implies that ethnic heritage, religious traditions, and experiences with discrimination are unimportant. Jewish individuals, for example, seem more likely to celebrate Passover than non-Jewish people, and African Americans likely have different experiences with discrimination than do White Americans. A more useful approach, then, may be to identify the circumstances under which stereotypes should be considered to be potentially applicable and the ways that stereotyperelevant information might be most effectively utilised. Yet the implicit assumption that stereotypes are inherently unjust and inaccurate remains strong. Even a great deal of social cognition research continues to be based on the assumption that stereotypes serve primarily to increase cognitive efficiency by limiting information-processing requirements. People are certainly more likely to apply stereotypes to judgements of individuals when circumstances, such as time pressure and distractions, limit their mental capacity (Hamilton & Trolier, 1986). Nevertheless, it seems unlikely that stereotypes as they are now defined always (and it is possible that they do not even usually) involve less effortful information processing (Oakes & Turner, 1990). Perceivers may expend considerable cognitive resources developing elaborate cognitive representations, seeking and incorporating new information, and utilising existing information about groups that are important to them, such as work groups, in-groups, gender groups, and sports teams (Ryan, Robinson, & Hausmann, in press). My purpose in this chapter is to consider these types of interesting theoretical questions, that is, the types of questions that arise when one considers that stereotypes may sometimes be accurate and useful. However, this task is difficult, without first considering how one might assess accuracy. I therefore begin by discussing the choice of an appropriate accuracy criterion—an issue that has been the focus of considerable debate in both the person perception (Cronbach, 1955; Funder, 1995; Gage & Cronbach, 1955; Kenny, 1991; Oakes & Reynolds, 1997; Swann, 1984) and group perception literature (Judd & Park, 1993; Ryan & Bogart, 2001; Stangor, 1995). I then discuss various methods for determining the extent to which individuals’ perceptions of a group correspond to the criterion. The remainder of the chapter focuses on what I believe to be some of the most promising and important theoretical issues that are only beginning to be addressed in accuracy research.
78
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
DETERMINING THE ACTUAL CHARACTERISTICSOF A GROUP: THE ACCURACY CRITERION The most frequently used method for determining a group’s actual characteristics has been to ask a sample of (or all) individual group members to rate themselves as individuals on the stereotype-relevant attributes of interest (Diekman, Eagly, & Kulesa, 2002; Judd, Ryan, & Park, 1991; Madon, Jussim, Keiper, Eccles, Smith, & Palumbo 1998; McCauley & Stitt, 1978; Nisbett & Kunda, 1985; Ryan, 1996; Ryan & Bogart, 2001). Researchers then compute the mean and variability of these individual self-ratings to determine the stereotypicality and variability of the group as a whole. So, for example, to determine the actual aggressiveness of men, one might ask each of 100 men to indicate whether he is aggressive and, if so, the extent to which he is aggressive. The researcher can compute the percentage of men who consider themselves to be aggressive, the mean self-rating, the range (i.e., the difference between the lowest and highest self-ratings in the group), and the standard deviation of these self-ratings. The mean and percentages provide measures of the group’s actual stereotypicality, whereas the range and standard deviation provide measures of the group’s actual variability (Park et al., 1991) on the aggressiveness dimension. Two potential sources of bias are of concern when self-report data are used. The first source is sample selection. If one is to argue that the sample from which the self-ratings were gathered is representative of the larger group of interest, self-ratings must be gathered from a probability (e.g., randomly selected—not haphazardly selected) sample or from all of the individual group members. In the example described above, there is no reason to assume that the 100 men are representative of all men unless they constitute a probability sample of all men. The second source of bias is the self-report judgements themselves. As Judd and Park (1993) have pointed out, a lack of knowledge, memory lapses, the context and format in which the self-ratings were gathered, and social desirability may influence selfreports (Schwartz, 1999). Individuals may be unaware of the extent to which they possess some attributes, such as agreeableness or neuroticism. And even when individuals are asked more factual questions, for example, questions concerning the frequency of a specific behaviour (e.g., asking students whether or how often they drink alcohol), responses may be influenced by an inability to remember all instances of the behaviour as well as by beliefs about what constitutes a socially desirable response. The extent to which such biases affect conclusions regarding stereotype accuracy remains unclear. Only a few researchers have systematically examined potential biases. Judd et al. (1991) examined differences in accuracy conclusions as a function of trait versus attitude items, using
STEREOTYPE ACCURACY 79
accuracy criteria that were derived from the self-ratings of probability samples of group members. Judd et al. reasoned that the attitude items may have provided a better measure of the group’s actual characteristics than the trait items—individuals would seem better able to identify their preferences than to determine, for example, the degree to which they are analytical. The results indicated that although perceivers overestimated stereotypic traits more than stereotypic attitudes, the same in-group-out-group differences in accuracy emerged for both types of items. In other words, although the magnitude of the effects was greater for trait than for attitude items, the same theoretical conclusions could have been drawn from either type of item. A more serious concern is social desirability (Judd & Park, 1993; Ryan, Park, & Judd, 1996b). Individuals may be reluctant to rate themselves in ways that they perceive to be negative or socially undesirable. So far, however, it appears that conclusions concerning stereotype accuracy when accuracy criteria are derived from group members’ self-ratings may not be as strongly affected by attribute valence as one might expect. Two studies in which self-report data were used examined effects as a function of positive versus negative items. One study examined in-group and outgroup stereotypes among Black/African American and White American college students (Ryan, 1996). The results indicated that the tendency for Blacks (vs Whites) to overestimate the group stereotypes did not depend on attribute valence. Similarly, the tendency for Blacks (vs Whites) to underestimate the variability of the groups was about equally true for positive and negative attributes. More recently, new group members demonstrated decreased stereotype overestimation over time and, although the effect was greater for negative than for positive attributes, decreased stereotype overestimation was evident in both types of attributes (Ryan & Bogart, 2001). Similarly, the overall tendency to overestimate the outgroup more than the in-group stereotype was evident on both positive and negative attributes. Again, then, although there were differences due to attribute valence, the theoretical conclusions were the same for both types of items. An alternative to gathering self-report data would be to administer a standardised test (e.g., to assess intelligence or personality) to a probability sample or to all group members (Madon et al., 1998). Standardised tests seem likely to yield more objective assessments of a group’s actual characteristics, at least partly because they rely on what Biernat (1995) has referred to as common-rule or externally anchored scales. Such scales minimise the problem of shifting standards (Biernat, Manis, & Nelson, 1991), that is, the tendency to define attribute dimensions differently depending on the group that is judged. Judging a person (including oneself) to be highly verbal, for example, may mean something different if the person is female than if the person is male, whereas SAT scores are defined
80
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
identically regardless of group membership. But standardised tests have other limitations. Standardised tests do not exist for many attributes that are contained in stereotypes. And standardised tests can be costly to administer. In addition, responses to standardised tests may be influenced by other sources of bias. For example, the administration of some tests requires greater involvement of an interviewer whose own expectations and biases, interpersonal skills, and level of expertise may influence responses. Similarly, participants’ performance may be affected by their expectations and concerns about the outcomes of the tests. Finally, standardised tests, for example, intelligence tests, may be culturally biased (Helms, 1992) or have other conceptual limitations (Neisser et al., 1996). Another method for determining a group’s actual characteristics is to use public records, such as census data (McCauley & Stitt, 1978; Wolsko, Park, Judd, & Wittenbrink, 2000, Experiment 2). Such data have obvious advantages. There is an attempt to gather the data systematically from all members of the population; the data are readily available; and the data are less susceptible to the shifting standards problem (Biernat, 1995; Biernat et al., 1991). But public records are not without their limitations. US census data, for example, are gathered for characteristics that tend to be relatively well defined, such as occupation and household size, as opposed to attitudes or personality traits that may be more clearly stereotypic of a group. One also needs to consider how the data contained in public records were originally gathered. There has been considerable debate as to whether the US census can be adequately implemented, given such factors as the size of the population, its vast geographic distribution, and the mobility of the population. In 1990, approximately 25 million, or 1 in 10, people in the US were not properly counted (Anderson & Fienberg, 2000). Moreover, omissions occurred disproportionately among minority groups, whereas erroneous enumerations occurred disproportionately among nonminorities. Such problems undermine the validity of census data for some of the very groups whose stereotypes are of greatest concern. Finally, many public records, including census data, are based on individual self-reports. As such, they are subject to the limitations of self-reports described above. Some public records rely primarily on the judgements of people, usually considered experts, who must identify instances of the event, determine whether alleged instances actually occurred and, if so, whether they should be reported. Such records often provide useful accuracy criteria, although they have their own sets of limitations. Consider the use of crime statistics to assess the degree to which African Americans are aggressive or, more precisely, engage in criminal activities. African Americans may be more likely than Whites to be initially identified as criminals because they are targeted for greater scrutiny (Cole, 1999; Harris, 1999). Crime statistics may also be influenced by subsequent judgements that are made by police, victims, prosecutors, defence attorneys, judges, and juries. Another type of
STEREOTYPE ACCURACY 81
public record, that is, records of academic performance (Ashton & Esses, 1999) may also result from expert judgements (e.g., teacher evaluations). Again, although they may provide useful accuracy criteria, it is important to remember that they may be influenced by factors such as teachers’ expectations and the social climate in which students’ academic performance was assessed. An interesting development in stereotype accuracy research has been the use of effect sizes derived from meta-analyses as accuracy criteria (Hall & Carter, 1999; Swim, 1994). Such criteria seem likely to increase in importance for two reasons. First, such data have been gathered for a variety of attributes that are relevant to many of the stereotypes that are important to researchers, for example, gender differences in verbal and mathematical abilities (Swim, 1994). Second, meta-analytic effect sizes are derived from scientific studies in which specific procedures have been systematically implemented to minimise bias (Eagly & Diekman, 1997). But the results from meta-analyses may have other limitations. As Swim (1994) indicates, for example, one must consider the circumstances under which the original data from which the effect sizes are computed were gathered. Such data may have been gathered in artificial settings from college students and may not therefore represent everyday behaviours or the larger population of interest. Summary It should be obvious from the discussion above that a perfect accuracy criterion does not exist. Researchers must therefore consider the limitations associated with the accuracy criteria they used when interpreting their conclusions. Ultimately, accuracy research will be most informative when it is based on multiple criteria, the potential biases associated with each criterion are identified, and similarities and differences in research conclusions are systematically examined as a function of the accuracy criterion. Findings that replicate when different criteria are used are more compelling, whereas findings that vary as a function of the criterion may provide important methodological and theoretical insights. It must be emphasised that determining the actual characteristics of particular groups tells us little about the sources or consequences of those characteristics (cf. Ashton & Esses, 1999; Eagly & Diekman, 1997), or the susceptibility of such characteristics to change. In addition to the measurement error that may influence scores on any measure, group characteristics may emerge as a result of social, cultural, and/or biological factors. To say, for example, that a group that is relatively communal on some measure is inherently so, is to ignore the fact that such characteristics may result from differences in social experiences and economic resources (Eagly & Steffen, 1984). And although an understanding of actual group
82
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
differences may be useful to perceivers who interact with the members of different groups (Lee & Duenas, 1995) or to those who make decisions that affect the group as a whole (e.g., policy makers), perceiving variability within groups also appears to be important (Bogart, 1998; Lambert, 1995; Park & Hastie, 1987; Ryan et al., 2000; Ryan et al., 1996a). ASSESSING CORRESPONDENCE BETWEENPERCEIVED AND ACTUAL GROUPCHARACTERISTICS The second major methodological issue concerns the method of comparing perceived with actual group characteristics. In other words, assuming that an acceptable accuracy criterion is available, how might one assess the degree to which individuals’ perceptions of the group correspond to it? Three methods are considered: signed perceived-actual discrepancies, absolute perceived-actual discrepancies, and within-subject sensitivity correlations. Signed perceived-actual discrepancies The method that directly addresses traditional accuracy questions, that is, the accentuation of intergroup differences and intragroup similarities (Corneille & Judd, 1999; Doise, Deschamps, & Meyer, 1978; Krueger & Clement, 1994; Tajfel 1969; Tajfel & Wilkes, 1963), involves computing difference scores for which the sign or direction of each difference is retained. Consider the hypothetical data in Table 1 for two perceivers who were asked to estimate the percentage of group members that possessed each of two stereotypic and two counterstereotypic attributes. The third column of the table contains signed discrepancy scores (i.e., the perceived minus the actual values) for each attribute. Positive discrepancies indicate that the perceiver overestimated, and negative discrepancies indicate that the perceiver underestimated, the group’s actual percentage on the given dimension. Each perceiver’s overall signed discrepancy accuracy score is computed by subtracting the mean perceived-actual discrepancy on counterstereotypic attributes from the mean perceived-actual discrepancy on stereotypic attributes. If the mean accuracy score for a sample of participants were statistically significant and positive, participants’ stereotypes could be characterised as exaggerations of the group’s actual characteristics (Judd & Park, 1993). One could also assess overgeneralisation, that is, the extent to which participants underestimate the actual variability of group members (Judd & Park, 1993). For example, participants might be asked to indicate where the highest and lowest individual group members fall on each of the same stereotype-relevant dimensions. The difference between the highest and
STEREOTYPE ACCURACY 83
TABLE 1 Perceived and actual percentage estimates for two hypothetical perceivers
lowest estimates, that is, the range, provides a measure of perceived group variability (Park et al., 1991). One could then compare each participant’s perceived range with the actual range derived from a probability sample of group members’ self-ratings on the same dimensions. More negative discrepancies would reflect greater stereotype overgeneralisation. But interpreting discrepancies is more complicated than this example thus far suggests. Perceived-actual discrepancies may reflect stereotype inaccuracy or methodological problems that are known as response language confounds (Cronbach, 1955; Judd & Park, 1993; Ryan et al., 1996b). For example, there is a systematic tendency for people to underestimate ranges (Judd et al., 1991) and overestimate standard deviations (Judd et al., 1991; Nisbett & Kunda, 1985). (Perceived standard deviations are typically computed by the researcher from frequency distributions generated by participants, Park et al., 1991.) Thus, the task that is used to assess perceptions (indicating where the lowest and highest individual group members lie on an attribute dimension in the case of the perceived range versus distributing group members across an attribute dimension in the case of the standard deviation) influences participants’ responses, as do participants’ actual perceptions of the group. A negative range discrepancy might, therefore, reflect stereotype overgeneralisation (i.e., the underestimation of actual group variability), which would be of conceptual interest to stereotyping researchers. But it may also reflect a judgement-extremity bias, that is, a tendency to provide variability
84
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
judgements that are too low. The latter response language confound has nothing to do with stereotypes—one could have asked participants to estimate the range of any distribution and a significant overall negative discrepancy would likely result. In either case, there is evidence of inaccuracy. But if one wants to draw conclusions about stereotype accuracy, one needs evidence that such discrepancies assess stereotype inaccuracy rather than inaccuracy that results simply from the type of task that was used to assess it. The latter type of inaccuracy, which results from the task itself, is confounded with stereotype inaccuracy (hence the term response language confound). Two additional response language confounds may also influence discrepancies: judgement-elevation and judgement-positivity biases (Judd & Park, 1993; Ryan et al., 1996b). The former refers to the tendency to provide systematically higher or lower mean ratings on the response scale. In other words, to the extent that perceivers have a tendency to use the higher (or lower) end of the response scale in their ratings of a group’s central tendency as compared with the sample of group members from whom the accuracy criteria were derived, the resulting discrepancies would be larger, indicating inaccuracy in perceivers’ perceptions. A researcher might find, for example, that perceivers rate a group higher on stereotypic attributes than the group members rate themselves. This discrepancy may reflect stereotype exaggeration or a much less interesting judgement-elevation response bias. The judgement-positivity bias refers to the tendency to provide systematically higher mean ratings on positive than on negative attributes. This bias is problematic when researchers compare perceivers’ perceptions of the positivity of a group with the actual positivity of the group (e.g., with the overall positivity of group members’ self-ratings). A researcher might, for example, find that perceivers judge a group more negatively, on average, than the group members judge themselves, resulting in overestimation of the group’s negativity. Such a discrepancy may reflect prejudice. But it may also reflect a tendency for perceivers to provide more negative (or less positive) judgements in general, regardless of what or whom they were asked to judge. How, then, can researchers use discrepancies to assess stereotype accuracy? Researchers simply cannot presume to reach conclusions about absolute levels of accuracy. In other words, one cannot conclude that a given stereotype is accurate or inaccurate, on average, even when discrepancies are computed for a variety of attributes. Researchers can, however, reach conclusions regarding relative degrees of accuracy. For example, in the case of a negative perceived-actual range discrepancy, one can conclude that overall, participants underestimated the range of the group and, although it is not clear whether this difference reflects stereotype inaccuracy or a response language confound, one can say that
STEREOTYPE ACCURACY 85
more negative discrepancies reflect greater inaccuracy. One can then examine whether theoretically relevant variables that would be expected to be associated with stereotype accuracy or inaccuracy, but not with response language confounds, are correlated. For example, perceivers might be asked to judge two or more target groups, such as men and women. The researcher could then examine whether judgements of men are more accurate than judgements of women. If perceivers from two different subject groups were to provide these judgements (e.g., male and female subjects), one could examine whether male perceivers are more accurate than female perceivers. Finally, if the same subject and target groups are included, as in my example of males and females judging men and women, one could examine whether in-group judgements (e.g., men judging men and women judging women) are more accurate than out-group judgements (e.g., men judging women and women judging men). Absolute perceived-actual discrepancies One might also compare perceived to actual group values by examining the absolute value of the same discrepancies, ignoring the sign that indicates whether the perceived value is higher or lower than the actual value. Hypothetical absolute discrepancies for the same two perceivers described above are reported in the fourth column of Table 1. Because the sign or direction of the difference is disregarded, this accuracy measure does not indicate whether people overestimate or underestimate the group’s actual characteristics—larger discrepancies simply reflect greater inaccuracy. Thus, not only are these discrepancies subject to the same response language confounds described above, they also do not tell us about the types of accuracy that have been of primary interest to stereotyping researchers. Nevertheless, absolute discrepancies may be important to examine, inasmuch as they are potentially distinct from signed discrepancies. Note, for example, that the signed discrepancies in Table 1 indicate that the second perceiver is more accurate than the first, whereas the absolute discrepancies indicate that the two perceivers are equally (in) accurate. Within-subject sensitivity correlations A third method for comparing perceived to actual group values involves computing for each participant the correlation between perceived and actual characteristics of the group. Each within-subject correlation is computed across attributes so that the sample size for each correlation is based on the number of attributes for which that particular participant has complete data. Higher sensitivity correlations indicate greater accuracy— that is, greater sensitivity to differences between attributes in the group’s
86
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
actual characteristics. Consider once again the hypothetical data in Table 1. The sensitivity correlation for each perceiver is reported in the last column. Each correlation is computed by correlating the perceived values in the first column with the actual values in the second column. (Note that I first reverse-scored the perceived and actual values for counterstereotypic attributes so that the magnitude of the correlation would not be confounded with the stereotypic-counterstereotypic attribute difference.) As the data indicate, Perceiver #1 is much more sensitive than is Perceiver #2 to the fact that the group is more athletic than it is streetwise and more streetwise than it is educated. (Remember that these actual data are strictly hypothetical.) An advantage of sensitivity correlations is that they are not influenced by the types of response language confounds that make it difficult to interpret discrepancies. If a mean correlation is significantly greater than zero, one can conclude that the group is perceived accurately or, more precisely, that the perceivers in the study are sensitive to between-attribute differences in the actual characteristics (e.g., the mean or range) of the target group. The disadvantage, however, is that sensitivity correlations do not address traditional questions concerning stereotype overestimation (i.e., exaggeration) and variability underestimation (i.e., overgeneralisation). Such traditional questions require one to examine the signed discrepancies between perceived and actual values. Consider again the hypothetical data for Perceiver #1 in Table 1. If 10 were added to each of the perceived values on stereotypic attributes and subtracted from each of the perceived values for counterstereotypic attributes, the signed discrepancy score would substantially increase to 60. This discrepancy score may reflect greater stereotype exaggeration or a stronger judgement extremity bias (i.e., a stronger tendency to use the extremes of the response scale). However, the sensitivity correlation would not be affected. It would remain at 1.00 (recall that counterstereotypic attributes are reverse-scored before the correlation is computed). Note that in some accuracy studies researchers have averaged across perceivers for each attribute and then correlated these mean values with the actual group values for each attribute (e.g., Eagly & Diekman, 1997; Swim, 1994). In other cases, researchers have compared the magnitudes and statistical significance of two separately computed correlations: the correlation between perceivers’ judgements of individuals and those individuals’ group membership (e.g., the correlation between teachers’ judgements of students and students’ ethnicity) and the correlation between the same individuals’ self-ratings or performance on a standardised test and their group membership (e.g., Madon et al., 1998). These methods assess accuracy at an aggregate level; they do not assess the accuracy of individual perceivers’ stereotypes. It is possible for there to be relatively
STEREOTYPE ACCURACY 87
high accuracy at an aggregate level, even when individual perceivers are inaccurate. Relationships among signed discrepancies,absolute discrepancies, and within-subjectsensitivity correlations There is no necessary relationship between discrepancies, whether signed or absolute, and within-subject sensitivity correlations. More positive signed central tendency discrepancies on stereotypic attributes indicate greater overestimation of the stereotype. Similarly, more negative signed range discrepancies reflect greater underestimation of group variability. In contrast, sensitivity correlations assess the extent to which perceivers detect differences between attributes in the actual central tendency or range of the group. Discrepancy and correlation measures may therefore yield different conclusions. Indeed, one would reach very different conclusions concerning the relative accuracy of Perceivers #1 and #2 in Table 1, depending on the accuracy measure used. The signed discrepancies indicate that Perceiver #1 overestimated the stereotype more than Perceiver #2; the absolute discrepancies indicate that the two perceivers are equally accurate or inaccurate; and the sensitivity correlations indicate that Perceiver #1 is more accurate than Perceiver #2. In practice, however, the extent to which signed discrepancies, absolute discrepancies, and sensitivity correlations differ from each other is unclear. Signed discrepancies and sensitivity correlations have sometimes yielded different conclusions (e.g., Ryan, 1996). And only one study has assessed absolute discrepancies as well as signed discrepancies and sensitivity correlations (Ryan & Bogart, 2001). The general conclusions for signed and absolute discrepancies in this study, although not identical, were highly similar. The conclusions from discrepancy and sensitivity correlation measures were also similar, although much more so for stereotypicality than for variability measures. Theoretical conclusions from this research are described later in the chapter. For now, let us examine the empirical relationships among the various accuracy measures that were used in this study. The study (Ryan & Bogart, 2001) examined longitudinal changes in the accuracy of new group members’ in-group and out-group stereotypes. The new members of four college sororities judged their in-group sorority and three out-group sororities four times during their first year of membership. Perceived stereotypicality was assessed using the percentage estimation task; participants estimated the percentage of group members that they believed possessed stereotypic versus counterstereotypic attributes. Perceived variability was assessed using the range task; participants indicated where the highest and lowest group members fell on the same stereotype-relevant dimensions. Larger differences between judgements of the highest and
88
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
TABLE 2 Correlations among accuracy measures
N=81. For stereotypicality, lower signed and absolute discrepancies and higher sensitivity correlations indicate greater accuracy. For variability, higher signed discrepancies and sensitivity correlations and lower absolute discrepancies indicate greater accuracy. *p< .05, **p<.01, ***p<.001.
lowest group members (i.e., larger ranges) reflect greater perceived variability. New members’ judgements were then compared with full group members’ aggregated self-ratings. The correlations among the accuracy measures (partialling out subject group) for judgements of one target group at one point in time are reported in Table 2. First, consider the three measures of stereotypicality accuracy. Higher signed discrepancies indicate greater stereotype overestimation, that is, a stronger tendency to overestimate the group mean on stereotypic versus counterstereotypic attributes. Higher values for this signed discrepancy thus indicate greater inaccuracy. Higher absolute discrepancies also indicate greater inaccuracy. In contrast, higher sensitivity correlations indicate greater accuracy. Note that greater accuracy as indicated by sensitivity correlations is moderately correlated with lower accuracy (i.e., greater stereotype overestimation) as assessed by signed discrepancies, which is the opposite of what one would expect. However, the sensitivity correlations are highly negatively correlated with absolute discrepancies, indicating that greater accuracy as assessed by sensitivity correlations is strongly associated with greater accuracy as assessed by absolute discrepancies. I examined the same set of correlations for each of the other three target groups in this study (Ryan & Bogart, 2001) at each of the four points in time at which data were gathered—a total of 16 correlation matrices. The results were similar overall. The correlations between signed discrepancies and absolute discrepancies were largely positive, although they ranged from •0.45 to 0.73. The correlations between absolute discrepancies and sensitivity correlations varied from •0.35 to •0.91. However, the correlations between signed discrepancies and sensitivity correlations
STEREOTYPE ACCURACY 89
ranged all the way from •0.61 to 0.84—sometimes stereotype overestimation was strongly correlated with lower sensitivity, as one would expect, whereas in other cases, stereotype overestimation was strongly correlated with greater sensitivity. Now consider the three measures of variability accuracy in the bottom half of Table 2. For these measures, higher absolute discrepancies indicate greater inaccuracy, whereas higher signed discrepancies (i.e., less variability underestimation) and higher sensitivity correlations indicate greater accuracy. Note that signed discrepancy scores were virtually identical to absolute discrepancy scores, which results from the fact that the signed discrepancies were nearly all negative in these data. Analyses of the absolute discrepancies would thus not provide any additional information concerning stereotype accuracy than that already provided by signed discrepancies. Interestingly, however, both signed and absolute discrepancies were uncorrelated with sensitivity correlations, suggesting that variability discrepancies and sensitivity correlations tap different types of accuracy. I also examined this set of correlations for each of the other three target groups at each of the four points in time at which data were gathered. The magnitude of the relationship between absolute and signed variability discrepancies showed little variation; the correlations ranged from •0.95 to •1.0. Again, the strong tendency for people to underestimate the ranges of social distributions resulted in virtually identical signed and absolute discrepancies. In addition, the lack of correlation between discrepancies and sensitivity correlations was evident in 15 of the 16 matrices. Discrepancies were weakly correlated with sensitivity correlations in only one case. In this single case, greater accuracy as assessed by discrepancies was associated with greater accuracy as assessed by sensitivity correlations, r(81)=0.24 and •0.27, ps<.05, for the signed and absolute discrepancies, respectively. Finally, the correlations in Table 2 indicate that overestimation of stereotypicality (i.e., stereotype exaggeration) is correlated with variability underestimation (i.e., overgeneralisation). However, these correlations are not very strong. The relationship between these two types of accuracy across the 16 correlation matrices ranged from the moderate correlations reported in Table 2 to correlations that were virtually zero. These results are quite consistent with a great deal of research indicating that perceived stereotypicality is often only moderately correlated or even uncorrelated with perceived group variability (Park et al., 1991; Ryan et al., 2000). In sum, it appears that (a) the relationships among the three stereotypicality measures vary in both direction and strength, (b) signed and absolute range discrepancies assess identical types of accuracy as a result of the rather strong tendency for people to underestimate ranges, (c) signed and absolute range discrepancies assess a type of accuracy that is
90
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
different from that of the range sensitivity correlations, and (d) stereo typicality and variability accuracy are not strongly correlated and may sometimes be completely independent. Conclusions from accuracy research may thus differ depending on the stereotype component that is examined (i.e., stereotypicality or variability) as well as the method that is used to compare perceived to actual values. Clearly, researchers need to employ those accuracy measures that provide the best examination of the theoretical issues they wish to examine. If one wishes to reach conclusions concerning stereotype exaggeration, for example, then one needs to examine signed stereotypicality discrepancies. However, multiple measures will ultimately be most informative. Each measure has certain methodological and conceptual limitations and little is known about the conditions under which they yield different conclusions. EMPIRICAL RESEARCH AND THEORETICALSIGNIFICANCE OF ACCURACY RESEARCH Until quite recently, the primary goal of stereotype accuracy research was to address traditional assumptions, for example, that particular cultural stereotypes (e.g., of Blacks in the USA) are exaggerations of a group’s actual characteristics (e.g., McCauley & Stitt, 1978). Such research indicates that stereotypes may be more accurate than one might expect, and has therefore been extremely important in challenging long-standing assumptions about the inherently biased and unjust nature of stereotypes. Nevertheless, there are methodological problems (e.g., response language confounds) and conceptual intricacies (e.g., whether one is talking about a group’s stereotypicality or variability) that make it difficult to draw conclusions about the overall accuracy of a particular cultural stereotype. And even if one were able to do so, it would tell us little about the causes or consequences of that accuracy. A more fruitful approach is to consider the ways in which stereotype accuracy research can improve our understanding of stereotyping processes. Some of these theoretical issues are considered next. Moderators of intergroup differences in stereotypes A variety of factors are believed to moderate intergroup differences in perceptions (Hewstone, Rubin, & Willis, 2002) and virtually all of them are based on implicit assumptions about the relative accuracy of perceptions. Cognitive theories, for example, posit that people are more likely to expend limited cognitive resources on understanding their own group than the out-group (Hamilton & Trolier, 1986; Park et al., 1991; Tajfel, 1969; Tajfel & Turner, 1979), suggesting that out-groups are
STEREOTYPE ACCURACY 91
perceived less accurately than in-groups. And members of higher-status and more powerful groups often exhibit greater in-group bias than do members of lower-status and less powerful groups (Hewstone et al., 2002), suggesting that higher-status and more powerful people have less accurate perceptions (Ryan, 1996). Although these types of assumptions about relative group differences in stereotype accuracy lie at the heart of most stereotyping theories, only a handful of studies have attempted to examine them empirically. I discuss seven such studies, focusing on the consequences of cognitive processes, status and power, and social ideology for group differences in stereotype accuracy. Judd et al. (1991) examined the out-group homogeneity effect, that is, the tendency to perceive out-groups as less variable than in-groups. Virtually all explanations for the effect have assumed that stereotypes of out-groups are less accurate than those of in-groups. Such explanations have included a relative lack of familiarity with out-groups (Linville et al., 1989), a failure to update and revise cognitive representations of out-groups (vs in-groups) as new information about individual members is encountered (Park & Hastie, 1987; Park et al., 1991), and memory retrieval processes that are less comprehensive in the case of out-groups than in-groups (Park et al., 1991). An obvious question, then, is whether perceptions of outgroup variability really are less accurate. Judd et al. (1991) examined this question by comparing business and engineering majors’ perceptions of their in-groups and out-groups with accuracy criteria that were derived from the aggregated self-ratings of random samples of group members. The data revealed a remarkably consistent pattern across measures, indicating greater accuracy in judgements of in-groups than out-groups. Moreover, participants who exhibited larger out-group homogeneity effects also exhibited greater sensitivity to between-attribute differences in the actual variability of the ingroup than the out-group. In short, perceptions of the in-group were indeed more accurate than those of the out-group and this was truer for people who demonstrated stronger out-group homogeneity effects. Other research has emphasised the influence of social status, power, and social experiences on in-group-out-group differences in stereotype accuracy (Ryan, 1996). Such factors seem unlikely to have played much of a role in the Judd et al. (1991) study, because the groups were business and engineering majors. In contrast, Blacks and Whites in the US differ in status and power and seem likely to have qualitatively different kinds of racerelated experiences. Blacks as a group have lower status and power. They are more likely to encounter resistance to their opinions, behaviours, and attitudes and are more likely to be targets of prejudice and discrimination than are Whites. At the same time, Blacks’ outcomes often depend on their daily interactions with Whites, for example, in employment and educational settings, whereas Whites’ outcomes do not
92
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
depend on their interactions with Blacks in the same way or to the same degree. The importance of ethnicity to Blacks in daily life seems likely to cause Blacks to accentuate intergroup differences and intragroup variability relative to Whites (cf. Brown & Simon, 1987), that is, to overestimate the stereotypicality and underestimate the variability of Blacks and Whites. At the same time, however, the importance of ethnicity and of knowing what Whites are like seems likely to cause Blacks as a group to be attentive to differences in the degree to which various attributes characterise both their own group and the White majority out-group. In other words, although Blacks would be expected to demonstrate greater discrepancies between their perceptions and the actual characteristics of the groups, they would also be expected to be equally sensitive to between-attribute differences in the actual characteristics of the two groups. These expectations were supported by an empirical study of stereotype accuracy among Black and White college students (Ryan, 1996). Students’ perceptions of Black and White first-year college students were compared with accuracy criteria derived from self-reports provided by random samples of group members. The results indicated that all participants generally overestimated the stereotypicality and underestimated the variability of both target groups. But because these differences were greater for Black than for White participants, Blacks exhibited less accuracy. However, sensitivity correlations indicated that Blacks were as accurate in their judgements of Whites as in their judgements of Blacks, whereas Whites were much more accurate in judgements of their own group than the Black out-group. A belief in the importance of ethnicity and a need to understand the White target group apparently led Blacks to detect betweenattribute differences in the actual characteristics of the groups equally well, whether they were judging their own group or the out-group. Indeed, the importance in everyday life of understanding the White target group appeared to result in all participants perceiving Whites more accurately (see also Ryan, 1995, Study 2). Of course, social status, power, and social experiences were not directly examined in this study (Ryan, 1996). And, in any case, it would have been difficult to tease apart their effects. This research does suggest, however, that a stronger social, as opposed to cognitive, perspective is needed to understand in-group-out-group differences in ethnic and racial stereotypes. Blacks and Whites have rather different race-related experiences, which seem likely to result in rather different perspectives and thus different stereotypes. This study further indicates that researchers cannot simply assume that weaker stereotypes are necessarily better or more accurate. The White (vs Black) participants in this study held equally weak stereotypes about Blacks and Whites, exhibiting less overestimation of group stereotypicality and less underestimation of within-group variability in judgements of both groups. Yet sensitivity correlations revealed that they
STEREOTYPE ACCURACY 93
were much better able to detect between-attribute differences, and were thus much more accurate, in judgements of their own group than the outgroup as compared with the Black participants. Other research has similarly suggested that Blacks and Whites have different experiences and thus different perspectives about ethnicity, which may influence in-group and out-group stereotypes (Judd, Park, Ryan, Brauer, & Kraus, 1995) and thus stereotype accuracy. This work suggests that Blacks as a group have adopted a multicultural view in which both positive and negative differences between groups are acknowledged and generally appreciated (cf. Hewstone, 1996). This perspective is similar to that of appreciating the positive and negative characteristics of individuals, which sometimes help and sometimes hinder daily interactions, depending on circumstances. In contrast, Whites as a group appear to have adopted a colour-blind view based on egalitarian notions that all people are the same and it is therefore wrong to make any judgements based on group membership. Wolsko and his colleagues (2000 Experiment 2) directly examined whether this difference in intergroup ideology influences stereotypicality accuracy. They manipulated participants’ endorsement of a colour-blind versus a multicultural perspective by instructing White college students that intergroup harmony could best be achieved either by recognising the similarities and equality of all individuals in the nation (the colour-blind perspective) or by appreciating diversity and accepting each group’s positive and negative qualities (the multicultural perspective). All participants then estimated the percentage of Blacks and Whites in the US that they believed possessed each of 16 stereotype-relevant attributes. Participants’ judgements were compared with accuracy criteria derived from various public sources, including the US Bureau of the Census, the US Department of Education, and the US Department of Health and Human Services. The results indicated that, although the multiculturalism instructions resulted in stronger stereotypes about both groups relative to the colourblind condition, they also resulted in greater stereotype accuracy and less ethnocentrism. Specifically, participants underestimated the stereotypicality of both Blacks and Whites—that is, they judged the groups less stereotypically than the groups actually were. But those who had received the multiculturalism rather than the colour-blind instructions were less likely to do so and were thus more accurate. Participants in the multiculturalism condition were also less likely to overestimate the prevalence of positive versus negative attributes when judging Whites than Blacks as compared with those in the colour-blind condition. Importantly, then, this research provides experimental evidence that a social ideology focusing on group differences, rather than similarities, can cause people to perceive ethnic groups more accurately and more positively even though
94
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
their stereotypes may also be stronger. These findings further underscore a primary point of the present chapter—one cannot assume that stronger stereotypes are necessarily less accurate (or more negative). One study has also directly examined the role of power in stereotype accuracy. Guinote, Judd, and Brauer (2002) were interested in understanding why the members of less powerful groups perceive greater variability in the out-group than the in-group (Lorenzi-Cioldi, Eagly, & Stewart, 1995). Explanations for this effect have largely focused on the perceiver’s social position. For example, less (vs more) powerful perceivers are thought to be motivated to differentiate out-group members, because information about powerful groups is important to their daily interactions (Fiske, 1993). One implication of this explanation is that less powerful people have more accurate stereotypes than do more powerful people (Ryan, 1996). However, Guinote and her colleagues (2002) have argued that power may affect perceptions of group variability for one of three reasons: the perceiver’s power may affect her perceptions of group variability (a perceiver effect as described above), the group’s power may affect how variable it is perceived to be (a target group effect), and the power of a group may affect the actual variability of its members such that more powerful groups may actually be more variable. One group may actually be more variable than another for a variety of reasons, an obvious one being that the members of more powerful groups have more varied social opportunities and are therefore less restricted in the attributes they develop and express (Eagly & Steffen, 1984). Guinote and her colleagues (2002) used a minimal group paradigm to tease apart these explanations. Participants were randomly assigned to low- or high-power groups on the basis of a test that ostensibly determined whether one was better able to generate or evaluate solutions. Participants were told that members of the low-power groups (i.e., “workers”) would work on solving tasks and that members of the high-power groups (i.e., “judges”) would judge the quality of the solutions that were generated. Participants were further told that workers’ financial compensation would be partially determined by judges’ evaluations. Each group then completed an alleged training task after which the members of each group were asked to provide brief oral self-descriptions to other participants. All participants were videotaped while completing the training task and providing their selfdescriptions. Participants then judged the variability of their own group and the out-group. A second group of participants who did not participate in the study were subsequently asked to judge the variability of the same two groups after viewing the videotaped group tasks and selfpresentations. Half of the participants in this second group were told about the ostensible worker-judge distinction between the two groups and half were not.
STEREOTYPE ACCURACY 95
The results indicated that low-power group members perceived greater variability among higher-power out-groups (Guinote et al., 2002). Importantly, however, this effect appeared to be due to differences in the actual variability of the groups, because the second group of participants who were not members of either group also judged the high-power groups to be more variable than the low-power groups. And they did so regardless of whether they knew about the power relationship that had existed between the groups. The tendency to judge the high-power groups as more variable could not have been due only to the low power of the perceivers, because the effect also occurred among perceivers who were not members of the low-power group. Similarly, the effect was unlikely to be due only to a tendency for high-power groups to be perceived as more variable, because even participants who did not know about the power relationship judged the high-power groups as more variable. Instead, those who were in more powerful positions actually were more variable. Presumably, they had an opportunity to exhibit more variable traits and behaviours and for this reason were perceived to be more variable. Three additional studies have examined the accuracy of stereotypes about gender differences, using meta-analytic effect sizes (Hall & Carter, 1999; Swim, 1994) and self-ratings from a representative sample of Americans’ social and political attitudes (Diekman et al., 2002) as indicators of actual gender differences. Interestingly, this research has not revealed systematic differences either as a function of the perceiver’s gender or as a function of in-group versus out-group. Hall and Carter reported that women were more accurate than men in their beliefs about gender differences in nonverbal behaviour, but not in their beliefs about any of the five other categories of behaviours examined, including cognitive performance and personality. Diekman and her colleagues reported a systematic tendency for all participants to underestimate men’s, but not women’s, support for policies that favour women’s interests. Although none of these studies directly examined variability accuracy, they are consistent in their findings of overall accuracy in gender stereotypes and in their failure to find differences due to perceiver’s gender or in-group versus out-group. These studies suggest, then, that the tendency to perceive men as more variable than women (Lorenzi-Cioldi et al., 1995) may result at least partly from greater actual variability among men than among women. In sum, it appears that people tend to have more accurate stereotypes about in-groups than out-groups. However, intergroup differences in stereotype accuracy depend on factors, such as social status, power, social experiences, and social ideology. Members of lower-status and less powerful groups may perceive greater variability among higher-status and more powerful group members not only because of their importance in daily interactions, but also because those group members actually are more variable. The research further indicates that stronger stereotypes may result
96
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
from differences in social experiences and ideology (e.g., multiculturalism) and that stronger stereotypes are not necessarily less accurate than weaker stereotypes; indeed, sometimes they are more accurate. Finally, it appears that people on the whole perceive gender differences relatively accurately and that women are no more accurate, overall, than are men. Thus, the tendency to perceive greater variability among men than among women may result at least partly from greater actual variability among men than among women. Stereotype change and development The processes underlying stereotype change (Hewstone, 1996) and development (Mackie, Hamilton, Susskind, & Rosselli, 1996) have also figured prominently in the stereotyping literature. Stereotypes have been thought to develop for a variety of reasons, including faulty personality characteristics (Adorno, Frenkel-Brunswik, Levinson, & Sanford, 1950), limited information processing ability (Hamilton & Trolier, 1986), biased perceptual processes (Tajfel, 1969), and socioeconomic inequalities (Sherif & Sherif, 1953). This research has yielded important insights. For example, the formation of social categories often results in biased perceptual processes (Corneille & Judd, 1999; Doise et al., 1978; Krueger & Clement, 1994; Tajfel 1969; Tajfel & Wilkes, 1963). But virtually all theories of stereotype development start with the assumption that stereotypes are inaccurate. To my knowledge, only one study (Ryan & Bogart, 1997; 2001) has empirically assessed stereotypes and stereotype accuracy as they develop over time. This study suggests that under some circumstances, stereotypes may become more accurate as they develop. Ryan and Bogart (1997) conducted a detailed study of stereotype development and change among the new members of four college sororities. Participants judged the stereotypicality, variability, and positivity of their new in-group and three new out-groups four times during their first year of membership. The results revealed that new members perceived their out-groups more stereotypically than their ingroups at every wave. And although perceived group stereotypicality generally decreased during the year, there was no change in the in-groupout-group difference. In other words, the in-group-out-group difference in perceived stereotypicality was established early and remained relatively stable during the year. In contrast, out-groups were perceived to be more variable than in-groups in the first wave of data—an in-group homogeneity effect. Participants also judged their in-groups and out-groups, but especially their out-groups, to be increasingly less variable during the year. Participants thus perceived both groups as increasingly less variable during the year. But because the reduction was greater for the out-group, what was initially in-group homogeneity changed to (non-significant) out-group
STEREOTYPE ACCURACY 97
homogeneity. Participants also exhibited in-group bias at each wave. However, somewhat surprisingly, participants judged the groups to be increasingly less positive during the year, and this was especially true for the in-group. In-group bias thus decreased over time even though all new members were becoming increasingly involved in their in-groups. New members’ self-ratings further indicated that they perceived themselves, that is, as individuals, to be more similar to their in-groups during the year. Existing stereotyping literature suggests that the accuracy of these new members’ in-group and out-group stereotypes would not change or would decrease. After all, in-groups and out-groups, but especially out-groups, were perceived to be less variable over time; and in-groups and out-groups, but especially in-groups, were perceived less positively over time. Stereotypes have been characterised as erroneous because they have been largely thought of as overgeneralisations about the negative characteristics of a group—the sorts of changes that were evident in these data. The changes in the accuracy of new members’ perceptions, however, suggest that a different perspective is required to understand the developmental process that occurred. Ryan and Bogart (2001) subsequently compared the same new members’ perceptions with accuracy criteria derived from the self-reports that had been provided by the full members of each group at each of the same four points in time. The mean signed discrepancies and sensitivity correlations are presented in Table 3. First, consider the results for stereotypic accuracy. In-groups were judged more accurately than out-groups on both discrepancy and sensitivity correlation measures and although the in-groupout-group differences did not change over time, stereotypicality judgements generally became more accurate. These differences are consistent with the differences that existed in new members’ perceptions. In-groups were judged less stereotypically and more accurately than out-groups. Neither the in-group-out-group difference in perceived stereotypicality nor the ingroup-out-group difference in stereotypicality accuracy changed over time. And perceptions of group stereotypicality generally decreased and became more accurate over time as new members became more involved in their groups. In short, then, new members’ stereotypicality judgements became weaker and more accurate as they developed over time, and there was no change in the in-group-out-group difference. Now consider the accuracy of new members’ variability judgements. The relevant means are reported in the bottom half of Table 3. The signed discrepancies have been reverse-scored so that higher values indicate greater underestimation of the group’s actual variability. (As indicated earlier, the signed discrepancies were virtually all negative.) Note that the variability sensitivity correlations are much lower than the stereotypicality sensitivity correlations, a finding that is consistent with other accuracy research (Judd et al., 1991; Ryan, 1996). People are much better able to
98
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
TABLE 3Mean discrepancies and sensitivity correlations
Adapted from Ryan and Bogart (2001). Lower discrepancies and higher sensitivity correlations indicate greater accuracy.
detect differences between attributes in the actual stereotypicality than in the actual variability of groups. More importantly, the longitudinal analyses revealed an interaction indicating that new members were increasingly less likely to underestimate in-group variability during the year, but increasingly more likely to underestimate out-group variability. The sensitivity correlations further indicated that new members’ judgements of group variability generally became more accurate over time; there was no change over time in the in-group-out-group difference. Overall, then, new members perceived the in-group more accurately over time on every accuracy measure, although the in-group had been judged as less variable and less positive over time. New members perceived their outgroups more accurately over time on three of the four accuracy measures, even though out-groups had also been judged to be less variable and less positive during the year. Interestingly, the greater reduction in out-group than in-group variability judgements over time was accompanied by greater underestimation of out-group variability (i.e., decreasing accuracy) as one might expect, but also less underestimation of in-group variability judgements (i.e., increasing accuracy). Again, the latter result seems surprising in light of the fact that new members judged their in-groups as increasingly less variable and less positive over time. The overall pattern of change is consistent with the group socialisation perspective (Ryan et al., in press). According to this perspective, new members of existing groups undergo a period of socialisation that is characterised by a relatively high degree of uncertainty. New members are uncertain about many things, for example, the skills and behaviours that are associated with group membership; indeed, learning the required skills
STEREOTYPE ACCURACY 99
and behaviours is often considered the primary purpose of the socialisation phase. But new members also want to be fully accepted by the group and they are uncertain about whether and how it may occur. In other words, they are uncertain about what is required to “fit in”. The most direct means of managing this uncertainty is to gather information about the characteristics of the group and its members so that uncertainty about the group is reduced and understanding of acceptance criteria is increased. Such information may be acquired from formal training and orientation programmes, written documents about the group’s history and purpose, and direct interactions with group members. In short, then, a primary task for new members is the development of cognitive structures that represent information about the group—that is, stereotypes. The uncertainty about being accepted by the group leads new members to focus initially on understanding the group as a whole. New members’ group-level impressions serve as structures that enable them to organise the information they encounter throughout the socialisation period and beyond. The desire to reduce uncertainty and the attendant focus on the group as a whole results in decreased variability (cf. Park & Hastie, 1987; Ryan et al., 2000; Ryan et al, 1996a), decreased positivity (uncertainty is generally unpleasant), and increased accuracy (Ryan & Bogart, 2001; Ryan et al., in press). Indeed, without a group-level structure, the diversity within the group may even be less apparent later (e.g., beyond the socialisation phase) when one’s primary focus turns to understanding individual group members. In short, although focusing on the group as a whole may lead to decreased variability in the short term (e.g., during socialisation), it may ultimately lead to increased perceived variability when that initial focus represents an effort to develop an organised and refined impression of the group (Ryan et al., in press). In sum, stereotype development may not always involve the development of inaccurate and unjust perceptions of a group. And to the extent that the inaccuracy assumption continues to influence the field, researchers may fail to examine important theoretical questions or reach premature conclusions about the nature of the processes that are involved. It seems possible and especially important to identify strategies that groups might use, such as advocating a multicultural ideology (Wolsko et al., 2000), to promote the development of more accurate and potentially more useful stereotypes. But research concerning the characteristics of groups that influence stereotype development and accuracy remains sparse. Individual differences in stereotype accuracy The correlates of individual differences in stereotype accuracy have received greater attention. Once again, the results of this research are not always as one might expect. Some findings are consistent with the notion that
100 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
stereotypes are often unjust. People who are more strongly affiliated with the in-group (Judd & Park, 1993) or perceive greater variability in the ingroup than the out-group (Judd et al., 1991) tend to have more accurate ingroup than out-group stereotypes. Ryan’s (1996) research with Black and White college students indicated that people who perceived groups more (vs less) positively had more accurate stereotypes. Hall and Carter (1999) examined a variety of potential correlates of individual differences in the accuracy of perceived gender differences, defining actual gender differences as meta-analytic effect sizes for each of 77 gender-relevant attributes. Accuracy was assessed via within-subject sensitivity correlations. This research indicated that higher social dominance orientation (Pratto, Sidanius, Stallworth, & Malle, 1994), higher personal need for structure (cf. Schaller, Boyd, Yohannes, & O’Brien, 1995), lower need for cognition (Cacioppo, Petty, Feinstein, & Jarvis, 1996), and lower social sensitivity (Hall & Carter, 1999; Rosenthal, Hall, DiMatteo, Rogers, & Archer, 1979) predicted less accuracy in perceived gender differences. In general, then, the results were as one might expect. People who were more attuned to their social environments and more receptive to new information concerning the group appeared to have more accurate stereotypes. Interestingly, however, there is thus far little evidence that increased familiarity with a group leads to greater accuracy. Ryan (1996) examined the relationship between familiarity with the out-group and the accuracy of out-group judgements. Of the eight correlations examined, only one was significant. Blacks who were more familiar with the White out-group were more sensitive to between-attribute differences in the variability of Whites. Ryan and Bogart (2001) found evidence of increased accuracy over time as new members became more familiar with their in-group and out-group. But the only significant correlations between familiarity and accuracy that emerged were three of the eight correlations at wave four. (None of the same eight correlations was significant at any of the three other waves.) And one of these three correlations was the opposite of what one would expect. That is, new members who were more familiar with the in-group were more likely to overestimate the stereotype of the in-group. Increased familiarity during new members’ socialisation was also virtually unrelated to changes in stereotype accuracy. Although these results may seem surprising, they are actually consistent with other work indicating that familiarity per se appears to play a minor role in intergroup perceptions (e.g., Park, Ryan, & Judd, 1992; Ryan & Bogart, 1997). My own view is that the type of cognitive structures that one develops, and motivations, such as those examined by Hall and Carter (1999; cf. Stangor, 1995), play more important roles. Simply encountering more information seems unlikely to lead to greater accuracy unless it is perceived as new and potentially useful in some way. Otherwise, there
STEREOTYPE ACCURACY 101
would seem to be little point in expending the cognitive resources that would be required to incorporate the information into one’s cognitive representation of the group. To my knowledge, these are the only studies that have examined individual differences in stereotype accuracy. There are other potential correlates of accuracy, such as empathy, the motivation to be accurate, and perceived threat, as well as questions as to whether the relationships described here hold when other types of stereotypes are examined. As with some of the research described above, future research may support some of our long-standing assumptions about stereotype accuracy. But like the research concerning the role of familiarity, it also seems likely that some research will cause researchers to rethink ideas about the factors that predict greater stereotype accuracy and thus some ideas about the nature of stereotyping processes. A particularly important issue for future research concerns the consequences of stereotype accuracy for group-relevant judgements and interactions with group members. The discussion thus far, as well as implicit assumptions about the likely importance of accuracy for daily social functioning, suggest that it is better to have a more accurate stereotype of a group. However, there is thus far extremely little empirical research that has examined whether stereotype accuracy results in more accurate group-relevant judgements, or more productive or more harmonious interactions with individual group members. These types of questions are certainly important. But they are also difficult to examine. Consequences of stereotype use Stereotyping researchers have traditionally focused on the negative consequences of applying a stereotype to an individual group member. And there have been good reasons for doing so. Stereotypes, by definition, often do not accurately characterise a given individual group member; stereotypes are beliefs about the characteristics of a group of people. And the more variable group members actually are, the less likely the stereotype (i.e., the stereotypicality component of the stereotype) is to characterise a particular group member. Thus, to the extent that group members vary on stereotypic attribute dimensions, even the use of an accurate stereotype may have a relatively high probability of resulting in an inaccurate judgement of an individual. In addition, for some types of judgements, the use of a stereotype— regardless of its accuracy—may result in what most would consider bias. Consider perceivers (e.g., personnel staff) who gather information about an individual (e.g., a job applicant) so that a judgement (e.g., an employment decision) can be made. Stereotypes may influence the process by which perceivers gather information about the individual, influencing the
102 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
type of information that is ultimately available for making the judgement. Even if one assumes that the information people gather is not distorted and that such information is relevant to the judgement, the information may not represent the full range of relevant information that could have been obtained. Ryan et al. (2000), for example, gave participants identical information describing ambiguously aggressive interactions between boys whose ethnicity was White or Black as indicated by drawings that accompanied brief written descriptions of the interactions. Participants were asked to generate three questions that they would ask to determine what had actually occurred. Participants who perceived Blacks to be less variable, as indicated by a measure that had been completed 2 to 4 months earlier, were more likely to generate questions that asked about the aggressiveness of the Black versus the White actors in the scenarios. Stereotypes thus influenced the manner in which perceivers sought information about individual group members, even though participants’ responses were relatively unconstrained and the instructions should have caused them to consider a range of possibilities. In other situations, whether a judgement of an individual is accurate may be less important than other aspects of the judgement, for example, its consequences. More accurate judgements do not necessarily result in more harmonious or more productive interactions with individual group members. And more harmonious or more productive interactions with individuals may be more important than the accuracy of one’s judgements about them. Nevertheless, there may be some circumstances in which the use of an accurate stereotype may result in a more accurate judgement and this would be important to know for both practical and theoretical reasons. Consider, for example, an employer who holds the belief that all people are the same and that employees’ religious group memberships are therefore irrelevant. The employer does not take into account significant religious holidays when scheduling meetings or setting deadlines, placing the employee in the difficult position of either ignoring the holiday or requesting that adjustments be made. Such a failure to recognise or acknowledge group membership and the characteristics associated with it, such as ethnic heritage, experiences with discrimination, and religious traditions, may be considered offensive, exacerbate intergroup tensions, and/or result in a less accurate and less useful judgement of an individual. Understanding the circumstances under which stereotype use may actually be appropriate and helpful would also provide a more complete theoretical understanding of stereotyping processes. But determining whether greater stereotype accuracy predicts more accurate judgements of individuals is not as simple as it may sound. The path from group-level beliefs to individual-level judgements is complicated. Whether more accurate stereotypes lead to more accurate judgements of
STEREOTYPE ACCURACY 103
individuals depends on a variety of factors, including the diagnosticity of individuating information that is available to the perceiver, the relevance of the stereotype to the judgement, and the manner in which stereotype information is combined with individuating information to form the judgement. Certainly, a great deal of research has established that perceivers are more likely to use stereotypes when relevant individuating information is unavailable or ambiguous, the stereotype is potentially relevant to the judgement, and a judgement is required (Ryan et al., 1996a). Presumably, under these types of circumstances, people who perceive the group more accurately will judge an individual group member more accurately. However, assuming that the individual is a stereotype-consistent group member, this hypothesis suggests that more accurate perceivers are likely to rely on their stereotypes to the same extent as, or even more than, less accurate perceivers. This would seem to be a bit of a paradox. People who are less likely to accept and use stereotypes might be expected to have more accurate stereotypes (Hall & Carter, 1999). If, however, such people were less likely to use their stereotypes in these situations, their judgements of an individual would be equally inaccurate or perhaps even less accurate. In other words, people who have more accurate stereotypes could be less accurate in their judgements of an individual, because they underutilise group-level information. The notion that underutilisation of group-level information may lead to less accurate judgements of individual group members is not a popular one. But it has been considered as a possibility in at least one study. Recall that Wolsko et al. (2000) told participants either that a multicultural or a colourblind perspective would improve intergroup relations. The results indicated that people who received the multiculturalism instructions were less likely to underestimate group stereotypicality; they were more accurate. A separate experiment further indicated that participants given the multicultural and colour-blind instructions were equally likely to utilise individuating information to judge individual members of an ethnic group (i.e., Hispanics). But those acting under the multicultural instructions also used group-level information, whereas those acting under the colour-blind perspective appeared to ignore the group-level information. In other words, people operating under a colour-blind ideology may actually undemtilise relevant stereotype-based information as compared with people operating under a multicultural perspective. Other research concerning stereotype accuracy and judgements of individuals comes from Jussim and his colleagues (Jussim, Eccles, & Madon, 1996; Madon et al., 1998; Smith, Jussim, Eccles, VanNoy, Madon & Palumbo, 1998). These researchers have conducted a great deal of empirical work indicating that the prevalence of self-fulfilling prophecies has been overestimated. Self-fulfilling prophecies occur when erroneous
104 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
expectancies, which are often based on stereotypes, influence interactions with a target individual, causing the individual to behave in a way that confirms the expectancy. Jussim et al.’s conclusions are largely based on research examining interactions between teachers and students in classroom settings, the type of setting in which self-fulfilling prophecies have been believed to be most pervasive. Yet this research indicates that teachers’ perceptions of students tend to be accurate; self-fulfilling prophecies tend to be small, as do biased stereotype-based judgements. It is not possible to provide a comprehensive review of all the factors that are relevant to our understanding of, or to reach firm conclusions about, the role of stereotype accuracy in stereotype use, particularly given the limited empirical work. It appears, however, that people often arrive at more accurate judgements than might be expected and that inaccurate judgements may result from either under- or over-utilisation of group-level information. It thus seems likely that the field will be best served by considering not only that stereotypes may result in inaccurate judgements of individuals regardless of the accuracy of the stereotype, but also that there may be circumstances in which one might reach a more accurate and/or more useful judgement if an accurate stereotype were applied in a reasonable and appropriate manner. CONCLUSION I have focused on the potential for stereotypes to be accurate and useful, arguing that researchers cannot fully understand stereotyping processes without considering and empirically examining this possibility. I wish to emphasise, however, that I am not arguing that stereotypes are always or even usually accurate or that stereotype use is not a problem. Stereotypes are often likely to be exaggerations and overgeneralisations that cause problems for the members of stereotyped groups. Under some circumstances, however, the perceiver’s impression of a group may be the result of a concerted effort to develop an accurate differentiated cognitive structure that facilitates the organisation and effective use of grouprelevant information. It seems important not to assume that the former is always the case. Restricting scientific discourse and research to the biases and negative consequences associated with stereotypes ultimately serves only to restrict our understanding of the very processes we seek to understand and improve. Consider the findings presented in this chapter that conflict with traditional assumptions about the nature of stereotypes. People may perceive more variability among more powerful groups because the groups actually are more variable (Guinote et al., 2002). People who have stronger stereotypes sometimes also have more accurate stereotypes (Ryan, 1996; Wolsko et al., 2000). Stereotype development is sometimes characterised
STEREOTYPE ACCURACY 105
by increased accuracy (Ryan & Bogart, 2001). And greater familiarity with a group appears not to result in greater accuracy (Ryan, 1996; Ryan & Bogart, 2001). Nevertheless, the notion that stereotypes can be accurate is a politically and affectively charged idea (Oakes & Reynolds, 1997; Stangor, 1995). And there is perhaps a greater than normal potential for accuracy research to be misinterpreted and misused. Many people will not appreciate the sorts of measurement and conceptual limitations that plague accuracy research. And many will not understand that identifying a characteristic of a group does not by itself indicate the source (s) of the characteristic, its consequences, or methods of achieving change, if change is considered desirable. But stereotype accuracy research has the potential to support or challenge existing theory and stimulate new ideas about the nature of stereotyping processes. It therefore seems important to pursue these ideas while being aware of the potential pitfalls and taking steps to ensure that accuracy findings are fairly represented and used. REFERENCES Adorno, T.W., Frenkel-Brunswik, E., Levinson, D.J., & Sanford, R.N. (1950). Theauthoritarian personality.New York: Harper & Row. Allport, G.W. (1954). The nature of prejudice.Reading, MA: Addison-Wesley. Anderson, M., & Fienberg, S.E. (2000). Partisan politics at work: Sampling and the 2000 census. Political Science and Politics, 33,795–799. Ashmore, R.D., & Del Boca, F.K. (1981). Conceptual approaches to stereotypes and stereotyping. In D.L.Hamilton (Ed.), Cognitive processes in stereotyping and intergroupbehavior(pp. 1–35). Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Ashton, M.C., & Esses, V.M. (1999). Stereotype accuracy: Estimating the academic performance of ethnic groups. Personality and Social Psychology Bulletin, 25, 225–236. Biernat, M. (1995). The shifting standards model: Implications of stereotype accuracy for social judgment. In Y.Lee, L.J.Jussim, & C.R.McCauley (Eds.), Stereotype accuracy: Towardappreciating group differences(pp. 87–114). Washington, DC: American Psychological Association. Biernat, M., Manis, M., & Nelson, T.E. (1991). Stereotypes and standards of judgment. Journalof Personality and Social Psychology, 60,485–499. Bogart, L.M. (1998). The relationship of stereotypes about helpers to help-seeking judgments, preferences, and behaviors. Personality and Social Psychology Bulletin, 24,1264–1275. Cacioppo, J.T., Petty, R.E., Feinstein, J.A., & Jarvis, W.B.G. (1996). Dispositional differences in cognitive motivation: The life and time of individuals varying in need for cognition. Psychological Bulletin, 119,197–253. Cole, D. (1999). No equal justice: Race and class in the American criminal justice system.New York: New Press.
106 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Corneille, O., & Judd, C.M. (1999). Accentuation and sensitization effects in the categorization of multifaceted stimuli. Journal of Personality and Social Psychology,77, 927–941. Cronbach, L.J. (1955). Processes affecting scores on “understanding of others” and “assumed similarity”. Psychological Bulletin, 52,177–193. Diekman, A.B., Eagly, A.H., & Kulesa, P. (2002). Accuracy and bias in stereotypes about the social and political attitudes of women and men. Journal of Experimental Social Psychology,38,268–282. Doise, W., Deschamps, J.C., & Meyer, G. (1978). The accentuation of intracategory similarities. In H.Tajfel (Ed.), Differentiation between social groups: Studies in the socialpsychology of intergroup relations(pp. 159–168). London: Academic Press. Eagly, A.H., & Diekman, A.B. (1997). The accuracy of gender stereotypes: A dilemma for feminism. International Review of Social Psychology, 10,11–30. Eagly, A.H., & Steffen, V.J. (1984). Gender stereotypes stem from the distribution of women and men into social roles. Journal of Personality and Social Psychology, 46,735–754. Fiske, S.T. (1993). Controlling other people: The impact of power on stereotyping. AmericanPsychologist, 48,621–628. Funder, D.C. (1995). On the accuracy of personality judgment: A realistic approach. Psychological Review, 102,652–670. Gage, N.L., & Cronbach, L.J. (1955). Conceptual and methodological problems in interpersonal perception. Psychological Review, 62,411–422. Guinote, A., Judd, C.M., & Brauer, M. (2002). Effects of power on perceived and objective group variability: Evidence that more powerful groups are more variable. Journal ofPersonality and Social Psychology, 82,708–721. Hall, J.A., & Carter, J.D. (1999). Gender-stereotype accuracy as an individual difference. Journal of Personality and Social Psychology, 77,350–359. Hamilton, D.L., & Trolier, T.K. (1986). Stereotypes and stereotyping: An overview of the cognitive approach. In J.F.Dovidio & S.L.Gaertner (Eds.), Prejudice, discrimination, andracism(pp. 127–163). Orlando, FL: Academic Press. Harris, D. (1999). Driving while black: Racial profiling on our nation’s highways. American Civil Liberties Union Report. Helms, J.E. (1992). Why is there no study of cultural equivalence in standardized cognitive ability testing? American Psychologist, 47,1083–1101. Hewstone, M. (1996). Contact and categorization: Social psychological interventions to change intergroup relations. In C.N.Macrae, C.Stangor, & M.Hewstone (Eds.), Stereotypes andstereotyping(pp. 323–368). New York: Guilford Press. Hewstone, M., Rubin, M., & Willis, H. (2002). Intergroup bias. Annual Review of Psychology,53,575–604. Judd, C.M., & Park, B. (1993). Definition and assessment of accuracy in social stereotypes. Psychological Review, 100,109–128. Judd, C.M., Park, B., Ryan, C.S., Brauer, M., & Kraus, S. (1995). Stereotypes and ethnocentrism: Diverging interethnic perceptions of African American and White American youth. Journal of Personality and Social Psychology, 69,460– 481.
STEREOTYPE ACCURACY 107
Judd, C.M., Ryan, C.S., & Park, B. (1991). Accuracy in the judgment of in-group and out-group variability. Journal of Personality and Social Psychology, 61, 366–379. Jussim, L., Eccles, J., & Madon, S. (1996). Social perception, social stereotypes, and teacher expectations: Accuracy and the quest for the powerful selffulfilling prophecy. In M.P. Zanna (Ed.), Advances in experimental social psychology, 28, (pp. 281–388). San Diego, CA: Academic Press. Kenny, D.A. (1991). A general model of consensus and accuracy in interpersonal perception. Psychological Review, 98,155–163. Krueger, J., & Clement, R.W. (1994). Memory-based judgments about multiple categories: A revision and extension of Tajfel’s accentuation theory. Journal of Personality and SocialPsychology, 67,35–47. Lambert, A.J. (1995). Stereotypes and social judgment: The consequences of group variability. Journal of Personality and Social Psychohgy, 68,388–403. Lee, Y., & Duenas, G. (1995). Stereotype accuracy in multicultural business. In Y.Lee, L.J. Jussim, & CR McCauley (Eds.), Stereotype accuracy: Toward appreciating group differences(pp. 157–186). Washington, DC: American Psychological Association. Lee, Y., Jussim, L.J., & McCauley, C.R. (Eds.) (1995). Stereotype accuracy: Towardappreciating group differences.Washington, DC: American Psychological Association. Linville, P.W., Fischer, G.W., & Salovey, P. (1989). Perceived distributions of characteristics of in-group and out-group members: Empirical evidence and a computer simulation. Journalof Personality and Social Psychology, 57,165– 188. Lorenzi-Cioldi, F., Eagly, A.H., & Stewart, T. (1995). Homogeneity of gender groups in memory. Journal of Experimental Social Psychology, 31,193–217. Mackie, D.M., Hamilton, D.L., Susskind, J., & Rosselli, F. (1996). Social psychological foundations of stereotype formation. In C.N.Macrae, C.Stangor, & M.Hewstone (Eds.), Stereotypes and stereotyping(pp. 41–78). New York: Guilford Press. Madon, S., Jussim, L., Keiper, S., Eccles, J., Smith, A., & Palumbo, P. (1998). The accuracy and power of sex, social class, and ethnic stereotypes: A naturalistic study in person perception. Personality and Social Psychology Bulletin, 24, 1304–1318. McCauley, C., & Stitt, C.L. (1978). An individual and quantitative measure of stereotypes. Journal of Personality and Social Psychology, 36,929–940. Neisser, U., Boodoo, G,, Bouchard, T.J., Jr, Boykin, A.W., Brody, N., & Ceci, S.J., et al (1996). Intelligence: Knowns and unknowns. American Psychologist, 51, 77–101. Nisbett, R.E., & Kunda, Z. (1985). Perception of social distributions. Journal of Personality andSocial Psychology, 48,297–311. Oakes, P.J., & Reynolds, K.J. (1997). Asking the accuracy question: Is measurement the answer? In R.Spears, P.J.Oakes, N.Ellemers, & S.A.Haslam (Eds.), The social psychologyof stereotyping and group life.Oxford, UK: Blackwell. Oakes, P.J., & Turner, J.C. (1990). Is limited information processing the cause of social stereotyping? In W.Stroebe & M.Hewstone (Eds.), European Review of Social Psychology,1, (pp. 111–135). Chichester, UK: Wiley.
108 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Park, B., & Hastie, R. (1987). The perception of variability in category development: Instanceversus abstraction-based stereotypes. Journal of Personality and Social Psychology, 53,621– 635. Park, B., Judd, C.M., & Ryan, C.S. (1991). Social categorization and the representation of variability information. In W.Stroebe & M.Hewstone (Eds.), European Review of SocialPsychology, 2, (pp. 211–245). Chichester, UK: John Wiley. Park, B., Ryan, C.S., & Judd, C.M. (1992). Role of meaningful subgroups in explaining differences in perceived variability for in-groups and out-groups. Journal of Personality andSocial Psychology, 63,553–567. Pratto, F., Sidanius, J., Stallworth, L.M., & Malle, B.F. (1994). Social dominance orientation: A personality variable predicting social and political attitudes. Journal of Personality andSocial Psychology, 67,741–763. Rosenthal, R., Hall, J.A., DiMatteo, M.R., Rogers, P.L., & Archer, D. (1979). Sensitivity tononverbal communication: The PONS test.Baltimore: Johns Hopkins University Press. Ryan, C.S. (1995). Motivations and the perceiver’s group membership: Consequences for stereotype accuracy. In Y.Lee, L.J.Jussim, & C.R.McCauley (Eds.), Stereotype accuracy:Toward appreciating group differences(pp. 189– 214). Washington, DC: American Psychological Association. Ryan, C.S. (1996). Accuracy of black and white college students’ in-group and outgroup stereotypes. Personality and Social Psychology Bulletin, 22,1114–1127. Ryan, C.S., & Bogart, L.M. (1997). Development of new group members’ in-group and out-group stereotypes: Changes in perceived group variability and ethnocentrism. Journal ofPersonality and Social Psychology, 73,719–732. Ryan, C.S., & Bogart, L.M. (2001). Longitudinal changes in the accuracy of new group members’ in-group and out-group stereotypes. Journal of Experimental Social Psychology,37,118–133. Ryan, C.S., Bogart, L.M., & Vender, J.P. (2000). Effects of perceived group variability on the gathering of information about individual group members. Journal of Experimental SocialPsychology, 36,90–101. Ryan, C.S., Judd, C.M., & Park, B. (1996a). Effects of racial stereotypes on judgments of individuals: The moderating role of perceived group variability. Journal of ExperimentalSocial Psychology, 32,71–103. Ryan, C.S., Park, B., & Judd, C.M. (1996b). Assessing stereotype accuracy: Implications for understanding the stereotyping process. In C.N. Macrae, C.Stangor, & M.Hewstone (Eds.), Stereotypes and stereotyping(pp. 121–157). New York: Guilford Press. Ryan, C.S., Robinson, D.R., & Hausmann, L.R.M. (in press). Group socialization, uncertainty reduction, and the development of new members’ perceptions of group variability. In V.Yzerbyt, C.M.Judd, & O.Corneille (Eds.), The psychology of groupperception: Contributions to the study of homogeneity, entitativity, and essentialism.Hove, UK: Psychology Press. Sagar, H.A., & Schofield, J.W. (1980). Racial and behavioral cues in Black and White children’s perceptions of ambiguously aggressive acts. Journal of Personality and SocialPsychology, 39,590–598.
STEREOTYPE ACCURACY 109
Schaller, M., Boyd, C, Yohannes, H., & O’Brien, M. (1995). The prejudiced personality revisited: Personal need for structure and formation of erroneous group stereotypes. Journalof Personality and Social Psychology, 68,544–555. Schwarz, N. (1999). Self-reports: How the questions shape the answer. American Psychologist,54,93–105. Sherif, M., & Sherif, C.W. (1953). Groups in harmony and tension.New York: Harper. Simon, B., & Brown, R. (1987). Perceived intragroup homogeneity in minoritymajority contexts. Journal of Personality and Social Psychology, 53,703–711. Smith, A.E., Jussim, L., Eccles, J., VanNoy, M., Madon, S., & Palumbo, P. (1998). Self-fulfilling prophecies, perceptual biases, and accuracy at the individual and group levels. Journal of Experimental Social Psychology, 34,530–561. Stangor, C. (1995). Content and application inaccuracy in social stereotyping. In Y.Lee, L.J. Jussim & C.R.McCauley (Eds.), Stereotype accuracy: Toward appreciating group differences(pp. 275–292). Washington, DC: American Psychological Association. Swann, W.B., Jr. (1984). Quest for accuracy in person perception: A matter of pragmatics. Psychological Review, 91,457–477. Swim, J.K. (1994). Perceived versus meta-analytic effect sizes: An assessment of the accuracy of gender stereotypes. Journal of Personality and Social Psychology, 66,21–36. Tajfel, H. (1969). Cognitive aspects of prejudice. Journal of Social Issues, 25,79–98. Tajfel, H., & Turner, J.C. (1979). An integrative theory of intergroup conflict. In W.G.Austin & S.Worchel (Eds.), The social psychology of intergroup relations (pp. 33–47). Monterey, CA: Brooks/Cole. Tajfel, H., & Wilkes, A.L. (1963). Classification and quantitative judgement. British Journal ofSocial Psychology, 54,101–114. Wolsko, C., Park, B., Judd, C.M., & Wittenbrink, B. (2000). Framing interethnic ideology: Effects of multicultural and color-blind perspectives on judgments of groups and individuals. Journal of Personality and Social Psychology, 78,635– 654.
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY, 2002, 13, 111–153
The psychology of system justification and thepalliative function of ideology John T.Jost Stanford University, and the Radcliffe Institute forAdvanced Study at Harvard University, USA Orsolya Hunyady University of Debrecen, Hungary, and theUniversity of California at Berkeley, USA
In this chapter, we trace the historical and intellectual origins of system justification theory, summarise the basic assumptions of the theory, and derive 18 specific hypotheses from a system justification perspective. We review and integrate empirical evidence addressing these hypotheses concerning the rationalisation of the status quo, the internalisation of inequality (outgroup favouritism and depressed entitlement), relations among ego, group, andsystem justification motives (including consequences for attitudinal ambivalence, selfesteem, and psychological well-being), and the reduction ofideological dissonance.Turning to the question of why people would engage in system justification—especially when it conflicts with other interests and motives—we propose that system-justifying ideologies serve a palliative function in that they reduce anxiety, guilt, dissonance, discomfort, and uncertainty for those who are advantaged and disadvantaged. The primary excusatory function of ideology, therefore, is to provide people, both as victims and pillars of the posttotalitarian system, with the illusion that the system is in harmony with the human order and the order of the universe. (Vaclav Havel, 1991, p. 134) © 2002 European Association of Experimental Social Psychology http://www.tandf.co.uk/journals/pp/02699931.html DOI: 10.1080/10463280240000046
PSYCHOLOGY OF SYSTEM JUSTIFICATION 111
Throughout the world, most or all of the available wealth, power, privilege, and prestige is enjoyed by a minority of citizens. In the United States, for instance, the richest 1% controls almost half of the country’s total financial wealth, and the top 20% possesses 94% of the nation’s net wealth (Wolff, 1996, pp. 10–11). European nations have grown somewhat more egalitarian over the course of the 20th century, but even in France, England, and Sweden, the richest 1% still hold 20% or more of the total national wealth, and the top 20% control much more than half (Wolff, 1996, pp. 21–25). Economists calculate a gini coefficient to measure the degree to which the distribution of income and wealth is skewed disproportionately in favour of the wealthy, and while there are important differences across countries, the results are fairly similar in the US, Canada, UK, Germany, Italy, Switzerland, Holland, Sweden, Norway, Israel, Australia, and Japan (O’Higgins, Schmaus, & Stephenson, 1990). In all of these nations, there is a wide gap between rich and poor, and the spread of global capitalism seems to be increasing the gap (e.g., Marshall, 2000; Wolff, 1996). Despite the visibility of a relatively small number of protestors at recent meetings of the World Trade Organisation in both the United States and Italy, the existence of widespread economic inequality does not currently pose a significant threat to the legitimacy or stability of the capitalist system or to that of major national governments in North America, Europe, or Asia. On the contrary, most people seem to find ways of tolerating and even justifying social and economic disparities as fair, legitimate, necessary, and inevitable. Social scientists typically point to the role of ideology in maintaining popular support for the system by explaining, justifying, and rationalising inequality in such a way that people are seen as deserving the outcomes and treatment they receive (e.g., Jackman, 1994; Lane, 1962; Major, 1994; Sidanius & Pratto, 1999; Tyler & McGraw, 1986).
Correspondence should be addressed to John T.Jost, Radcliffe Institute for Advanced Study, Harvard University, 38 Concord Ave., Cambridge, MA 02138, USA. Email:
[email protected] We wish to acknowledge the essential contributions made by Diana Burgess, Mauricio Carvallo, Elizabeth Haines, Maria Cristina Jimenez, Aaron Kay, Yephat Kivetz, Brenda Major, Brett Pelham, Oliver Sheldon, Bilian Ni Sullivan, and Erik Thompson to the research summarised in this chapter. Preparation of the manuscript was facilitated by Lea Richards, and helpful suggestions for revision were made by Miles Hewstone, Gerold Mikula, Larissa Myaskovsky, Jennifer Overbeck, Wolfgang Stroebe, and several anonymous reviewers. We are also grateful for incisive comments we received at Stanford, UCLA, the University of California at Berkeley, and the University of Michigan, where portions of this esearch were presented.
112 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Stereotypes of the working class (or immigrants or Gypsies) as lazy, irresponsible, and unintelligent allow people to blame these groups for their own poverty and to deflect blame from the system. Ideological beliefs associated with individualism, meritocracy, belief in a just world, and the Protestant work ethic presumably serve the same function (e.g., Furnham, 1990; Kluegel & Smith, 1986; Lerner & Miller, 1978; Weiss, 1969). The use of stereotypes and other ideological devices to preserve the legitimacy of the existing social system is a major focus of system justification theory. Although there are several influential precursors discussed in the next section, system justification theory largely originated with an article by Jost and Banaji (1994), which sought to account for the consensuality of social stereotypes across perceiver groups and the prevalence of outgroup favouritism among members of disadvantaged groups. It was argued that many common forms of stereotyping and intergroup behaviour could not be explained in terms of prevailing theories, which tended to stress either ego-justifying motives to maintain or enhance individual self-esteem or group-justifying motives to maintain or enhance collective self-esteem and/or positive group distinctiveness. Drawing on such diverse sources as marxism-feminism, cognitive dissonance theory, justice research, and social identity theory, Jost and Banaji proposed the existence of a system-justifying motive, whereby people seek to maintain or enhance the legitimacy and stability of existing forms of social arrangements. The most provocative aspect of this argument was that members of disadvantaged groups would themselves engage in system justification (at least under some circumstances), even at the expense of their immediate personal or collective interests or esteem. Since the publication of Jost and Banaji’s (1994) article, research on system justification theory has addressed a much wider set of concerns, many of which are reviewed here. These include the tendency for people to subjectively enhance the desirability of anticipated events (whether good or bad) as they become more likely (Kay, Jimenez, & Jost, 2002); the tendency for members of disadvantaged groups to accept and legitimise their own situations (Haines & Jost, 2000; Jost, 2001); implicit as well as explicit cognitive, affective, and behavioural biases in favour of higherstatus groups (Jost, 2001; Jost, Pelham, & Carvallo, 2002; Nosek, Banaji, & Greenwald, 2002); the depressed entitlement effect among women and other disadvantaged groups (Blanton, George, & Crocker, 2001; Jost, 1997; Major, 1994; Pelham & Hetts, 2001); attitudinal ambivalence directed at one’s low-status group (Glick & Fiske, 2001; Jost & Burgess, 2000); consequences for self-esteem and psychological well-being among members of disadvantaged groups who support the system and oppose egalitarian reforms (Jost & Thompson, 2000); and the surprising degree of ideological support for the social system and its authorities provided by members of disadvantaged groups (Jost, Pelham, Sheldon, & Sullivan,
PSYCHOLOGY OF SYSTEM JUSTIFICATION 113
2003b). All of these various phenomena pertain, in one way or another, to the antecedent conditions, manifestations, and/or consequences of the system justification motive. OVERVIEW OF THE CHAPTER In this chapter, we first trace the origins of the system justification perspective and its major theoretical influences, and we contrast it with other prominent theories of ideology, justice, and intergroup relations. Then we summarise the basic assumptions of system justification theory and derive several specific hypotheses that are unique to the theory. Next, we review and integrate conclusions from a number of empirical studies (most of which have been published elsewhere) that address the hypotheses of system justification theory. Finally, we address the question of why people would engage in system justification, especially when it conflicts with individual and group interests. We speculate that system-justifying ideologies serve a palliative function in that they reduce anxiety, guilt, dissonance, discomfort, and uncertainty for people who are in positions that are either advantaged or disadvantaged. ORIGINS OF THE SYSTEM JUSTIFICATIONPERSPECTIVE System justification theory originated in an effort to integrate and expand upon several bodies of substantive work, including social identity theory, just world theorising, cognitive dissonance theory, marxist-feminist theories of ideology, and social dominance theory. In this section, we summarise major similarities and differences between system justification theory and other prominent theories, before moving on to summarise predictions that follow uniquely from a system justification perspective. Our theory should be thought of as a compliment, a complement, and in some ways also a corrective to its theoretical predecessors. Social identity theory The first substantive influence was that of social identity theory, especially its attempt to link patterns of stereotyping, prejudice, and intergroup relations to sociostructural variables such as the perceived legitimacy and stability of the system (e.g., Tajfel, 1981; Tajfel & Turner, 1986; Turner & Brown, 1978). Tajfel (1981, p. 156) explicitly considered the social (as well as cognitive) functions of stereotyping, arguing that: outgroup social stereotypes tend to be created and widely diffused in conditions which require: (i) a search for the understanding of
114 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
complex, and usually distressful, large-scale social events; (ii) justification of actions, committed or planned, against outgroups; (iii) a positive differentiation of the ingroup from selected outgroups at a time when such differentiation is perceived as becoming insecure and eroded; or when it is not positive, and social conditions exist which are perceived as providing a possibility for a change in the situation (p. 156). Thus, Tajfel described a link between stereotyping and group ideologies, but the justification function he emphasised had to do with using stereotypes to justify discrimination against and resistance to outgroup members. This fits with the overall prominence (beginning with the minimal group paradigm) accorded to the phenomenon of ingroup bias in social identity theory (see also Jost, 2001). With regard to the situation facing low-status groups, the key question for social identity theorists is how their members seek to overcome “threatened identities” arising from their position in the hierarchy (e.g., Tajfel & Turner, 1986). The theory emphasises identity-related motives to move “from social stability to social change” (Tajfel, 1981) or from “passive acceptance to collective protest” (Wright, Taylor, & Moghaddam, 1990) or from “social reality to social resistance” (Spears, Jetten, & Doosje, 2001) whenever circumstances allow for these possibilities. Social identity theory’s forte is accounting for situations of intergroup conflict in which sides are highly polarised and antagonistic and boundaries between groups are clear and distinctive (e.g., Brown, 2000; Hewstone, 1989; Turner, Hogg, Oakes, Reicher, & Wetherell, 1987). Because the theory locates all social behaviour on a continuum ranging from “interpersonal” to “intergroup” behaviour (e.g., Tajfel, 1981; Tajfel & Turner, 1986), it fails to account adequately for the fact that unequal social systems are maintained because people support them even when a different system would serve their own personal and group interests better (see Jost, 1995). In part, this is because people find it difficult to imagine “cognitive alternatives,” as Tajfel and Turner (1986) propose (but do not explain), and in part it is because of social psychological motives to justify and rationalise the way things are. Thus, Jost and Banaji (1994) argued that a system justification perspective provides a better and more complete account of outgroup favouritism among low-status groups than a social identity perspective does. A system justification perspective also helps to understand why it is such a difficult personal and collective task to overcome social stability, passive acceptance, and the apparent demands of “social reality”.
PSYCHOLOGY OF SYSTEM JUSTIFICATION 115
The belief in a just world The second major substantive influence came from the field of justice research, and it did not fit very well with social identity theory’s emphasis on group-serving biases in stereotyping, attribution, and social perception (e.g., Hewstone, 1989; Hogg & Abrams, 1988; Tajfel, 1981). Specifically, work on the “tolerance of injustice” among the disadvantaged (e.g., Martin, 1986; Tyler & McGraw, 1986) seemed to contradict the notion that individuals and groups defend their interests and identities. Similarly, the fact that people want to believe in a “just world” in which people “get what they deserve and deserve what they get” (e.g., Lerner, 1980) may be consistent with self-interest for people who are advantaged, but it leads to self-blame and the internalisation of inferiority among the disadvantaged (e.g., Miller & Porter, 1983). System justification theory follows up on the possibility that people are motivated to believe that outcomes and arrangements are fair, legitimate, and deserved, but it rejects the idea that the “belief in a just world” is a universal need arising (solely or primarily) from the desire to perceive that one has control over one’s environment (Lerner, 1980). We also disagree with the notion that genuine justice (rather than justification) concerns underlie the belief in a just world. Rather, our theory stresses processes of ideological persuasion and social learning that lead people to rationalise the way things are (e.g., Bem & Bem, 1970; Jost & Banaji, 1994; Kluegel & Smith, 1986; Tyler & McGraw, 1986). And, although Lerner and Miller (1978) commented that insufficient attention had been given to individual differences in the tendency to believe in a just world, almost all of the research in this area over the past two decades has focused on the connection between individual differences in just world beliefs and victimblaming tendencies (e.g., Furnham & Procter, 1989; Rubin & Peplau, 1975). System justification theory reflects the influence of just world theorising, but we consider a much broader set of causes (dispositional, situational, cultural) and consequences (for ideology, justice, and intergroup relations) of the socially validated belief that the status quo is legitimate and necessary. Cognitive dissonance theory The most prominent social psychological theory of justification and rationalisation processes is cognitive dissonance theory (e.g., Festinger, 1957). Although system justification theory is strongly influenced by dissonance theory (see especially Jost et al., 2003b; Kay et al., 2002), it differs in at least three significant ways. First, dissonance theory is often interpreted as an ego justification theory, in so far as efforts at dissonance reduction are seen as driven by the desire to preserve a positive image of
116 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
the self following acts of hypocrisy (Aronson, 1992; Greenwald & Ronis, 1978; Steele & Liu, 1983). By contrast, we propose that when people reduce ideological dissonance, they defend the legitimacy of the system in order to maintain a positive image of that system, and this may come at the expense of a positive self-image or a positive group image. A second difference is that most dissonance theorists assume that people must feel personally responsible for the aversive consequences of an action in order to justify it (e.g., Cooper & Fazio, 1984; Wicklund & Brehm, 1976), whereas system justification theory suggests that people justify the status quo, even when they have no direct responsibility for it (Kay et al., 2002). A third difference is that dissonance theory stresses cognitive consistency (e.g., Abelson, Aronson, McGuire, Newcomb, Rosenberg, & Tannenbaum, 1968), whereas system justification theory stresses motives to imbue the system with legitimacy and fairness, even if such beliefs actually create dissonance, conflict, and ambivalence (Jost & Burgess, 2000). Marxist-feminist theories of ideology The marxist-feminist analysis of ideology, especially as it developed throughout the 20th century (e.g., Elster, 1983; Gramsci, 1971; Lukács, 1971; MacKinnon, 1989), emphasises the cognitive dimensions of oppression and system preservation. It builds on Marx and Engels’ (1846/ 1978) observation that, “The class which has the means of material production at its disposal, has control at the same time over the means of mental production, so that, thereby, generally speaking, the ideas of those who lack the means of production are subject to it” (p. 172). This insight concerning the function of dominant ideology as an instrument of social control has been enormously influential in sociology and political science, and it has proved useful even in critiques of Communist systems (e.g., Havel, 1991). Jost (1995) argued that the marxian-feminist analysis of “false consciousness” is useful for understanding why the disadvantaged sometimes hold attitudes and beliefs that play some role in their own subjugation, and Jost and Banaji (1994) applied this concept to analyse consensual stereotyping and outgroup favouritism in particular. Social dominance theory Social dominance theory (unlike system justification theory) makes the evolutionary assumption that “all social systems will converge toward the establishment of stable, group-based social hierarchies” (Sidanius & Pratto, 1993, p. 177). One of the mechanisms by which human beings are said to maintain unequal relations between groups is the diffusion of “legitimising myths”, which are “hierarchy enhancing” rather than “hierarchy attenuating” (Sidanius & Pratto, 1999). Pratto, Sidanius,
PSYCHOLOGY OF SYSTEM JUSTIFICATION 117
Stallworth, and Malle (1994) developed an individual difference scale measuring “social dominance orientation” (SDO), and it has been widely used to predict hierarchy-enhancing attitudes, behaviours, and even career choices (e.g., Altemeyer, 1998; Sidanius & Pratto, 1999). Thus, the SDO scale, like the belief in a just world, provides a reasonably good measure of individual differences in system-justifying tendencies (Jost & Burgess, 2000). However, Jost and Thompson (2000) argued that conceptual and operational definitions of social dominance orientation have generally confounded two concepts that system justification theorists are at pains to distinguish, namely the desire for ingroup superiority (group justification) and the desire to preserve existing hierarchical arrangements (system justification). In several studies, they found that a (correlated) two-factor solution of the SDO scale fitted the data better than a one-factor solution. Scale items are listed in Table 1, along with loadings from two studies on the factors of “group-based dominance” (GBD; similar to group justification) and “opposition to equality” (OEQ; a type of system justification). Jost and Thompson found that GBD was positively related to ingroup favouritism for European Americans and African Americans alike. By contrast, OEQ was related positively to self-esteem and ingroup favouritism for European Americans, but it was related negatively to ingroup favouritism and self-esteem for African Americans. As discussed below, these findings are consistent with a system justification analysis of conflicts among ego, group, and system justification motives. HYPOTHESES DERIVED FROM SYSTEMJUSTIFICATION THEORY A system justification view integrates and builds on various ideas from social identity theory, just world research, dissonance theory, marxismfeminism, and social dominance theory. At the most basic level, we postulate the existence of a system justification motive, whereby people justify and rationalise the way things are, so that existing social arrangements are perceived as fair and legitimate, perhaps even natural and inevitable. Most of our theoretical and empirical efforts have been focused on this motive, because it has been neglected relative to two other motives long appreciated by social psychologists—ego justification and group justification. The hypotheses and findings we review in this chapter address the rationalisation of the status quo, the internalisation of inequality (including outgroup favouritism and depressed entitlement), relations among ego, group, and system justification, and the reduction of ideological dissonance.
118 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
TABLE 1 Factor loadings from the standardised solution of a confirmatory analysis of the correlated two-factor model of the social dominance orientation (SDO) scale
Data are adapted from Jost and Thompson (2000, Studies 1 and 2), aggregating across European and African Americans. “GBD” indicates group-based dominance items; “OEQ” indicates opposition to equality items (reverse-scored).
Rationalisation of the status quo System justification theory makes several predictions concerning the rationalisation of the status quo in general. These concern the tendency for people to elevate the desirability of anticipated events as their likelihood increases (Kay et al., 2002; McGuire & McGuire, 1991), the tendency for people to use stereotypes to justify status differences between groups (Hoffman & Hurst, 1990; Jost, 2001; Jost & Banaji, 1994)—especially when the system is under threat (Jost, Overbeck, Guermandi, Rubini, Mosso, & Kivetz, 2003a)—and tendencies for members of disadvantaged groups to accept and even legitimise their own powerlessness (Haines &
PSYCHOLOGY OF SYSTEM JUSTIFICATION 119
Jost, 2000; Jost, 2001). In so far as people are motivated to imbue the status quo with legitimacy, stability, and rationality, they should exhibit system justification tendencies with respect to judgements of desirability, affect, stereotyping, and memory. Specific hypotheses may be summarised as follows: (H1) People will rationalise the status quo by judging likely events to be more desirable than unlikely events, (a) even in the absence of personal responsibility, (b) whether those events are initially defined as attractive or unattractive, and (c) especially when motivational involvement is high rather than low. (H2) People will use stereotypes to rationalise social and economic status differences between groups, so that the same target group will be stereotyped differently depending on whether it is perceived to be high or low in status. (H3) People will defend and justify the social system in response to threat by using stereotypes to differentiate between high- and lowstatus groups to a greater degree. (H4) Providing explanations (or pseudo-explanations) for status or power differences between groups will (a) increase the use of stereotypes to rationalise differences, and (b) lead members of disadvantaged groups to express more positive (relative to negative) affect. (H5) Over time, members of disadvantaged groups will misremember explanations for their powerlessness as being more legitimate than they actually were. Internalisation of inequality For members of advantaged groups, rationalising the status quo also means rationalising their own position of advantage, which is consistent with the expression of ingroup favouritism. For members of disadvantaged groups, however, one of the (unintended) consequences of rationalising the status quo is the internalisation of inequality. That is, to the extent that one subscribes to the legitimacy of the system and its outcomes, one accepts blame or responsibility for being in a state of disadvantage (e.g., Jost & Banaji, 1994; Lane, 1962; Miller & Porter, 1983).1 We focus on two examples of the internalisation of inequality: ingroup vs outgroup favouritism and depressed entitlement. Ingroup vs outgroup favouritism. From the standpoint of system justification theory, outgroup favouritism on the part of low-status group members (like ingroup favouritism on the part of high-status group members) both reflects and contributes to ideological support for the system (Jost & Banaji, 1994). In this sense, it expresses a genuine,
120 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
internalised sense of inferiority, akin to false consciousness (Jost, 2001; Jost et al., 2002). Departing from at least some interpretations of social identity theory, we argue that outgroup favouritism is neither the result of demand characteristics (Mullen, Brown, & Smith, 1992) nor an insincere, self-presentational display of deference (e.g., Spears et al., 2001). By linking outgroup favouritism to ideological factors (and responses to ideological threats), a system justification perspective also suggests that outgroup favouritism is not merely the result of “halo effects” or consistency biases. Hypotheses concerning ingroup-outgroup favouritism may be stated as follows: (H6) Members of low-status groups will exhibit outgroup favouritism even on (a) open-ended, non-reactive, qualitative measures, and (b) implicit, nonconscious cognitive, affective, and behavioural measures. (H7) As the perceived legitimacy of the system increases, (a) members of high-status groups will exhibit increased ingroup favouritism, and (b) members of low status groups will exhibit increased outgroup favouritism. (H8) As system justification tendencies increase, (a) members of high-status groups will exhibit increased ingroup favouritism, and (b) members of low-status groups will exhibit increased outgroup favouritism. Depressed entitlement. Numerous studies conducted over several decades have found that women believe that they deserve less money for their work than men do (e.g., Callahan-Levy & Messé, 1979; Major, 1994; Major, McFarlin, & Gagnon, 1984). From a system justification perspective, this is another (largely implicit) example of the internalisation of inferiority
1
Hofstede (1997) has described an analogous tendency associated with “power distance”, which he defines as “the extent to which the less powerful members of institutions and organizations within a country expect and accept that power is distributed unequally” (p. 28). The thrust of Hofstede’s analysis is on cross-cultural differences with respect to power distance (which, like the belief in a just world and opposition to equality [a subscale of social dominance orientation], we see as a kindred concept to system justification). Although cross-cultural differences are beyond the scope of this chapter, it is worth noting that in Hofstede’s (1997, p. 42) research countries in which Romance languages are spoken (Spain, Portugal, Italy, and France) scored higher on power distance in general than did countries in which Germanic languages are spoken (Germany, England, Holland, Scandinavia). Future research would do well to reconcile these findings with observations made by others (e.g., Tiraboschi & Maass, 1998, p. 408) that Catholic countries (such as Italy) are generally more egalitarian than Protestant countries (such as Germany and the US).
PSYCHOLOGY OF SYSTEM JUSTIFICATION 121
(Jost, 1997). Several additional hypotheses concerning depressed entitlement may be derived from system justification theory (see Blanton et al., 2001; Pelham & Hetts, 2001): (H9) Members of disadvantaged groups (not just women) will exhibit a depressed sense of entitlement relative to members of advantaged groups, even in explicitly egalitarian environments. (H10) Members of disadvantaged groups will be more likely to exhibit depressed entitlement (relative to members of advantaged groups) for past work that has already been completed than for future work that has not yet been completed. Relations among ego, group, and systemjustification motives We use the term ego justification to refer to the tendency to develop and maintain a favourable self-image and to feel valid, justified, and legitimate as an individual actor (see Jost & Banaji, 1994). Group justification, which is the primary (but not sole) focus of social identity theory, captures the desire to develop and maintain favourable images of one’s own group and to defend and justify the actions of fellow ingroup members. According to system justification theory, motives for ego, group, and system justification are consistent and complementary for members of high-status or advantaged groups. That is, believing that the social system is structured fairly to reward the worthy and punish the unworthy is also perfectly consistent with motives to feel and assert that one is a good and worthwhile person and that one’s social group is valued and respected. For members of low-status or disadvantaged groups, by contrast, ego, group, and system justification motives are often in conflict with one another. In this case, the tendency to accept the fairness and legitimacy of the social system is at odds with motives for the enhancement of individual or collective self-esteem. What this means is that the situation faced by members of disadvantaged groups is rife with the potential for conflicts or crises among the self, the group, and the system (see Jost, Burgess, & Mosso, 2001). It also means that the disadvantaged are most likely to engage in system justification when competing motives for ego justification or group justification are low in salience or strength. From these basic theoretical assumptions concerning the existence of a system justification motive and relations among ego, group, and system justification motives, a number of additional hypotheses may be derived concerning attitudinal ambivalence (e.g., Glick & Fiske, 2001; Jost & Burgess, 2000), self-esteem, and well-being (Jost & Thompson, 2000):
122 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
(H11) Members of low-status groups will exhibit greater ambivalence towards their own group than will members of high-status groups. (H12) Members of low-status groups will exhibit increased ambivalence towards their own group as system justification is increased. (H13) Members of high-status groups will exhibit decreased ambivalence towards their own group as system justification is increased. (H14) System justification will be associated with (a) increased selfesteem for members of advantaged groups, and (b) decreased selfesteem for members of disadvantaged groups. (H15) System justification will be associated with (a) decreased depression for members of advantaged groups, and (b) increased depression for members of disadvantaged groups. (H16) System justification will be associated with (a) decreased neuroticism for members of advantaged groups, and (b) increased neuroticism for members of disadvantaged groups. The reduction of ideological dissonance By drawing on the logic of dissonance theory, it may be argued that members of disadvantaged groups should have the strongest system justification needs, at least under certain circumstances (Elster, 1983; Jost et al., 2003b; Lane, 1962). Just as suffering paradoxically increases commitment to the sources of one’s suffering through dissonance-reduction mechanisms (e.g., Wicklund & Brehm, 1976), members of disadvantaged groups may exhibit enhanced levels of system justification relative to members of advantaged groups (Jost et al., 2003b), especially as objective circumstances worsen (Glick & Fiske, 2001). Thus, a theoretical hybrid of cognitive dissonance and system justification perspectives leads to the following counter-intuitive predictions: (H17) When individual and group needs and interests are low in salience or strength, members of disadvantaged groups will provide stronger support for the social system and its authorities than will members of advantaged groups, in so far as the former will have a stronger need than the latter to reduce ideological dissonance through system justification. (H18) System justification levels will be higher in societies in which social and economic inequality is more extreme rather than less extreme.
PSYCHOLOGY OF SYSTEM JUSTIFICATION 123
EMPIRICAL STUDIES OF SYSTEM JUSTIFICATIONHYPOTHESES In this section, we review and integrate empirical studies addressing the 18 hypotheses derived above from the basic tenets of system justification theory. Most of these studies have been published elsewhere in a variety of sources (e.g., Blanton et al., 2001; Glick & Fiske, 2001; Haines & Jost, 2000; Jost, 1997, 2001; Jost & Burgess, 2000; Jost et al., 2002, 2003b; Jost & Thompson, 2000; Kay et al., 2002; Pelham & Hetts, 2001). Space limitations obviously prohibit a comprehensive review of methods and findings from all of these studies; rather, we cover the highlights in so far as they relate to the hypotheses of interest. Rationalisation of the status quo Coming to terms with the inevitable . According to system justification theory, the human capacity for rationalisation lends considerable stability and support to the social system. That is, people’s remarkable ability to accommodate formerly unwelcome outcomes may help to explain why social and political systems are successful at retaining cooperation and consent and why social change is so difficult to accomplish (e.g., Elster, 1983; Jost, 1995; Lane, 1962). McGuire and McGuire (1991) argued for the existence of both “sour grapes” and “sweet lemon” types of rationalisations, such that people derogate anticipated events as their likelihood decreases and enhance the subjective value of anticipated events as their likelihood increases. Dissonance researchers have also studied rationalisation processes, but they have largely confined their studies to cases in which (a) people are personally responsible for the outcomes they justify, and (b) the rationalisation occurs post hoc, that is, only after a choice or behaviour has occurred. Kay et al. (2002) hypothesised that people would engage in a rationalisation of the existing state of affairs, whether or not they were personally responsible for bringing it about and whether they stood to gain or lose. They argued that the legitimation needs of the system would be best served by people anticipating likely outcomes and rationalising them in advance (H1). Kay et al. (2002) investigated the anticipatory rationalisation of the status quo in two studies, one of which addressed reactions to the US Presidential Election of 2000 between George W.Bush and Al Gore. In the week immediately prior to the election, we administered a survey to 288 adult respondents in which we (a) manipulated the perceived likelihood that one (or the other) candidate would win, and (b) measured the subjective desirability of each possible outcome. Data were analysed according to a Partisanship (three levels: Republicans, Democrats, and non-
124 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
partisans) by Outcome Likelihood (five levels ranging from strong likelihood of a Gore victory to strong likelihood of a Bush victory) between-subjects factorial design. Not too surprisingly, Republicans preferred Bush, and Democrats preferred Gore in general. But in both cases effects of partisanship were qualified by higher-order interactions with outcome likelihood. Republican and Democratic respondents rated both candidates to be significantly more desirable as the perceived likelihood of their winning the election increased, as indicated by linear contrast tests (see Figure 1). Non-partisans failed to show any significant rationalisation tendencies, presumably because they were not sufficiently motivationally invested in the outcome of a Bush-Gore election. These results were replicated in a follow-up experiment in which college students were found to rationalise large tuition increases (an unattractive outcome) and decreases (an attractive outcome) as they were seen as more likely to occur, providing further support for the existence of “sour grapes” and “sweet lemons” forms of rationalisation (see Kay et al., 2002). Stereotyping as rationalisation. Groundbreaking experimental research by Eagly and Steffen (1984) and Hoffman and Hurst (1990) suggested that people use stereotypes as a way of rationalising the unequal distribution of social roles. System justification theory builds on the theme that stereotypes arise (in part) to justify social and economic differences between groups (Jost & Banaji, 1994), proposing that the same target group (whether it is an ingroup or an outgroup) will be stereotyped differently depending on whether it is perceived to be high or low in status (H2). Jost (2001) described an experimental paradigm in which perceived socioeconomic success could be manipulated experimentally in the context of real-world group memberships (see sample materials in Table 2). A study by Jost and Burgess (2000, Study 1) made use of this paradigm, leading students at the University of Maryland to believe that alumni from their school were either more or less socioeconomically successful than were alumni from a rival school, the University of Virginia. This simple procedure has been successful (in several studies using a variety of groups) at leading members of “real” groups to explain and justify either the “success” or “failure” of their own group relative to a salient outgroup by showing ingroup favouritism or outgroup favouritism, respectively, on achievement-related traits. Importantly, neutral observers show similar patterns of stereotyping: the high-status target group is generally stereotyped as intelligent, hard-working, and competent, whereas the lowstatus target group is stereotyped as friendly, honest, and likeable (see Jost, 2001). We interpret this evidence as indicating that one function of stereotyping is to explain, justify, and rationalise unequal social outcomes (see also Conway, Pizzamiglio, & Mount, 1996; Glick & Fiske, 2001; Jackman & Senter, 1983; Jost & Banaji, 1994; Ridgeway, 2001).
PSYCHOLOGY OF SYSTEM JUSTIFICATION 125
Figure 1. (A) Desirability ratings of a Bush Presidency. (B) Desirability ratings of a Gore Presidency. Adapted from Kay et al. (2002, Study 1). For ratings of both Bush and Gore, higher numbers indicate greater judged desirability. On the X-axis, values range from least to most likely that Bush (A) or Gore (B) would be elected. Linear contrast tests indicated that Democrats and Republicans rated both presidencies (preferred and non-preferred) to be more desirable as their likelihood increased (p•. 07 in all cases), whereas nonpartisans did not.
Increased stereotypic rationalisation in response to system threat. Jost et al. (2003a) argued that people would be especially likely to use stereotypes to bolster support for the status quo following an ideological attack on the system. More specifically, we predicted that people would show increased stereotypic differentiation in response to a system-level threat (H3). For members of high-status groups, this is equivalent to scapegoating members of lower-status groups under conditions of system threat, as authoritarianism theory would predict (e.g., Adorno, Frenkel-Brunswik, Levinson, & Sanford, 1950; Altemeyer, 1998). A novel prediction that
126 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
TABLE 2 Sample materials for manipulating perceived socioeconomic status in experimental studies involving real-world groups
These materials were used in studies reported by Jost and Burgess (2000) and Jost (2001).
arises from system justification theory is that members of low-status groups would engage in self-scapegoating in order to provide ideological support (through the rationalisation of inequality) for the system at a time when it appears to be vulnerable. This possibility was investigated in a study conducted in Israel concerning differences between Ashkenazi Jews of European descent, who are relatively high in social and economic status, and Sephardic Jews of Middle Eastern and African descent, who occupy a much lower-status position both socially and economically (Jost et al., 2003a). Survey respondents were sampled from public trains in the Tel Aviv area and asked about their beliefs concerning the characteristics of Ashkenazim and Sephardim. Just prior to reporting these beliefs, respondents were exposed to either a “high system threat” message or a “low system threat” message. 2 Mean levels of ingroup and outgroup favouritism (aggregated across the traits of intelligent, ambitious, responsible, hard-working, calm, openminded, and valuing education) are presented in Figure 2 as a function of group membership and system threat. Under conditions of low threat, both
PSYCHOLOGY OF SYSTEM JUSTIFICATION 127
groups exhibited mild ingroup favouritism, claiming that their own group was slightly better in terms of these (largely achievement-related) characteristics. But under conditions of high threat, Ashkenazi respondents displayed greater ingroup favouritism, whereas Sephardic respondents displayed outgroup favouritism. Much as ego-related and group-related threats have been demonstrated to increase ingroup favouritism in prior research (Branscombe & Wann, 1994; Fein & Spencer, 1997; Grant & Brown, 1995; Oakes & Turner, 1980), we have found that a system-related threat increases consensual stereotypic differentiation between high- and low-status target groups. Placating the powerless. Haines and Jost (2000) argued that in so far as people are generally motivated to rationalise the status quo, they would accept and bolster even relatively placebic explanations for power differences between groups. Specifically, we hypothesised that exposing members of powerless groups to explanations for their powerlessness would increase the use of stereotypes to rationalise differences (H4a) and lead members of disadvantaged groups to express more positive (relative to negative) affect (H4b). We also considered the possibility that they would imbue those explanations with increased legitimacy over time (H5). These predictions were assessed in an experimental study involving students from Hunter College (in New York City) who believed (in some conditions) that students from a nearby college (Brooklyn College) held power over them. Specifically, in the power difference conditions they were told that Brooklyn College students would (a) evaluate their abilities at a task that required distinguishing authentic suicide notes from fakes, and (b) decide when they could leave the experiment. In other words, Hunter College students were outcome-dependent relative to Brooklyn College students (e.g., Goodwin, Operario, & Fiske, 1998). In this context, we investigated the effects of legitimate and illegitimate explanations for power on measures of affect, stereotyping, and memory. Based on pre-testing, we found that expertise, experience, and ability were perceived as the most legitimate reasons why one group would have power over another group (e.g., French & Raven, 1959). Therefore, these reasons 2The
high system threat message included the following passage: “These days, many people in Israel feel disappointed with the nation’s condition. Many citizens feel that the country has reached a low point in terms of social, economic, and political factors. People do not feel as safe and secure as they used to, and there is a sense of uncertainty regarding the country’s future.” By contrast, the low system threat message included the following: “These days, despite the difficulties the nation is facing, many people in Israel feel safer and more secure relative to the past. Many citizens feel that the country is relatively stable in terms of social, economic, and security factors. There is a sense of optimism regarding Israel’s future and an understanding that this is the only place where Israeli people can feel secure.”
128 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 2. Stereotypical ingroup and outgroup favouritism as a function of group membership and system threat. Adapted from Jost et al. (2003a, Study 5), aggregating across the following traits: intelligence, ambition, responsibility, industriousness, calmness, open-mindedness, and valuing of education. Positive scores indicate ingroup favouritism and negative scores indicate outgroup favouritism. The interaction between group membership and system threat was significant, p<.05.
were given as the bases of power mentioned in the “legitimate explanation” condition. Raters indicated that being friends with people in power as a way of acquiring power was the most illegitimate basis for having power over another group, and so this reason was used in the “illegitimate explanation” condition.3 A “no explanation” condition was also included in which participants were told nothing about the reasons for the power differences. Haines and Jost (2000) calculated the ratio of positive to negative affect expressed by students after learning about the power differences (or not) and, if there were power differences, after hearing a legitimate explanation, or an illegitimate explanation, or no explanation at all. Results indicated that there were no statistically reliable differences between the “legitimate explanation” condition and the “illegitimate explanation” condition on the ratio of positive to negative affect. When these two conditions were combined, they were successful in producing a significant improvement in affect over the “no explanation” condition. In other words, people felt
PSYCHOLOGY OF SYSTEM JUSTIFICATION 129
better after hearing an explanation, and it did not matter whether it was a legitimate explanation or an illegitimate explanation, supporting (H4b).4 We also found that Hunter College students rated the outgroup of Brooklyn College students as significantly more intelligent and more responsible in general when that group had power over them than when they did not (see Table 3), providing additional support for (H2). As Pepitone (1950) argued long ago, people are motivated to think that those who have power over them are qualified, deserving, and benign. The stereotyping data mirrored the affect data: Providing an explanation (whether legitimate or illegitimate) significantly increased the degree to which the outgroup was perceived as intelligent, responsible, and having a “right to judge” them. These data, which are summarised in Table 3, support hypothesis (H4a) that people are more likely to apply stereotypes that justify the use of power when they are provided with an explanation for the power differences than when they are not. At the conclusion of the experiment, participants were asked to recall the reason that was given just one hour earlier for the power differences between Brooklyn College and Hunter College. Four possible responses were given in a multiple-choice format: (a) “Because they are upper division Psychology students and work in a suicide prevention unit”, (b) “Because the principal investigator graduated from Brooklyn College and knows people from Brooklyn College”, (c) “We weren’t told”, and (d) “I don’t remember”. The percentage of research participants within each condition selecting each of the first three responses is presented in Figure 3. Consistent with (H5), people misremembered the reasons for the power differences as being morelegitimate than they actually were. In the “no explanation” condition, for instance, people mistakenly chose the legitimate explanation 33.3% of the time, and none of them chose the illegitimate explanation. In the “illegitimate explanation” condition, people
3 In the legitimate explanation condition, participants were told that the Brooklyn College students were “upper division Psychology students who worked in a suicide prevention unit”, and in the illegitimate explanation condition, they were told that the outgroup was given power because “the principal investigator had graduated from the school and had friends there”. 4 Readers may wonder how we square this finding that people did not differentiate between legitimate and illegitimate explanations with the hypothesis (H7) that the degree of perceived legitimacy should affect patterns of intergroup relations. These two ideas may be reconciled by drawing on the concept of attributional ambiguity (see Major & Schmader, 2001, p. 191). It follows from a system justification perspective that under conditions of ambiguity people will tend to exaggerate the legitimacy of the status quo, in the manner suggested by hypotheses (H4) and (H5). When circumstances are unambiguously perceived as illegitimate, however, they should lead to a qualitatively different pattern of responding, as suggested in (H7).
130 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
TABLE 3 Effects of power differences and explanation on stereotypical beliefs about the outgroup
Data are adapted from Haines and Jost (2000). Higher means indicate that the outgroup was rated to be more intelligent, more responsible, and as having more of a “right to judge.” (Standard deviations are listed in parentheses.) The “No Power” condition differed from the conditions in which power differences existed (multivariate F [1, 187]=3.39, p<.02); the presence of a power differential between the two groups led participants to rate the more powerful outgroup to be significantly more intelligent (p<.06) and more responsible (p<.01) but not as having more of a right to judge their work (n.s.). The “No Explanation” condition differed from the conditions in which an explanation was given for the power differences (multivariate F [1, 141]=4.79, p<.005); providing an explanation (whether legitimate or illegitimate) significantly increased the favourability of the outgroup stereotype for all three ratings (p<.05). a Means for the “Explanation” condition are collapsed across “Legitimate” and “Illegitimate” explanation conditions, which did not differ from one another.
falsely chose the legitimate explanation 30.2% of the time. This evidence suggests that people are relatively willing to give others—especially authority figures—the “benefit of the doubt” with regard to legitimacy, and even to actively imbue the proceedings with increased legitimacy (Haines & Jost, 2000). Internalisation of inequality Outgroup favouritism on open-ended, non-reactive measures. One potential objection to interpreting outgroup favouritism as indicative of system justification (e.g. Jost, 2001; Jost & Banaji, 1994; Jost & Burgess, 2000) is that such evidence may reflect demand characteristics or impression management rather than genuine internalisation on the part of low-status group members (e.g., Mullen et al., 1992; Scott, 1990; Spears et al., 2001). One way of addressing this issue is to utilise open-ended, nonreactive, qualitative measures of ingroup and outgroup favouritism, as suggested by hypothesis (H6a). In a study reported by Jost (2001) in which perceived socioeconomic success was manipulated with the use of experimental materials illustrated
PSYCHOLOGY OF SYSTEM JUSTIFICATION 131
Figure 3. Memory for initial reasons given for the power differences between groups. Adapted from Haines and Jost (2000), omitting “Don’t remember” responses. Values indicate percentage of participants (within each condition) choosing each reason for the power differences between groups. Correct responses for each condition are indicated above the appropriate bar. The results of a chisquare test of independence indicated that people consistently misremembered explanations as being more legitimate than they actually were ( 2=32.45, p<.0001, df=2, n=107).
in Table 2, Yale University students were asked to explain why members of their ingroup were either more or less successful than alumni from Stanford University. Two independent judges coded and content analysed the openended responses according to whether they focused on the ingroup (Yale) or the outgroup (Stanford) and whether they were favourable or unfavourable (or neutral) about that group. When Yale students were assigned to the high socioeconomic success condition, explanations making reference to characteristics of the ingroup tended to be very favourable (e.g., “Yale admits students with better records who are innately more driven”). When they generated explanations pertaining to the outgroup, these tended to be unfavourable (e.g., “Because Stanford is a sport scholarshipgranting school, they are going to get athletes that are not as intelligent as the students who get in regularly”). Results are illustrated in Figure 4. When Yale students were assigned to a position of low socioeconomic success, however, the results were very different. Under these conditions, explanations involving the ingroup were generally unfavourable in nature (e.g., “Yale students are too idealistic, and usually have impractical or false
132 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 4. Percentage of favourable and unfavourable attributions generated about the ingroup and the outgroup as a function of relative ingroup status. Adapted from data reported by Jost (2001). Values indicate percentage of participants (within each condition) generating favourable and unfavourable attributions about the ingroup (Yale students) and the outgroup (Stanford students) as a function of low or high perceived socioeconomic success of the ingroup (relative to the outgroup).
imaginations about real world life”), and explanations involving the outgroup were favourable (e.g., “Stanford is a more selective school, so it has smarter people”). Thus, members of low-status groups displayed strong outgroup favouritism in offering their own (unconstrained) attributions for the socioeconomic success differences (see Jost, 2001), suggesting that these rationalising opinions were sincerely generated. Outgroup favouritism on implicit, non-conscious measures. Three studies conducted by Jost et al. (2002) directly examined the internalisation of status inequality as a non-conscious form of system justification (H6b). The first study made use of the “Implicit Association Test” (IAT) developed by Greenwald, McGhee, and Schwartz (1998). Investigating a pre-existing status difference between (high-status) Stanford University students and (lower-status) San Jose State University students, we found that both groups were faster to associate Stanford-related stimuli with academic stereotypes (such as “successful”, “intelligent”, “ambitious”) and to associate San Joserelated stimuli with extracurricular stereotypes (such as “friendly”, “fashionable”, and “fun-loving”). For an affective measure of implicit ingroup vs outgroup favouritism, ingroup and outgroup stimuli
PSYCHOLOGY OF SYSTEM JUSTIFICATION 133
TABLE 4 Correlations between implicit ingroup favouritism (affective measure) and implicit self-esteem/implicit stereotyping as a function of group status
Data are adapted from Jost et al. (2002, Study 1). Values are bivariate correlation coefficients (Pearson rs). Results indicate that Stanford students who had higher implicit self-esteem and held stronger implicit stereotypes exhibited increased implicit ingroup favouritism on the affective measure. San Jose students who had lower implicit self-esteem and held stronger implicit stereotypes exhibited increased implicit outgroup favouritism. *p<.05, **p<.005
were paired with either pleasant words (such as “glory”, “warmth”, and “gold”) or unpleasant words (such as “poison”, “filth”, and “agony”). We found that Stanford students exhibited significant levels of ingroup favouritism, but San Jose students did not. In fact, 36% of the San Jose students had overall reaction times that reflected automatic, non-conscious preferences for the higher-status Stanford outgroup (compared with only 16% of Stanford students who showed outgroup favouritism on this measure). In addition, we found that implicit favouritism towards Stanford on the affective measure was positively correlated with an IAT measure of implicit self-esteem for Stanford students, but it was negatively correlated with implicit self-esteem for San Jose students (see Table 4), providing at least some evidence of the internalisation of inferiority at the level of nonconscious self-evaluation and group-evaluation (see also Nosek et al., 2002). For the lower-status group of San Jose students, internalisation of implicit stereotypes also predicted outgroup favouritism on the affective measure. For Stanford students, by contrast, there was a positive (nonsignificant) correlation between implicit stereotyping and ingroup favouritism (see Table 4). In a second study, Jost et al. (2002) sought to determine whether members of disadvantaged groups would exhibit outgroup favouritism on an indirect behavioural measure, contrary to theories of similarity, homophily, and social identification. American-born UCLA students of either Latino, Asian, or European descent were given the opportunity to choose an interaction partner for a “getting acquainted study” by pairing themselves up with strangers whose last names indicated that they were members of different ethnic groups. Each student received a photocopy of a sign-up sheet that had ostensibly been filled out by 24 other students. The
134 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
sign-up sheet contained the first initials and surnames (in different handwriting styles) of 24 fictitious students. We manipulated the apparent ethnicity of the bogus interaction partners by using eight common surnames from each of three ethnic groups. Students were asked to choose three different time slots (presumably on the basis of scheduling considerations) and rank their chosen time slots from most preferred to least preferred. In support of (H6b), members of low-status groups (Latinos and Asian Americans) preferred to interact with members of a higher-status outgroup (Whites) rather than with fellow ingroup members. As illustrated in Figure 5, Latinos and Asians chose to interact with ingroup members only 20–23% of the time, which was significantly below the 33% standard that one would expect on the basis of chance. In addition, members of all three groups revealed preferences for interacting with White interaction partners at least 40% of the time.5 A third study demonstrated that parents exhibit implicit gender biases in favour of males in the naming of their children and the advertisement of their births, providing further support for the hypothesis (H6) that inequality is internalized and does not merely reflect demand characteristics or impression management (Jost et al., 2002). Effects of perceived legitimacy on ingroup and outgroup favouritism. System justification theory suggests an interaction hypothesis (H7), such that perceived legitimacy should increase ingroup favouritism on the part of high-status group members and outgroup favouritism on the part of lowstatus group members.6 Using materials reproduced in Table 2, Jost and Burgess (2000, Study 1) manipulated perceived socioeconomic success differences between the ingroup (University of Maryland students) and a rival outgroup (University of Virginia students) and measured how fair or unfair, how justifiable or unjustifiable, and how legitimate or illegitimate those differences were ( =.71). The interaction hypothesis was supported. As can be seen in Table 5, perceived legitimacy was associated with increased ingroup favouritism for those who were assigned to high-status groups and with decreased ingroup favouritism (or increased outgroup favouritism) for those who were assigned to low-status groups (see also Jost, 2001).
5
An alternative explanation for these findings is that Latino and Asian American students may have had more familiarity with European Americans than with members of their own group and therefore would have preferred to interact with them in our “getting acquainted” study. However, the familiarity interpretation does not account for all of our data. It seems that Latinos and Asian Americans (but not European Americans) showed a preference for not interacting with members of their own group (see Figure 5). They did not avoid members of the other minority group, although these minority groups would have been relatively unfamiliar, compared with Whites.
PSYCHOLOGY OF SYSTEM JUSTIFICATION 135
Figure 5. Percentage of Latino, Asian, and White participants choosing interaction partners of each ethnic group. Adapted from Jost et al. (2002, Study 2). Numbers indicate percentage of participants (within each ethnic group) who chose interaction partners from each of the three target groups. Members of all three groups selected White interaction partners at least 40% of the time, which was significantly above the 33% standard that one would expect on the basis of chance (p<.05).
Effects of system justification on ingroup and outgroup favouritism. A conceptually related hypothesis is that system justification in general (as well as perceived legitimacy in particular) should be associated with increased ingroup favouritism on the part of high-status group members and increased outgroup favouritism on the part of low-status group members (H8). Jost and Thompson (2000) validated a scale for measuring economic system justification; items are listed in Table 6. This scale was administered to Northern Italians (high status) and Southern Italians (low status) in the context of a study on ingroup and outgroup favouritism (Jost et al., 2003a, Study 1). As can be seen in Figure 6, increased system justification was associated with increased ingroup favouritism among Northerners and increased outgroup favouritism among Southerners (see also Jost & Thompson, 2000).
6 This hypothesis differs from the main effect prediction of Turner and Brown (1978), who argued that because of status insecurity “[g]roups with illegitimate status relations would display more ingroup bias than those with legitimate status relations” (p. 210), regardless of the status of the ingroup.
136 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
TABLE 5 Correlations between perceived legitimacy and ingroup/outgroup favouritism as a function of group status
Data are adapted from Jost and Burgess (2000, Study 1). Values are bivariate correlation coefficients (Pearson rs). Results indicate that the perceived legitimacy of the system had opposite effects on ingroup/outgroup favouritism for members of high- and low-status groups. This was true for both achievement-related (intelligent, hard-working, and skilled at verbal reasoning) and socioemotional (friendly, honest, and interesting) attributes. *p<.05, **p<.01
Depressed entitlement Jost (1997) conducted an experimental replication of the depressed entitlement effect (e.g., Callahan-Levy & Messé, 1979; Major et al., 1984) to see if women in an explicitly feminist environment (Yale college in the 1990s) would exhibit the internalisation of inequality by reporting that they deserved to be paid less than men did for their own work (H9). In this study, 132 Yale undergraduates (68 men and 64 women) were required to perform some written work, listing the advantages and disadvantages of home shopping by computer. After evaluating the quality and determining the payment for several other examples of written work, the students were asked to evaluate their own written work on the same scales.7 Results are summarised in Figure 7. Yale women rated their own written work as less sophisticated, less original, and less insightful than Yale men rated their own work. When asked how much money they deserved to be paid for their work, women paid themselves on average $1.51 (or 18%) less than men paid themselves. Furthermore, the gender difference on payment remained statistically significant even after controlling for gender differences on the evaluation dimensions. The work was subsequently evaluated by two external, independent judges who were unaware of the author’s gender and of the hypotheses of the study. The judges perceived no differences between work that had been written by female authors and work that had been written by male authors. Thus, there were no actual differences between the work of men and women, only a difference in perceived deservingness. Pelham and Hetts (2001) extended the system justification analysis of depressed entitlement effects by demonstrating that members of other disadvantaged groups (not just women) would also feel that they deserved
PSYCHOLOGY OF SYSTEM JUSTIFICATION 137
TABLE 6 Items on the Economic System Justification Scale
This scale was used in research reported by Jost and Thompson (2000, Study 4, =. 73). Items followed by “(–)” are reverse-scored.
less than others, especially on difficult (as opposed to easy) tasks. Specifically, they found that people who were employed in low-paying jobs, regardless of their gender, believed that their work on difficult tasks was worth less than did people who were employed in higher-paying jobs. This evidence suggests that people internalise and adapt to economic inequality, apparently rationalising their own state of relative disadvantage and lowering their expectations accordingly. In an integration of cognitive dissonance and system justification theories, Blanton et al. (2001) hypothesised that members of disadvantaged groups would be more likely to exhibit depressed entitlement for past work than for future work (H10), in so far as effort justification applies to the former but not the latter case. Blanton et al. found that women “paid themselves” less money than men did in the “past work” condition, but not in the “future work” condition. Participants in this study were also given the opportunity to learn the average earnings of either men or women who participated in this study. In the “past work” condition, 72% of 7 Specifically, for the measure of payment deservingness, they were asked: “If you were an employer in charge of paying authors for their thought-listing contribution based on quality, how much (from $1 to $15) would you pay the author of the thoughts you listed?”
138 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 6. Effects of economic system justification on ingroup and outgroup favouritism among Northern and Southern Italians. Adapted from Jost et al. (2003a, Study 1). Lines are regression slopes for the relation between economic system justification scores (see items in Table 3) and ingroup/outgroup favouritism, aggregated across the following traits: efficient, responsible, productive, active, dominant, educated, ambitious, intelligent, emotional, honest, friendly, extraverted, religious, and happy. The interaction between group membership and economic system justification was significant, F(1, 154)=13.47, p<.001.
women chose to find out what other women had earned rather than what men had earned, but in the “future work” condition, this preference was reversed. When considering future outcomes, 72% of women preferred to find out what men earned, and only 28% wanted to know what women earned.8 It appears that these women realised that they could not change the past, so they accepted and rationalised it (e.g., Kay et al., 2002); they even rejected the opportunity to find out whether men had earned more than they had in the “past work” condition. With regard to undecided future outcomes, which are less in need of justification, most women chose to
8
Men, by contrast, overwhelmingly preferred the social comparison information about their own group, regardless of whether they were in the “past work” or “future work” condition.
PSYCHOLOGY OF SYSTEM JUSTIFICATION 139
Figure 7. Quality ratings and payment allocations made by men and women for their own work. Adapted from data reported by Jost (1997). Higher numbers indicate higher quality ratings of sophistication, originality, insightfulness, and higher dollar amounts rendered by men and women concerning their own written work. All ratings were made on 15-point scales (including a payment scale ranging from $1$15), and all gender comparisons differ at p<.10 or better. The gender difference on payment remained statistically significant even after controlling for gender differences on the evaluation dimensions, F (1, 129)=4.04, p<.05.
learn what men earn, possibly so they could minimise gender discrimination in the future. Ingroup ambivalence W.E.B.DuBois wrote in 1903 about African American experience that, “One ever feels his two-ness—an American, a Negro; two souls, two thoughts, two unreconciled strivings; two warring ideals in one dark body, whose dogged strength alone keeps it from being torn asunder” (DuBois, 1903/1995, p. 45). In an effort to understand conflicts between competing needs for group and system justification, Jost and Burgess (2000) hypothesised that low-status group members should exhibit stronger ingroup ambivalence than members of high-status groups (H11). It was also predicted that for members of psychologically meaningful groups (for whom at least moderate levels of group justification motives would be present), ambivalence towards the ingroup would be increased for
140 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
TABLE 7Correlations between perceived legitimacy and ambivalence as a function of groupstatus
Data are adapted from Jost and Burgess (2000, Study 1). Formulae for calculating the three models of ambivalence are described by Priester and Petty (1996). “SIM” refers to the “Similarity Intensity Model”, “CRM” refers to the “Conflicting Reaction Model”, and “GTM” refers to the “Gradual Threshold Model”. Z-score tests for the difference between correlations are two-tailed. *p• .05, **p•.005
members of low-status groups as system justification motives were increased (H12) and decreased for members of high-status groups as system justification motives were increased (H13). These hypotheses were addressed in two studies relayed by Jost and Burgess (2000). In the experiment involving University of Maryland students (described above), participants who were led to believe that their group occupied a relative position of either high or low socioeconomic success (see Table 2) rated their own group on a series of unipolar scales that contained both positively worded and negatively worded judgements (e.g., intelligent vs unintelligent, lazy vs hard-working, friendly vs unfriendly). Conceptually, a maximally ambivalent person would be one who reports (on one item) that the group is extremely intelligent and (on another) that the group is extremely unintelligent. For three measures of ambivalence reviewed by Priester and Petty (1996),9 Jost and Burgess found that ambivalence towards the (same) ingroup was significantly higher for people who were assigned to the lowstatus condition than for people assigned to the highstatus condition. Furthermore, perceived legitimacy was associated with increased ambivalence on the part of low-status groups and with decreased ambivalence on the part of high-status groups (see Table 7). Thus, (H11) and (H12) were both supported. In a second study, men and women read about a female plaintiff who was suing her university for gender discrimination and therefore posing a challenge to the overarching social system. Results indicated that women experienced a significant degree of emotional conflict and ambivalence,10 especially to the extent that they endorsed system-justifying beliefs. As can be seen in Table 8, emotional ambivalence scores were correlated positively with just world beliefs and social dominance orientation among women respondents, but they were negatively correlated with social dominance orientation among men and uncorrelated with just world beliefs, in general support of (H12) and (H13). As DuBois (1903) observed, ingroup
PSYCHOLOGY OF SYSTEM JUSTIFICATION 141
ambivalence seems to be one consequence of the psychological conflict that exists for members of low-status (but not high-status) groups between group and system justification motives (Glick & Fiske, 2001; Jost & Burgess, 2000). Self-esteem, depression, and neuroticism To the extent that members of disadvantaged groups are also faced with a conflict between ego and system justification needs, they should suffer in terms of psychological well-being (see Jost et al., 2001). Consistent with this analysis, Quinn and Crocker (1999) found that ideological endorsement of the Protestant work ethic was associated with a decrease in well-being among overweight women (as measured by self-esteem, depression, and anxiety), but it was associated with an increase in psychological well-being among normal weight women. Research by Jost and Thompson (2000) indicates that providing ideological support for inequality is associated with similar psychological disadvantages for members of low-status groups and advantages for members of high-status groups. A scale measuring general opposition to equality (see items in Table 1) was administered to European American and African American respondents, along with measures of well-being. Results from regression analyses are summarised in Table 9. The evidence to date generally supports hypotheses (H14)–(H16), which state that system justification has
9 Following Priester and Petty (1996), “dominant” (D) and “conflicting” (C) attitudinal components were identified for each pair of trait ratings (e.g., intelligent vs unintelligent). For example, if the ingroup received a “6” for the rating of “intelligent” and a “3” for the rating of “unintelligent”, then D=6 and C=3. The three formulae for calculating ambivalence (SIM, CRM, and GTM) are as follows: for the Similarity Intensity Model: Ambivalence (SIM)=3C–D; for the Conflicting Reactions Model: Ambivalence (CRM)=2C; for the Gradual Threshold Model: Ambivalence (GTM)=5(C+1).5–D1/C+1. An ambivalence score was calculated for each individual research participant and for each attribute pair (e.g., intelligent/ unintelligent) according to the above three formulae. Overall ambivalence towards the ingroup was then calculated by taking the means of the ambivalence scores across each of the six attribute dimensions. Thus, ambivalence was calculated in a within-participant fashion for each trait independently, before aggregating across traits and participants, so that ambivalence is a property of individual respondents rather than a property of the group as a whole. 10 Participants were asked to answer seven questions about positive and negative feelings towards the female plaintiff. An index of positive feelings towards the plaintiff ( =.94) was created by averaging responses to four items (respect, support, pride, and sympathy). An index of negative feelings ( =.89) was created by averaging three items (anger, unfairness, shame). The same three measures of ambivalence (SIM, CRM, GTM) were calculated on the basis of these indices of positive and negative feelings towards the plaintiff.
142 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
TABLE 8 Correlations between social dominance orientation/belief in a just world and ambivalence as a function of group status
Data are adapted from Jost and Burgess (2000, Study 2). Formulae for calculating the three models of ambivalence are described by Priester and Petty (1996). “SIM” refers to the “Similarity Intensity Model”, “CRM” refers to the “Conflicting Reaction Model”, and “GTM” refers to the “Gradual Threshold Model”. Z-score tests for the difference between correlations are two-tailed. *p•.05, **p•.005
opposite effects for members of high- and low-status groups on variables of self-esteem, depression, and neuroticism (see also Jost et al., 2002; Jost & Thompson, 2000). The reduction of ideological dissonance Jost et al. (2003b) presented evidence from five US national surveys indicating that members of disadvantaged groups show enhanced levels of system justification (H17). In Study 1, for instance, low-income respondents and African Americans were more likely than others to support limitations on the rights of citizens and media representatives to criticise the government. In Study 2, low-income Latinos were more likely to trust in government officials and to believe that “the government is run for the benefit of all” than were high-income Latinos. Results from Study 3, which are illustrated in Figure 8, indicated that low-income respondents were more likely than high-income respondents to believe that large differences in pay are necessary to “get people to work hard” and “as an incentive for individual effort”. In Study 4, African Americans living in the south (compared to African Americans living in the north) possessed lower income levels but endorsed meritocratic belief systems to a greater extent. And finally, in Study 5, low-income respondents and African Americans were more likely than others to believe that economic inequality is legitimate and necessary. In most cases, these effects retained significance even after controlling for education and other variables. Although future research is needed to measure the process of dissonance arousal directly, these results suggest that those who are most disadvantaged provide the
PSYCHOLOGY OF SYSTEM JUSTIFICATION 143
TABLE 9 Standardised regression coefficients for the relation between opposition to equality and self-esteem, depression, and neuroticism as a function of group status
Data are adapted from Jost and Thompson (2000, Studies 1, 3, and 4). Values are standardised regression coefficients ( s). All analyses include group-based dominance as a control variable.+p<.10, *p<.05, **p<.01 a Data are taken from Study 1. bData are taken from Study 3. cData are taken from Study 4.
strongest degree of ideological support for the system, at least under certain conditions.11 y Gender research summarised by Glick and Fiske (2001) similarl suggests that system justification may increase as objective circumstances worsen (H18). The ambivalent sexism inventory (which includes subscales d of “hostile” and “benevolent” sexism) was administered to men an e women in 19 different countries. Mean scores at the country level wer found to be negatively correlated with indices of gender development (women’s education, longevity, and standard of living relative to that of d men) and gender empowerment (women’s representation in business an government). Women were most likely to endorse sexism against women (especially “benevolent” forms of sexism) in those societies in which theyy were most disadvantaged, and in many cases, they were even more likel than men to endorse benevolent sexism. Glick and Fiske also found that n “when men in a nation more strongly endorsed sexist ideologies, wome followed suit, providing strong correlational evidence of system n justification” (p. 114). Indeed, average within-country correlations betwee sexism scores of men and women exceeded .80, suggesting that consensual ideologies exist to rationalise gender inequality around the world.
11 Hofstede (1997, p. 30) similarly found that people in low-status occupations and with lower educational levels scored higher in general on “power distance” than did people in high-status occupations and people with higher educational levels.
144 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 8. Agreement that large differences in income are necessary as a function of respondent income. Adapted from Jost et al. (2003b, Study 3). Values are combined percentages of respondents who believed that differences in pay were either “absolutely necessary” or “probably necessary” to “get people to work hard” and as “an incentive for individual effort”. For both items, regression analyses yielded negative linear effects of family income on the belief that income inequality is necessary (p<.01).
THE PALLIATIVE FUNCTION OF SYSTEMJUSTIFYING IDEOLOGY If system justification leads to outgroup favouritism, depressed entitlement, ingroup ambivalence, heightened neuroticism and depression, and lowered self-esteem on the part of low-status group members, why would they ever engage in system justification? In the remainder of this chapter, we explore the possibility that, despite these disadvantages, endorsing systemjustifying ideologies serves to make people feel better in other ways. Specifically, it has been suggested that ideology convinces people that the world is controllable, fair, and just (Lerner & Miller, 1978; Major, 1994; Olson & Hafer, 2001), and it allows people to be more satisfied with their own situation and with the system as a whole (Kluegel & Smith, 1986; Lane, 1962). As Havel (1991) puts it, ideology provides people “with the illusion that the system is in harmony with the human order and the order of the universe” (p. 134).
PSYCHOLOGY OF SYSTEM JUSTIFICATION 145
The power of meritocratic ideology One type of ideology (meritocracy) is particularly effective in placating people in democratic, free market, post-totalitarian systems. Meritocratic ideology refers to the conviction that ability and hard work lead to success and, conversely, that if people are not successful or if they fail in some ways, it is because they have not worked hard enough or they do not have the necessary abilities. Subscribing to a meritocratic ideology serves to increase the confidence and the esteem of those who are privileged and to ease their consciences (Chen & Tyler, 2001; Montada, Schmitt, & Dalbert, 1986). At the same time, such an ideological rationalisation may also convince those who are unsuccessful that they have (or at least had) a fair chance to succeed, which may make it easier for them to accept inequality (cf. Lane, 1962). Consistent with this possibility, Kluegel and Smith (1986, pp. 280–283) found that poor people reported more positive emotion, less guilt, and greater satisfaction when they felt responsible for their situation than when they made external (system-blame) attributions for their poverty. Research by Jost et al. (2003b, Study 5) further indicates that ideology is related to satisfaction in terms of one’s job, one’s financial situation, and life in general. We examined the effects of demographic status characteristics (socioeconomic status and race) and ideological beliefs concerning the legitimacy of economic inequality (e.g., “large differences in income are necessary for America’s prosperity”) and meritocracy (e.g., how important ambition, ability, and hard work are for “getting ahead in life”) on satisfaction. As mentioned above, African Americans and people who were lower in socioeconomic status were more likely than others to believe that socioeconomic differences were necessary and legitimate, although they were not more likely to endorse meritocracy in this study (see Table 10). The endorsement of meritocratic ideology was associated with greater economic satisfaction for all respondents. That is, the more people believed that hard work, ability, and motivation lead to success, the more they reported being satisfied with their own economic situation, regardless of whether they were rich or poor. Although this evidence is correlational, it suggests that meritocratic ideology serves a palliative function by making people feel better about their own situation, whatever that situation happens to be. Future research would do well to address this possibility more directly, especially with better measures of system-justifying ideologies and the use of experimental methods that afford causal inferences.
146 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
TABLE 10 Maximum likelihood estimates of regression coefficients for socioeconomic status, meritocratic ideology, perception of inequality, and economic satisfaction
Data are adapted from Jost et al. (2003b, Study 5). Entries are unstandardised regression coefficients arising from a model (n=788) in which variables of socioeconomic status and race were used to predict legitimation of inequality, endorsement of meritocratic ideology, and economic satisfaction. Gender was entered as a control variable; no significant gender effects were obtained. Race was dummy-coded as follows: Black=0, White=l. Model fit statistics: 2(44)=117.2, RMSEA=0.046, GFI=0.98, AGFI=0.96, IFI=0.94. *p<.05, **p<.005 (both onetailed)
A coping perspective on the psychology ofsystem justification Why would system-justifying ideologies make people feel better and more satisfied with their situation? We believe that a stress and coping perspective helps to answer the complex question of why people would support and justify a system that keeps them in a state of disadvantage (see also Miller & Kaiser, 2001; Schmader, Major, & Gramzow, 2001). Specifically, we argue that people engage in system justification (and other forms of rationalisation) in order to cope with and adapt to unjust or unpleasant realities that appear to be inevitable (see also Dalbert, 1997; Kay et al., 2002; Kieselbach, 1997). Although belief systems stand in a rather complicated relation to stress and coping in general (Lazarus & Folkman, 1984), we offer one attempt to describe this relationship in the specific case of system justification. We argue that in addition to causing certain kinds of stress for members of lowstatus or stigmatised groups (Jost & Thompson, 2000; Quinn & Crocker, 1999), system justification is related to stress and coping in at least three important ways. First, system justification, as a set of beliefs and assumptions about the existing social system, serves a stress-preventing function by allowing the individual to feel that the social context is stable,
PSYCHOLOGY OF SYSTEM JUSTIFICATION 147
understandable, predictable, consistent, meaningful, and just (see also Janoff-Bulman, 1992; Kluegel & Smith, 1986; Lane, 1962; Lerner, 1980; Montada et al., 1986). Second, system justification may be considered as a coping resource in that it not only reduces stress via the primary appraisal process by preventing the individual from perceiving certain types of stress, but also via the secondary appraisal process by fostering a sense of control and hope, which facilitates decisions about how to cope once stress is detected (see Lazarus & Folkman, 1984). Third, system justification can be viewed as a kind of coping activity or response to the many different stressors that are experienced by members of both low- and high-status groups as a consequence of their unequal positions in society. According to this analysis, system justification has the potential to be involved at any stage of the stress and coping process (Lazarus & Folkman, 1984), as a stressor, a perceptual factor in the primary appraisal process, a background coping resource affecting the secondary appraisal process, part of the coping activity itself, or an outcome of the coping process. It might seem paradoxical to suggest that system justification is both a cause of stress and an attempt to cope with stress, but this is not the case. We propose that in order to minimise or avoid certain kinds of stress, such as the stress that comes from perceiving that one is a victim of discrimination (e.g., Branscombe, Schmitt, & Harvey, 1999), people are willing to pay other psychological costs, such as those that follow from blaming themselves for their own misfortune (e.g., Janoff-Bulman, 1992; Miller & Porter, 1983), and these system-justifying responses may then lead to other kinds of stress (e.g., Jost & Thompson, 2000; Quinn & Crocker, 1999). By adopting a coping perspective, we do not necessarily assume that system justification provides an inherently adaptive coping strategy (see Jost, 1995), but it does seem to offer some measure of consolation to those who are disadvantaged as well as advantaged (see also Lane, 1962). The research summarised in this chapter has suggested some of the social psychological costs and benefits of engaging in system justification, but more work is needed to spell out these implications fully. CONCLUDING REMARKS What we have argued is that there is a socially acquired motive to justify and rationalise the existing social system. The operation of this motive has been demonstrated on measures of stereotyping, ideology, deservingness, desirability, and even memory. System justification occurs even on openended, nonreactive, qualitative measures and on implicit (as well as explicit) cognitive, affective, and behavioural measures. Paradoxically, the system justification motive is sometimes strongest among those who are the most disadvantaged, presumably because they have the most ideological dissonance to resolve.
148 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
System justification often has opposite social psychological effects on members of advantaged and disadvantaged groups. For members of highstatus or advantaged groups, system justification is generally associated with ingroup favouritism, increased self-esteem, and decreased ambivalence, depression, and neuroticism. For members of low-status groups, by contrast, system justification is generally associated with outgroup favouritism, ingroup ambivalence, decreased self-esteem, increased depression, and increased neuroticism. We have argued that, despite these potential costs, system-justifying ideologies serve a palliative function in that they make people feel better about their own situation. System justification may reduce dissonance and uncertainty especially (but not exclusively) among members of advantaged groups. From a coping perspective, there are many reasons why one might accept the potential costs that come from embracing system-justifying ideologies. These include the denial of discrimination, the perception of control, and the preservation of hope. We have argued that people engage in system justification in an attempt to cope with circumstances that they cannot change. It is an open question for future research as to whether this ideological resolution, which is no doubt adaptive in some respects, outweighs the personal, social, and political costs that are associated with it. REFERENCES Abelson, R.P., Aronson, E., McGuire, W.J., Newcomb, T.M., Rosenberg, M.J., & Tannenbaum, P.H. (Eds.) (1968). Theories of cognitive consistency: A sourcebook.Chicago, IL: Rand-McNally. Adorno, T.W., Frenkel-Brunswik, E., Levinson, D.J., & Sanford, R.N. (1950). Theauthoritarian personality.New York: Harper. Altemeyer, R.A. (1998). The other ‘authoritarian personality’. Advances in Experimental SocialPsychology, 30,47–91. Aronson, E. (1992). The return of the repressed: Dissonance theory makes a comeback. Psychological Inquiry, 3,303–311. Bem, S.L., & Bem, D.J. (1970). Case study of a nonconscious ideology: Training the woman to know her place. In D.J.Bem (Ed.), Beliefs, attitudes, and human affairs.Belmont, CA: Brooks/Cole Publishing Co. Blanton, H., George, G., & Crocker, J. (2001). Contexts of system justification and system evaluation: Exploring the social comparison strategies of the (not yet) contented female worker. Group Processes and Intergroup Relations, 4,127– 138. Branscombe, N.R., Schmitt, M.T., & Harvey, R.D. (1999). Perceiving pervasive discrimination among African Americans: Implications for group identification and well-being. Journal of Personality and Social Psychology,77, 135–149.
PSYCHOLOGY OF SYSTEM JUSTIFICATION 149
Branscombe, N.R., & Wann, D.L. (1994). Collective self-esteem consequences of outgroup derogation when a valued social identity is on trial. European Journal of Social Psychology,24, .641–657. Brown, R.J. (2000). Group processes: Dynamics within and between groups (2nd ed.). Oxford: Blackwell. Callahan-Levy, C.M., & Messé, L.A. (1979). Sex differences in the allocation of pay. Journal ofPersonality and Social Psychology, 37,433–446. Chen, E.S., & Tyler, T.R. (2001). Cloaking power: Legitimizing myths and the psychology of the advantaged. In A.Y.Lee-Chai & J Bargh (Eds.), The use and abuse of power: Multipleperspectives on the causes of corruption.Philadelphia: Psychology Press. Conway, M., Pizzamiglio, M.T., & Mount, L. (1996). Status, communality, and agency: Implications for stereotypes of gender and other groups. Journal of Personality and SocialPsychology, 71,25–38. Cooper, J., & Fazio, R.H. (1984). A new look at dissonance theory. In L.Berkowitz (Ed.), Advances in experimental social psychology(Vol. 17, pp. 229–266). New York: Academic Press. Dalbert, C. (1997). Coping with an unjust fate: The case of structural unemployment. SocialJustice Research, 10,175–189. DuBois, W.E.B. (1995). The souls of black folk: Essays and sketches.New York: Signet Classics. (Original work published 1903) Eagly, A., & Steffen, V. (1984). Gender stereotypes stem from the distribution of women and men into social roles. Journal of Personality and Social Psychology, 46,735–754. Elster, J. (1983). Sour grapes: Studies in the subversion of rationality.Cambridge, UK: Cambridge University Press. Fein, S., & Spencer, S.J. (1997). Prejudice as self-image maintenance: Affirming the self through negative evaluations of others. Journal of Personality and Social Psychology, 73,31–44. Festinger, L. (1957). A theory of cognitive dissonance.Evanston, IL: Row, Peterson. French, Jr, J.R.P., & Raven, B.H. (1959). The bases of social power. In D.Cartwright (Ed.), Studies in social power(pp. 150–167). Ann Arbor, MI: Institute for Social Research. Furnham, A. (1990). The Protestant work ethic: The psychology of work-related beliefs andbehaviours.London: Routledge. Furnham, A., & Procter, E. (1989). Beliefs in a just world: Review and critique of the individual difference literature. British Journal of Social Psychology, 28, 365–384. Glick, P., & Fiske, S.T. (2001). An ambivalent alliance: Hostile and benevolent sexism as complementary justifications for gender inequality. American Psychologist, 56,109–118. Goodwin, S.A., Operario, D., & Fiske, S.T. (1998). Situational power and interpersonal dominance facilitate bias and inequality. Journal of Social Issues, 54,677–698. Gramsci, A. (1971). Selections from the prison notebooks.New York: International Publishers.
150 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Grant, P.R., & Brown, R. (1995). From ethnocentrism to collective protest: Responses to relative deprivation and threats to social identity. Social Psychology Quarterly, 58,195–211. Greenwald, A.G., McGhee, D.E., & Schwartz, J.L.K. (1998). Measuring individual differences in implicit cognition: The implicit association test. Journal of Personality andSocial Psychology, 74,1464–1480. Greenwald, A.G., & Ronis, D.L. (1978). Twenty years of cognitive dissonance: Case study of the evolution of a theory. Psychological Review, 85,53–57. Haines, E.L., & Jost, J.T. (2000). Placating the powerless: Effects of legitimate and illegitimate explanation on affect, memory, and stereotyping. Social Justice Research, 13,219–236. Havel, V. (1991). The power of the powerless. In V.Havel (Ed.), Open letters(pp. 125–214). London: Faber & Faber. Hewstone, M. (1989). Causal attribution: From cognitive processes to collective beliefs.Cambridge, MA: Basil Blackwell. Hoffman, C., & Hurst, N. (1990). Gender stereotypes: Perception or rationalization? Journal ofPersonality and Social Psychology, 58,197–208. Hofstede, G. (1997). Cultures and organizations: Software of the mind.New York: McGraw-Hill. Hogg, M.A., & Abrams, D. (1988). Social identifications: A social psychology of intergrouprelations and group processes.London: Routledge. Jackman, M.R. (1994). The velvet glove: Paternalism and conflict in gender, class, and racerelations.Berkeley, CA: University of California Press. Jackman, M.R., & Senter, M.S. (1983). Different, therefore unequal: Beliefs about trait differences between groups of unequal status. Research in Social Stratification and Mobility,2,309–335. Janoff-Bulman, R. (1992). Shattered assumptions: Toward a new psychology of trauma.New York: Free Press. Jost, J.T. (1995). Negative illusions: Conceptual clarification and psychological evidence concerning false consciousness. Political Psychology, 16,397–424. Jost, J.T. (1997). An experimental replication of the depressed entitlement effect among women. Psychology of Women Quarterly, 21,387–393. Jost, J.T. (2001). Outgroup favoritism and the theory of system justification: An experimental paradigm for investigating the effects of socio-economic success on stereotype content. In G. Moskowitz (Ed.), Cognitive social psychology: The Princeton symposium on the legacy andfuture of social cognition(pp. 89– 102). Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Jost, J.T., & Banaji, M.R. (1994). The role of stereotyping in system-justification and the production of false consciousness. British Journal of Social Psychology, 33,1–27. Jost, J.T., & Burgess, D. (2000). Attitudinal ambivalence and the conflict between group and system justification motives in low status groups. Personality and Social Psychology Bulletin,26,293–305. Jost, J.T., Burgess, D., & Mosso, C. (2001). Conflicts of legitimation among self, group, and system: The integrative potential of system justification theory. In J.T.Jost & B.Major (Eds.), The psychology of legitimacy: Emerging perspectives on ideology, justice, andintergroup relations.New York: Cambridge University Press.
PSYCHOLOGY OF SYSTEM JUSTIFICATION 151
Jost, J.T., Overbeck, J., Guermandi, G., Rubini, M., Mosso, C, & Kivetz, Y. (2003a). System-justifying functions of complementary status stereotyping: Evidence from Italy, Israel,England, and the United States.Manuscript submitted for publication. Jost, J.T., Pelham, B.W., & Carvallo, M.R. (2002). Non-conscious forms of system justification: Cognitive, affective, and behavioural preferences for higher status groups. Journal of Experimental Social Psychology, 38,586–602. Jost, J.T., Pelham, B.W., Sheldon, O., & Sullivan, B. (2003b). Social inequality and the reduction of ideological dissonance on behalf of the system: Evidence of enhanced system justification among the disadvantaged. European Journal of Social Psychology, 32,1–24. Jost, J.T., & Thompson, E.P. (2000). Group-based dominance and opposition to equality as independent predictors of self-esteem, ethnocentrism, and social policy attitudes among African Americans and European Americans. Journal of Experimental Social Psychology,36,209–232. Kay, A., Jimenez, M.C, & Jost, J.T. (2002). Sour grapes, sweet lemons, and the anticipatory rationalization of the status quo. Personality and Social Psychology Bulletin, 28,1300–1312. Kieselbach, T. (1997). Unemployment, victimization, and perceived injustices: Future perspectives for coping with occupational transitions. Social Justice Research, 10,127–151. Kluegel, J.R., & Smith, E.R. (1986). Beliefs about inequality: Americans’ view of what is andwhat ought to be.Hawthorne, NJ: Aldine de Gruyter. Lane, R.E. (1962). Political ideology: Why the American common man believes what he does.New York: Free Press. Lazarus, R.S., & Folkman, S. (1984). Stress, appraisal and coping.New York, Springer Publishing Company. Lerner, M.J. (1980). The belief in a just world: A fundamental delusion.New York: Plenum Press. Lerner, M.J., & Miller, D.T. (1978). Just world research and the attribution process: Looking back and ahead. Psychological Bulletin, 85,1030–1051. Lukács, G. (1971). History and class consciousness.Cambridge, MA: MIT Press. MacKinnon, C.A. (1989). Toward a feminist theory of the state.Cambridge, MA: Harvard University Press. Major, B. (1994). From social inequality to personal entitlement: The role of social comparisons, legitimacy appraisals, and group memberships. Advances in ExperimentalSocial Psychology, 26,293–355. Major, B., McFarlin, D., & Gagnon, D. (1984). Overworked and underpaid. Journal ofPersonality and Social Psychology, 47,1399–1412. Major, B., & Schmader, T. (2001). Legitimacy and the construal of social disadvantage. In J.T. Jost & B.Major (Eds.), The psychology of legitimacy: Emerging perspectives on ideology,justice, and intergroup relations(pp. 176– 204). New York: Cambridge University Press. Marshall, R. (Ed.) (2000). Back to shared prosperity: The growing inequality of wealth andincome in America.London: ME Sharpe. Martin, J. (1986). The tolerance of injustice. In J.M.Olson, C.P.Herman, & M.P.Zanna (Eds.) Relative deprivation and social comparison/The Ontario symposium (Vol. 4). Hillsdale, NJ: Lawrence Erlbaum Associated Inc.
152 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Marx, K., & Engels, F. (1846/1978). The German ideology. In R.C.Tucker (Ed.), The Marx-Engels reader(pp. 146–200). New York: WW Norton. McGuire, W.J., & McGuire, C.V. (1991). The content, structure, and operation of thought systems. In R.S.Wyer Jr & T.K.Srull (Eds.), Advances in social cognition(Volume IV, pp. 1–78). Hillsdale, NJ: Lawrence Erlbaum Associated Inc. Miller, C.T., & Kaiser, C.R. (2001). A theoretical perspective on coping with stigma. Journal ofSocial Issues, 57,73–92. Miller, D.T., & Porter, C.A. (1983). Self-blame in victims of violence. Journal of Social Issues,39,139–152. Montada, L., Schmitt, M., & Dalbert, C. (1986). Thinking about justice and dealing with one’s own privileges: A study of existential guilt. In H.W.Bierhoff, R.L.Cohen, & J.Greenberg (Eds.), Justice in social relations(pp. 125–143). New York: Plenum Press. Mullen, B., Brown, R., & Smith, C. (1992). Ingroup bias as a function of salience, relevance, and status: An integration. European Journal of Social Psychology, 22,103–122. Nosek, B.A., Banaji, M.R., & Greenwald, A.G. (2002). Harvesting implicit group attitudes and beliefs from a demonstration website. Group Dynamics, 6,101– 115. O’Higgins, M., Schmaus, G., & Stephenson, G. (1990). Income distribution and redistribution: A microdata analysis for seven countries. In T.M.Smeeding, M.O’Higgins, & L.Rainwater (Eds.), Poverty, inequality and income distribution in comparative perspective(pp. 20–56). New York: Harvester Wheatsheaf. Oakes, P.J., & Turner, J.C. (1980). Social categorization and intergroup behaviour: Does minimal intergroup discrimination make social identity more positive? European Journal ofSocial Psychology, 10,295–301. Olson, J., & Hafer, C.L. (2001). Tolerance of personal deprivation. In J.T.Jost & B.Major (Eds.), The psychology of legitimacy: Emerging perspectives on ideology, justice, andintergroup relations(pp. 157–175). New York: Cambridge University Press. Pelham, B.W., & Hetts, J.J. (2001). Underworked and overpaid: Elevated entitlement in men’s self-pay. Journal of Experimental Social Psychology, 37, 93–103. Pepitone, A. (1950). Motivational effects in social perception. Human Relations, 3, 57–76. Pratto, F., Sidanius, J., Stallworth, L.M., & Malle, B.F. (1994). Social dominance orientation: A personality variable predicting social and political attitudes. Journal of Personality andSocial Psychology, 67,741–763. Priester, J.R., & Petty, R.E. (1996). The gradual threshold model of ambivalence: Relating the positive and negative bases of attitudes to subjective ambivalence. Journal of Personality andSocial Psychology, 71,431–449. Quinn, D.M., & Crocker, J. (1999). When ideology hurts: Effects of belief in the Protestant Ethic and feeling overweight on the psychological well-being of women. Journal ofPersonality and Social Psychology, 77,402–414. Ridgeway, C. (2001). The emergence of status beliefs: From structural inequality to legitimizing ideology. In J.T.Jost & B.Major (Eds.), The psychology of
PSYCHOLOGY OF SYSTEM JUSTIFICATION 153
legitimacy: Emergingperspectives on ideology, justice, and intergroup relations (pp. 257–277). New York: Cambridge University Press. Rubin, Z., & Peplau, L.A. (1975). Who believes in a just world?Journal of Social Issues, 31,65– 89. Schmader, T., Major, B., & Gramzow, R.H. (2001). Coping with ethnic stereotypes in the academic domain: Perceived injustice and psychological disengagement. Journal of SocialIssues, 57,93–111. Scott, J. (1990). Domination and the arts of resistance.New Haven, CT: Yale University Press. Sidanius, J., & Pratto, F. (1999). Social dominance: An intergroup theory of social hierarchy andoppression.New York: Cambridge University Press. Spears, R., Jetten, J., & Doosje, B. (2001). The (il) legitimacy of ingroup bias: From social reality to social resistance. In J.T.Jost & B.Major (Eds.), The psychology of legitimacy: Emergingperspectives on ideology, justice, and intergroup relations.New York: Cambridge University Press. Steele, C.M., & Liu, T.J. (1983). Dissonance processes as self-affirmation. Journal ofPersonality and Social Psychology, 41,831–846. Tajfel, H. (1981). Human groups and social categories.Cambridge: Cambridge University Press. Tajfel, H., & Turner, J.C. (1986). The social identity theory of intergroup behaviour. In S. Worchel & W.G.Austin (Eds.), The psychology of intergroup relations(pp. 7–24). Chicago: Nelson-Hall. Tiraboschi, M., & Maass, A. (1998). Reactions to perceived deprivation in ingroup and outgroup: A cross-cultural comparison. European Journal of Social Psychology, 28,403–421. Turner, J.C., & Brown, R. (1978). Social status, cognitive alternatives, and intergroup relations. In H.Tajfel (Ed.), Differentiation between social groups (pp. 201–234). London: Academic Press. Turner, J.C, Hogg, M.A., Oakes, P.J., Reicher, S., & Wetherell, M.S. (1987). Rediscoveringthe social group: A self-categorization theory.Oxford: Basil Blackwell. Tyler, T.R., & McGraw, K.M. (1986). Ideology and the interpretation of personal experience: Procedural justice and political quiescence. Journal of Social Issues, 42,115–128. Weiss, R. (1969/1988). The American myth of success: From Horatio Alger to Norman VincentPeale.Urbana: University of Illinois Press. Wicklund, R.A., & Brehm, J.W. (1976). Perspectives on cognitive dissonance.Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Wolff, E.N. (1996). Top heavy: The increasing inequality of wealth in America and what can bedone about it.New York: The New Press. Wright, S.C., Taylor, D.M., & Moghaddam, F.M. (1990). Responding to membership in a disadvantaged group: From acceptance to collective protest. Journal of Personality andSocial Psychology, 58,994–1003.
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY, 2002, 13, 155–184
Leader endorsement in social dilemmas: Comparingthe instrumental and relational perspectives Mark Van Vugt University of Southampton, UK David De Cremer University of Maastricht, The Netherlands
This chapter examines the role of leadership in overcoming social dilemmas within groups. First, based on prior theorising and research we present two alternative perspectives on leader endorsement in social dilemmas, an instrumental and a relational perspective. Next, we systematically compare these perspectives in a series of experiments investigating leadership in social dilemmas created within small groups in the laboratory. The results of our studies suggest that when their personal identity is salient, group members more strongly endorse leaders who are perceived to be instrumental in solving the freerider problem. In contrast, when a social identity is salient, members more strongly endorse leaders who fulfil their relational needs. Based on these findings we propose a differential needs model of leader endorsement in social dilemmas. “Leadership is getting someone to do what they don’t want to do, to achieve something that they want to achieve” —Tom Landry (legendary American football coach) Leadership is commonly defined as a process of influence to attain important group, organisational, and societal goals (Bass, 1990; Chemers, 2001; Haslam, 2001; Hollander, 1985; Yukl, 1989). In order to achieve these goals leaders must ensure that conflicts, which frequently emerge between the self-interest of individual members and the collective interest of the group, are resolved. Such conflicts are better known as © 2002 European Association of Experimental Social Psychology http://www.tandf.co.uk/journals/pp/02699931.html DOI: 10.1080/10463280240000055
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 155
socialdilemmas (Dawes, 1980, Komorita & Parks, 1995, Wilke, 1991). Good leadership is necessary in order to manage these dilemmas, which occur at every level of society. In work teams, for example, an important role of the manager is to ensure that all team members contribute towards the completion of a group task. In sports, a team manager must be able to motivate players to put in their best performance to beat other teams. On a larger scale, political leaders and authorities must secure the cooperation of citizens for the maintenance of important public goods, such as schools and hospitals, and natural resources, such as land and water (Van Vugt, Snyder, Tyler, & Biel, 2000). Without some form of leadership many groups and organisations would not be able to deal adequately with social dilemmas, because in the absence of leaders and authorities freeriding in groups would be too widespread (Olson, 1965, Yamagishi, 1986). Consistent with this view, B.M.Bass, a leadership theorist, recently stated that successful leadership involves “the moving of followers beyond their self-interests for the good of the group, organisation, or society” (Bass, 1997, p. 130). Appointing a leader seems a sensible solution to deal with social dilemmas in groups, but it is not a straightforward solution. Leadership is a complex political process involving continuous negotiations between group members about what kind of leaders they desire. For example, groups must decide where the leader should come from (e.g., from inside or outside the group), how they should be assigned to the group (e.g., by election or appointment), what power base they should have (e.g., reward, coercive or legitimate power), what should be their leadership style (e.g., task or relation-oriented) and personal attributes (e.g., a highly skilled versus highly committed leader), and, finally, how they can be replaced if necessary (Bass, 1990; French & Raven, 1959; Hollander, 1985; Levine & Moreland, 1998; Yukl, 1989). Furthermore, once leaders are in place, group members must decide whether to cooperate with their directives (Lippitt & White, 1968; Tyler & Degoey, 1995; Van Vugt & De Cremer, 1999). In this dynamic process, Address correspondence to: Mark Van Vugt, Department of Psychology, University of Southampton, Highfield, Southampton, SO17 IBJ, UK, Email:
[email protected] We are indebted to Margaret Foddy, Henk Wilke, and the reviewers for their helpful comments on previous versions of this chapter. The experiments reported in this paper were part of the second author’s PhD thesis, which received the 2000 BPS Social Psychology Section Outstanding PhD award. Portions of this chapter were written while the first author was on sabbatical leave at the Free University of Amsterdam (with support from the Netherlands Science Foundation (NWO; B 57– 205).
156 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
group members will presumably not be focused solely on the perceived instrumentality of leadership in resolving the social dilemma at hand. They might also consider the consequences of having a leader for the social climate in the group and the quality of interpersonal relationships between group members (cf. goal achievement vs group maintenance; Cartwright & Zander, 1968). It is important to distinguish between two kinds of solutions to social dilemmas—individual and structural solutions (Messick & Brewer, 1983; Van Vugt et al., 2000). Individual solutions involve group members’ voluntary efforts to produce valuable goods for their group. The vast majority of social dilemma research has been devoted to studying socialpsychological determinants of voluntary cooperation, such as communication, trust, and the development of prosocial norms (for overviews, see Komorita & Parks, 1994; Liebrand, Messick, & Wilke, 1992; Schroeder, 1995). Over the past two decades, however, researchers have become interested in the emergence of structural solutions to social dilemmas, the main question being when groups opt for a change in the structural features of groups in order to resolve the social dilemma. Structural solutions that have been empirically investigated include the introduction of contribution and distribution rules (Sato, 1987; Van de Kragt, Orbell, & Dawes, 1983), reward and punishment systems (Komorita, Parks, & Hulbert, 1992; Yamagishi, 1986), systems of exclusion (Kerr, 1999), formal authorities (Tyler & Degoey, 1995), and leadership (Foddy & Crettenden, 1994; Foddy & Hogg, 1999; Messick, Wilke, Brewer, Kramer, Zemke & Lui, 1983; Rutte & Wilke, 1984; Samuelson, Messick, Rutte, & Wilke, 1984; Van Vugt & De Cremer, 1999; Wilke, 1991; Wit, Wilke, & Van Dijk, 1989). In this chapter, we concentrate on the role of leadership because it is potentially the most viable solution to social dilemmas within small groups (Levine & Moreland, 1998; Messick & Brewer, 1983; Van Vugt & De Cremer, 1999). We investigate two interrelated aspects of leadership in social dilemmas, leadership emergence and leadership influence. First, we examine why group members want to voluntarily assign a leader to the group and what type of leader they choose. Second, we study the influence of different leader types on the voluntary cooperation of members. Together we refer to the voluntary acceptance and cooperation with leaders as leader endorsement. Our main research aim is to demonstrate that the endorsement of leadership in social dilemmas is a function of the prevailing needs within the group. If members are concerned primarily about their shortterm personal welfare, they will endorse leaders who are believed to be instrumental in solving the social dilemma task. In contrast, if members assign greater priority to the long-term group welfare, they will endorse leaders whose primary goal is to strengthen the relationships between group members. In
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 157
this regard, our research is inspired by the traditional leadership literature, which draws a distinction between the instrumental and relational roles of leadership in groups (Bass, 1990; Cartwright & Zander, 1968; Hemphill, 1961; Yukl, 1989). Extending this work we argue that the importance of these two leadership functions in social dilemmas is contingent upon the prevailing needs of the group. LEADERSHIP AS A SOLUTION TOSOCIAL DILEMMAS Social dilemma is the generic term for two special classes of social conflicts, the resource dilemma and the public good dilemma (Komorita & Parks, 1994; Messick & Brewer, 1983). A resource dilemma involves a potential conflict between a group of people over the distribution of a finite resource. Many natural resources have this property (Van Vugt et al., 2000). A public good dilemma entails a potential conflict between group members over the contributions necessary to create a commonly shared good, for example, a group facility or a successful team performance (Stroebe & Frey, 1982). Because public good dilemmas are relatively more common in small groups, the present research focuses on these types of conflicts (although our conclusions may speak to both dilemma types). At the heart of the public good dilemma lies the freerider problem (Kerr, 1983; Olson, 1965). In creating public goods, group members must decide whether to cooperate by making a contribution or to freeride on the contributions of others. Freeriding is personally more attractive, but if it is too widespread the group may fail to secure the good, which leaves every group member worse off. Cooperation is thus the most sensible strategy from a collective viewpoint. Yet members may be reluctant to cooperate because (a) it is tempting to freeride, and (b) even if they cooperate there is a risk that they are being exploited by other group members (“the sucker’s pay-off”; Komorita & Parks, 1994). Hence, both motives of greed and fear can explain the emergence of freeriding in social dilemmas (Kerr, 1983). One of the main tasks of a group leader in a social dilemma is to prevent freeriding. In a fully cooperative group task each member will be inclined to cooperate spontaneously, and therefore the demands on leadership are relatively straightforward. The main leadership function is to coordinate the often diverse efforts of group members and bring them together. In a fully competitive group task (“zero-sum game”) the role of leaders is also fairly clear. They must primarily serve as an arbitrator to mediate between the individuals (or groups) with opposing interests. Social dilemmas, however, are mixed-motive conflicts in the sense that for individuals there are incentives both to freeride—to enhance their personal welfare—as well as to cooperate for the group. Here the role of leadership becomes more complicated. In order to achieve the group goals, leaders must deter and
158 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
punish freeriding. At the same time, however, they should contribute to a positive group climate to ensure that members, particularly those with a cooperative inclination, enjoy being in this group and are tempted to stay and contribute to the group’s welfare (Van Vugt, Jepson, & De Cremer, 2001). Social dilemmas thus provide an ideal laboratory to test hypotheses about the different roles and functions of leadership in groups. Instrumental perspective on leadership insocial dilemmas Rational decision-making theories, such as game theory, rational choice theory, and social exchange theory (Hardin, 1968; Luce & Raiffa, 1957; Olson, 1965; Thibaut & Kelley, 1959) postulate that individuals in social dilemmas are primarily concerned about their short-term self-interest. According to the notion of self-interest, group members endorse leaders when they are instrumental in providing favourable outcomes. Hence, they will cooperate more with leaders who provide material rewards for cooperation and punishments for freeriding. Note that a leader who achieves group success will also be endorsed, according to this perspective, because all members benefit materially if the leader successfully resolves a social dilemma. Group members will therefore primarily look for information about leader characteristics which suggests that they are capable of solving the freerider problem by modifying members’ selfish behaviours. This instrumental perspective on leadership in social dilemmas has received considerable empirical support (Messick et al., 1983; Rutte & Wilke, 1984; Samuelson & Messick, 1986; Samuelson et al., 1984). For example, Messick et al. (1983) showed that group members were more likely to choose a leader if their group had previously failed to collectively sustain a common resource pool. Furthermore in applied research on a water shortage in California (Tyler & Degoey, 1995) it was found that residents who perceived the shortage as more threatening were more likely to defer control over their personal water use to the water authorities. Finally, in a public good experiment, Yamagishi (1986) found that the introduction of a sanctioning regime to punish freeriders increased the average contribution level in groups. However, these findings do not tell the whole story about the role of leadership in social dilemmas. First, if group members are solely concerned about material outcomes, they would not hesitate to give up complete decisional freedom to a leader so as to eliminate the freerider problem. In fact, even in a collective crisis individuals are quite reluctant to vote for an autocratic leadership regime (Rutte & Wilke, 1984, 1985; Samuelson, 1993; Tyler & Degoey, 1995). When given the choice between various structural solutions, group members tend to prefer democratic solutions, such as majority and unanimity rules, above autocratic leadership (Rutte &
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 159
Wilke, 1985)—perhaps due to a feared loss of personal control or concerns about the possibility of corruption and exploitation by the leader. Second, there are no straightforward effects of the use of reward and punishment schemes by leaders in social dilemmas. For example, in one study it was found that a weak sanctioning regime resulted in a lower level of cooperation than when there was no sanctioning at all (Tenbrunsel & Messick, 1999). Furthermore, Yamagishi (1986) showed that a sanctioning regime had more effect on members with a low trust in other people’s voluntary cooperation than on members with high trust in others. Finally, in a recent field study on water conservation we found that when water authorities sanctioned excessive water use—by installing water meters in properties—this intervention had more effect on residents who identified weakly with their residential community than residents with a strong community identification (Van Vugt, 2001). These results suggest a more group-based view on the role of leadership in social dilemmas. The emergence of leadership and the subsequent influence of leaders may be influenced by the prevailing motives and needs within a particular group. Sometimes group members are more focused on the material outcomes that leaders could potentially provide, judging their ability to solve the problem of freeriding. At other times, however, they may look for their leaders to satisfy other important, non-instrumental needs. A relational perspective on leadership insocial dilemmas An alternative motive for leader endorsement in social dilemmas is the extent to which the leader is capable of fulfilling the relational needs of group members. People join groups for a multitude of different reasons, only some of which are instrumental (e.g., the achievement of some specific group goal). An important alternative motive for group membership is that it allows individuals to fulfil a desire to establish positive social relationships with other people. Group membership gives people a sense of identity and belonging, which is regarded by many theorists as essential for the survival and psychological well-being of an individual (Baumeister & Leary, 1995; Brewer, 1979; Deci & Ryan, 2000; Tajfel & Turner, 1979; Turner, Hogg, Oakes, Reicher, & Wetherell, 1987). For example, social identity theory (Tajfel & Turner, 1979) assumes that people’s sense of who they are, their identity, is shaped in part by the social groups with which they are associated. This aspect of people’s self-concept is described as one’s social identity (Tajfel, 1972, p. 273): “that part of an individual’s self-concept which derives from his [or her] knowledge of his [or her] membership of a social group (or groups) together with the value and emotional significance attached to that membership”. Social identity theory argues that people seek a positive social identity, which is achieved
160 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
through a positive distinctiveness from relevant other groups. Whereas social identity theory was originally formulated as a theory to describe the intergroup dynamics underlying social identity processes, more recent adaptations of this theory have focused more on the intragroup dynamics of social identity (Hogg & Abrams, 1988; Turner et al., 1987). Recently, for example, social identity researchers have started to investigate the emergence of leadership in groups (Hains, Hogg, & Duck, 1997; Hogg, 2001; Turner & Haslam, 2000). These insights are useful for understanding the different roles of leadership in social dilemmas. Within social dilemmas, the emergence and effectiveness of leadership may be dictated, in addition to their perceived instrumentality, by the perceived influence on the relational needs of members (i.e., identity and belongingness needs). Leaders can fulfil these needs by developing pleasant social relationships with group members and by creating an encouraging social climate so that members enjoy their group membership and wish to stay. Leaders who facilitate these goals will be perceived as legitimate (French & Raven, 1959). Hence, members will cooperate voluntarily with them in solving social dilemmas without the promise of specific material rewards or the threat of punishments (Lind & Tyler, 1988; Tyler, 2000; Tyler & Dawes, 1993; Tyler & Lind, 1992). A differential needs model of leadership What specific factors determine whether instrumental or relational needs are more salient in the endorsement of leaders in social dilemmas? Following an instrumental perspective on leadership in social dilemmas, leaders’ primary role is outcome-directed. Leaders must help to increase the material pay-offs for the group and its members and, to ensure this, they must demonstrate the ability to detect and punish freeriding. In contrast, a relational perspective asserts that to secure acceptance and cooperation from group members, leaders must show a concern about the social relationships within the group. This differential needs model thus suggests that impressions about the role of leaders may differ between members, as well as between groups, depending on their dominant needs and motives. A key difference between group members presumably lies in the extent to which they consider themselves to be part of the group. Either they perceive themselves essentially as a unique individual, in which case their personal identity is more salient, or as member of a group, in which case their social identity is more salient (Turner et al., 1987). When their social identity is salient, group members tend to believe that they have very much in common with other group members, both in terms of opinions, values, shared goals, and interests (Spears, Oakes, Ellemers, & Haslam, 2000). The salience of a social identity thus blurs the distinction
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 161
between an individual’s self-interest and the group’s interest—they are perceived as overlapping—which effectively solves the social dilemma conflict (Brewer, 1979; De Cremer & Van Vugt, 1999). Furthermore, a salient social identity enhances depersonalised trust in other group members’ cooperative intentions (Kramer & Brewer, 1984). Both mechanisms, a cooperative goal and depersonalised trust, may explain why group members exhibit greater voluntary cooperation in social dilemmas when a social identity is activated (Brewer, 1979; Brewer & Kramer, 1986; De Cremer & Van Vugt, 1999; Kramer & Brewer, 1984). Conversely, when a personal identity is salient, individuals perceive little communality between their attitudes and interests and those of the other group members. Hence, they are less likely to cooperate spontaneously and they expect no cooperation from others either. The moderating role of identity The attitudinal and behavioural differences between social and personal identifiers yield important implications for the perceived role of leadership in social dilemmas. First, lacking in trust, personal identifiers probably see a greater urgency to voluntarily accept a leader as a solution to the social dilemma. Moreover, they will be looking for evidence in leaders which shows that they are capable of overcoming the freerider problem in their group. Finally, the primacy of self-interest over the group interest dictates that the behaviour of personal identifiers is shaped by the expected material rewards and punishments received from the leader. Accordingly, personal identifiers will be more accepting of and influenced by instrumental leadership in overcoming a social dilemma. What about social identifiers? Because the conflict between their selfinterest and the collective interest is absent, or at least less intense, they are more optimistic that the social dilemma can be resolved through the voluntary contributions of themselves and others in the group. Hence, they presumably perceive less need to move from an unstructured group setting to a situation with a group leader. Yet, when a leader is already there they will assign less weight to the specific instrumental qualities that the leader brings to the group. After all, they have an intrinsic motivation to cooperate with the leader in securing the collective good. What social identifiers presumably care about more is the possible impact of leadership on the psychological experience of their group membership, in other words, their social identity. Both intragroup and intergroup aspects of leadership may be important in fostering a positive social identity, thus fulfilling the needs of social identifiers. Within the group the leader must be seen to contribute to a pleasant social climate in order to be influential. Consistent with the groupvalue model (Tyler & Lind, 1992) group members base their
162 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
judgements about a leader on the quality of their interactions with them. The perceived quality of these interactions is influenced primarily by the behavioural style of a leader. If a leader treats members respectfully, for example, by being fair and by keeping promises, this will positively influence the psychological experience of group membership, hence foster an individual’s social identity (De Cremer & Van Knippenberg, 2002). Furthermore, a leader who provides encouragement and support as opposed to threats and punishments is also likely to strengthen an individual’s social identity. Intergroup qualities of leadership may also determine whether a leader is able to fulfil the relational needs of group members. Yet, whereas impressions about the intragroup leader qualities are primarily shaped by the behavioural style of a leader, impressions about intergroup qualities are probably derived primarily from the personal attributes of a leader, both cognitive and motivational attributes. Consistent with a social identity model of leadership (Hains et al., 1997; Hogg, 2001), leaders have been found to be more influential to the extent that they represent the group prototype, which comprises “fuzzy sets of attributes that define and prescribe attitudes, feelings, and behaviours that characterise one group and distinguish it from other groups” (Hogg, 2001, p. 187). Prototypes are heavily influenced by the specific comparative intergroup context. In addition to the social-cognitive leader attributes, the intergroup context may also influence members’ expectations about the motivational attributes of leaders. In particular, members would expect their leader to express a strong desire to belong to this group rather than to some other group in order to maximise the distinctiveness from other groups. Leaders can achieve this by showing to members that they care about the group, like its members, and are committed to stay in the group, particularly when things are going wrong. The perceived strength of leader’s groupcommitment is presumably an important determinant of leader endorsement when members’ social identity is salient, perhaps more so than the stereotypical leadership skills that a leader brings to the group (e.g., communication and coordination skills). Prediction We can now formulate our main research hypothesis. We have offered two alternative needs perspectives on the role of leadership in social dilemmas. According to an instrumental model the primary role of a leader in social dilemmas is to solve the freerider problem. In contrast, the relational model asserts that a leader’s primary function is to strengthen the social relationships between group members. These models are not mutually exclusive. With many other leadership theories (Bass, 1990; Cartwright & Zander, 1968; Hollander, 1985; Yukl, 1989), we believe that both
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 163
instrumental and relational functions are probably important in managing social dilemmas within groups. Yet the relative weight of these leadership functions may vary with the dominant needs of group members, which are contingent upon their self-definition. Members who perceive themselves and others primarily as unique individuals—personal identifiers—will be more focused on the instrumental qualities of leaders, evaluating them in terms of their effectiveness in modifying individuals’ self-interested behaviours. Conversely, when individuals see themselves as group members—social identifiers—they will concentrate more on the relational qualities of a leader: Is the leader able to maintain pleasant social relationships with group members and promote a positive social identity? These differential needs, activated by the salience of one’s identity, will influence various stages of leader endorsement in social dilemmas (a) the voluntary acceptance of leadership as a solution to a social dilemma, (b) the negotiation process within groups about the preferred type of leader, and finally (c) the cooperation with the leader’s directives. RESEARCH PARADIGM AND PROCEDURE To systematically investigate the claims made by the differential needs model we conducted a series of laboratory experiments in which we used a step-level public goods task (Van de Kragt et al., 1983) to stimulate a small group social dilemma. A step-level public good is an investment task in which group members are endowed with a monetary sum, which they can either keep to themselves or invest in a collective good (Komorita & Parks, 1994). This good, usually an extra monetary bonus, becomes available only if a sufficient number of people contribute a sufficient amount towards the provision of the good.1 Yet, and herein lies the crux of the dilemma, when the good is provided it becomes available to all group members regardless of whether they contributed or not. This is known as the non-excludability criterion of public goods (Olson, 1965). Consequently, it is in each member’s immediate self-interest to contribute nothing in the hope that others will contribute. This is essentially the freerider problem that is inherent to public good dilemmas. The participants in our experimental studies, all undergraduate students, conducted these public good tasks in small groups consisting of six members each. We used only ad-hoc groups. Group members were informed that they were linked with each other via computers, but, in reality, all the computer messages had been pre-programmed. Before the public good task started, members received information about their particular group as well as the purpose of the study. In half of the conditions, they were told that the study intended to compare group decision making within groups from several different
164 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
universities, all in southern England, and that the results of their group would be compared with those of other universities participating in the research. This was the social identity condition, because it highlighted group membership by drawing an intergroup comparison (i.e., social competition; Turner, 1975). Others were told that the aim of the study was to investigate individual decision making within groups, and no reference was made to a between-university comparison. This was the personal identity condition, because it de-emphasised group membership by highlighting an interpersonal comparison between members. The same identity manipulation was used in all the studies we report here (for a similar procedure, see Kramer & Brewer, 1984, Exps. 1 and 2). Within this experimental context, we investigated the moderating role of social identity on the various stages of leader endorsement in social dilemmas. In the first part of our research we concentrated on the emergence of leadership in public good dilemmas. In the second part we studied the actual influence of leaders on the contribution decisions of group members. EMERGENCE OF LEADERSHIP IN GROUPS When does leadership emerge in public good dilemmas? When are group members willing to give up their personal freedom to a leader, and what kind of leader do they prefer? Preference for a group leader We predicted that the voluntary acceptance of a group leader to solve the dilemma would be determined primarily by instrumental needs. Members would be more likely to defer to a group leader if they thought their group was unable to provide the good through voluntary contributions (De Cremer, 2000; Messick et al., 1983; Samuelson et al., 1984). Like any structural solution, the introduction of leadership in groups entails costs, both financial and psychological, associated with the appointment of a leader. Such transition costs (Samuelson & Messick, 1995; Van Vugt, 1997) tend to lead to a preference for the status quo unless the status quo is clearly undesirable. Accordingly, we predicted that there would be a stronger preference to accept a group leader if the group had failed previously to secure the good. 1
There are two types of public goods, the step-level and the continuous public good (Komorita & Parks, 1994). Unlike in a step-level good, in a continuous public good the size of the good is variable and depends on the total amount of contributions received from group members. In this research we studied both types of public good tasks.
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 165
Furthermore, the preference for a leader would be stronger when a personal identity rather than a social identity was activated in group members. Compared to social identifiers, personal identifiers would perceive a greater need for a structural change. These hypotheses were tested in a first experiment in which individuals, in groups of six, were asked to complete a step-level public good task, introduced as a group investment game (Van Vugt & De Cremer, 1999, Exp. 1). At the start of the game, each group member would receive an endowment of £3. They were told that the group would receive a £30 bonus (£5 per member), provided that a sufficient number of members made an investment (steplevel good; Van De Kragt et al., 1983).2 We then activated members’ identity by stressing either an intergroup comparison (social identity) or interpersonal comparison (personal identity). The post-experimental questionnaire results revealed that this manipulation was successful. Participants in the social identity condition identified more strongly with their group than participants in the personal identity condition. Next, individuals were asked if they wanted to contribute their £3 to the group and were subsequently provided with false feedback about the group’s success (or failure) to secure the good. Afterwards, in preparation for the second round of contribution sessions, they were asked if they wanted to appoint a group leader. Consistent with predictions, group members exhibited a stronger leadership preference if the group previously failed to secure the good. Furthermore, regardless of outcome feedback, personal identifiers displayed a stronger preference for appointing a group leader than did social identifiers, which was consistent with their contribution pattern. Fewer people contributed in the personal identity condition (70%) than in the social identity condition (88%). Moreover, personal identifiers considered their fellow group members to be less trustworthy, less cooperative, and less fair than did social identifiers. What kind of leader? We also investigated whether personal and social identifiers would differ in the type of leader they preferred for their group. The differential needs model would predict that personal identifiers more strongly preferred a leader instrumental in solving the dilemma, whereas for social identifiers the relational qualities of a leader would dominate. Accordingly, after expressing whether they wanted a group leader or not, participants were told that, regardless of preferences, each group would have a leader in place for the second contribution round. Each member would have an input in the kind of leader that was to be assigned to their group. Accordingly, they were asked to give preference ratings to six different leadership prototypes, which represented a wide range of leadership
166 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
options in work organisations (Bass, 1990; Yukl, 1989). The leader type that received the highest overall group rating would be assigned to the group. Based on the leadership literature, these prototypes were then divided into three pairs of more or less contrasting leader types, which were accompanied by a short description: (i) Democratic versus autocratic leader. The democratic group leader was described as “a leader who will ask each member of your group informally about their intended contribution decision, and then make a decision about which group members should contribute their endowment”. In contrast, the autocratic leader was portrayed as “a leader who decides for the group which group members should contribute their endowment, without consulting the group”. (ii) Elected versus appointed leader. The person to serve as the group leader was someone who would either be “chosen by the majority of the group members via a vote” (elected) or “appointed by the experimenter” (appointed). (iii) Internal versus external leader. The internal leader was described as “a person from Southampton University” and the external as “a person from one of the other universities participating in this experiment”. As can be seen in Table 1, group members preferred leaders who, by virtue of either a personal attribute (from inside the group), their assignment to the group (elected), and source of power (democratic) were more strongly embedded within the group. Indeed these three leader types were perceived as more legitimate than external, appointed, or autocratic leader types. Yet there were some interesting differences in preference as a result of the salience of one’s identity. Social identifiers displayed a stronger preference than personal identifiers for an internal leader and an elected leader, whereas personal identifiers displayed a stronger preference for an external and appointed leader. A possible interpretation of these findings is that elected leaders and leaders from inside the group would facilitate the social interactions between the leader and group, thus fostering a positive social climate. Finally, social and personal identifiers displayed an equally strong dislike for an autocratic leader. This corroborates the findings of other social dilemma research by showing that members have a desire to keep
2 We also manipulated task difficulty by telling half of the group members that their group would need at least five contributors to reach the bonus (difficult task), whereas in the other half of the conditions the group needed just two contributors (easy task). Because there were no predictions regarding the impact of this variable on leader endorsement and it did not influence any of the results we will not discuss it here.
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 167
TABLE 1 Preferences for different leader types in social dilemmas as a function of member’s identity
Ratings were made on a 7-point scale (1=no preference for leader; 7=very strong preference). Means with a different subscript differ significantly at p<.01 in a rowwise comparison. From Van Vugt and De Cremer (1999, Exp. 1), with permission ©1999 American Psychology Association.
some control over their contribution decisions (Rutte & Wilke, 1985; Samuelson, 1993). Conclusions We can conclude from this first study that both instrumental and relational motives play a role in the emergence of leadership in social dilemmas. Instrumental rather than relational concerns determine whether group members want to move from a leaderless group to a group with a leader. In determining what kind of group leaders members wish to appoint, however, other non-instrumental motives become important. Overall, there is a preference for leader types that, due to either a personal attribute (from inside the group), the way they have been assigned to the group (through election), or their power base (democractic), are regarded as more legitimate. Furthermore, the preference for a legitimate leader type is particularly strong when a social identity is activated among group members, perhaps because social identifiers are primarily concerned about the quality of leader-member relationships. THE INFLUENCE OF LEADERSHIP ON COOPERATION A second goal of our research programme was to determine the effectiveness of leadership in managing public good dilemmas in groups. Do groups perform better in solving social dilemmas when a leader is in place? Which leader types are more effective in fostering and sustaining
168 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
cooperation and to what extent does their effectiveness vary with the needs of the group? These questions are important because, as we have seen, group members are quite reluctant to voluntarily accept a leader who takes all the investment decisions for the group—an autocratic leader. Therefore, leaders must rely on other tactics to ensure that the group is successful. Based upon a differential needs model of leadership, we anticipated that the influence of leaders would vary with the prevailing needs of group members, thought to be activated by their level of identity. Whereas personal identifiers would be more sensitive to information about the perceived instrumentality of a leader in tackling free riding, social identifiers would be more strongly influenced by the perceived relational qualities of a leader. We expected that these impressions were derived primarily from the leader’s power basis within the group, their style of leadership, and their personal attributes and characteristics. Power base of group leader Early research on power and influence identified several key power bases of leaders in groups (French & Raven, 1959), such as reward, coercive, legitimate, and relevant power. Leaders who control one or more of these power bases are thought to have more influence on group members. Following this taxonomy, we can distinguish between leaders that have either an instrumental or a relational power base. Instrumental leaders reward members if they cooperate and punish if they do not cooperate. Conversely, relational leaders exercise influence by being liked and respected by group members. Following a differential needs model, we predicted that leaders with an instrumental power base would be more effective in inducing cooperation, particularly when a personal identity was activated and members would be focused on their immediate outcomes. In contrast, leaders with a relational power base should have little or no influence on personal identifiers, but they should be able to influence the contribution decisions of social identifiers who would be more focused on the quality of the leader-member relationships. To test this we conducted a second experiment with a similar procedure to the first study (Van Vugt & De Cremer, 1999, Exp. 2). Participants played eight sessions of an investment game, and before each session they received an endowment of £3. A bonus would be provided to all members if the entire group contributed £12 or more. Unlike the first experiment, however, members could now invest any part of their endowment (from zero to three pounds). Hence, the average personal contribution per session should be at least £2 to reach the bonus. After these instructions, we induced members’ level of identity (social vs personal) using the same procedure as in the first study.
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 169
Subsequently, group members played four contribution sessions. After each session they received feedback that their group had failed to provide the bonus. Following the fourth session they were told that, as a result of their failure, a group leader would be appointed for the subsequent sessions. This leader was somebody from within the group who would monitor the contribution decisions and communicate to members via the computer. At this point, the leader introduced him/herself via an email message on the screen. In half of the conditions, the leader was given an instrumental power base. This was their message to the group. I have to make sure your group will receive the bonus in the forthcoming sessions. However, I do not believe that each group member will contribute enough voluntarily. In the next sessions I will penalise the least contributing member per session. That is, in each session the group member contributing the least amount will get a fine of £2.20. This amount will be subtracted from the amount of money he or she will have earned by the end of the sessions. Because people who do not contribute affect the group’s success, I think a punishment is the best thing to ensure that they will contribute enough next time. In the other half of the conditions, the leader was given a relational power base. They sent the following message: I have to make sure your group will receive the bonus in the forthcoming sessions. I trust each of you to contribute enough of your endowment to the provision of the good. If the group fails, however, I will send an encouraging message to group members who contributed little to ask them to contribute sufficiently the next time. I will not punish anyone, but I will try to give you support and explain things if necessary. You can trust me that everyone will be treated equally and with respect. After this initial message from the leader, the contribution sessions continued. After each session, group members again received feedback from the leader that their group had failed to secure the bonus. Instrumental leaders subsequently sent messages stating that they had penalised the least contributing member(s) of that particular session, whereas relational leaders sent encouraging messages to group members. After the eighth session, the experiment was interrupted and members were asked to complete some final questions. The pattern of contributions, which is displayed in Table 2, confirmed our prediction in part. When a personal identity was activated among group members they responded only to an instrumental leader. Yet they
170 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
TABLE 2 Contributions as a function of members’ identity and the power base of the leader
The range of possible contributions varies from 0 to £3. Means with a different subscript differ significantly at p<.01. From Van Vugt and De Cremer (1999, Exp. 2) with permission ©1999 American Psychological Association.
failed to respond to a leader with a relational power base. For social identifiers, however, it made no difference whether the leader was instrumental or relational. They were equally cooperative with a leader who simply encouraged and praised them as with a leader who would penalise them for contributing little to their group. These results were supported by additional psychological measurements, administered after the investment task. First, across all conditions members rated a relational leader as more legitimate than an instrumental leader (e.g., more trustworthy, competent, fair, and honest; Tyler, 1997). The interactions with the group leader also influenced members’ selfevaluations, in particular their specific state self-esteem. We used several items of the Rosenberg (1979) scale and adapted them to the specific task situation (“After participating in these contribution sessions do you feel sure of yourself?” “…do you feel satisfied with yourself?’ “…do you feel proud of what you have accomplished?”).3 As can be seen in Table 3, the state self-esteem of social identifiers was significantly higher when being supervised by a relational leader than by an instrumental leader. Interestingly, the reverse effect was found for personal identifiers who reported a higher state self-esteem under an instrumental leader than a relational leader. This suggests that leaders may, at least temporarily, boost the psychological well-being of a group member if they appeal to their dominant need, be it instrumental or relational.
3
Meta-analytic research has found evidence that when measuring temporary changes in self-esteem, specific state measures are better predictors than global trait self-esteem scales (Rubin & Hewstone, 1998).
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 171
TABLE 3 Differences in state-specific self-esteem as a function of members’ indentity and the power base of the leader
Higher scores indicate a higher state-specific self-esteem (1–7). Means with a different subscript differ significantly at p<.05. Adapted from Van Vugt and De Cremer (1999, Exp. 2), with permission ©1999 American Psychological Association.
Leadership style Impressions about the relational qualities of a group leader derive not only from their power base, but also from their perceived behavioural style, the way they treat and interact with the members of their group. Group members base their evaluation of a group leader, among other things, on the perceived fairness of their decision-making procedures (De Cremer & Van Knippenberg, 2002; Tyler, 1997, 2000; Tyler & Lind, 1992). For example, do leaders keep their promises, do they treat people with dignity and respect, and do they allow them to have a say in the decision-making process? Concerns about issues of procedural fairness are believed to be largely non-instrumental. According to the group-value model (Tyler & Lind, 1992; Tyler & Smith, 1998), a fair leadership style matters because it communicates information about the quality of social relationships in a group, and, in particular, a person’s standing within the group. A fair decision-making procedure signifies to individuals that they are respected and valued members of their group. According to an instrumental perspective on leadership, whether a leader acts in a procedurally fair or unfair manner should have no or, at least very little, impact on members’ cooperation as long as it does not directly influence their material outcomes. However, a fair or unfair procedure, such as keeping or breaking a promise, should have an impact if group members are concerned with the social relationships within the group. Hence, following the logic underlying the differential needs model we predicted that social identifiers would be more sensitive than personal identifiers to a fair or unfair leadership style. This hypothesis was tested in a third experiment (De Cremer & Van Vugt, 2002, Exp. 1), using a fairly similar procedure. First, we induced group members’ level of identity (social, personal) with the same manipulation as in the previous studies. Thereafter, they played eight
172 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
sessions of an investment game in groups of six. For each session, the endowment was £3, any amount of which they could invest in the group. Unlike the previous studies, however, the size of the bonus was variable rather than fixed. The size of the bonus depended upon the sum of contributions from the entire group. This amount would be multiplied by two and then distributed equally among all members, regardless of how much each contributed. This investment game is known as a continuous public good dilemma (rather than the step-level good we used before), and it meets the definitions of a social dilemma (Komorita & Parks, 1994). A second difference with the previous study was that there would be a group leader from the beginning of the task. The leader was chosen via a random selection procedure. The group leader, who did not participate in the contribution sessions, was given an instrumental power base: The leader would be able to penalise the least contributing member per session by £2. We manipulated leadership style as follows. Before the start of the investment task, group members were told that they would get an overview of their earnings when the entire task had ended. If they did not agree with the amount they had received they would be able to discuss this matter with the leader. Subsequently, the contribution sessions started, and after each session the leader sent an email, stating which member received the penalty for that session. After the first block of four sessions, the leader made an announcement about the earlier promise. In the fair leader style condition, the leaders reasserted that they would keep the promise to allow each of the group members the opportunity to discuss their earnings at the end of the task with them. In the unfair leaderstyle condition, the leaders withdrew their promise by stating that they had decided against discussing the earnings with each group member. They did not give a reason for this. Subsequently, the second block of four contribution sessions started. After the eighth session, the task ended and group members received some final questions concerning their impressions of the task and the group leader. The contribution patterns, displayed in Table 4, confirmed our prediction about the impact of leadership style. As can be seen, only social identifiers were affected by the leadership style. Their contributions were substantially higher when leaders kept their promise than when they broke it. In fact, they were substantially higher than in any other condition. For personal identifiers, however, there was no such effect. Their contributions were unaffected by leadership style. These results were corroborated by the effects of leadership style on group member’s specific state self-esteem, for which we used the same three-item measure as in the previous experiment. The state self-esteem of social identifiers was substantially higher after a procedurally fair treatment by the group leader, whereas there was little effect of treatment
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 173
TABLE 4 Contributions as a function of members’ identity and leadership style
The range of possible contributions varies from £0 to £3. Means with a different subscript differ significantly at p<.01. From De Cremer and Van Vugt, 2002, Exp. 1, with permission ©2002 Academic Press.
on the state self-esteem of personal identifiers. Interestingly, after controlling statistically for differences in reported state self-esteem, the influence of a fair versus unfair treatment on cooperation was substantially reduced for social identifiers. One possible interpretation of this result is that the positive effect of a fair treatment on the relational needs of these members enhanced their mood, which in turn increased their willingness to contribute. Personal attributes of leader Whereas the power base and behavioural style of leaders primarily provide information about their intragroup functioning, the personal attributes of leaders yield information about both their intragroup and intergroup leadership qualities. From a stereotypical point of view on leadership (Lord & Maher, 1991), leaders are expected to possess a set of skills and abilities that are cognitively associated with good leadership. One of these stereotypical leader skills considered necessary to achieve a group goal is to be a good communicator. Yet when a particular intergroup context is salient, the expectations concerning leadership may change. For example, members may want a leader who best fulfils the group’s prototype (Hogg, 2001). Similarly they may want a leader for whom membership of the group is important and who expresses a strong group commitment. We expected that the perceived importance of these motivational leader attributes would be influenced by how strongly members identify with their group. Consistent with the differential needs model, when a social identify was activated group members would cooperate more with leaders who showed a strong group commitment. Conversely, personal identifiers would cooperate more with leaders with stereotypically good leadership skills, such as good communication, planning and organisation skills— skills necessary to deal with the freerider problem. Accordingly, we
174 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
predicted that social identifiers would cooperate more with a highly committed group leader, irrespective of their specific leadership skills, whereas personal identifiers would cooperate more with a highly skilled leader, irrespective of their group commitment. This prediction was tested in two studies. As an initial test, we examined the role of leadership commitment in the experiment we described previously (De Cremer & Van Vugt, 2002, Exp. 1). In that particular study we manipulated, in addition to the style of leadership, the personal attributes of the group leader in an orthogonal between-participants design. Recall from the previous study that the experimenter selected the group leader from the six available group members (yet the focal participant was never elected). In one condition, the leader was somebody who showed a strong group commitment. This information was based on the fictitious scores of members on a self-reported three-item commitment scale, which we administered before the start of the task (e.g., “How committed do you feel to this group?”, “To what extent do you identify with this group?” 1=not at all, 7=very strongly). In the high-committed condition the chosen leader was reported to have an average score of 6 on the 7-point commitment scale, thus showing high group commitment. In the lowcommitted condition the leader had an average score of just 2, showing lowgroup commitment. Subsequently, we examined how this information affected the cooperation with leaders during the eight contribution sessions. In accordance with the differential needs prediction we found that only social identifiers were influenced by information about the leader’s group commitment. Averaging across the contribution sessions, they contributed more with a high-committed leader—an average of £2.12 per session—than they did with a low-committed leader—an average of £1.68. For personal identifiers there was no difference in cooperation with a high-committed (£1.52) or a low-committed group leader (£1.62). The effect of leader commitment was independent from the impact of fair or unfair treatment, which we discussed earlier. In the next experiment we provided a more direct test of this prediction by contrasting a leader with high group commitment with a leader who possessed strong leadership skills. Accordingly, we conducted a fourth and final experiment (De Cremer & Van Vugt, 2002, Exp. 2), using the same investment task as in the third experiment, albeit with a total of six rather than eight contribution sessions. Before the start of the task, we manipulated members’ level of identity using the standard procedure. Subsequently, we asked each member to select a leader from within their group of six. To facilitate the selection process, we asked each person to answer a series of questions regarding their leadership skills and group commitment.
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 175
Eight leadership skills questions were taken from an existing questionnaire (Ritchie & Moses, 1983) containing a list of established predictors of managerial success (e.g., “How well do you communicate verbally?”, “How good are your organisational and planning skills?”). Thereafter, they completed an extended eight-item group commitment scale (e.g., “How committed do you feel to this group?”, “How much do you identify with this group?”. On the computer screen they then saw how each group member, including themselves, was ranked on the skills and commitment scales. This information was non-veridical and the focal participant occupied the fourth position on both scales. Subsequently, in half of the conditions the experimenter asked members to elect the person they preferred to lead the group during the investment task (i.e., the leader selection would be based on a majority vote). In the other half of the conditions, the experimenter was said to appoint the group leader. Depending on the condition, either the person who ranked first on the leadership skills scale and fourth on the commitment scale (skilled leader) or the person who ranked first on the commitment scale and fourth on the skills scale (committed leader), was elected (appointed) as group leader. This leader would monitor the contributions of group members and penalise the least contributing member per session with a small fine (£0.50). The contribution patterns, which are displayed in Table 5, were in line with what we expected based on the differential needs hypothesis. We found that the contributions made by personal identifiers were systematically lower when they were supervised by a committed leader than by a skilled leader. In contrast, social identifiers contributed more under a committed group leader than under a skilled group leader. TABLE 5 Contributions as a result of members’ identity and the personal attributes of the leader
The range of possible contributions varies from £0 to £3. Means with a different subscript differ significantly at p<.01. From De Cremer and Van Vugt, 2002, Exp. 2, with permission ©2002 Academic Press.
Finally, we examined whether there were any differences that were related to the way leaders had been assigned to the group. Overall, group members cooperated more with an elected leader (an average of £2.21)
176 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
than with an appointed leader (an average of £1.76). This result was not influenced by the salience of an individual’s identity. Conclusions These findings provide further support for a differential needs model of leadership in social dilemmas. When leaders are in place, members are more likely to endorse them if leaders appeal to the dominant needs of group members. Leaders who have a relational power base, a fair leadership style, and are highly committed to their group have a greater influence when a social identity is activated among group members. However, when a personal identity is activated leaders must possess an instrumental power base and specific leadership skills in order to be influential. CONCLUSIONS AND DISCUSSION Taken together these experiments provide support for a differential needs model of leader endorsement in social dilemmas. Group members endorse leaders either because they are perceived to be instrumental in solving the freerider problem or because they help to establish positive social relationships within the group. Consistent with a social identity perspective (Tajfel & Turner, 1979; Turner et al., 1987), the relative importance of these two needs for group members was influenced by the salience of their group membership. When a personal identity was salient the instrumental aspects of leadership were regarded as more important, whereas the relational aspects gained in importance when a social identity was salient. In this chapter, we examined three stages of the leader endorsement process in social dilemmas: (a) the voluntary acceptance of leadership as a viable solution, (b) negotiations about the preferred type of leadership, and finally (c) cooperation with leaders in trying to solve the dilemma. First, regarding the acceptance of leadership as a structural solution to public goods we found that personal identifiers were more in favour of the structural change than social identifiers. Furthermore, without a leader they contributed less to the provision of public goods than did social identifiers and they rated their fellow members as less cooperative and trustworthy. These results suggest that there is a trade-off between social identity as an individual-psychological solution and leadership as a structural solution to social dilemmas. Group members opt for the leader solution when they perceive that the voluntary cooperative efforts of their group are insufficient to provide the common good. Yet, when due to the salience of their social identity group members have more trust in each other’s voluntary contribution, they will deem a group leader as unnecessary. Hence,
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 177
our research shows that a salient social identity, at least when activated by a symbolic intergroup competition, effectively serves as a leadership substitute (Kerr & Jermier, 1978). Further research is needed to determine whether other kinds of identity manipulations, such as accentuating a common fate or a shared similarity (Haslam, 2001), can produce similar results. Furthermore, future research should establish other factors that could serve as leadership substitutes. One candidate is members’ social valueorientation (De Cremer & Van Vugt, 1999; Van Lange, 2000). This personality variable distinguishes between people who are either more focused on their personal outcomes (proselves) or on the outcomes of the group (prosocials) in social dilemmas. It may well be that a prosocial orientation operates in the same manner as a salient social identity, with prosocial people exhibiting a weaker preference for adopting a group leader than proself people. Second, we examined the influence of members’ level of identity on their preferences for particular leader prototypes. In accordance with the differential needs model we found that when a social identity was salient, group members concentrated more on the relational attributes of a leader. That is, more than personal identifiers, social identifiers preferred a leader from inside the group as well as a leader who was elected by the members rather than appointed by the experimenter. This suggests that social identifiers cared more about the fit between the characteristics of the leader and those of the group, which is important for maintaining a positive group distinctiveness (Hogg, 2001). An alternative, instrumental explanation is that social identifiers were more concerned than personal identifiers with exercising control over the leader, hence their preference for an elected and an in group leader. Yet, it this were true, we would also expect them to have a greater preference for a democratic leader. In fact, both personal and social identifiers displayed a strong preference for democratic leadership above autocratic leadership. There may well be relational as well as instrumental motives behind the preference for a leader who would allow members to have a voice in the decision making process. Following the differential needs model, personal identifiers may prefer to have voice, mainly because they can control their personal outcomes. Conversely, having voice may also serve a relational function for social identifiers, because it communicates a symbolic message of respect and acceptance by the group (group-value model; Tyler & Lind, 1992). As a third stage in the leader endorsement process we examined the extent to which group members cooperated with the group leader in solving the social dilemma. To test predictions derived from the differential needs model we investigated the influence of members’ identity on their cooperation with various group leaders, who differed in (a) their power
178 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
base, (b) their style of leadership, and (c) their personal attributes. These leader manipulations enabled us to distinguish quite clearly between instrumental and relational motives for leader endorsement. First, we found that when a personal identity was salient, group members responded more strongly to leaders with an instrumental power base—who penalised the least contributing member—than leaders with a relational power base—who could only offer praise and support (French & Raven, 1958). Yet when a social identity was salient, a relational leader was at least as influential. Second, when a social identity was salient, group members responded more strongly to a leader who was procedurally fair by keeping an earlier promise. Yet a consistent, fair leadership style had no influence when a personal identity was salient among group members. These two findings extend and complement research on the group-value model (Tyler, 1997, 2000; Tyler & Lind, 1992; Tyler & Smith, 1998). The group-value model postulates that the quality of treatment by a leader or authority provides important relational information to individuals. Fair and considerate treatment signifies to individuals that they are respected members of their group. This strengthens an individual’s identification with their group, which in turn affects their self-esteem. Our results are consistent with this model by showing that high-quality treatment more strongly affected the feelings and behaviours of members for whom group membership was actually important to their self-definition (i.e., social identifiers). They cooperated more with a leader who communicated relational information and they reported a higher situation-specific selfesteem after interacting with this leader. Conversely, when group membership was not all that important to members’ self-definition (i.e., personal identifiers) the quality of treatment they received had little influence on their feelings and behaviours. As predicted by the differential needs model they were influenced more strongly by a leader who could directly influence their outcomes in the dilemma. In addition to their power base and leadership style, we found that the personal attributes of a group leader also mattered in the decision whether to cooperate with a leader or not. For personal identifiers a group leader was more influential if they were believed to possess strong leadership skills. In contrast, social identifiers cooperated more with a leader who showed a strong group commitment. This provides further evidence for a differential needs model, because it suggests that members’ expectations about what constitutes good leadership varies with the salience of their identity. When their social identity is activated members are less influenced by a leader who possesses stereotypical leadership skills than a leader who shows a strong desire to belong to the group. This suggests that relational impressions about group leaders are not just shaped by their intragroup functioning (in terms of power base and leadership style), but also by their intergroup functioning. Does the
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 179
leader play a role in establishing a positive between-groups identity? In this regard, this result fits nicely with recently developed social identity perspectives on leadership (Hogg, 2001; Turner & Haslam, 2000). These models postulate that leadership influence is a function of the cognitive fit between the characteristics of the leader and the characteristics of the group, in other words, their prototypicality. The most prototypical member of a group with high-identifying members is indeed more likely to be someone who exhibits a strong group commitment than in a group with low-identifying members. How do our findings contribute to the broader leadership literature? The distinction we made between the instrumental and relational functions of leadership is reminiscent of a large body of theory and research on leadership. For example, the classic Ohio State studies on leadership, conducted in the 1940s and 1950s (Hemphill, 1961), showed that the two primary leadership functions in small groups are task orientation (i.e., initiating structure) and relationship orientation (i.e., consideration). Other researchers have made essentially the same distinction, but with the use of slightly different labels. For example, Cartwright and Zander (1968) differentiated between goal achievement and group maintenance as the two basic functions of leadership. Our findings suggest that these functions are not entirely independent from each other. Leaders can solve social dilemmas either through a focus on the completion of the specific task or on the strengthening of the group. The extent of their influence, however, depends upon the match between their primary objective and the dominant needs within the group. If group members perceive an intense conflict between the personal interests of group members and the broader group interest, the leader must concentrate on the dilemma task at hand, solving the freerider problem, to be influential. Yet if group members perceive little or no conflict between personal and group interests, the group leader’s primary objective is to foster and maintain pleasant social relationships between group members. This group-based view on leadership is consistent with the work of other leadership theorists. For example in his work on followership Hollander (1964) argues that the analysis of leaders must include a careful consideration of the group that they represent: “It is therefore important that the leader, by his behaviour manifest a loyalty to the needs and aspirations of group members. These things must matter to him in ways that are accessible to view because such evidence of good faith and sincere interest serve to elicit greater acceptance of influence” (Hollander, 1964, p. 231). The social dilemma paradigm proved to be a suitable environment for investigating the different functions of leadership in groups. In a social dilemma there is a conflict between the material needs and the relational needs of group members. Members who try to maximise their
180 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
personal outcomes will harm the interests of the group to which they belong. Relational concerns thus play a role in restricting the relentless pursuit of self-interest and they promote the viability of the group. Leaders can solve this conflict of needs either by appealing directly to members’ self-interest, for example by introducing reward and punishments, or by an appeal to the overarching group interest. The latter suggests a more constructive way to solve social dilemmas, but will only be effective if group members care enough about their group membership. Our research has so far concentrated on the role of leaders in fulfilling the instrumental and relational needs of group members. Although these two needs are clearly important in shaping leader endorsement, there are probably other needs that leaders should also fulfil in order to be endorsed. An alternative need for individuals in most groups is the desire for personal autonomy and control (Deci & Ryan, 2000). These autonomy needs could explain why group members are quite reluctant to endorse autocratic forms of leadership in social dilemmas, although they are potentially highly effective structural solutions (see also Rutte & Wilke, 1985). In addition, individuals may wish to fulfil their intellectual and competence needs in groups, which may require a more transformational leadership style (Bass, 1990)—a leader who challenges members with high standards and stimulates the expression of ideas. Cooperation in social dilemmas presumably results from the satisfaction of a multitude of different needs, which can be more or less salient in any particular group context. Future research should investigate when particular needs are more salient than others so that leaders can try to satisfy these needs, and thus exercise influence in groups. REFERENCES Bass, B.M. (1990). Handbook of leadership.New York: The Free Press. Bass, B.M. (1997). Does the transactional-transformational leadership paradigm transcend organisational and national boundaries? American Psychologist,52, 130–139. Baumeister, R.F., & Leary, M.R. (1995). The need to belong: Desire for interpersonal attachments as a fundamental human motivation. Psychological Bulletin,117, 497–529. Brewer, M.B. (1979). In-group bias in the minimal intergroup situation: A cognitive-motivational analysis. Psychological Bulletin, 86,307–324. Brewer, M.B., & Kramer, R.M. (1986). Choice behavior in social dilemmas: Effects of social identity, group size and decision framing. Journal of Personality and Social Psychology, 3,543–549. Cartwright, D., & Zander, A. (1968). Group dynamics.New York: Harper & Row.
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 181
Chemers, M. (2001). Leadership effectiveness: An integrative review. In M.A.Hogg & R.S. Tindale (Eds.), Blackwell handbook of social psychology: Group processes (pp. 376–399). Oxford: Blackwell. Dawes, R.M. (1980). Social dilemmas. Annual Review of Psychology, 31,169–193. Deci, E.L., & Ryan, R.M. (2000). The what and why of goal pursuits: Human needs and the self-determination of behaviour. Psychological Inquiry, 11,227– 268. De Cremer, D. (2000). Leadership selection in social dilemmas—Not all prefer it: The moderating effect of social value orientation. Group Dynamics: Theory, Research andPractice, 4,330–337. De Cremer, D., & van Knippenberg, D. (2002). How do leaders promote cooperation? The effects of charisma and procedural fairness. Journal of Applied Psychology, 87,858–866. De Cremer, D., & Van Vugt, M. (1999). Social identification effects in social dilemmas: A transformation of motives. European Journal of Social Psychology, 29,871–893. De Cremer, D., & Van Vugt, M. (2002). Intergroup and intragroup dynamics of leadership in social dilemmas. Journal of Experimental Social Psychology, 38, 126–136. Foddy, M., & Crettenden, A. (1994). Leadership and group identity as determinants of resource consumption in a social dilemma. In U.Schulz, W.Albers, & U.Mueller (Eds.), Socialdilemmas and cooperation(pp. 207–232). Berlin: Springer-Verlag. Foddy, M., & Hogg, M. (1999). Impact of leaders on resource consumption in social dilemmas: The intergroup context. In M.Foddy, M.Smithson, S.Schneider, & M.A.Hogg (Eds.), Resolving social dilemmas: Dynamic, structural, and intergroup aspects.Philadelphia, PA: Psychology Press. French, J.R.P., & Raven, B. (1959). The bases of social power. In D.Cartwright (Ed.), Studiesin social power(pp. 118–149). Ann Arbor: University of Michigan, Research Center Group. Hains, S.C, Hogg, M.A., & Duck, J. (1997). Self-categorization and leadership: Effects of group prototypicality and leader stereotypicality. Personality and Social Psychology Bulletin,23,1087–1100. Hardin, G. (1968). The tragedy of the commons. Science, 162,1243–1248. Haslam, A. (2001). Psychology in organizations: The social identity approach.London: Sage. Hemphill, J.K. (1961). Why people attempt to lead. In L. Petrullo & B.M.Bass (Eds.), Leadership and interpersonal behaviour(pp. 201–215). New York: Holt. Hogg, M.A. (2001). A social identity theory of leadership. Personality and Social PsychologyReview, 5,184–200. Hogg, M.A., & Abrams, D. (1988). Social identifications: A social psychology of intergrouprelations and group processes.London: Routledge. Hollander, E.P. (1964). Leaders, groups, and influence.New York: Oxford University Press. Hollander, E.P. (1985). Leadership and power. In G.Lindzey & E.Aronson (Eds.), Thehandbook of social psychology(pp. 485–537). New York: Random House. Kerr, N.L. (1983). Motivation losses in small groups: A social dilemma analysis. Journal ofPersonality and Social Psychology, 45,819–828.
182 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Kerr, N.L. (1999). Anonymity and social control in social dilemmas. In M.Foddy, M. Smithson, S.Schneider, & M.A.Hogg (Eds.), Resolving social dilemmas: Dynamic,structural, and intergroup aspects(pp. 103–120). Philadelphia, PA: Psychology Press. Kerr, S., & Jermier, J.M. (1978). Substitutes for leadership: Their meaning and measurement. Organisational Behaviour and Human Performance, 22,375– 403. Komorita, S.S., & Parks, C.D. (1994). Social dilemmas.Dubuque, IA: Brown & Benchmark. Komorita, S.S., Parks, C.D., & Hulbert, L. (1992). Reciprocity and the induction of cooperation in social dilemmas. Journal of Personality and Social Psychology, 62,607–617. Kramer, R.M., & Brewer, M.B. (1984). Effects of group identity on resource use in a simulated commons dilemma. Journal of Personality and Social Psychology, 46,1044–1057. Levine, J.M., & Moreland, R.L. (1998). Small groups. In D.Gilbert, S.T.Fiske, & G.Lindzey (Eds.), The handbook of social psychology(pp. 415–469). New York: McGraw Hill. Liebrand, W.B.G., Messick, D.M., & Wilke, H. (1992). Social dilemmas: Theoretical issues andresearch findings.Oxford: Pergamon Press. Lind, E.A., & Tyler, T.R. (1988). The social psychology of procedural justice.New York: Plenum Press. Lippitt, R., & White, K.K. (1968). Leader behaviour and membrane reaction in three social climates. In D.Cartwright, & A.Zander (Eds.), Group dynamics.New York: Harper & Row. Lord, R.G., & Maher, K.J. (1991). Leadership and information processing: Linking perceptionsand performance.Winchester, MA: Unwin Hyman. Luce, R.D., & Raiffa, H. (1957). Games and decisions: Introduction and critical survey.London: John Wiley & Sons. Messick, D.M., & Brewer, M.B. (1983). Solving social dilemmas: A review. In L.Wheeler & P. Shaver (Eds.), Review of personality and social psychology(Vol. 4, pp 11–44). Beverly Hills, CA: Sage. Messick, D.M., Wilke, H., Brewer, M.B., Kramer, R.M., Zemke, P.E., & Lui, L. (1983). Individual adaptations and structural change as solution to social dilemmas. Journal ofPersonality and Social Psychology, 44,294–309. Olson, M. (1965). The logic of collective action.Cambridge, MA: Harvard University Press. Ritchie, R.J., & Moses, J.L. (1983). Assessment center correlates of women’s advancement into middle-management. Journal of Applied Psychology, 68, 227–231. Rosenberg, M. (1979). Conceiving the self.New York: Basic Books. Rubin, M., & Hewstone, M. (1998). Social identity theory’s self-esteem hypothesis: A review and some suggestions for clarification. Personality and Social Psychology Review, 2,40–62. Rutte, C.G., & Wilke, H.A.M. (1984). Social dilemmas and leadership. European Journal ofSocial Psychology, 14,105–121. Rutte, C.G., & Wilke, H.A.M. (1985). Preference for decision structures in a social dilemma situation. European Journal of Social Psychology, 15,367–370.
LEADER ENDORSEMENT IN SOCIAL DILEMMAS 183
Samuelson, C.D. (1993). A multi-attribute approach to structural change in resource dilemmas. Organisational Behaviour and Human Decision Processes, 55,298–324. Samuelson, C.D., & Messick, D.M. (1986). Inequities in access to and use of shared resources in social dilemmas. Journal of Personality and Social Psychology, 51,960–967. Samuelson, C.D., & Messick, D.M. (1995). When do people want to change the rules for allocating shared resources? In D.Schroeder (Ed.), Social dilemmas: Perspectives onindividuals and groups(pp. 143–162). New York: Praeger. Samuelson, C.D., Messick, D.M., Rutte, C.G., & Wilke, H.A.M. (1984). Individual and structural solutions to resource dilemmas in two cultures. Journal of Personality and SocialPsychology, 47,94–104. Sato, K. (1987). Distribution of the cost of maintaining common resources. Journal ofExperimental Social Psychology, 23,19–31. Schroeder, D.A. (1995). Social dilemmas: Perspectives on individuals and groups.London: Praeger. Spears, R., Oakes, P.J., Ellemers, N., & Haslam, A. (2000). The social psychology ofstereotyping and group life.Oxford: Blackwell. Stroebe, W., & Frey, B.S. (1982). Self-interest and collective action: The economics and psychology of public goods. British Journal of Social Psychology, 21,121– 137. Tajfel, H. (1972). La categorisation sociale [social categorisation]. In S.Moscovici (Ed.), Introduction à la psychologie sociale(pp. 272–302). Paris: Larouse. Tajfel, H., & Turner, J.C. (1979). An integrative theory of intergroup conflict. In W.G.Austin & S.Worchel (Eds.), The social psychology of intergroup relations (pp. 33–48). Monterey, CA: Brooke/Cole. Tenbrunsel, A., & Messick, D.M. (1999). Sanctioning systems, decisions frames, and cooperation. Administrative Science Quarterly, 44,684–707. Thibault, J., & Kelley, H.H. (1959). The social psychology of groups.New York: John Wiley. Turner, J.C. (1975). Social comparison and social identity: Some prospects for intergroup behaviour. European Journal of Social Psychology,5, 5–34. Turner, J.C., & Haslam, A. (2000). Social identity, organizations, and leadership. In M.E. Turner Ed.), Groups at work: Advances in theory and research.Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Turner, J.C, Hogg, M.A., Oakes, P.J., Reicher, S.D., & Wetherell, M. (1987). Rediscoveringthe social group: a self-categorization theory.Oxford: Blackwell. Tyler, T. (2000). Why do people cooperate with groups? Support for structural solutions to social dilemma problems. In M.Van Vugt, M.Snyder, T.R.Tyler, & A.Biel (Eds.), Cooperation in modern society: Promoting the welfare of communities, states, andorganisations(pp. 64–82). London: Routledge. Tyler, T.R.Dawes R.M. (1993). Fairness in groups: Comparing the self-interest and social identity perspectives. In B.A.Melkers & J.Baron (Eds.), Psychological perspectives onjustice: Theory and applications(pp. 87–108). New York: Cambridge University Press. Tyler, T., & Smith, H.J. (1998). Social justice and social movements. In D.Gilbert, S.T.Fiske, & G.Lindzey (Eds.), The handbook of social psychology(pp. 595– 632). New York: McGraw Hill.
184 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Tyler, T.R. (1997). The psychology of legitimacy: A relational perspective on voluntary deference to authorities. Personality and Social Psychology Review, 1,323–345. Tyler, T.R., & Degoey, P. (1995). Collective restraint in social dilemmas: Procedural justice and social identification effects on support for authorities. Journal of Personality and SocialPsychology, 69,482–497. Tyler, T.R., & Lind, E.A. (1992). A relational model of authority in groups. In M. Zanna (Ed.), Advances in experimental social psychology(Vol. 25, pp. 115– 191). New York: Academic Press. Van de Kragt, A.J.C, Orbell, J.M., & Dawes, R.M. (1983). The minimal contribution set as a solution to public goods problems. American Political Science Review, 77,112–122. Van Lange, P.A.M. (2000). Beyond self-interest: A set of propositions relevant to interpersonal orientations. European Review of Social Psychology, 11,297– 331. Van Vugt, M. (1997). When the privatization of public goods may fail: A social dilemma approach. Social Psychology Quarterly, 60,355–367. Van Vugt, M. (2001). Community identification moderating the impact of financial incentives in a natural social dilemma. Personality and Social Psychology Bulletin, 27,1440–1449. Van Vugt, M., & De Cremer, D. (1999). Leadership in social dilemmas: The effects of group identification on collective actions to provide public goods. Journal of Personality and SocialPsychology, 76,587–599. Van Vugt, M., Jepson, S.F., & De Cremer, D. (2001). Why autocratic leadership might fail insolving public good dilemmas: The importance of group stability.Unpublished manuscript, University of Southampton. Van Vugt, M., Snyder, M., Tyler, T., & Biel, A. (2000). Cooperation in modern society:Promoting the welfare of communities, states, and organisations.London: Routledge. Wilke, H.A.M. (1991). Greed, efficiency and fairness in resource management situations. European Review of Social Psychology, 2,165–187. Wit, A.P., Wilke, H.A.M., & Van Dijk, E. (1989). Attribution of leadership in a resource management situation. European Journal of Social Psychology, 19, 327–338. Yamagishi, T. (1986). The structural goal/expectation theory of cooperation in social dilemmas. In E Lawler (Ed.), Advances in group processes(Vol. 3, pp. 51–87). Greenwich, CT: JAI Press. Yukl, G.A. (1989). Leadership in organisations.Englewood Cliffs, NJ: Prentice Hall.
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY, 2002, 13, 185–216
Gender-related inequalities in the division of familywork in close relationships: A social psychologicalperspective Esther S.Kluwer Utrecht University, The Netherlands Gerold Mikula University of Graz, Austria
We offer a social psychological perspective on gender-related inequalities in close relationships and integrate two lines of research that have focused on the intrapersonal perceptions and interpersonal consequences respectively of the gendered division of labour. We start with a brief summary of research on genderrelated inequalities in the division of labour and discuss prior explanations and paradigms. We then address the extent to which spouses consider the division of labour as fair and the factors that contribute to perceptions of (un)fairness. Central to our argument is the distributive justiceframework (Major, 1987, 1993; Thompson, 1991) which claims that fairness judgements are affected by people’s wants and values, comparison standards, and justifications. In the following section, we address the question of whether dissatisfaction over the division of labour causes relationship conflict and how couples manage these conflicts. The main argument is that conflict over the division of labour generally comprises an asymmetrical conflict structure (with wives as complainants and husbands as defenders of the status quo), which elicits asymmetrical conflict interaction patterns (i.e., demand/ withdraw interaction) that result in asymmetrical outcomes (i.e., status quo maintenance). Finally, we summarise the main conclusions and address contributions to theory and research as well as directions for further research. © 2002 European Association of Experimental Social Psychology http://www.tandf.co.uk/journals/pp/02699931.html DOI: 10.1080/10463280240000064
186 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
INTRODUCTION Over the past decades, the division of labour in close relationships has received increasing attention in the social sciences. This interest was instigated by the increased participation of women in the paid labour force and the fact that these changes did not lead to parallel changes in the division of family work (housework and child care). Women still do a much larger share of the family work than men, regardless of their employment status, both in Western Europe (e.g., Kalleberg & Rosenfeld, 1990; Kluwer, Heesink, & Van de Vliert, 1996, 1997; Mikula & Freudenthaler, 1999; Mikula, Freudenthaler, Brennacher-Kröll, & Brunschko, 1997a) and in the United States (for reviews, see Menaghan & Parcel, 1990; Shelton & John, 1996; Steil, 1997). In developed countries, women perform two thirds to three quarters of the family work (United Nations, 1995). At the same time, women’s share in the labour force in Western Europe, the United States, and other developed countries has increased significantly over the past decades, from 33% in 1970 to 42% in 1990 (United Nations, 1995). The amount of time men invest in family work has risen over the past decades, but not enough to compensate for women’s increased labour force participation. Notwithstanding the clear evidence of gender inequality in the division of domestic labour, a large majority of women do not regard the division of work as unfair (Thompson, 1991) or dissatisfying (Major, 1987, 1993). If mere gender inequality is not a sufficient condition of feelings of injustice or discontent, what determines perceptions of justice when it comes to the division of labour? Furthermore, when spouses do consider the division of labour as unfair or dissatisfying, do they express their discontent and engage in conflict? And if they do, how are these conflicts managed and do couples create change through their mutual interactions? It is these and similar questions that are central to the current review. In this review, we integrate two lines of social psychological research that have focused on justice perceptions and on marital conflict over the division of labour respectively. Our goal is twofold—to develop a richer understanding of the determinants of perceived justice of the division of
Address correspondence to: Esther Kluwer, Utrecht University, Department of Social & Organisational Psychology, P.O.Box 80140, 3508 TC Utrecht, The Netherlands. Email:
[email protected] We acknowledge Miles Hewstone, Wolfgang Stroebe, and three anonymous reviewers for their valuable comments on a previous draft.
DIVISION OF FAMILY WORK 187
Figure 1. Conceptual model.
labour, and to develop a better understanding of the consequences of discontent over the division of labour in terms of conflict management and outcomes. According to our conceptual model (see Figure 1), the division of labour can result in both evaluative and affective responses (cf. Grote & Clark, 2001; Mikula et al., 1997a). A main evaluative response is the perceived (in) justice of the labour distribution and an important affective response is (dis) satisfaction with the division of labour. Social justice research has shown that these responses are related: Feelings of injustice generally cause discontent (Adams, 1965; Buunk & VanYperen, 1991; Cropanzano & Greenberg, 1997; Mikula et al., 1997a; Sprecher, 1986; Tyler & Smith, 1998; Walster, Walster, & Berscheid, 1978). Part 1 of this review focuses on the intrapersonal determinants of perceived justice of the division of labour. Central to our argument is the distributive justice framework (Major, 1987, 1993; Thompson, 1991) which claims that fairness judgements are affected by people’s wants and values, comparison standards, and justifications. Part 2 focuses on the interpersonal consequences of these personal experiences and addresses how couples deal with discontent and conflict over the division of labour. We start with a brief summary of research on gender-related inequalities in the division of labour and discuss prior explanations and paradigms. Finally, we summarise the main conclusions, address contributions to theory and research, and provide some directions for further research.1 Previous theoretical explanations In their endeavour to describe and explain the gendered division of labour, four main theoretical perspectives have emerged (see Kluwer, 1998a; Layte, 1998; Mikula, 1998; Shelton & John, 1996). One of the earliest approaches, the relative resource approach, argued that income, education, and occupational status are exchanged for domestic labour. One version of this model claims that the spouse who holds more power and authority in
188 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
the marital dyad (traditionally the husband) can minimise his or her participation in housework (Blood & Wolfe, 1960). Another version is that the husband’s resources negatively affect his participation in family work because these resources increase the value of his working time rather than the value of his domestic time (Geerken & Gove, 1983). Empirical findings relevant to this model have been inconsistent (e.g., Coverman, 1985; Deutsch, Lussier, & Servis, 1993; Layte, 1998; Steil, 1997). The demand-response model argues that men participate more in family work when there is greater need and when they are more available. According to this model, men do less family work than women because they spend more time on outside employment. This model is supported by research that showed a positive relationship between husbands’ participation on the one hand and wives’ working hours and the number and age of children on the other hand (Barnett & Baruch, 1987; Coverman, 1985; Deutsch et al., 1993). However, there is also evidence to suggest that there is no such relationship (e.g., Berk, 1985; Brines, 1993; England & McCreary, 1987). The gender role ideology model asserts that beliefs and attitudes towards gender roles are responsible for the division of domestic work (e.g., Layte, 1998). Egalitarian ideologies have been found to be positively associated with husbands’ involvement in child care (Deutsch et al., 1993; Greenstein, 1996). However, other studies showed no relationship between gender role ideology and involvement in family work (Barnett & Baruch, 1987; Coverman, 1985; Crouter, Perry-Jenkins, Huston, & McHale, 1987). More recently, the life course perspective has focused on the implications of the timing, sequencing, and duration of life events such as marital timing and childbearing for task allocation (Avioli & Kaplan, 1992; Coltrane & Ishii-Kuntz, 1992; Pittman & Blanchard, 1996). Women who married later tended to have smaller families, which was subsequently related to smaller contributions to family work (Pittman & Blanchard, 1996). In addition, men in couples that delay childbearing tend to be more involved in family work (Coltrane & Ishii-Kuntz, 1992). In sum, there is some support for each of the models that have been developed to explain the gendered division of labour, but the strength of the associations between various factors and the division of labour is rather weak (Shelton & John, 1996). The division of labour tends to be more unequal when gender differences in outside employment and income are larger, when the normative climate in the social environment prescribes a
1 The terms justice and fairness are used interchangeably in the present chapter in line with prevailing traditions in the field. In addition, the division of labour, family work or domesticlabour include both housework and child care tasks, unless mentioned otherwise.
DIVISION OF FAMILY WORK 189
sex-typed allocation of work, when spouses have more traditional gender role ideologies, and when couples have children at a younger age. Ultimately, however, mere gender explains the largest amount of variance in individual contributions to family work. A social psychological perspective Most prior research has concentrated on describing gender-related inequalities and exploring possible causes of inequality, usually from a sociological perspective. Much less is known about the perception and consequences of gender-related inequality in close relationships. In the main part of this review, we will first consider the extent to which spouses perceive (in)justice, the factors that contribute to perceptions of (in)justice, and moderators of the relationship between gender-related inequality and perceived justice (Part 1). The research reported in this section is grounded primarily in social psychological theories of justice (e.g., Mikula, 1993, 2001a; Törnblom, 1992; Tyler & Smith, 1998) that provide a solid theoretical framework for the study of the division of labour and inform us about relevant variables and processes. Social justice research has only just begun to explore the role of justice in close relationships, focusing primarily on distributive justice (Mikula & Lerner, 1994). The role of procedural justice in close relationships has been largely ignored until very recently (e.g., Kluwer, Chin-A-Fat, & Wissink, 2002a; Kluwer, Heesink, & Van de Vliet, 2002b). Second, we will consider the consequences of dissatisfaction with the division of labour for marital quality, psychological well-being, and marital conflict (Part 2). The research reported in this section is grounded primarily in theories of social conflict. Research on marital conflict over the division of labour offers a new perspective by recognising that the division of labour is actively (re)negotiated on a continuous basis, instead of a static agreement between spouses. It offers a structural/behavioural explanation for the fact that gender-related inequalities tend to persist in close relationships. The main argument is that conflict over the division of labour generally comprises an asymmetrical conflict structure (with wives as complainants and husbands as defenders of the status quo), which elicits asymmetrical conflict interaction patterns that result in asymmetrical outcomes. With the exception of Pruitt and his colleagues (Peirce, Pruitt, & Czaja, 1993; Pruitt, 1995), conflict research has largely ignored asymmetrical conflicts, implicitly assuming that conflict parties have equal complaints against each other.
190 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
PART 1: GENDER-RELATED INEQUALITY ANDPERCEPTIONS OF (IN) JUSTICE Early social psychological theories of justice accord a central role to distributive justice, that is the justice of the distribution of outcomes or resources (for reviews, see Mikula, 200 la; Törnblom, 1992; Tyler & Smith, 1998). A basic principle of distributive justice claims that people have a right to be treated equally with others who are like them. This principle holds for the distribution of positive outcomes such as goods, benefits, and rewards as well as for negative outcomes such as costs, burdens, and duties. Given the importance of equality as a basic rule of justice, one might expect inequality in the division of domestic labour to produce feelings of injustice, especially among women. However only a minority of respondents regards the division of labour as (somewhat) unfair (for a review, see Mikula, 1998). The lack of perceived injustice among women has puzzled many scholars and stimulated them to study the determinants of perceptions of (in)justice regarding the division of family labour. The evidence obtained in these studies, and recent conceptua lisations of the perception of injustice (e.g.,Major, 1987,1993;Mikula, 1993, in press; Mikula, Petri, & Tanzer, 1990), suggest that the preconditions of feelings of injustice are more multifarious and complex than the mere violation of equality. The following section will review the relevant research. Central to our argument is the distributive justice framework (e.g., Major, 1987, 1993; Thompson, 1991) which claims that fairness judgements are affected by people’s wants and values, comparison standards, and justifications. Determinants of perceived (in)justice Mikula (1998) reviewed research on the perception of justice of the division of household labour. He grouped the variables that have been studied as possible determinants of perceptions of (in)justice into four categories: Work load and division of labour, characteristics of the social system, individual differences, and cognitions. The main conclusions of this review can be summarised as follows. Work load and division of labour. Measures of men’s and women’s employment hours outside the home show low or insignificant correlations with justice perceptions (e.g., Blair & Johnson, 1992; DeMaris & Longmore, 1996; Demo & Acock, 1993; Greenstein, 1996; John, Shelton, & Luschen, 1995; Lennon & Rosenfield, 1994; Mikula & Freudenthaler, 1999). Measures of the division of family work are more consistently correlated with perceptions of (in)justice. Generally speaking, unbalanced divisions of labour are regarded as less fair than balanced divisions. Absolute measures, such as the hours spent doing family work, are less predictive of
DIVISION OF FAMILY WORK 191
justice perceptions than the relative size of men’s and women’s contributions to family work (e.g., DeMaris & Longmore, 1996; Hawkins, Marshall, & Meiners, 1995; John et al., 1995; Sanchez, 1994). In addition, the size of men’s contribution to family work shows higher correlations with perceived justice than the size of women’s contributions (e.g., Demo & Acock, 1993; John et al., 1995; Sanchez, 1994), and contributions to female-type tasks (e.g., cooking, cleaning) are better predictors of perceived justice than contributions to other tasks and global measures of domestic labour (e.g., Blair & Johnson, 1992; DeMaris & Longmore, 1996; John et al., 1995; Lennon & Rosenfield, 1994; Perry-Jenkins & Folk, 1994). System characteristics. Only a few studies focused on characteristics of the family or household system as predictors of perceptions of justice. Women with outside employment perceive the division of family work as less fair than housewives (e.g., Mikula & Freudenthaler, 1999). Marital status and the number of children seem to be uncorrelated with perceived justice (e.g., John et al., 1995; Lennon & Rosenfield, 1994; Mikula & Freudenthaler, 1999). No significant correlations were found between individuals’ earnings and their justice perceptions (e.g., DeMaris & Longmore, 1996; John et al., 1995; Lennon & Rosenfield, 1994; Ward, 1993). Individual differences. The most frequently studied individual difference variable is gender. Women generally perceive the division of family labour as less fair than men (e.g., Blair, 1993; DeMaris & Longmore, 1996; Lennon & Rosenfield, 1994; John et al., 1995; Mikula & Freudenthaler, 1999; Robinson & Spitze, 1992; Sanchez, 1994; Ward, 1993). In addition, younger, more educated, and White people typically perceive the division of work as less fair than older, less educated, and non-White people (e.g., DeMaris & Longmore, 1996; Hawkins et al., 1995; John et al., 1995; Lennon & Rosenfield, 1994; Sanchez, 1994; Ward, 1993). The evidence for the correlation between gender ideology and perceptions of justice is mixed. Some studies found significant but weak and inconsistent support for the prediction that people with egalitarian ideologies evaluate the unbalanced division of family work as more unfair than those with traditional ideologies (e.g., Blair & Johnson, 1992; DeMaris & Longmore, 1996; John et al., 1995; Greenstein, 1996; Mikula et al., 1997a). Other studies reported no significant correlations (e.g., Lennon & Rosenfield, 1994; Mederer, 1993; Ward, 1993). Cognitions. More recently, studies have increasingly considered various cognitions about the division of family work, such as appraisals, attributions, comparative judgements, and justifications, as possible predictors of the perception of justice. Since most of these studies have been guided by the distributive justice framework of Thompson (1991) and Major (1993), the relevant evidence will be reviewed after a discussion of this framework.
192 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
The distributive justice framework Building upon Major’s (1987) theoretical and empirical work on gender differences in entitlement, Thompson (1991) and Major (1993) proposed a very similar distributive justice framework to explain women’s sense of fairness (Thompson, 1991) and paradoxical contentment (Major, 1993) with the unequal division of family work. While Thompson (1991) focused on women’s relative sense of fairness, Major (1993) argued that women report relative contentment with an unequal division of labour because it does not violate their sense of entitlement.2 Both authors assumed three factors to contribute to perceptions of entitlement, and fairness: Wants and values, referring to the outcomes people desire or value, comparison standards, referring to the standards people use to judge their outcomes, and justifications, referring to procedures, attributions, or rules that legitimise the outcome. Wants and values. Major (1987, 1993) and Thompson (1991) argued that to understand fairness judgements, we should take into account the individual’s aspirations, expectations, and desires about the division of labour. Several theories support this proposition. According to Crosby (1982), wanting is a necessary condition for feelings of deprivation or injustice. Social exchange theorists (e.g., Kelley & Thibaut, 1978) have shown that discrepancies between aspirations or expectations and the status quo lead to dissatisfaction. Similarly, role theory infers that discrepancies between what is considered desirable and what is actually possible lead to role strain and dissatisfaction (Araji, 1977; Chassin, Zeiss, Cooper, & Reaven, 1985). Hence, individuals feel dissatisfied or unfairly treated when they feel they are not receiving the rewards that they want or value. Aside from time and tasks as valued outcomes, interpersonal or symbolic outcomes of family work should be considered (Thompson, 1991). Taking care of family members is a valued outcome for many women, which may lighten the relative burden of these tasks. In addition, women may want their husband to contribute to family work because they value the fact that their husband takes care of the children, rather than because it would decrease the actual work load. In addition, the husband’s responsiveness to
2
The concept of entitlement is related to the concepts of justice and fairness but must be distinguished. Entitlement refers to ideas about what people have a right to get by virtue of who they are and what they have done. Justice and fairness refer to the judgement of whether people actually do or do not get what they are entitled to. While the fulfillment of somebody’s entitlement will evidently be judged as just, judgements of injustice may presuppose more than the mere perception that somebody’s entitlement has been violated (see below).
DIVISION OF FAMILY WORK 193
the wife’s needs, and his appreciation of her work, are assumed to be a valued outcome for women. Comparison standards. Like most theories of distributive justice, Major (1987, 1993) and Thompson (1991) assumed that feelings of (in)justice are primarily derived from comparison processes. Comparison standards may entail normative or ideological standards (e.g., DeMaris & Longmore, 1996; Freudenthaler & Mikula, 1998) or feasibility standards (Mederer, 1993). In addition, the distributive justice model accords a central role to comparisons that are socially derived. Consistent with equity theory (Adams, 1965; Homans, 1974; Walster et al., 1978), Major (1987, 1993) and Thompson (1991) argued that when wives perceive inequality between the ratio of their own outcomes and inputs and the ratio of outcomes and inputs of their spouse, they feel distressed. This type of social comparison with the spouse is referred to as relational comparison (Buunk & VanYperen, 1991; Hatfield, Traupmann, Sprecher, Utne, & Hay, 1985). In general, perceived inequity results in lower levels of satisfaction than perceived equity (Buunk & VanYperen, 1991; Sprecher, 1986; VanYperen & Buunk, 1991; Walster et al., 1978). Interestingly, this holds not only for the disadvantaged but also for those who are advantaged compared to their spouse. According to social comparison theory (Festinger, 1954), however, individuals tend to compare themselves with similar others, and women may find other women more similar than their husband, especially on a gendered topic like the division of labour. Major (1987, 1993) therefore suggested that wives compare their inputs and outcomes in the relationship primarily with those of similar others in a reference group—family members, friends, and acquaintances of the same gender. This type of comparison with same-gender others is referred to as referential comparison (Buunk & VanYperen, 1991). Referential comparisons may serve the purpose of maintaining a positive view of the self and the situation (Wood, 1989): Women who feel that they are better off than other women (i.e., downward comparison) will be more satisfied and perceive the division of labour as more fair (Buunk & VanYperen, 1991). Justifications. Major (1987, 1993) and Thompson (1991) argued that perceptions of fairness are also affected by the perceived justifiability or legitimacy of the unequal division of labour. There are various reasons or circumstances that women regard as acceptable justifications for the gendered division of labour (see Major, 1993; Mikula, 1998; Thompson, 1991). For example, justifications are that the wife has more time available for family work, that she is more competent than her partner, and that she enjoys family work more than the husband. An important way to justify an unequal or negative outcome is the belief that the outcome resulted from the application of a fair procedure. The literature on procedural justice has shown that people judge the same
194 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
outcome as more fair under conditions of high procedural fairness than under conditions of low procedural justice (see Lind & Tyler, 1988; Van den Bos, Lind, Vermunt, & Wilke, 1997). Early studies showed that, regardless of the outcome, individuals judge a procedure as more fair when they can exert influence on the procedure (Thibaut & Walker, 1975; Walker, Latour, Lind, & Thibaut, 1974). Procedural justice is enhanced by the opportunity to express one’s views and opinions and present information relevant to the decision (i.e., voice). Accordingly, voice in decision making results in greater perceived justice (Lind, Kanfer, & Earley, 1990) and satisfaction (Folger, Rosenfield, Grove, & Corkran, 1979; Thibaut & Walker, 1975) with the outcome. Voice is seen as fair because it helps individuals to control their outcome (Thibaut & Walker, 1975) and people value voice because it implies that their opinion is valued, which fulfils their need for self-esteem and self-identity (Lind & Tyler, 1988; see also Lind et al., 1990). Until now, procedural justice has been studied in experimental, legal, and organisational settings (Lind & Tyler, 1988; see Brockner & Wiesenfeld, 1996, for a review). Surprisingly, only a few recent studies have tested these effects in the domain of close relationships (Hawkins et al., 1995; Kluwer et al., 2002a, 2002b). Empirical tests of the distributive justice framework Even though their theoretical model has been quite influential in the literature on justice in the family area, empirical evidence for Major’s (1993) and Thompson’s (1991) distributive justice framework is very limited. Prior studies that used this approach primarily focused on structural or ideological factors (Blair & Johnson, 1992; DeMaris & Longmore, 1996; Greenstein, 1996; Mederer, 1993). Hawkins and colleagues (1995) conducted the first comprehensive test of the distributive justice framework with 234 dual-earner wives in the United States. Women were 40 years old on average, their professional work load was 38 hours per week on average. The authors found support for the prediction that outcome values and justifications are relevant to women’s perceptions beyond the perceived division of labour. But, generally speaking, only few of the variables that were considered in the analyses turned out to be significantly related to women’s perceptions of justice. Women reported more justice when they felt their family work was appreciated by their partner and the division of labour had been worked out together with the partner. Kluwer et al. (2002b), Freudenthaler and Mikula (1998), and Mikula and Freudenthaler (2002) provided further tests of predictions derived from the distributive justice framework. These studies will be described in more detail below. Kluwer et al. (2002b) conducted a three-wave longitudinal survey among 293 couples making their transition to parenthood. At the first wave,
DIVISION OF FAMILY WORK 195
husbands were 31 years of age on average. Wives were 29 years on average and were pregnant of their first child. During the second and third wave, the first child was 5.5 and 14.5 months of age on average. Kluwer and colleagues tested predictions derived from the distributive justice framework among both husbands and wives. They tested the impact of (a) discrepancies between the actual and desired division of labour as perceived by spouses, (b) the outcomes of referential and relational comparisons, and (c) procedural justice on perceived fairness, both cross-sectionally and longitudinally. Multiple regression analyses showed that discrepancies between the actual and the desired division of labour, social comparison outcomes, and procedural justice explained over 30% of the variance in women’s fairness perceptions. For husbands, discrepancies between the actual and desired division of labour and social comparisons explained over 25% of the variance in perceived fairness. Support was also found for the long-term impact of the variables in the model on husbands’ and wives’ perceptions of fairness across time. Surprisingly, husbands’ fairness perceptions were not affected by procedural justice, suggesting that procedural justice is more important for women than for men. This is in line with Kluwer and colleagues (2002a), who showed that the relationship between procedural justice and marital satisfaction was much stronger for women than for men. There is evidence that women are more process-oriented whereas men seem to be more outcome-oriented (Major & Deaux, 1982) and that procedural justice affects women more than men (Sweeney & McFarlin, 1997). Furthermore, women may concentrate on procedural issues because this appears a feasible way of reaching outcomes in domains that are generally dominated by men (cf. Thoits, 1987). Freudenthaler and Mikula (1998) also tested predictions derived from Major’s (1993) and Thompson’s (1991) distributive justice framework. They tested the impact of unfulfilled wants, the outcomes of social and normative comparisons, and the lack of justification on the perceived violation of entitlement. In addition, they integrated the distributive justice framework with the two-factor model of relative deprivation (Crosby, 1982) and the attribution-of-blame model of judgements of injustice (Mikula, 1993, in press). The three models have much in common, but differ in some of their propositions. Most importantly, the distributive justice framework equates the perception of injustice with the perception of violated entitlement, while the other models propose additional preconditions of perceptions of injustice aside from the perceived violation of entitlement. Crosby’s (1982) two-factor model of relative deprivation proposes the existence of unfulfilled wants (i.e., discrepancies between actual and desired outcomes) as a further presupposition of feelings of deprivation or injustice. Obviously, there is much overlap between the concepts of “unfulfilled
196 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
wants” in Crosby’s (1982) model and “wants and values” in Major’s (1993) and Thompson’s (1991) distributive justice framework. However, the two-factor model of relative deprivation proposes a direct effect of unfulfilled wants on perceived injustice, in addition to the effect on the perceived violation of entitlement, while the distributive justice framework assumes an indirect effect of unfulfilled wants on the perception of injustice via the perception of violated entitlement. The attribution-of-blame model of judgements of injustice of Mikula (1993, in press) conceptualises judgements of injustice as an attribution of blame for the violation of a person’s entitlement. Thus, the model proposes the attribution of blame as an additional precondition of feelings of injustice aside from the violation of entitlement. Attributions of blame in turn presuppose the attribution of responsibility and perceived lack of justification for the violation of entitlement (cf. Heider, 1958; Shaver, 1985). The main difference between the distributive justice framework and the attribution of blame model exists with respect to the role of attributions of responsibility and the perceived lack of justification. Attributions of responsibility are not considered in the distributive justice framework but constitute a major or precondition of attributions of blame in Mikula’ s model. The perceived lack of justification is important in both models, though at different locations. The distributive justice framework assumes a lack of justification to contribute to the perceived violation of entitlement. The attribution of blame model regards it as part of the attribution of blame. In sum, the attribution-of-blame model proposes direct effects of the perceived violation of entitlement and the attribution of blame on the perception of injustice and indirect effects of the perceived violation of entitlement, attributions of responsibility, and the lack of justification through their impact on the attribution of blame. The comprehensive model that integrates the distributive justice framework, the two-factor model of relative deprivation and the attribution-of-blame model of judgements of injustice makes the following predictions (see Figure 2): (1) Unfulfilled wants, the violation of entitlement, and the attribution of blame contribute directly to the perception of injustice, (2) unfulfilled wants, the outcomes of social and normative comparisons, and the lack of justification are indirectly related to justice evaluations as they affect the perception of violation of entitlement, and (3) the perceived violation of entitlement, the attribution of responsibility, and the lack of justification contribute indirectly to justice evaluations as they affect the attribution of blame. Freudenthaler and Mikula (1998) tested the predictions derived from the comprehensive model with data from a sample of 132 employed women who lived together with an employed partner. Employed women were chosen because the division of housework seems to be more of a problem
DIVISION OF FAMILY WORK 197
for women in dual-earner couples. The mean age of the women in the sample was 34 years, their professional work load was 36 hours per week on average. The predictions were tested by three separate regression analyses and one path analysis (see Figure 2). Three separate multiple regression analyses provided good support for the proposed predictors of the perceived violation of entitlement, blaming the partner, and perceived injustice. Unfulfilled wants, comparison outcomes, and lack of justification significantly contributed to the perceived violation of entitlement and explained 57% of variance. The violation of entitlement, the attribution of responsibility, and the perceived lack of justification significantly contributed to the attribution of blame and explained 47% of variance. Unfulfilled wants, the perceived violation of entitlement, and the attribution of blame significantly contributed to and explained 49% of the variance in women’s justice evaluations. The comprehensive model was tested in a LISREL8 structural modelling analysis (see Figure 2). The findings provide good support for the direct and indirect effects of the comprehensive model. In addition, the comprehensive model turned out to have a better fit than various alternative models. Mikula and Freudenthaler (2002) focused on one single aspect of the distributive justice framework: Social comparisons. They considered the outcomes of three different kinds of social comparison: Relational comparisons between oneself and the partner, and referential comparisons between oneself and other women and between the partner and other men. The comparison outcomes were assessed by ratings like “How much household labour are you doing compared with your partner (compared with other women)?”, “How much household labour is your partner doing compared with other men?”. The impact of the social comparison outcomes on women’s justice evaluation of the division of labour was tested among a representative sample of 2200 Austrian women (Mikula & Freudenthaler, 1999), the sample of 132 employed women described earlier (Freudenthaler & Mikula, 1998), and another convenience sample of 148 employed women (Mikula, Freudenthaler, Hörtner & Foufas, 2001, Study 3). The mean age of the women [in the latter sample] was 34 years and their professional work load was 36 hours per week on average. Multiple regression analyses showed that the outcomes of comparisons between oneself and the partner and between the partner and other men primarily mattered for the justice evaluations. Women regarded the division of household labour as less fair when they felt they did more housework compared with their partner and when they felt their partner did less housework compared with other men. The outcome of the comparison between oneself and other women had less impact than the other two comparisons in the Mikula and Freudenthaler (1999) study and had no impact at all in the other two studies.
198 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 2. Path model predicting (a) violation of entitlement from unfulfilled wants, comparison outcomes, and lack of justification; (b) attribution of blame from violation of entitlement, attribution of responsibility, and lack of justification; and (c) perceived injustice from unfulfilled wants, violation of entitlement, and attribution of blame: standardised path and regression coefficients. Goodness-of-fit index=.98; adjusted goodness-of-fit index=.91, comparative Figure index=1.00; 2 (8, N=132)=10.20, p=.25. All coefficients shown are significant (<.05). Adapted from Freudenthaler and Mikula (1998), with permission.
Further analyses of the Freudenthaler and Mikula (1998) data and the Mikula et al. (2001) data considered the frequencies of making different kinds of comparisons in addition to the outcomes of the comparisons.3 (Frequencies were measured by ratings like “How often do you think about, or compare, how much housework you do and how much housework your partner does?”.) Women considerably differed in the frequencies of making different kinds of comparisons. Some women predominantly compared themselves with their partner, some predominantly compared themselves with other women, and some women predominantly compared their partner with other men. Regression analyses
DIVISION OF FAMILY WORK 199
revealed that the frequencies of making different kinds of comparisons did not directly contribute to women’s justice evaluations. However, they moderated the extent to which women relied on the outcomes of the various comparisons in their justice evaluations. Women who predominantly compared themselves with their partner relied more on the outcome of this comparison than on the outcome of the comparison between the partner and other men. Women who predominantly compared their partner with other men relied more on the outcome of this comparison than on the outcome of the comparison between oneself and the partner. Finally, women who made both kinds of comparisons to the same extent relied on the outcomes of both comparisons in their justice evaluations.4 Another finding is also worth mentioning: Women who predominantly compared themselves with their partner reported more injustice than those who predominantly compared their partner with other men. In sum, the data are in line with the distributive justice framework and suggest that women use different social comparisons to construe a certain picture of their situation and relationship. Summary Research on the perception of (in)justice with regard to the division of family work has yielded a number of important findings that have advanced our knowledge of factors that contribute to perceived (in)justice. Most importantly, the actual division of labour, and absolute measures of work load, are by no means the main determinants of perceptions of (in) justice. System characteristics and individual difference variables, aside from gender, do not make major contributions to perceived injustice either. Cognitions about family work and its division, such as attributions, comparative judgements, and justifications, proved to be much more relevant to how women perceive and evaluate the gender-related inequalities in the division of domestic labour. In support of this conclusion, studies that focused primarily on cognitive variables accounted for much larger proportions of variance in perceived fairness than did other studies (e.g., Freudenthaler & Mikula, 1998; Hawkins et al., 1995; Kluwer et al., 2002b; Mikula & Freudenthaler, 2002). The findings suggest that women partly shape their view of their situation by the social
3 Frequencies of making different kinds of comparisons were not measured in the survey with the representative sample of Mikula and Freudenthaler (1999). 4 We considered only the comparisons between oneself and the partner and between the partner and other men, and neglected the comparison between oneself and other women in these analyses, because the latter comparison had no impact on women’s justice judgments in the two studies.
200 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
comparisons and attributions they make, and by the justifications they provide for the situation. PART 2: CONSEQUENCES OF GENDERRELATEDDISCONTENT AND PERCEIVED INJUSTICE The previous part focused on the intrapersonal evaluation and experience of the division of labour. In this part, we will address the interpersonal consequences of these personal experiences. Individuals who perceive the division of family work as less fair, report less satisfaction with the division of labour (Mikula et al., 1997a), less marital satisfaction and stability, and more marital conflict (e.g., Mederer, 1993; Mikula & Freudenthaler, 1999; Perry-Jenkins & Folk, 1994; Ward, 1993). Perceptions of justice appear to be a more important predictor of marital quality than the division of labour itself (Blair, 1998; Greenstein, 1996; Mikula & Freudenthaler, 1999). Furthermore, there is evidence for a significant association between perceived injustice and depressive symptoms (Lennon & Rosenfield, 1994) and general unhappiness and distress (Robinson & Spitze, 1992) among wives. Some of these relationships tend to be stronger for women than for men (Mikula et al., 1997a) or exist only among women (Perry-Jenkins & Folk, 1994; Robinson & Spitze, 1992; Ward, 1993). Aside from these important findings, we know very little about the specific issues regarding the division of labour that cause relationship conflict, how couples manage their conflicts over the division of labour, and how couples hinder or create change through their mutual interactions. Prior research has failed to recognise that spouses actively negotiate and renegotiate the division of labour on a continuous basis. However, traditional values are increasingly rejected, leaving couples deprived of a set of rules and norms to guide their behaviour. Indeed, couples with egalitarian gender role ideologies tend to report more marital conflict and negotiation than couples in traditional marriages (e.g., Buunk, Kluwer, Schuurman, & Siero, 2000; Sillars & Kalbflesch, 1989; Thoits, 1987; VanYperen & Buunk, 1991). They experience more uncertainty and conflict about gender roles because these roles are subject to change. Marital conflict over the division of labour Marital conflict exists when at least one spouse feels obstructed or irritated by the other spouse (cf. Deutsch, 1973; Holmes & Murray, 1996). The conflict issue is the subject matter around which obstruction or irritation is experienced. It is a subjective experience that does not necessarily have an objective basis. Accordingly, the research reviewed next focused on the
DIVISION OF FAMILY WORK 201
TABLE 1 Self reported mean conflict frequency for husbands and wives (unpublished data)
All variables were measured on 7-point Likert scales (1=never, 7=very often). Standard deviations are in parentheses.
subjective perceptions of imbalance and subsequent feelings of discontent, not on the objective imbalance in the division of labour or gender inequality per se. Several studies among both parents and non-parents have shown that couples have significantly more conflict over housework than over paid work or child care and that the wife’s discontent with housework is the primary source of marital conflict over the division of labour (Kluwer et al., 1996, 1997, 2000). For example, Table 1 shows the mean reported conflict frequency for a sample of 314 Dutch couples with a first child under 2 years of age (unpublished data). Wives reported more conflict over housework than over any of the other issues. A closer look at the data revealed that the main issues for wives were the unequal division of housework, the fact that the partner did not carry out housework properly, disagreement about how household chores should be done, and the fact that the partner did not carry out housework without being asked to do so. Husbands reported somewhat less conflict. Although they reported more conflict over housework than over child care or paid work, conflict over housework did not exceed conflict over the other issues. Husbands’ main conflict issue was being pressured by the partner into doing housework. These results are consistent with other research (Kluwer et al., 1996, 1997, 2000). Housework is generally repetitive, rather disliked, and often regarded as a burden or duty (Berheide, 1984; Mikula, Freudenthaler, BrennacherKröll, & Schiller-Brandl, 1997b; Robinson & Spitze, 1992; Steil, 1997). Women therefore tend to prefer doing less rather than doing more (Kluwer et al., 1996). Child care tasks are new for first-time parents and they may
202 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
be regarded as more pleasant and rewarding than housework. In addition, wives may feel it is not appropriate to complain about the division of child care tasks because it may be interpreted as not caring about the child (Mederer, 1993), or they may not complain because the gendered division of child care tasks enables them to fulfil their role as a mother. In general, couples expect the baby to create a lot more work, but their expectations do not prepare them sufficiently for what is actually coming (Belsky, Lang, & Huston, 1986; Hackel & Ruble, 1992; see Kluwer, 2000). Unpublished longitudinal data showed that it was typically household tasks, such as dish washing, laundry, grocery shopping, and cleaning, that increased more than couples expected before childbirth and that wives usually carried the burden of these extra tasks (Kluwer, 1997). This violation of expectations may be an additional explanation for the finding that it is typically the wife’s discontent with housework that causes conflict over the division of labour. The division of labour: An asymmetricalconflict issue Demand/withdraw interaction. The foregoing showed that the wife’s discontent with housework is the primary source of marital conflict over the division of labour. In line with this, a vignette study by Kluwer et al. (2000, Study 1) among 121 husbands and 141 wives with a child under 2 years of age showed that most conflicts over the division of labour have an asymmetrical conflict structure: The wife (the complainant) wants her husband to contribute more to housework, while the husband (the defendant) wants to maintain the status quo. During conflicts with an asymmetrical structure, complainants are likely to pressure the other for change, while defendants will avoid a discussion that may lead to a change in their own behaviour (cf. Klinetob & Smith, 1996; Peirce et al., 1993; Pruitt, 1995). In close relationships, this type of interaction has been labelled demand/withdrawinteraction: One spouse (usually the wife) attempts to engage in a discussion, resorting to pressures and demands, while the other (usually the husband) attempts to avoid conflict and withdraws from the discussion (e.g., Christensen & Heavey, 1990; Heavey, Layne, & Christensen, 1993). According to the foregoing, the structure of the conflict determines the roles men and women take during the interaction. Some authors have argued that men are the primary beneficiaries of the traditional marriage, are more likely to have structured the relationship to their liking and subsequently have little or less interest in changing the status quo (Heavey et al., 1993; Jacobson, 1989; Thoits, 1987). Women tend to be less satisfied with the status quo of the relationship (Buunk & VanYperen, 1989) and engagement in conflict is their means of changing the relationship according to their desires (Steil, 1997).
DIVISION OF FAMILY WORK 203
In addition, research has shown that women have a general tendency to demand and men to withdraw in marital conflict. Developmental theories claim that men have developed a self differentiated from others, whereas women have developed a self in relation to others (Cross & Madson, 1997; Chodorow, 1978; Miller, 1986). Subsequent socialisation has reinforced men’s achievement orientation and need for autonomy and women’s relationship orientation and need for connectedness (Eagly, 1987; Maccoby & Jacklin, 1987; Miller, 1986). This gender difference permeates social interactions and it predicts that women will pursue conflict in their search for connectedness, whereas it predicts that men will withdraw in pursuit of their autonomy (cf. Gray-Little & Burks, 1983). Other explanations have focused on the evolutionary basis for gender differences in conflict behaviour (e.g., Buss, 1989) and gender differences in physiological arousal (Gottman & Levenson, 1988). To summarise, we expected women to show an overall tendency to be demanding and men to be withdrawing in marital conflict, but primarily when the wife wants to change the status quo. This proposition was tested in three vignette studies on conflict over the division of labour among a total of 347 husbands and 364 wives (the majority being couples) with a first child under 2 years of age (Kluwer, 1998b; Kluwer et al., 2000). In all studies, respondents were presented a conflict situation in which one spouse (the complainant) was dissatisfied with the contribution of the other spouse to paid work, housework, or child care, while the other spouse (the defendant) was satisfied with the status quo. The gender of the complainant and the defendant varied but the vignettes were otherwise identical. Respondents wrote down how they thought they and their spouse would respond to this situation. Independent judges rated the extent to which the answers represented demand/withdraw interactions on a 4-point scale. Figure 3 presents the collapsed data of two of these studies (Kluwer, 1998b; Kluwer et al., 2000, Study 1) and shows that spouses reported more wife-demand/husband-withdraw interaction than husband-demand/ wife-withdraw interaction during conflict over the division of labour, but only when the wife desired a change in her spouse’s contribution. These results were replicated by Kluwer et al. (2000, Study 2) and the data are consistent with Kluwer et al. (1997) who found that the wife’s discontent over housework was positively related to wife-demand/husband-withdraw interaction, which, in turn, predicted destructive conflict outcomes. The data are also consistent with Mikula et al. (1998) who found that women responded more accusingly than men in the role of those suffering from an unjust incident. Status quo maintenance versus change. Our next research question was whether defendants would be more successful than complainants in reaching their goal (status quo maintenance versus change respectively),
204 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 3. Demand/withdraw interaction as a function of complainant. Adapted from Kluwer (1998b) and Kluwer et al. (2000a, Study 1), with permission.
because this would explain why the gendered division of labour persists despite women’s attempts to achieve change (cf. Thoits, 1987). We expected that defenders of the status quo were more likely to reach their goal than complainants during conflict over the division of labour. Our reasoning was three-fold. First, defendants have a major advantage over complainants, because the status quo is on their side. Complainants depend on the cooperation of the defendant, that is, whether the defendant is willing to change his or her behaviour (cf. Babcock, Waltz, Jacobson, & Gottman, 1993; Kelley, 1979; Keltner & Robinson, 1997; Pruitt, 1995). Second, research on human information processing and decision making has found evidence for statusquo bias, the tendency for people to disproportionally stick with the status quo (e.g., Ritov & Baron, 1992). People generally prefer the current state of affairs because changing the status quo involves the need to formulate a new arrangement, transition costs, ambivalence, risk, and uncertainty (cf. Homans, 1974; Schweitzer, 1994). Third, theory on loss aversion postulates that changing the status quo entails gains and losses across different dimensions (Kahneman & Tversky, 1984; Ritov & Baron, 1992). Since losses weigh more heavily than commensurate gains, people will be inclined to favour the current state of affairs. In addition, because defendants face a potential loss while complainants face a potential gain, complainants will yield more easily than defendants (see De Dreu, Carnevale, Emans, & Van de Vliert, 1995). Following from the above, we expected a status quo effect: Defenders of the status quo are more likely to reach their goal than complainants during conflict over the division of labour. Thoits (1987) argued that women are
DIVISION OF FAMILY WORK 205
less successful than men in changing the status quo because of their power disadvantage, which is especially an economic disadvantage (cf. Blood & Wolfe, 1960). However, one can also argue that women are more successful in their attempt to change the status quo than men in the role of complainant because of a behavioural advantage. Demanding behaviour is likely to increase the chance of goal accomplishment (Babcock et al., 1993; Noller, Feeney, Bonnell, & Callan, 1994). Obviously, withdrawing behaviour on the part of the complainant strongly decreases the chance that the spouse will change his or her behaviour, thus enhancing status quo maintenance. Since women are more likely to demand than men, especially when they desire change, and men are more likely to withdraw than women, even when they desire change, we expected female complainants to be more successful in their attempt to change the status quo than male complainants. We tested this proposition in a study using the vignettes described previously (Kluwer et al., 2000, Study 2). Both members of 128 couples with a first child under 2 years of age were presented a vignette and asked to rate the likelihood of various conflict outcomes on a 7-point scale. Figure 4 shows that defendants were thought more likely to reach their goal than complainants during conflict over the division of labour, and that wives were thought more likely to reach their goal than husbands in the role of complainant. Further analyses showed that this effect occurred when the conflict concerned family work, which is considered the genderstereotypical domain of women. Kluwer (1998b) found further support for the status quo effect and showed a negative association between wife-demand/husband-withdraw interaction and the chance that the husband would contribute more to family work. This supports the notion that those in favour of the status quo are more likely to reach their goal due to their withdrawal. In addition, a longitudinal study among couples going through their transition to parenthood showed that the way spouses negotiated the division of labour did not contribute to actual changes in the division of labour in the long run (Kluwer, 1998a). In summary, the results imply that an asymmetrical conflict structure elicits asymmetrical behaviours such as demand/withdraw interaction and result in status quo maintenance. These findings are consistent with research on asymmetrical conflict outside the domain of close relationships. Kluwer and Nauta (2001) conducted two role-playing studies on collaboration and the division of work in student project groups, comparing symmetrical and asymmetrical conflict structures. They showed that complainants demonstrated more forcing and less problem solving than defendants in asymmetrical conflicts, whereas conflict parties showed matching behaviours in symmetrical conflicts. In addition, asymmetrical conflicts were more likely to end in a win/lose outcome than
206 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 4. Conflict outcomes as a function of complainant. Adapted from Kluwer et al. (2000a, Study 2), with permission.
symmetrical conflict. Conflict behaviours in symmetrical conflict primarily affected symmetrical outcomes, such as an integrative agreement and stalemate, whereas the behaviours of conflict parties in asymmetrical conflict primarily affected asymmetrical win/lose outcomes. Together, these results imply that asymmetrical conflict structures elicit asymmetrical behaviours and result in win/lose outcomes, while symmetrical conflict structures elicit matching behaviours and result in symmetrical outcomes. Gender-role ideology The last topic of this section concerns the moderating role of gender-role ideology in the relationship between discontent over the division of labour and marital interaction. Some authors have suggested that women would rather do more work to avoid conflict than push their husbands to contribute more to family work (Berheide, 1984; Hochschild, 1989; Pleck, 1985; see also Mederer, 1993). Wives may be unwilling to risk potentially negative consequences of conflict engagement. A plausible explanation for conflict avoidance lies in spouses’ gender-role ideology—how a person identifies himself or herself with regard to gendered roles. Research has revealed that less conflict engagement and open negotiation takes place in traditional marriages because there is consensus about roles, rules, and norms within the relationship (Sillars & Kalbflesch, 1989; cf. VanYperen & Buunk, 1991). Contesting traditional roles is difficult and may be considered socially inappropriate or unnatural (DeVault, 1990; West & Zimmerman, 1987). In addition, women in traditional relationships are
DIVISION OF FAMILY WORK 207
less inclined to confront their spouse because they feel less powerful (cf. Berheide, 1984; Hochschild, 1989; Mederer, 1993; Pleck, 1985) or because they feel discouraged by their traditional husband. Spouses with egalitarian gender-role ideologies need to explicitly negotiate gender roles because these roles have not yet been clearly established both within society and within the relationship. Spouses in egalitarian relationships tend to perceive each other as equals and engage in comparisons with the spouse (Buunk & VanYperen, 1989; VanYperen & Buunk, 1991), which results in feelings of deprivation and dissatisfaction among egalitarian women (Buunk et al., 2000). Thus, egalitarian women will ventilate their discontent more often in order to structure the relationship to their liking. Indeed, Greenstein (1996) found that the relationship between perceptions of unfairness and marital quality was stronger for women with egalitarian gender-role attitudes than for women with traditional views. Accordingly, we expected gender-role ideology to moderate the relationship between wives’ discontent over the division of labour and conflict avoidance. We tested this hypothesis in a survey among both members of 495 Dutch couples who were expecting their first child (Kluwer et al., 1997). The questionnaire measured, among other things, discontent over the division of labour, conflict avoidance during conflicts over the division of labour, and gender-role ideology. As expected, genderrole ideology moderated the relationship between discontent over the division of labour and conflict avoidance. Wives with a traditional ideology and wives with a traditional husband were more likely to avoid conflict when they experienced high levels of discontent over the division of labour than wives with egalitarian ideologies and wives with an egalitarian husband. These findings imply that traditional spouses tend to reinforce their division of labour even when the wife is dissatisfied. Thus, although traditional relationships may appear to be more peaceful than egalitarian relationships (cf. Sillars & Kalbflesch, 1989), this peacefulness may partially result from conflict avoidance. Logically, when discontent is not expressed and dealt with, the status quo remains unchanged, leaving the frustrated party discontented. Although egalitarian couples experience more conflict and negotiation, this also gives them the opportunity to work out their differences, provided that this is done in mutually integrative ways, as conflict can facilitate the reconciliation of spouses’ interests (Rubin, Pruitt, & Kim, 1994). SUMMARY AND CONCLUSIONS In this review, we offered a social psychological perspective on genderrelated inequalities in close relationships and integrated two lines of
208 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
research that have focused on the intrapersonal determinants of justice perceptions and on the interpersonal consequences of these personal experiences. In this last section, we will provide a summary of the findings and draw the main conclusions. It will further address the contributions of our research and point out directions for future research. Gender-related inequality and perceptionsof (in)justice The research on the determinants of perceptions of gender-related (in) justice can be summarised in three conclusions. First, notwithstanding the clear evidence of gender inequality in the division of domestic labour, a large majority of respondents does not regard the division of work as unfair. This shows that mere gender inequality is not a sufficient condition for feelings of injustice. Second, research on the determinants of women’s perceptions of injustice revealed that cognitions about domestic work and its division account for the largest proportion of variance of perceived justice, while measures of work load, system characteristics, and individual difference variables other than gender make a much smaller contribution. Whether or not women regard the division of labour as unfair depends on the interpersonal outcomes they get from family work, on the social comparisons they make and the respective comparison outcomes, and on their attributions of and justifications for the unbalanced division of labour. Third, the available evidence supports the usefulness of the distributive justice framework of Major (1993) and Thompson (1991) to understand women’s sense of (un)fairness regarding the division of domestic labour. More generally speaking, the reviewed studies lend support to social psychological models that emphasise the mediating role of cognitive processes in the perception of injustice, such as Crosby’s version of relative deprivation theory (Crosby, 1982) and the attributionof-blame model of judgements of injustice (Mikula, 1993, 2001). The reported findings suggest that it may be promising to develop a comprehensive theoretical model that integrates the distributive justice framework with complementary propositions of other models of perceptions of injustice. This research advanced our understanding of the role of social comparisons in perceptions of (in)justice. The finding that social comparison processes strongly shape justice perceptions is in line with distributive justice theories. Apparently, it is the outcome of the comparisons that directly affects perceived justice. The particular comparison referent with whom people choose to compare themselves or their spouse proved to be particularly important as it had an additional effect on the extent to which people perceive their situation as positive or negative. The complex interplay of the effects of the choice of comparison referent and the comparison outcomes on the perception of (in)justice
DIVISION OF FAMILY WORK 209
evidently deserves more attention in future research. In addition, individual difference variables that affect the predominance of making particular kinds of comparisons deserve to be studied systematically. The evidence that procedural elements contribute to perceived (in)justice of the division of family labour is another important finding. Since procedural justice has barely been considered in research on close relationships thus far, the evidence not only points to a hitherto neglected factor that contributes to justice evaluations of the division of domestic labour, but also to a new and promising research setting for procedural justice research. Until now, procedural justice research has paid little attention to how individual factors affect justice processes. The possibility that procedural justice may be particularly relevant to women or, more generally speaking, to those parties who have less power than their counterparts should be pursued in future studies. Consequences of gender-related discontent andperceived injustice The research on the consequences of gender-related dissatisfaction and injustice can be summarised in four conclusions. First, conflict over the division of labour in close relationships revolves around housework rather than paid work or child care, and originates from wives’ rather than husbands’ discontent. Accordingly, the typical conflict situation has an asymmetrical structure in which the wife is discontent with her husband’s contribution to housework, while her husband wants to maintain the status quo. This asymmetrical conflict structure supports the gender linkage in behaviour in close relationships—wives’ tendency to demand and husbands’ tendency to withdraw. Hence, the second conclusion, that the destructive wife-demand/husband-withdraw interaction pattern is a typical reaction to conflict over the division of labour. The asymmetrical conflict structure also affects the outcomes of conflict over the division of labour. The thirdconclusion is that, in general, those in favour of the status quo are more likely to reach their goal than those who desire change, but wives are more successful in accomplishing change than husbands when the conflict issue concerns their own genderstereotypical domain (i.e., family work). Fourth, traditional women are more inclined to avoid conflict over the division of labour when they experience high levels of discontent than egalitarian women. This research fills an important void in the literature on the division of labour by viewing men and women as active participants in the establishment, maintenance, or change of gender-based roles in close relationships, rather than viewing the division of labour as a fixed arrangement between spouses. Instead of a focus on sociological models that is more common in prior research, the present research illuminates the
210 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
psychological processes that underlie the construction of gender-based roles. The contribution of this research to the close relationships area is twofold. First, the division of family work is among the top six issues that cause relationship conflict and yet we know very little about the dynamics of these conflicts (cf. Kluwer, 2000). “Hot” issues have been relatively understudied in the domain of close relationships. Second, the need to negotiate gender roles is becoming an important feature of close relationships as our society is gradually becoming less gender-segregated and relationships are becoming more egalitarian. Women will increasingly perceive the traditional division of labour as unfair and demand changes in their husbands’ contributions. Couples need to explicitly negotiate gender roles, because they often lack a blueprint of rules and norms as they divide family work and combine their work and family responsibilities. This research contributes to the literature on conflict and negotiation by illuminating the impact of asymmetrical conflict structures on social interaction and outcomes. Aside from the work of Pruitt and colleagues (e.g., Peirce et al., 1993; Pruit, 1995), conflict research has largely neglected the distinction between complainants and defendants, implicitly assuming that both conflict parties have equal complaints against each other. The impact of asymmetrical conflict structures is present in any social interaction where one party desires change while the other party benefits from the status quo, such as interactions between employees and employers, minorities and majorities, or even between countries. Future research should further extend conflict research to asymmetrical conflicts. Finally, the present research also has several practical implications. First, this research showed that the division of housework is an issue that couples particularly struggle with. It also showed the importance of integrative communication about the division of labour and mutual decision making. Wives, in particular, need to have a “voice” in the decisions on how family work is divided because it enhances their satisfaction and perceptions of fairness in the household. Second, it is not conflict that is the problem, but rather couples’ response to conflict issues. Wife-demand/husband-withdraw interaction is a particular pitfall when couples manage conflict over the division of labour. When couples learn to recognise this interaction pattern in their relationship, they may be able to break through their stereotypical roles and bend wife-demand/husband-withdraw interactions into more constructive modes of behaviour. Limitations A few limitations to the foregoing require mention. First, the conclusions are based on data gathered through self-report methods. Although selfreports are an important source of information on subjective perceptions,
DIVISION OF FAMILY WORK 211
opinions, and feelings, research has questioned the accuracy of spouses as reporters on their own relationships. This concern is greater with conflict behaviour (cf. Kluwer, De Dreu, & Buunk, 1998). Ultimately, a combination of both self-report and, for example, observational methods would be the best approach to study conflict processes in close relationships. Second, as is common in research on the division of labour in close relationships, the majority of the reviewed studies have used correlational designs. Accordingly, conclusions about causal effects should be made with caution. Experimental studies are needed to corroborate the assumed causal relations. Conclusion We can conclude that the division of family work has proved to be a timely, ecologically valid, and challenging research setting for the study of justice perceptions and social conflicts. The research reviewed in this article has contributed to the cross-fertilisation of research on division of labour and research on social justice and social conflict respectively. Social psychological theories of justice and conflict have enriched the study of gender-related inequalities in the division of labour and improved our understanding of the factors that contribute to perceptions of injustice. In turn, social justice theories have been enriched, for example by the study of procedural justice in the domain of close relationships, and social conflict research can profit from the insights in asymmetrical conflict situations. REFERENCES Adams, J.S. (1965). Inequity in social exchange. Advances in Experimental Social Psychology, 2,267–299. Araji, S.K. (1977). Husbands’ and wives’ attitude-behaviour congruence on family roles. Journal of Marriage and the Family, 39,309–320. Avioli, P.S., & Kaplan, E. (1992). A panel study of married women’s work patterns. Sex Roles,26,227–242. Babcock, J.C, Waltz, J., Jacobson, N.S., & Gottman, J.M. (1993). Power and violence: The relationship between communication patterns, power discrepancies, and domestic violence. Journal of Consulting and Clinical Psychology, 61,40–50. Barnett, R.C., & Baruch, G.K. (1987). Determinants of fathers’ participation in family work. Journal of Marriage and the Family, 49,29–40. Belsky, J., Lang, M., & Huston, T.L. (1986). Sex typing and division of labour as determinants of marital change across the transition to parenthood. Journal of Personality and SocialPsychology, 50,517–522. Berheide, C.W. (1984). Women’s work in the home: Seems like old times. In B.B.Hess & M.B. Sussman (Eds.), Women and the family: Two decades of change(pp. 37–55). New York: Harworth.
212 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Berk, S.F. (1985). The gender factory.New York: Plenum Press. Blair, S.L. (1993). Employment, family, and the perception of marital quality among husbands and wives. Journal of Family Issues, 14,189–212. Blair, S.L. (1998). Work roles, domestic roles, and marital quality: Perceptions of fairness among dual-earner couples. Social Justice Research, 11,313–336. Blair, S.L., & Johnson, M.P. (1992). Wives’ perceptions of fairness of the division of household labour: The intersection of housework and ideology. Journal of Marriage and the Family, 54,570–581. Blood, R.O., & Wolfe, D.M. (1960). Husbands and wives.Glencoe, IL: Free Press. Brines, J. (1993). The exchange value of housework. Rationality and Society, 5,302– 340. Brockner, J., & Wiesenfeld, B.M. (1996). An integrative framework for explaining reactions to decisions: Interactive effects of outcomes and procedures. Psychological Bulletin, 120,189– 208. Buss, D.M. (1989). Conflict between the sexes: Strategic interference and the evocation of anger and upset. Journal of Personality and Social Psychology, 56, 735–747. Buunk, B.P., Kluwer, E.S., Schuurman, M., & Siero, F.W. (2000). The division of labour among egalitarian and traditional women: Differences in discontent, social comparison, and false consensus. Journal of Applied Social Psychology, 30,759–779. Buunk, B.P., & Van Yperen, N.W. (1989). Social comparison, equality, and relationship satisfaction: Gender differences over a ten-year period. Social Justice Research, 3,157–180. Buunk, B.P., & Van Yperen, N.W. (1991). Referential comparisons, relational comparisons and exchange orientation: Their relation to marital satisfaction. Personality and SocialPsychology Bulletin, 17,710–718. Chassin, L., Zeiss, A., Cooper, K., & Reaven, J. (1985). Role perceptions, self-role congruence, and marital satisfaction in dual-worker couples with preschool children. Social PsychologyQuarterly, 48,301–311. Chodorow, N. (1978). The reproduction of mothering: Psychoanalysis and the sociology ofgender.Berkeley: University of California Press. Christensen, A., & Heavey, C.L. (1990). Gender and social structure in the demand/ withdraw pattern of marital conflict. Journal of Personality and Social Psychology, 59,73–81. Coltrane, S., & Ishii-Kuntz, M. (1992). Men’s housework: A life-course perspective. Journal ofMarriage and the Family, 54,43–57. Coverman, S. (1985). Explaining husband’s participation in domestic labour. SociologicalQuarterly, 26,81–97. Cropanzano, R., & Greenberg, J. (1997). Progress in organizational justice: Tunneling through the maze. In C.L.Cooper & I.T.Robertson (Eds.), International review of industrial andorganizational psychology(pp. 317–372). New York: Wiley. Crosby, F. (1982). Relative deprivation and working women.New York: Oxford University Press. Cross, S.E., & Madson, L. (1997). Models of the self: Self-construals and gender. PsychologicalBulletin, 122,5–37.
DIVISION OF FAMILY WORK 213
Crouter, A.C., Perry-Jenkins, M., Huston, T.L., & McHale, S.M. (1987). Processes underlying father involvement in dual-earner and single-earner families. Developmental Psychology, 23,431–440. De Dreu, C.K.W., Carnevale, P.J.D., Emans, B.J.M., & Van de Vliert, E. (1995). Outcome frames in bilateral negotiation: Resistance to concession making and frame adoption. European Review of Social Psychology, 6,97–125. DeMaris, A., & Longmore, M.A. (1996). Ideology, power, and equity: Testing competing explanations for the perception of fairness in household labour. Social Forces, 74,1043– 1071. Demo, D.H., & Acock, A.C. (1993). Family diversity and the division of domestic labor: How much have things really changed? Family Relations, 42,323–331. Deutsch, F.M., Lussier, J.B., & Servis, L.J. (1993). Husbands at home: Predictors of paternal participation in child care and housework. Journal of Personality and Social Psychology, 65,1154–1166. Deutsch, M. (1973). The resolution of conflict: Constructive and destructive processes.New Haven, CT: Yale University Press. DeVault, M.L. (1990). Conflict over housework: The problem that (still) has no name. In L. Kriesberg (Ed.), Research in social movements, conflict and change (pp. 189–202). Greenwood, CT: JAI Press. Eagly, A.H. (1987). Sex differences in social behaviour: A social-role interpretation.Hillsdale, NJ: Lawrence Erlbaum Associates Inc. England, P., & McCreary, L. (1987). Integrating sociology and economics to study gender at work. In A.H.Stromberg, L.Larwood, & B.A.Gutek (Eds.), Women and work: An annualreview, 2.Beverly Hills, CA: Sage. Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7, 117–140. Folger, R., Rosenfield, D., Grove, J., & Corkran, L. (1979). Effects of “voice” and peer opinions on responses to inequity. Journal of Personality and Social Psychology, 37,2253– 2261. Freudenthaler, H.H., & Mikula, G. (1998). From unfulfilled wants to the experience of injustice: Women’s sense of injustice regarding the lopsided division of household labour. Social Justice Research, 11,289–312. Geerken, M., & Gove, G.R. (1983). At home and at work: The family’s allocation of labour.Beverly Hills, CA: Sage. Gottman, J.M., & Levenson, R.W. (1988). The social psychophysiology of marriage. In P. Noller & M.A.Fitzpatrick (Eds.), Perspectives on marital interaction(pp. 182–200). Philadelphia: Multilingual Matters. Gray-Little, B., & Burks, N. (1983). Power and satisfaction in marriage: A review and critique. Psychological Bulletin, 93,513–538. Greenstein, T.N. (1996). Husband’s participation in domestic labour: Interactive effects of wives’ and husbands’ gender ideologies. Journal of Marriage and the Family, 58,585–595. Grote, N.K., & Clark, M.S. (2001). Perceiving unfairness in the family: Cause or consequence of marital distress? Journal of Personality and Social Psychology, 80,281–293. Hackel, L., & Ruble, D.N. (1992). Changes in the marital relationship after the first baby is born: Predicting the impact of expectancy disconfirmation. Journal of Personality and SocialPsychology, 62,944–957.
214 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Hatfield, E., Traupmann, J., Sprecher, S., Utne, M., & Hay, J. (1985). Equity and intimate relations: Recent research. In W.Ickes (Ed.), Compatible and incompatible relationships(pp. 91–117). New York: Springer-Verlag. Hawkins, A.J., Marshall, C.M., & Meiners, K.M. (1995). Exploring wives’ sense of fairness about family work. Journal of Family Issues, 16,693–721. Heavey, C.L., Layne, C., & Christensen, A. (1993). Gender and conflict structure in marital interaction: A replication and extension. Journal of Consulting and Clinical Psychology, 61,16–27. Heider, F. (1958). The psychology of interpersonal relations.New York: Plenum Press. Hochschild, A. (1989). The second shift: Working parents and the revolution at home.New York: Viking Press. Holmes, J.G., & Murray, S.L. (1996). Conflict in close relationships. In T.Higgins & A. Kruglanski (Eds.), Social psychology: A handbook of basic principles(pp. 622–654). New York: Guilford Press. Homans, G.C. (1974). Social behaviour: Its elementary forms(2nd ed.). New York: Harcourt , Brace, & World. Jacobson, N.S. (1989). The politics of intimacy. The Behaviour Therapist, 12,29– 32. John, D., Shelton, B.A., & Luschen, K. (1995). Race, ethnicity, gender and perceptions of fairness. Journal of Family Issues, 16,357–379. Kahneman, D., & Tversky, A. (1984). Choices, values, and frames. American Psychologist, 39,341–350. Kalleberg, A.L., & Rosenfeld, R.A. (1990). Work in the family and in the labour market: A cross-national, reciprocal analysis. Journal of Marriage and the Family, 52,331–346. Kelley, H.H. (1979). Personal relationships: Their structures and process.Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Kelley, H.H., & Thibaut, J.W. (1978). Interpersonal relationships: A theory of interdependence.New York: Wiley. Keltner, D., & Robinson, R.J. (1997). Defending the status quo: Power and bias in social conflict. Personality and Social Psychology Bulletin, 23,1066–1077. Klinetob, N.A., & Smith, D.A. (1996). Demand-withdraw communication in marital interaction: Tests of interspousal contingency and gender role hypotheses. Journal ofMarriage and the Family, 58,945–957. Kluwer, E.S. (1997). Taakverdeling tussen partners rond de komst van het eerste kind [The division of labour across the transition to parenthood]. Unpublished research report University of Groningen. Kluwer, E.S. (1998a). Marital conflict over the division of labour: When partners become parents. Unpublished doctoral dissertation, University of Groningen. Kluwer, E.S. (1998b). Responses to gender inequality in the division of family work: The status quo effect. Social Justice Research, 11,337–357. Kluwer, E.S. (2000). Marital quality. In R.M.Milardo & S.W.Duck (Eds.), Families asrelationships(pp. 59–78). Chichester, UK: Wiley. Kluwer, E.S., Chin-A-Fat, B.E.S., & Wissink, I. (2002a). Procedural and distributive justice in close relationships: The moderating role of gender. Manuscript in preparation.
DIVISION OF FAMILY WORK 215
Kluwer, E.S., De Dreu, C.K.W., & Buunk, B.P. (1998). Conflict in intimate versus nonintimate relationships: When gender role stereotyping overrides biased selfother judgment. Journal of Social and Personal Relationships, 15,637–650. Kluwer, E.S., Heesink, J.A.M., & Van de Vliert, E. (1996). Marital conflict about the division of household labour and paid work. Journal of Marriage and the Family, 58,958–969. Kluwer, E.S., Heesink, J.A.M., & Van de Vliert, E. (1997). The marital dynamics of conflict over the division of labour. Journal of Marriage and the Family, 59, 635–653. Kluwer, E.S., Heesink, J.A.M., & Van de Vliert, E. (2000). The division of labour in close relationships: An asymmetrical conflict issue. Personal Relationships, 7,263–282. Kluwer, E.S., Heesink, J.A.M., & Van de Vliert, E. (2002b). The division of labor across the transition to parenthood: A justice perspective. Journal of Marriage and the Family, 64,930– 943. Kluwer, E.S., & Nauta, A. (2001). It’s all a matter of (a) symmetry: The impact of conflict structure on conflict behaviour and outcomes. Manuscript in preparation. Layte, R. (1998). Gendered equity? Comparing explanations of women’s satisfaction with the domestic division of labour. Work, Employment & Society, 12,511–532. Lennon, M.C., & Rosenfield, S. (1994). Relative fairness and the division of housework: The importance of options. American Journal of Sociology, 100, 505–531. Lind, E.A., Kanfer, R., & Earley, P.C. (1990). Voice, control, and procedural justice: Instrumental and non-instrumental concerns in fairness judgements. Journal of Personalityand Social Psychology, 59,952–959. Lind, E.A., & Tyler, T.R. (1988). The social psychology of procedural justice.New York: Plenum Press. Maccoby, E.E., & Jacklin, C.N. (1987). Gender segregation in childhood. In H.W.Reese (Ed.), Advances in child development and behaviour, 20(pp. 239– 288). New York: Academic Press. Major, B. (1987). Gender, justice, and the psychology of entitlement. In P.Shaver & C. Hendrick (Eds.), Review of personality and social psychology: Vol. 7. Sex and gender(pp. 124–148). Beverly Hills, CA: Sage Publications Inc. Major, B. (1993). Gender, entitlement, and the distribution of family labour. Journal of SocialIssues, 49,141–159. Major, B., & Deaux, K. (1982). Individual differences in justice behaviour. In J.Greenberg &R. L.Cohen (Eds.), Equity and justice in social behaviour(pp. 43–76). New York: Academic Press. Menaghan, E.G., & Parcel, T.L. (1990). Parental employment and family life: Research in the 1980s. Journal of Marriage and the Family, 52,1079–1098. Mederer, H.J. (1993). Division of labour in two-earner homes: Task accomplishment versus household management as critical variables in perceptions about family work. Journal ofMarriage and the Family, 55,133– 145. Mikula, G. (1993). On the experience of injustice. European Review of Social Psychology, 4,223– 244.
216 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Mikula, G. (1998). Division of household labour and perceived justice: A growing field of research. Social Justice Research, 11,215–242. Mikula, G. (2001a). Social psychological perspectives on justice. In N.J.Smelser & P.B.Baltes (Eds.), International encyclopedia of the social and behavioural sciences.London: Pergamon Press. Mikula, G. (in press). Testing an attribution-of-blame model of judgments of injustice. European Journal of Social Psychology. Mikula, G., Athenstaedt, U., Heschgl, S., & Heimgartner, A. (1998). Does it only depend on the point of view? Perspective-related differences in justice evaluations of negative incidents in personal relationships. European Journal of Social Psychology, 28,931–962. Mikula, G., & Freudenthaler, H.H. (1999). Aufteilung von Familienarbeit als Gegenstand von Ungerechtigkeitswahrnehmungen und sozialen Konflikten: Zur Bedeutung von sozialen Vergleichen [Division of family work as a subject of perceptions of injustice and social conflicts: The role of social comparisons]. In W.Lutz (Ed.), Compendium of family studies inAustria 1999.Wien: Osterreichisches Institut fuer Familienforschung. Mikula, G., & Freudenthaler, H.H. (2002). Division of tasks and duties and the perception of injustice: The case of household chores. Psychologische Beitraege, 44,567–584. Mikula, G., Freudenthaler, H.H., Brennacher-Kröll, S., & Brunschko, B. (1997a). Division of household labour, perceived justice, and satisfaction in studenthouseholds. Basic andApplied Social Psychology, 19,275–289. Mikula, G., Freudenthaler, H.H., Brennacher-Kröll, S., & Schiller-Brandl, R. (1997b). Arrangements and rules of distribution of burdens and duties: The case of household chores. European Journal of Social Psychology, 27,189– 208. Mikula, G.Freudenthaler H.H.Hörtner G.Foufas C. (2001). Soziale Vergleiche undGerechtigkeitsbeurteilungen der Aufteilung der Haushaltsarbeit zwischen Partnern.[Social comparisons and evaluations of justice of the division of household labor between partners]. Paper presented at the 8th Tagung der Fachgruppe Sozialpsychologie. Würzberg, Germany, September, 23–26. Mikula, G., & Lerner, M.J. (1994). Justice in close relationships: An introduction. In M.J. Lerner & G.Mikula (Eds.), Entitlement and the affectional bond: Justice in closerelationships.New York: Plenum Press. Mikula, G., Petri, B., & Tanzer, N.K. (1990). What people regard as unjust: Types and structures of everyday experiences of injustice. European Journal of Social Psychology, 20,133–149. Miller, J.B. (1986). Toward a new psychology of women.Boston, MA: Beacon Press. Noller, P., Feeney, J.A., Bonnell, D., & Callan, V.J. (1994). A longitudinal study of conflict in early marriage. Journal of Social and Personal Relationships, 11, 233–252. Peirce, R.S., Pruitt, D.G., & Czaja, S.J. (1993). Complainant-respondent differences in procedural choice. International Journal of Conflict Management, 4,199– 222.
DIVISION OF FAMILY WORK 217
Perry-Jenkins, M., & Folk, K. (1994). Class, couples, and conflict: Effects of the division of labour on assessments of marriage in dual-earner families. Journal of Marriage and theFamily, 56,165–180. Pittman, J.F., & Blanchard, D. (1996). The effects of work history and timing of marriage on the division of household labour: A life-course perspective. Journal of Marriage and theFamily, 58,78–90. Pleck, J. (1985). Working wives/working husbands.Beverly Hills, CA: Sage. Pruitt, D.G. (1995). Process and outcome in community mediation. Negotiation Journal, 11,365–377. Ritov, L, & Baron, J. (1992). Status-quo and omission biases. Journal of Risk and Uncertainty,5, 49–61. Robinson, J., & Spitze, G. (1992). Whistle while you work? The effect of household task performance on women’s and men’s well-being. Social Science Quarterly, 73,844–861. Rubin, J.Z., Pruitt, D.G., & Kim, S.H. (1994). Social conflict: Escalation, stalemate, andsettlement.New York: McGraw-Hill. Sanchez, L. (1994). Gender, labor allocations, and the psychology of entitlement within the home. Social Forces, 73,533–553. Schweitzer, M. (1994). Disentangling status quo and omission effects: An experimental analysis. Organizational Behaviour and Human Decision Processes, 58,457–476. Shaver, K.G. (1985). The attribution of blame: Causality, responsibility, and blameworthiness.New York: Springer. Shelton, B.A., & John, D. (1996). The division of household labour. Annual Review ofSociology, 22,299–322. Sillars, A.L., & Kalbflesch, P.J. (1989). Implicit and explicit decision-making styles in couples. In D.Brinberg & J.Jaccard (Eds.), Dyadic decision-making(pp. 179– 215). New York: Springer-Verlag. Sprecher, S. (1986). The relationship between inequity and emotions in close relationships. Social Psychology Quarterly, 49,309–321. Steil, J.M. (1997). Marital equality. Its relationship to the well-being of husbands and wives.Thousand Oaks, CA: Sage Publications. Sweeney, P.D., & McFarlin, D.B. (1997). Process and outcome: Gender differences in the assessment of justice. Journal of Organizational Behaviour, 18,83–98. Thibaut, J.W., & Walker, L. (1975). Procedural justice: A psychological analysis.Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Thoits, P.A. (1987). Negotiating roles. In F.J.Crosby (Ed.), Spouse, parent, worker: On genderand multiple roles(pp. 11–22). New Haven, CT: Yale University Press. Thompson, L. (1991). Family work: Women’s sense of fairness. Journal of Family Issues, 12,181–196. Törnblom, K.Y. (1992). The social psychology of distributive justice. In K.R.Scherer (Ed.), Justice: Interdisciplinary perspectives(pp. 177–236). Cambridge: Cambridge University Press. Tyler, T.R., & Smith, H.J. (1998). Social justice and social movement. In D.Gilbert, S.T. Fiske, & G.Lindzey (Eds.), Handbook of social psychology(4th ed., pp. 595–629). New York: McGraw-Hill.
218 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
United Nations (1995). The world’s women 1995: Trends and statistics.New York: United Nations. Van den Bos, K., Lind, E.A., Vermunt, R., & Wilke, H.A.M. (1997). How do I judge my outcome when I do not know the outcome of others? The psychology of the fair process effect. Journal of Personality and Social Psychology, 72, 1034–1046. VanYperen, N.W., & Buunk, B.P. (1991). Sex-role attitudes, social comparison, and satisfaction with relationships. Social Psychology Quarterly, 54,169–180. Walker, L., LaTour, S., Lind, E.A., & Thibaut, J. (1974). Reactions of participants and observers to modes of adjudication. Journal of Applied Social Psychology, 4,295–310. Walster, E., Walster, G.W., & Berscheid, E. (1978). Equity: Theory and research.Boston: Allyn & Bacon. Ward, R.A. (1993). Marital happiness and household equity in later life. Journal of Marriageand the Family, 55,427–438. West, C., & Zimmerman, D.H. (1987). Doing gender. Gender and Society, 1,125– 151. Wood, J.V. (1989). Theory and research concerning social comparisons of personal attributes. Psychological Bulletin, 106,231–248.
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY, 2002, 13, 217–258
Does scientific thinking lead to success and sanity? Anintegration of attribution and attributional models Friedrich Försterling Ludwig-Maximilians-Universität München, Germany
A motivational model is introduced that integrates conceptions of attributional antecedents with those of attributional consequences, and suggests that it is functional to make realistic attributions. This model is applied to the analysis of achievement behaviour and depression. The use of a new experimental paradigm designed to test the model’s assumptions demonstrates that realistic attributions maximise performance. The model was also used to investigate whether depressives are more realistic than non-depressives. This research revealed that depressives draw more “pessimistic” conclusions from covariation information than non-depressives. In addition, depressives expect their partners to make more negative attributions about the depressives’ outcomes than their partners in fact do. However, these negative views could not be traced back to faulty inferences from covariation information. Finally, research is described showing that, in addition to depressogenic attributions, certain evaluative cognitions need to be present in order to trigger depressive mood reactions.
Models of the antecedents of causal attributions (i.e., attribution models) are based on rational principles in that they assume that individuals are
© 2002 European Association of Experimental Social Psychology http://www.tandf.co.uk/journals/pp/02699931.html DOI: 10.1080/10463280240000073
220 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
motivated to gain a realistic understanding of events. However, these models rarely consider the implications of causal cognitions for motivation, emotion, and action. By contrast, attributional conceptions of motivation (i.e., attributional models) are based on hedonistic principles. They specify the consequences of causal thoughts on motivation, emotion and action but neglect attributional antecedents and they do not adhere to the “naive scientist” metaphor. In this chapter, I will introduce an integrative motivational model that combines theoretical assumptions and research findings on attributional antecedents and consequences. It is based on the premise that it is functional (hedonism) to make realistic (rational) attributions. The model is applied to the analysis of achievement behaviour and depression.
ANTECEDENTS AND CONSEQUENCESOF ATTRIBUTIONS Antecedents of causal attributions Many of the most influential psychological theories assume that individuals are motivated to gain a realistic understanding of the causes of events and especially of the attributes of their social environment and their own traits. For instance, the desire to explain and understand is central to social comparison theory (Festinger, 1954), Kruglanski’s theory of lay epistemology (Kruglanski, 1980) and, especially, in attribution theory as articulated by Heider (1958) and Kelley (1967, 1973; for a summary, see Försterling, 2001). Attribution theory not only assumes that individuals are motivated to understand and explain, but postulates that the layperson can be conceived of as a “naive scientist” in that he or she uses tools that are comparable to the ones used by “professional” scientists. In the time of drive reduction theories and behaviourism during the 1950s, the conception of humans as “naive scientists” who are motivated to understand, predict, and control was revolutionary (see Weiner, 1992). Presently, however, this conception constitutes the dominant metaphor of psychological research in many areas of psychology. For instance, in
Address correspondence to: Friedrich Försterling, Department of Psychologie, Ludwig-Maximilians-Universität, Leopoldstr. 13, 80803 München, Germany, Email:
[email protected] The research reported here was supported by the Deutsche Forschungsgemeinschaft, Grant FOR 204/2–3. I would like to thank Miles Hewstone, Wolfgang Stroebe, Matthis Morgenstern, and three anonymous reviewers for valuable comments on an earlier draft of this chapter.
INTEGRATION OF ATTRIBUTION MODELS 221
contemporary approaches to learning (Rescorla, 1988; Shanks, Medin, & Holyoak, 1996), developmental (Carey & Gelman, 1991; Wilkening & Lamsfuß, 1993), and social psychology (see for instance, Trope & Liberman, 1996). All these areas of psychological research are guided by the question of how individuals arrive at hypotheses, how these hypotheses are examined and revised, and how these “naive” inferences overlap or diverge from statistical parameters such as correlation coefficients or F-ratios. Keywords such as “the scientist in the crib” (Gopnik, Meltzoff, & Kuhl, 1999) or “the animal as a naive scientist” (Rescorla, 1988) convincingly document that many areas of psychological research assume that starting (at least) at birth, it is a central concern of humans (and possibly also of animals) to develop and test theories about the personal domain. Attribution theorists (Heider, 1958; Kelley, 1967) have suggested that the process of making causal inferences can be described by the covariation principle (i.e., the method of difference Mill’s, 1872). This principle also constitutes the essence of how theories are tested in science and suggests that “an effect will be attributed to the cause that is present when the effect is present and absent when the effect is absent” (see Heider, 1958). For instance, we trace back the success (effect) of a person at a task to personal attributes (e.g., her high ability) and not to the entity (e.g., the ease of the task) when only this person (low consensus) succeeds at this task with high consistency and succeeds at other tasks as well (low distinctiveness). In this case, the effect covaries with the person and not with the entity. If, however, everybody succeeded at the task (high consensus) at which the person succeeds, with high consistency and the person fails at other tasks (e.g., success occurs with a high distinctiveness), the effect covaries with the entity and not with the person and we are—according to the covariation principle—inclined to (and in fact do) attribute the success to the entity (e.g., the ease of the task) and not to the person (e.g., her high ability). Empirical tests of Heider’s and Kelley’s covariation model as well as their revisions and refinements have shown that individuals use available covariation information and arrive at attributions that are predicted by the covariation principle (for a summary, see Försterling, 1989). Hence, we can summarise that many contemporary approaches to psychology in general and attribution theory in particular assume that individuals are motivated to gain cognitive mastery and that their attempts to understand and explain are characterised by the (possibly imperfect) use of the methods used in science. Consequences of causal attributions Compared to the large amount of research concerned with the process of acquiring and changing (causal) knowledge and hypotheses, we have many fewer insights about how the acquired (causal) knowledge determines and
222 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
influences our behaviour, motivation, and emotions. This is surprising, especially as many of the contemporary models of cognitive processes use an evolutionary perspective (e.g., Cosmides, 1989; Gigerenzer & Hug, 1992; for a summary, see Buss, 1995) and, from this perspective, it is obvious that knowledge by itself, without being translated into action, cannot have an advantage in selection. During the first decade of attribution research, interest in the cognitive processes of the “naive scientist” was also limited to the antecedents of causal perceptions and inferences and largely neglected the motivational, emotional, and behavioural consequences of causal cognitions. However, by the early 1970s the first attributional models of motivation, emotion, and behaviour were introduced (see Weiner, Frieze, Kukla, Reed, Rest, & Rosenbaum, 1971). Although these models were also guided by the metaphor of “the naive scientist”, and hence contrasted to the thus far dominant behaviouristic and drive-reduction theories, they still relied heavily on hedonistic mechanisms and were largely unconnected with the (rational) theories about the antecedents of (causal) cognitions. The attributional analysis of achievement behaviour. Heider’s (1958) and Kelley’s (1967) attribution models guided Weiner’s (Weiner, 1986; Weiner et al., 1971) research programmes on the cognitive, affective, and behavioural consequences of causal attributions in achievement contexts. Weiner’s attributional analysis of achievement behaviour, in turn, guided the attributional analysis of depression and learned helplessness in the late 1970s and the 1980s (Abramson, Seligman, & Teasdale, 1978; for a summary, see Meyer, 2000), the attributional analysis of social motivation in the 1990s (for a summary, see Weiner, 1995) and more recent models of achievement behaviour that refer to self-concepts of ability, especially Meyer’s (1984) and Dweck’s (1999; Dweck, Chiu, & Hong, 1995) theoretical conceptions. In his attributional analysis of achievement behaviour, Weiner (for a summary, see 1986) has shown that emotional reactions following success and failure (e.g., pride and shame) are determined by the locus dimension of causal attributions. Success that is traced back to ability (an internal factor) gives rise to pride more so than attributions to the ease of the task (an external factor). And following failure, the experience of the emotion of shame requires internal attributions as well. Causal explanations also determine changes of expectancy. Failure that is attributed to stable factors, such as low ability or the difficulty of the task, reduces perceived chances of succeeding at the respective task at a later point in time more so than failure that is thought to be caused by variable causes such as chance or lack of effort. Furthermore, Weiner (1986) assumes that attributionally mediated affects and expectancies influence subsequent achievement behaviour. Internal and stable attributions for failure and external variable ascriptions for success are thought to be especially detrimental for
INTEGRATION OF ATTRIBUTION MODELS 223
achievement outcomes as they maximise negative, and minimise positive, emotional reactions following the outcome and they minimise subsequent expectancies of success. By contrast, ascription of failure to external and variable causes and of success to internal stable ones should maximize positive emotions and success expectancies and minimise negative emotions following the outcome, and hence have a favourable impact on achievement behaviour. These assumptions clearly refer to hedonistic mechanisms, as the maximisation of expected affect is conceived of as the central motivational mechanism for achievement behaviour. In spite of the general hedonistic nature of the attributional model of achievement motivation, however, this approach has also referred to nonhedonistic principles. More specifically, Weiner et al. (1971) postulated that the phenomenon of task choice is guided by the desire to gain accurate information about the self. And Trope (1975; Trope & Brickman, 1975) introduced an experimental paradigm which unconfounded the hedonic and informational value of tasks of intermediate difficulty. He showed that task choice is motivated by the desire to reduce uncertainty regarding one’s ability and that individuals’ general preference for tasks of intermediate difficulty reflects the fact that outcomes at these tasks are maximally diagnostic of one’s ability. Failure at easy tasks and success at very difficult tasks is attributed to chance, and success at easy tasks and failure at very difficult tasks to task characteristics. Only outcomes at tasks of intermediate difficulty are attributed to and considered to be diagnostic of ability, and this difficulty level is in fact preferred. The attributional analysis of learned helplessness and depression. In their attributional reformulation of learned helplessness theory, Abramson et al. (1978) make assumptions similar to Weiner’s (1986) and propose that individuals who experience uncontrollability ask why they were unable to control a particular event. The answer to this “why” question (the causal attribution) should influence the subsequent expectancy of controllability and thus determine important aspects of helplessness and depression. The causal dimension of locus of control determines whether self-esteem problems arise as a consequence of failure. Failure traced back to internal factors may lead to doubts about self-worth (i.e., personal helplessness) whereas, given an external attribution for failure, self-esteem problems are not to be expected (universal helplessness). The dimension “stability over time” determines the temporal duration of helplessness. If an individual attributes uncontrollability to stable factors, the symptoms of helplessness should last for a long time because one does not anticipate any change in the unfortunate situation (i.e., chronic helplessness results). If unstable causes are used to explain noncontingency, the resulting helplessness should only last for a comparatively short time (i.e., temporary helplessness).
224 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
In addition to the dimensions of locus of control and stability, Abramson et al. (1978) introduce the dimension of generality with the poles labelled “global” and “specific”. If noncontingency is explained by a global factor, which is characterised by the perception that it influences a wide range of other situations, helplessness will spread to a larger area of life (i.e., global helplessness). However, if the failure is attributed to a specific cause, there should be no negative effects with regard to other tasks (i.e., specific helplessness). According to Peterson and Seligman (1984), the causal attributions that a person makes for uncontrollability depend on situational and personality factors. The situational factors are only briefly described as “the reality of bad events themselves” (p. 349). Peterson and Seligman (1984) introduce the example of the “reality” of the loss of a spouse which implies “stability” (the partner will not come back) and “globality” (this influences many areas of life). They do not, however, identify the crucial aspects of the situations that will lead to different attributions. The personality disposition that determines which causal explanation is used by an individual in a particular situation is described as “attributional style” or “explanatory style”. Abramson et al. (1978) postulate that an attributional style involving the use of internal, stable, and global attributions for negative events increases the probability of becoming depressed after the occurrence of such events. (Such a “depressogenic” attributional style should also include the use of the opposite causal attributions for success; namely, external, specific, and variable.) From these assumptions, Abramson et al. (1978) derive that both the prevention and therapeutic treatment of reactive depression should consist of reducing internal, stable, and global attributions for failure, and external, specific, and variable explanations for success. Weiner’s attributional analysis of achievement behaviour and Abramson et al.’s application of these ideas to learned helplessness and depression were models for more recent cognitive theories of motivation and emotion. For instance, Dweck (e.g., Dweck et al., 1995) suggests that the belief that one’s intelligence is stable and unmodifiable (i.e., an entity theory of intelligence) leads to negative consequences following failure (e.g., lack of persistence), whereas the assumption that one can modify and increase one’s intelligence (i.e., an incremental theory of intelligence) has a favourable impact on achievement behaviour.
INTEGRATION OF ATTRIBUTION MODELS 225
THE RELATION BETWEEN ATTRIBUTIONAL MODELS OF MOTIVATION AND CONCEPTIONS OFTHE ANTECEDENTS OF CAUSAL COGNITIONS Within the general S (Stimulus)—C (Cognition)—R (Reaction) framework of cognitive psychology, the (hedonistic) attributional models just described (i.e., Abramson et al., 1978; Dweck, 1999; Weiner, 1986) exclusively address the cognitive, affective, and behavioural consequences (R) of causal attributions (C) and largely neglect their situational antecedents (S) which are addressed by the (rational) models of, e.g., Heider and Kelley. Similarly, as already indicated, models of the antecedents of attributions also neglect the behavioural consequences of causal thoughts (see Försterling, 1986; Schuster, Rudolph, & Försterling, 1998). As we shall specify in the following sections, the full (S-C-R) sequence has not yet been explored by attribution or attributional research: Social perception is studied without reference to (social) action and (social) action is studied without reference to social perception. Or, to put it more formally: Investigations of the antecedent conditions of causal attributions typically vary as the independent variable situational cues (S; such as covariation information according to Kelley, 1967) and assess, as dependent variable, causal inferences (C; attributions). Attributional research, on the other hand, varies, as independent variables, causal cognitions (C) and assesses, as dependent variable, cognitive, emotional, or behavioural reactions (R). Integrations of the results of these two research approaches, however, hardly exist. In the following, we will show that the lack of integration between the (rational) models of the antecedents of attributions and the (hedonistic) conceptions of attributional determinants of motivation and emotion is possibly responsible for some shortcomings and inconsistencies of attributional models of action. In addition, we will introduce an integration of models concerning antecedent conditions with those concerning the consequences of attributions, and will describe a series of studies to test the integrative model. Shortcomings of attributional models ofmotivation, emotion, and action The lack of integration of attributional conceptions with models about the antecedents of attributions might explain why attributional models suggest that there is a certain way to think about successes and failures that is maladaptive and another way that is desirable and adaptive. For instance, Weiner’s attributional analysis of achievement behaviour suggests that attributions of failure to lack of ability trigger unfavourable consequences for the attributor: decreased expectancies of success, self-debilitating emotions such as shame and, as a consequence, lack of persistence resulting
226 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
in decreased performance. By contrast, attributions of failure to chance should result in desirable reactions following failure: comparatively high expectations of success, modulated negative emotions and, consequently, high persistence and performance. Similarly, Abramson et al. (1978) hypothesise that tracing back negative events to internal, stable, and global causes (such as lack of ability) can lead to helplessness and depression, whereas attribution of negative events to external, variable, and specific causes protects the individual from such reactions. And finally, Dweck et al. (1995) suggest that assuming that one’s intelligence is modifiable leads to positive consequences, whereas the naive theory that intelligence is stable typically has detrimental consequences. As reduced persistence, low performance levels, and helpless reactions are often conceived of as unproductive and disadvantageous, these reactions have been selected as targets of therapeutic or training interventions in clinical and educational practice. The attributional analyses of achievement behaviour and depressive reactions have in fact sparked training programmes designed to teach participants to give up the causal explanations that have been identified by the respective model as maladaptive and to foster ways of causal thinking that have been identified as adaptive (see Försterling, 1985, for a summary). It is expected that behavioural changes would accompany these cognitive modifications. More specifically, Weiner’s (1986) model suggests that, in order to reduce maladaptive achievement patterns, individuals should be taught to attribute failure not to low ability but, instead, to bad luck or lack of effort. According to Abramson et al. (1978), depressives should be taught to attribute failure to external, variable, and specific causes rather than to internal, stable, and global ones. The logic that guides these training or therapy studies might be summarised as follows: (A) Certain “non-desirable” reactions are caused by certain causal attributions. (B) Other reactions that are comparatively more “desirable” are caused by other causal attributions. (C) Therefore, it follows: If one wants to ameliorate the “non-desirable” and to foster the “desirable” reactions, one needs to eliminate the cognitions leading to the undesirable reactions and take on the cognitions that give rise to the “desirable” reactions. This logic implies that the consequence of a cognition (e.g., whether it leads to helplessness or persistence) is used to judge its “changeworthiness”. However, this logic, can also lead to untenable deductions. For instance, suppose a person feels depressive as soon as he believes that he has less than 100 dollars in his wallet. Further assume that that person feels happy as soon as he believes he has 100 dollars in his wallet. Now let us apply the logic outlined above to this scenario: (A) the belief that one does not have 100 dollars leads to the undesirable state of depression. (B) the belief that one does have 100
INTEGRATION OF ATTRIBUTION MODELS 227
dollars leads to happiness, (C) one should teach this person to always believe that he possesses 100 dollars in order not to feel depressed. Intuitively, however, it does not seem reasonable to always assume that one has 100 dollars, or, to return to attributional retraining studies, to avoid attributions to lack of ability. Under certain conditions it seems quite reasonable to assume that one’s wallet is empty (or to trace back failure to low ability). This seems to be the case when one does not in fact possess the respective finances (or abilities). An overestimation of one’s finances (or abilities) might tempt an individual into buying too many goods or might prevent her or him from going to the bank. And an overestimation of their abilities might lead individuals to engage in risky situations that they cannot master or to invest too much effort in unsolvable tasks, and to fail to ask for help. It is worth pointing out that the attributional theories described above do not specify conditions in which it might be desirable to make failure attributions to internal, stable, and global causes (e.g., lack of ability) and under which conditions the reactions that are triggered by, e.g., lack of ability attributions, have positive effects. They judge the appropriateness of a cognition alone on the basis of the consequences of the cognition on the behaviours that are addressed by the specific theory (e.g., depression or lack of persistence). In these models, the changeworthiness of a cognition is not judged on the basis of the extent to which a cognition reflects the situation or “reality” (e.g., whether there is money in the wallet or not, or whether one really possesses ability). In contrast to attributional models, theories about the antecedents of causal attributions assume that it is a basic desire of humans to strive to be accurately informed about the attributes of the self and that a realistic knowledge of the causes of events is highly adaptive. These theories are based on the view of humans as rational and not as (primarily) hedonistic. The attributional models were guided by these theories about attributional antecedents but, interestingly, they do not consequently take up the rational position and consider certain motivational, emotional, or behavioural consequences (e.g., low expectations of success or negative affect) following failure as, a priori, maladaptive (a hedonistic view). The same holds true for motivational conceptions that are based on selfperceptions of ability (e.g., Dweck et al., 1995; Kukla, 1972; Meyer, 1984;): The rational approach as advocated by the theories of the antecedents of attributions suggests that human judgements (should) closely parallel the judgements of scientists. Hence, as abilities with regard to a certain task are normally distributed, the numbers of individuals in a population who believe, respectively, that they possess above and below average ability should be equal. In addition, the rational approach would suggest that the (realistic) insight as to whether one is above or below average should be equally adaptive. However, following the (hedonistic)
228 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
attributional models, only the belief that one possesses high ability should be adaptive, and consequently, 100% of individuals should be convinced that they have such self-estimations, in order to assure adaptive reactions. THE INTEGRATIVE ATTRIBUTIONAL MODELOF MOTIVATION To take account of the above mentioned shortcomings and criticisms with regard to the attributional models, I have formulated an attributional model of motivation which integrates basic assumptions and findings from the theories of attributional antecedents with attributional theories (Försterling, 1986, 1994). This model assumes that it cannot be reasonable to generally judge an attribution (e.g., lack of ability) or a reaction that is triggered by this attribution (e.g., lack of persistence) as inappropriate or worthy of change, and a different attribution (e.g., lack of effort) and reaction (e.g., persistence) as adaptive and desirable. By contrast, the model assumes that it depends on the antecedent (covariation) information (i.e., the situation), whether or not a certain attribution and the resulting emotional and behavioural responses are adaptive or not. With these premises, the model takes up the rational metaphor of human beings that underlies the original models of the antecedents of attributions (Heider, 1958; Kelley, 1967) and assumes that humans are motivated to develop a realistic understanding of the causes of events and that such a realistic understanding leads to functional reactions. Hence, the central assumption of the model is that realistic attributions, (or, more generally, realistic “naive” theories) lead tofunctional reactions (see Figure 1). The veridicality of a causal attribution To determine whether a certain causal explanation is realistic or veridical, we use the already described models of covariation (see also Försterling, 1989, 1992; Heider, 1958; Kelley, 1967). We define an attribution as realistic when it is consistent with the antecedent covariation information and when it has not been falsified by such information. By contrast, we label an attribution as unrealistic when it constitutes an inference that violates the normative covariation principle. For instance: A pupil traces back his failure at a maths exam to the difficulty of the tasks chosen by the teacher. If the student had been successful at maths exams in the past (i.e., the failure occurred with low consistency) and at other exams as well (high distinctiveness), and if most other students were also unsuccessful at the focal exam (high consensus), our pupil’s attribution to task difficulty would be consistent with the predictions of the covariation principle and hence realistic according to our definition. However, an attribution of failure to task difficulty would be
INTEGRATION OF ATTRIBUTION MODELS 229
Figure 1. The relationship among antecedents, attributions, and reactions (from Försterling, 1986).
unrealistic if the pupil had failed maths exams in the past (high consistency of failure), if he was unsuccessful at other exams as well (low distinctiveness), and if the other pupils succeeded at the focal exam (low consensus). If such a configuration of covariation information were to exist for the given effect (failure) an attribution to the person (e.g., lack of knowledge or ability) would be more realistic. The use of the covariation principle as a normative model to assess the “veridicality” of lay persons’ causal hypotheses is quite similar to the way we use data in science to evaluate our hypotheses. As in science, we would not assume that there are naive (causal) hypotheses that are (like dogmas) generally right or wrong. However, when a causal hypothesis has been repeatedly falsified, we would expect the naive person—like the scientist— to give it up, and, if possible, come up with a new one. Covariation information can actually be conceived of as data relevant for our causal hypothesis. For instance, the hypothesis “my global and stable dispositions caused my failure at Task X” would be consistent with “data” indicating that I always failed at this and other tasks, whereas other individuals succeeded at X. By contrast, covariation information indicating that I succeeded at Task X before, others fail at Task X, and I succeed at other tasks, is highly inconsistent with the causal hypothesis that stable dispositions caused the failure. Hence, the application of the covariation principle as a normative model for the (non)veridicality of a “naive” hypothesis actually follows the same logic that scientists use when testing their hypotheses. Or, to put it still another way: If psychologists conceptualise individuals as naive scientists who use methods akin to those used in science, the scientific models can be used as norms for deciding to what extent a “naive” judgement deviates from them. Hence, we can label
230 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
judgements that do not deviate from the model “realistic” or “veridical”, and judgments that deviate “unrealistic” or “non-veridical”. It should be noted that the concept of veridicality is a relational concept which implies a “fit” between the antecedent situation (data, covariation information) and the cognition (hypothesis, attribution). Such a relational concept is not included—thus far—in attributional models. The functionality of reactions The second assumption of the integrative model is based on the idea that the ubiquitous desire for a realistic causal understanding of events must serve a function. It would seem improbable that evolution would have provided us with the desire and the ability to correctly perceive and explain one’s world, and the ability to draw causal inferences from information while using methods akin to those used by scientists, if this was an activity without further consequences. In fact, attributional models have clearly shown that important behavioural and emotional reactions are guided by our causal knowledge. Therefore, our model assumes that a causal understanding derived from the use of (naive) scientific methods (i.e., covariation based inferences) leads to reactions (i.e., behaviours and emotions) that are functional, adaptive, and appropriate (we use these three terms interchangeably). This idea has already been stated in early writings of attribution theorists. Kelley (1971) has stated that “the attributor is not simply an attributor, a seeker after knowledge. His latent goal in gaining knowledge is that of effective management of himself and his environment. He is not a pure ‘scientist’ then, but an applied one” (p. 220). As already indicated, and as depicted in Figure 1, we therefore assume that realistic (or veridical) causal judgements lead to functional reactions (emotions and behaviours). Furthermore, it logically follows from this assumption that a lack of a causal understanding or non-veridical or unrealistic causal judgements lead to dysfunctional reactions or behaviours. To illustrate, let us return to the pupil making a realistic or an unrealistic attribution for his failure at maths. If attributions of failure to the teacher’s selecting too difficult tasks were realistic, a complaint that might be triggered by this (realistic) attribution to task difficulty could be considered to be an appropriate reaction. However, if lack of ability were a realistic attribution, it would seem more adaptive for the pupil to experience regret (rather than anger) and to improve his efforts (rather than to complain). In addition, if the failure was clearly caused by lack of ability, which cannot be compensated with improved effort, it might also be appropriate to think about choosing—if possible—other courses instead of maths. These reactions would certainly not be adaptive if failure was caused by the
INTEGRATION OF ATTRIBUTION MODELS 231
difficulty of the tasks chosen by the teacher and not by relatively stable personal factors. The concepts of functionality, adaptiveness, and appropriateness are sometimes used in social psychology, and somewhat more often in clinical psychology. However, they are not clearly defined, and operationalisations of these concepts are generally lacking. In fact, evolutionarily oriented social psychologists have argued that (non-evolutionarily oriented) social psychologists rarely consider the function of the phenomena they study (see Buss, 1995). Buss (1995) argues that attribution researchers typically study phenomena like actor-observer differences, correspondence biases, or the influence of outcome attributions on subsequent expectancies without asking what the functions of such phenomena in the long run might be. Possibly the most prominent function of attributions that has been pointed out by social psychologists is their occasional self-serving nature. This mechanism has often been referred to for explaining errors and biases in the attribution process. For instance, the tendency to make internal attributions for success and external ones for failure has been conceived of as serving the function of maximising positive and minimising negative emotions following success and failure. However, we believe, that a fundamental and universal mechanism such as the desire to gain an accurate understanding of events must have functions that go beyond affective gains. We will not, however, attempt to give a formal definition of functionality. For the present purposes it is sufficient to postulate that making realistic attributions leads to goal attainment. For instance, it can lead to the successful completion of tasks in the achievement domain, and to the initiation and maintenance of social relationships. From an evolutionary perspective, it is also quite conceivable that a realistic understanding of the causes of events has survival value and that it maximises reproductive success. However, these aspects go beyond the scope of the present chapter. Differences between the integrative attributionalmodel and classical attributional models The presented integrative attributional model makes largely the same predictions as the classical attributional models. For instance, it assumes that we persist when we attribute failure to a variable cause and that we give up when we trace failure back to a stable cause. However, by integrating attribution and attributional models, while using the concepts of veridicality and functionality, the present model makes some additional assumptions, some of which go beyond and some of which are inconsistent with derivations made from classical attributional models.
232 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
The first important aspect of our model that goes beyond previous models is that it defines the concept of veridicality and nonveridicality as relational constructs that involve both the antecedent information and the attribution itself. Veridicality is not a function of the antecedent information or the attribution alone, but it is conceived of as a function of the “fit” between antecedent conditions and attributions. We will show in the subsequent sections how this concept can be operationalised and used to investigate questions of errors and biases in the attribution process. The second important aspect of the integrative model is that it hypothesises realistic appraisals (of the causes of events) to lead to adaptive, functional responses, whereas unrealistic inferences lead to dysfunctional reactions. In this respect our model generates—in part— different hypotheses from the classical attributional models. More specifically, it suggests that attributions to lack of ability, which have been conceived of in classical models as maladaptive and changeworthy, can lead to adaptive consequences when they are realistic. In this respect, the integrative model’s assumptions deviate from those of other attributional models. From the assumptions described above, some conclusions can be drawn with regard to the state of depression: If the state of depression is to be conceptualised as a dysfunctional reaction, the prediction that depression necessarily occurs when (important) negative events are explained with internal, stable, and global attributions would be inconsistent with our model, as it assumes that internal, stable, and global attributions are under certain circumstances realistic and should hence lead to functional reactions. Hence, our model suggests that depression should not necessarily occur when attributions to internal, stable, and global attributions are made, and therefore suggests that there are other determinants of depression than causal attributions. The assumption that realistic attributions should generally lead to functional reactions does not mean that there might not be exceptions to the rule. A person might hold the belief that she has high ability at roulette, therefore decides to spend much of her money gambling, and might indeed become rich. However, it is to be suspected that there are more individuals holding such a belief who will tragically lose their fortune than individuals who make a fortune. The integrative model, as already indicated, makes different assumptions about when “therapeutic” attributional retraining is indicated. The veridicality of a certain cognition (e.g., “my failure at this exam is due to lack of ability” or “I have got no money”) needs to be determined by analysing the antecedent conditions of this statement (e.g., by investigating how many other pupils failed or by looking into one’s wallet). It does not seem reasonable (and nobody suggested this) to use the consequences of thoughts to determine their veridicality (e.g., the fact that the thought of
INTEGRATION OF ATTRIBUTION MODELS 233
having low ability or not having money might result in depression is not an indication of the nonveridicality of the cognition). However, thus far, applications of attributional models use the consequences of cognitions to determine their changeworthiness and they do not make use of models of attributional antecedents to decide on the changeworthiness. For instance, attributional retraining studies teach individuals who easily give up following failure, to attribute negative outcomes to non-ability factors (for summaries, see Försterling 1985; Perry, Hechter, Menec, & Weinberg, 1993). Our argument is that such an approach might have negative side-effects when the attributions that apparently lead to “non-adaptive” behaviour are, in fact, realistic and when the target attributions of the therapy or training intervention would be unrealistic. The “maladaptive” consequences of unrealistic attributions of failure to non-ability factors, and adaptive aspects of realistic attributions for failure to, e.g., lack of ability, have neither been postulated in attributional models of motivation nor have they been experimentally demonstrated. The studies to be reported next address exactly this question. Evidence for the functionality of realisticattributions Before reporting a series of studies testing the core assumption of the integrative attributional model, that it is functional to make realistic attributions, we briefly refer to a (voluminous) literature that apparently also addresses this assumption. This is important, as this literature seems to suggest a conclusion that is diametrically opposed to the idea put forward by the integrative model, i.e., that veridical inferences lead to functional reactions. In a much-cited article, Taylor and Brown (1988) suggest that there is vast empirical evidence supporting the idea that it is highly adaptive to see oneself, one’s situation and the future in an unrealistically optimistic fashion. For instance, Taylor and Brown (1988) refer to the finding that most college students report that their chances of maintaining a satisfactory relationship are above average. The authors suggest that this expectation must be biased as not all individuals can have above-average chances of maintaining good relationships; by definition, 50% should have above and 50% should have below average relationships. In addition they suggest that college students have proved to be quite successful in the educational system and, hence, must have reacted quite healthily and functionally. From these findings, i.e., that “healthy” individuals have biased expectations, they conclude that positive biases must lead to psychological health. I will not attempt to comprehensively address the evidence that Taylor and Brown (1988) review to support their argument as most of the respective research does not address attributions or attributionally mediated reactions,
234 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
and as Taylor and Brown’s conclusions have been critically examined and qualified by other authors (see Colvin & Block, 1994; Försterling 1994). However, Taylor and Brown (1988) also discuss studies addressing the veridicality of depressives’ attributions. As these studies are relevant for the present model, I shall briefly refer to them. Attributional research on depression has asked whether the (internal, stable, and global) attributions (for failure) leading to depression are more realistic than the (external, variable, and specific) attributions preferred by non-depressives. The typical findings leading to the conclusion that causal attributions of depressed individuals are more realistic than those of nondepressed peers consist of the observation that depressed—as compared with non-depressed—individuals’ attributions are “balanced” (e.g., between success and failure and actors and observers). For instance, it was found that depressives attributed success as well as failure about equally strongly to their ability, whereas non-depressed individuals exhibited “selfserving” attributional tendencies, i.e., they attributed success more so than failure to their ability (see Campbell & Fairey, 1985; Kuiper, 1978; Rizley, 1978; see Taylor & Brown, 1988, for a summary). In addition, it was found that the attributions of depressive actors deviate less from the attributions of external observers than those of non-depressed actors (cf. Glass, McKnight, & Valdimarsdottir, 1993). Several authors have argued that balanced attributions cannot be used as a criterion for judging the veridicality of an attribution (Ackermann & DeRubeis, 1991; Alloy & Abramson, 1988; Colvin & Block, 1994; see also Försterling, 1986, 1994). Assume, for instance, that overweight individuals judge their weight to be lower than observing individuals of normal weight, whereas people of normal weight show less of a discrepancy (imbalance) between their own and observers’ judgements. According to the determination of realism by the criterion of balance, we would need to assume that overweight individuals are unrealistic, as they show a discrepancy between their own judgements and observers’ judgements. However, it is also conceivable that the observers make the error: They might overestimate the weight of the overweight group and realistically assess weight only in the normal range. Hence, when wanting to assess accuracy of weight estimates, the most appropriate index of realism or veridicality would not consist of comparisons of observers’ and actors’ judgements: Instead, comparisons of self-perceived weight with an objective (normative) criterion (i.e., weight in pounds) would have to be preferred on logical grounds. But is there such a “scale” for causal attributions? Above, we introduced a classification of an attribution as realistic or unrealistic on the basis of its antecedents. In the following, we will introduce studies on depressive realism that use this definition of attributional veridicality.
INTEGRATION OF ATTRIBUTION MODELS 235
As already indicated, one further central assumption of the attributional model of learned helplessness and depression poses an additional challenge to the integrative attributional model. The prediction that—under certain circumstances (i.e., when an important negative event occurs)—a tendency to make internal, stable, and global attributions will lead to depression, implies that a certain attribution (i.e., an internal, stable, and global one) necessarily leads to a dysfunctional reaction, regardless of its veridicality. However, it can be assumed that internal, stable, and global attributions are in some instances realistic. The integrative model predicts that such attributions should lead to functional reactions when they are realistic. Depression, however, can generally be conceived of as a dysfunctional reaction. Therefore, we have addressed in the research reported below the question whether internal, stable, and global attributions of negative outcomes always lead to depression or whether there are additional cognitive determinants of depression that need to be present before depression occurs (see also Brewin, 1985, for a discussion of the role of attributions in depression). Answers to this question are also relevant for applied (therapeutic) questions. For instance, if we knew the conditions under which one can trace back failure to internal, stable, and global causes without getting depressed, therapeutic interventions could be designed that might allow individuals to maintain realistic attributions and to take advantage of the positive implications of realistic self-assessments without reacting dysfunctionally. Empirical tests of the integrative model Before describing the empirical studies testing the integrative attributional model, we should briefly summarise the theoretical considerations discussed thus far: Neither the attributional analysis of achievement motivation (Weiner, 1986; Weiner et al., 1971) nor the attributional analysis of learned helplessness and depression (Abramson et al., 1978) includes antecedent conditions (covariation information) in its model. These models define (possibly due to the lack of integration of antecedent information) certain attributions (e.g., internal, stable, and global explanations for failure) as disadvantageous because they lead to reactions that are often conceived of as worthy of change (e.g., low expectations of success, and depression). This line of reasoning, however, seems inconsistent with the central assumption of attribution theories (see Heider, 1958; Kelley, 1967) which maintain that individuals are motivated and able to develop a realistic view of themselves, and which assume that such realistic self-knowledge has adaptive value. We have therefore introduced an integrative attributional model of motivation, which assumes that realistic attributions that are consistent with the available antecedent covariation information lead to functional
236 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
reactions, and unrealistic attributions lead to dysfunctional reactions. Empirical studies that test the relation between the veridicality of attributions, defined in this manner, and the functionality of reactions, do not yet exist (see, for an exception, Försterling & Rudolph, 1988). Studies that apparently show that depressives make more realistic attributions than non-depressives use a different criterion for defining veridicality (i.e., balance). In addition, studies investigating the relation between attributions and depression leave open the question of whether (realistic) internal, stable, and global attributions always lead to depression and when this might not be the case. In the following, we report studies that tested the integrative attributional model of motivation. These studies not only manipulate attributional antecedents and assess attributions (like attribution studies) or vary attributions and assess reactions (like attributional studies). By contrast, the reported research assesses or manipulates both antecedent conditions as well as attributions. In addition, indicators of behavioural reactions (e.g., their functionality) are assessed. The studies address the attributional analysis of achievement behaviour and depression. Applications of the model to achievement behaviour. Försterling and Morgenstern (2002) have tested the integrative model in achievement contexts. Studies designed to test attributional models of achievement behaviour thus far used situations in which participants experienced success or failure at one certain (hypothetical or real) task. Subsequently, attributions for the outcome were assessed or manipulated, and participants had to indicate their expectations of success for a further trial at this specific task or they had to work again on a task of the same type as the original task. These studies consistently showed that persons who attributed failure to internal, stable, and global causes (e.g., lack of ability) had relatively low expectations of success at the task on a future trial, they persisted for less time at the task, and performed worse than participants who attributed the original failure to, for instance, external and variable factors (e.g., chance; see Meyer, 1973) Försterling and Morgenstern (2002) have argued that this paradigm (i.e., manipulating outcome attributions at one specific task and then measuring performance at a second trial with this specific task) is not sensitive to detecting the disadvantages of unrealistic attributions (e.g., of failure to non-ability causes) and the advantages of realistic attributions (e.g., to low ability). Outside the laboratory, most tasks are characterised by the fact that several subtasks have to be performed within a limited amount of time in order to achieve success. For instance, a high-school diploma or a university degree typically consists of different areas, and time that is spent preparing maths might be lost for the preparation for the history exam. Moreover, a manager might need to negotiate, calculate tax, design marketing strategies, and supervise his staff. Suppose he fails in one of
INTEGRATION OF ATTRIBUTION MODELS 237
these aspects and spends too much time with it (e.g., tax), he might lose precious time for his negotiations and marketing strategies. Intuitively, it would seem beneficial for his long-range goals to give up quickly on the tax problem, search for the help of a tax consultant, and invest effort where he is more likely to achieve success. As already indicated, the empirical paradigms that have been used thus far in the motivational domain (attributions for performance at one task are assessed and related to subsequent performances at the same type of task) are not sensitive to the advantages of realistic attributions to lack of ability (both in general or a specific task). In these studies, the belief in one’s high ability should always be beneficial. Even individuals low in ability who maximise their effort and time at the experimental task should show performance maximisation under such instructions. However, unlike in most settings in real life, this paradigm does not penalise the participants for wasting time in a weak area instead of investing in their strengths. For these reasons, Försterling and Morgenstern (2002, 2003) developed a paradigm in which participants had to work for a fixed time on a task consisting of different subtasks requiring different abilities. It was expected that participants who attributed performances at the subtasks veridically would achieve better task performance than individuals attributing subtask performance unrealistically (optimistic or pessimistic). In a first study, Försterling and Morgenstern (2003) used a simulation experiment to assess whether laypersons share the view of our model that performance is maximised by realistic attributions and their behavioural consequences in situations in which overall performance is composed of the performances at subtasks. Participants were asked to imagine they were working on a task consisting of various subtasks requiring different abilities (a ring-toss game, working on a puzzle, solving anagrams and number sequences). The task consisted of maximising the sum of the correct answers across all subtasks within a limited time period that one could freely distribute across each of the five subtasks. Participants were further instructed to imagine that they had previously had the opportunity to gather information about the causes of their successes and failures at the various subtasks. For instance, and more specifically, they were asked to imagine that they had especially high ability with regard to some subtasks (i.e., to have consistently performed better than their peers at the subtask) and to have low ability for other subtasks (i.e., to have consistently performed worse than their peers at the subtask). As dependent variable, we assessed which strategies participants would use to maximise the overall result when experiencing failure at an item of a specific subtask. The strategies they could choose from were, for instance, to persist at the respective item, to immediately switch to another task, or to search for help. We found that participants considered it to be functional for the maximisation of overall performance to: (1) invest more
238 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
time on subtasks at which they previously performed better than their peers, and (2) to invest less time on subtasks at which they had previously performed worse than others. These findings indicate that the naive psychology of our experimental participants does not assume that an overall result is maximised by persisting at all subtasks, but only by persisting selectively at the subtasks for which one has high ability. Or, more generally, participants assumed that overall goal attainment was maximised by taking into account their low ability on some of the tasks and behaving accordingly. They did not assume that performance could be increased by approaching a task for which they had low ability as if they had high ability. In addition to this simulation study, Försterling and Morgenstern (2002, Study 2) also conducted a laboratory experiment in which realistic and unrealistic attributions were induced in participants and in which actual performance was used as the dependent variable. This laboratory experi ment was to a large extent a realisation of what participants had to imagine in the questionnaire study just described. In a first stage, participants worked on tasks from four different areas of an intelligence test (sentence completion, word selection, analogies, geometric figures). At this stage, the attributionally relevant information for the second stage was assessed: i.e., performances at the various subtasks (distinctiveness) in relation to their peers (consensus). In the second stage of the experiment (about 1 week later), half of the participants were realistically informed about their own (distinctiveness) and their peers’ (consensus) performances at each subtask area during pretest. For instance, a participant who had eight items correct in sentence completion, five at word selection, and two at analogies would be told just that. The remaining participants were informed unrealistically about their pretest performance. They were led to believe that they achieved the inverse value of their actual performance at Stage 1. For instance, a person who had eight items correct in sentence completion, five at word selection, and two at analogies was told that she had two items correct in sentence completion, five at word selection, and eight at analogies. In addition, both experimental groups were informed about the average number of correct solutions that were achieved by the other participants (consensus) in each of the four subtasks; participants received this (consensus and distinctiveness) information in tabular form at the beginning of the second stage of the experiment. Following the realistic or unrealistic feedback for pretest performance, participants had to work once more for 20 minutes at the various subtasks. However, in this second stage of the experiment, they were allowed to decide themselves how much time they wanted to spend with items of each of the four subtasks. They were instructed to try to maximise their overall performance (i.e., the sum of the correct solutions in all subtasks), and for
INTEGRATION OF ATTRIBUTION MODELS 239
Figure 2. Mean overall performance at pre- and posttest for the realistically and unrealistically informed groups (Försterling & Morgenstern, 2002, Study 2).
the person with the best performance, 50 German Marks (approximately US $23) were promised. As dependent variables, attributions for hypothetical failures at an item of each of the subtasks were assessed. In addition, the experimenter timed how many seconds participants spent on each subtask, and the overall performance (the sum of correctly solved items across all subtasks) was assessed as central dependent variable. Figure 2 shows that participants who were realistically informed about their performance configuration across subtasks at Stage 1 reached a higher overall performance level at Stage 2 than the unrealistically (inversely) informed group (there was a significant interaction of the factors Experimental Stage [pre- vs posttest] and Information [“realistic vs unrealistic”]). In addition, internal (correlational) analyses supported our speculations about the mechanisms responsible for the superior performance of the realistically informed group: Compared with the unrealistically informed participants, realistically informed individuals spent more time on tasks at which they had done well during the first stage of the experiment, and less time on tasks at which they had performed
240 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
poorly. The mean of the intra-individual correlations between the four test results at pretest and the time spent with the respective subtask during posttest was significantly higher for the realistically informed group (M=. 48) than for the unrealistically informed group (M=–.14). In addition, results of an analysis of covariance indicated that the ability-sensitive allocation of time was in fact responsible for the superior performance of the realistically informed group at posttest. When the intra-individual correlation (i.e., performance during pretest at the four subtasks with time spent at the respective subtask at posttest, determined for each individual participant) was partialled out from the analysis, the Information by Stage interaction for overall performance was no longer significant. In summary, our studies regarding the veridicality of attributions in achievement contexts used a new paradigm in which overall performance was defined as the sum of performances at subtasks. These studies yielded the following insights: The simulation study showed that in order to maximise overall performance, individuals planned more time in performance areas for which they had high self-perceived ability and less time in areas for which they perceived themselves to be less capable. The laboratory study showed that realistically-informed individuals in fact used such a strategy. Through the use of this strategy, they outperformed participants given unrealistic information about pretest performance; these latter participants were prevented from successfully applying this strategy, as they did not possess the necessary (realistic) attributionally relevant information. Hence, we can summarise that the studies in the achievement domain yielded results consistent with the integrative model: Realistic attributions—more so than unrealistic attributions—led to functional reactions resulting in better performance. Studies regarding the veridicality of depressives’attributions In several recent studies, we have applied our integrative attributional model to the analysis of the determinants of depression, and especially to the phenomenon of depressive realism. A central starting point of these studies was the assumption that depression is most likely a clearly dysfunctional reaction (see Hammen, 1999). If this was in fact so, it follows from our integrative model that depression should be tied to unrealistic attributions, as—according to our model—such attributions should lead to dysfunctional reactions. Försterling, Bühner and Gall (1998) therefore used the definition of the veridicality of an attribution as described above and an operationalisation of attributional veridicality similar to the one used by Försterling and Morgenstern (2002) to investigate the attributions of depressives compared to non-depressives. In contrast to the above-reported studies in the achievement domain, in which the functionality of a reaction was assessed
INTEGRATION OF ATTRIBUTION MODELS 241
as a dependent variable (i.e., overall performance), the studies to be reported next used—as quasi independent variables—already existing interindividual differences regarding the functionality of participants’ reactions, that is, their depression status. The first series of experiments investigated whether induced covariation information is processed differently by depressed and non-depressed individuals, and the second set of studies assessed self-perceived covariation information of both mood groups and investigated how perceived covariation information is related to attributions. As already indicated, we assumed that if the phenomenon of depressive realism extends to the operationalisation of realism as suggested in this chapter (i.e., as fit between attributions and covariation information), attributions of depressed individuals should be more affected by (induced) covariation or more closely related to (perceived) covariation information than the attributions of non-depressed individuals. However, if depression is conceived of as a dysfunctional reaction, our integrative attributional model suggests that depression should be triggered by unrealistic attributions (as it holds that unrealistic—rather than realistic—attributions lead to dysfunctional reactions). More specifically, the model predicts that depressives’ attributions should be less realistic than non depressives’ causal inferences (i.e., less affected by induced, and less consistent with selfperceived, covariation information). A third type of study is reported that assessed meta-attributional accuracy of depressives and non-depressives. In these studies both the attributions that depresssives’ partners make for participants as well as the attributions that participants expect their partners to be making are related. Again, the hypothesis of depressive realism and our integrative model make different suggestions: If depressives were more realistic, we would expect a more accurate prediction of their partner’s attributions, and if they were more unrealistic, we would expect depressives to hold less accurate beliefs about their partner’s attributions than non-depressives. A further concern of some of our studies on depression was whether attributions of failure to internal, stable, and global factors would, under all circumstances, lead to depression or whether, under other conditions, internal, stable, and global attributions for failure could be unaccompanied by this presumably dysfunctional reaction. This question is relevant for our integrative model as it assumes that “depressogenic” attributions should, at least occasionally, be realistic. If, however, a certain kind of (realistic) attribution were always to be associated with dysfunctional reactions, the prediction that realistic attributions lead to functional reactions would obviously be falsified. Processing of covariation information by depressed and nondepressedindividuals. The first studies tested the veridicality of depressives’ and non-depressives’ causal attributions (as defined by the integrative
242 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
model) by inducing covariation information and assessing attributions (Försterling et al., 1998, Studies 1–3). Participants were classified via median split of their test scores on the Beck Depression Inventory (BDI; Beck, Ward, Mendelson, Mock, & Erbaugh, 1961) as depressed and nondepressed. In addition, they were confronted with the hypothetical scenarios typically used to assess attributional style (Peterson, Semmel, von Baeyer, Abramson, Metalski, & Seligman, 1982), for instance, “imagine that you fail an important exam”. Further covariation information was provided in these hypothetical scenarios. One group received covariation information designed to lead to depressogenic (internal, stable, and global), failure attributions, and the other group received information designed to lead to anti-depressogenic (external, variable and specific) failure attributions. For instance, in order to induce depressogenic attributions, participants were asked to imagine that most other individuals who took the exam succeeded at it (low consensus), and that they also failed similar (high consistency) and dissimilar (low distinctiveness) exams in the past. To suggest antidepresso genic attributions, it was indicated that most other individuals also failed the exam (high consensus) and that they had succeeded at similar (low consistency) and dissimilar tasks (high distinctiveness) in the past. As dependent variable, we assessed to what extent depressive and non-depressive persons attributed the described events to internal, stable and global causes. If depressives were, in fact, more realistic than non-depressives, their causal judgements should be influenced more strongly by covariation information than nondepressives’ judgments. The integrative attributional model favours the contrary hypothesis, i.e., that non-depressives’ judgements should be more sensitive to covariation information, as depression is a dysfunctional reaction that should be associated with a tendency to make unrealistic attributions. Försterling et al. (1998, Experiments 1, 2, and 3) did not find significant differences with regard to the influence of covariation information on depressive or non-depressive college students’ attributions. Depressives generally showed a tendency to attribute failure more strongly to internal, stable, and global causes than non-depressives (i.e., the variable “depression” had a significant main effect on causal attributions), and covariation information also significantly influenced attributions in the expected direction (there was a significant main effect of covariation information, indicating that depressogenic covariation information led to internal, stable, and global failure attributions whereas antidepressogenic information led to external, variable, and specific attributions). An interaction between “depression” and “covariation information”, which would indicate that the two mood groups differentially process the presented covariation information was not found, however. This pattern of results was found for covariation information that was designed to lead to
INTEGRATION OF ATTRIBUTION MODELS 243
external and specific (Studies 1, 2, and 3) and/or to internal, stable, and global attributions (Studies 2 and 3); in addition, the studies included groups that received all three pieces of covariation information or only one or two pieces. These findings suggest that depressives and non-depressives do not differ with regard to the veridicality of their attributions, and hence neither support the phenomenon of depressive realism nor the prediction from our integrative model. The results might reflect a variety of facts. For instance, our “depressed” participants were not clinically depressed. Therefore, the assumption that the depressed participants truly suffered from a dysfunc tional reaction might not be correct. Furthermore, the covariation information that was presented in hypothetical scenarios might not have been convincing for the participants and/or participants might have had strong preconceptions about covariation information in the described situations. We therefore conducted a new set of studies in which covariation information was not experimentally manipulated but individually assessed, using both college students and clinical patients as participants. Self-perceived covariation information ofdepressive and non-depressive individuals andtheir relation to causal attributions Försterling et al. (1998, Study 4) and Försterling and Bühner (in press, Study 1 & 2) have followed up on these two possibilities and have investigated whether (clinically and non clinically) depressed and nondepressed individuals differ with regard to their implicit assumptions about covariation information. The investigation of self-perceived covariation information is also relevant for the question of why depressives and nondepressives make different attributions. For instance, Brewin and Furnham (1986) have found that depressives—more so than non-depressives—tend to assume, with regard to a certain negative event, high consistency and low consensus. These differential assumptions about covariation might sufficiently explain the differences of the two mood groups with regard to their attributions. To test these questions and possibilities, Försterling et al. (1998, Study 4) and Försterling and Bühner (in press, Studies 1 and 2) not only assessed participants’ attributions about several hypothetical situations, but also their self-perceived covariation information. For instance, participants were asked to imagine that they had failed an important exam. They were then asked to rate their attributions for the event (as in the studies reported above). To assess self-perceived covariation, they were asked how many other individuals had most likely failed at the task (perceived consensus), how they did at other exams (perceived distinctiveness), and how
244 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
successful they assumed they might have been previously at similar exams (perceived consistency). Results of all three studies revealed that depressives not only made more internal, stable, and global attributions for failure, but they also reported assumptions about covariation information that were entirely consistent with these attributions. They assumed—more so than non-depressives— that failures occurred with a low consensus (only few individuals failed at the respective tasks), high consistency (in the past, I have failed at similar exams), and low distinctiveness (I failed exams in different areas as well). These results point to the possibility that depressives might in fact have had different (covariation) experiences than non-depressives in that they might have consistently failed in various areas where others succeeded. However, whether differences between the mood groups in perceived covariation reflect actual experiences or biased memories or assumptions has not yet been investigated and the design of the present studies does not allow us to answer this question. A further question is whether differences in covariation information (which might reflect factual experiences) sufficiently explain the differences between the mood groups regarding their causal attributions. In all three studies, partialling out perceived covariation from the correlation between attributional style and depression did not result in a decrease of this correlation below chance level: Depressed individuals still made more internal, stable, and global attributions after perceived covariation information was partialled out from the analysis. This finding suggests not only that the subjectively perceived data basis (i.e., covariation information) is different for depressed and nondepressed individuals, but also that depressives and non-depressives draw different inferences from the subjectively perceived data basis. In order to determine whether the attributions of depressives or nondepressives more closely reflect self-perceived covariation information, we derived a composite “veridicality score” based on the “fit” between perceived covariation information and attributional ratings. More specifically, we formed difference scores between perceived covariation and attributional ratings. To illustrate, consider a participant who indicates, e.g., by marking a “10” on all of the rating scales that assessed perceived covariation, that a negative situation (i.e., failure in a certain situation) probably only happened to him (low consensus), that he probably failed this task before (high consistency), and that he probably also failed at different tasks (low distinctiveness). This participant would receive a score of “30” with regard to his self-perceived covariation. If this participant made an attribution consistent with his self-perceived covariation information (i.e., a “veridical” attribution according to our definition), we would expect him to trace back failure to an entirely internal, stable, and global cause by marking a “10” on all three rating scales on the
INTEGRATION OF ATTRIBUTION MODELS 245
attributional style questionnaire. As a consequence, he would receive an attributional index of “30”, and the difference (veridicality) score (perceived covariation minus attributional style) would be “0”. If, however, the individual with low perceived consensus and distinctiveness as well as high consistency (i.e., a covariation score of “30”) indicated in the attributional style questionnaire that failure was caused by an entirely external, variable, and specific cause (i.e., an attributional score of “3”) he would be considered to be making an attribution that was highly inconsistent with his self-perceived covariation information and a difference score of “+27” would result, indicating a tendency to bias causal inferences in an antidepressive direction (external, variable, and specific). By contrast, if an individual assumed that a bad event probably happened to most people (consensus rating of “1”), that it probably did not happen when he worked on this task before (consistency rating of “1”), and when working on other tasks (distinctiveness rating of “1”), it would be realistic to make external (“1”), variable (“1”), and specific (“1”) attributions. Consequently, a low difference score would result (3– 3=0). However, if this individual made an internal (“10”), stable (“10”), and global (“10”) attribution, the difference score would be high and negative (3–30=–27) reflecting a tendency to make causal inferences with a depressogenic bias. Note that the difference score allows us to distinguish realism (low deviation scores) from optimism (high positive difference scores), and pessimism (high negative difference scores). Figure 3 depicts the difference scores (aggregated across the eight situations) for very non-depressed (BDI scores from 0 to 3), non-depressed (BDI from 4 to 9), and dysphoric (BDI>9) participants. The figure reveals a tendency for depressed individuals to draw causal inferences from selfperceived covariation information that were biased towards internality, stability, and globality, whereas non-depressed and very non-depressed individuals’ causal explanations are, by comparison, in accordance with their self-perceived covariation information (difference scores close to “0”). These findings indicate that depressives are less (and not more) realistic than non-depressives when self-perceived covariation information is used as a criterion of the veridicality of an attribution. However, the difference score just reported has the disadvantage of being aggregated across eight situations. Hence, we cannot rule out that, for instance, a first individual has small negative deviations in each of the eight situations and hence receives a moderate (negative) deviation score. On the other hand, a second participant with large positive deviations in some of the situations and large negative deviations in the remaining ones would receive an even smaller deviation score than the first individual. To obtain a measure of congruence between participants’ attributions and their self-perceived covariation information that avoids this disadvantage we calculated intra-individual correlations between
246 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 3. Difference scores between self-perceived covariation ratings and attribution ratings for low, intermediate, and highly dysphoric participants (BDI=Beck Depression Inventory) (Försterling et al., 1998, Study 4).
attributions and perceived covariation information, individually for each participant across the eight hypothetical situations. A high intra-individual correlation reflects that a participant who, for instance, attributes failure at one situation to internal, stable, and global causes and at another situation to external, variable, and specific ones also assumes that the failure at the first situation covaries with the person (low consensus, high consistency, and low distinctiveness), whereas failure in the second situation covaries with the circumstances (high consensus, high distinctiveness, and low consistency). Negative and low intra-individual correlations, by contrast, imply that a person makes attributions that are inconsistent with selfperceived covariation information. This would be the case, for instance, when a person in a situation in which he traces failure back to his own person assumes that failure covaried with the task and/or circumstances (i.e., that many other persons failed, that he did not fail at this task in the past, and that he usually succeeds at other tasks) and when he makes attributions to the task while assuming covariation of failure with the person (i.e., that only he failed consistently at this as well as other tasks).
INTEGRATION OF ATTRIBUTION MODELS 247
We found that the mean intra-individual correlation coefficient between perceived covariation information and attributions significantly increased with depression: It was lowest for very non-depressed, higher for nondepressed, and highest for dysphoric individuals. These results reflect the fact that depressives were more realistic than non-depressives in the reported study, a finding that is consistent with the hypothesis of depressive realism and inconsistent with the prediction of the integrative model that suggests that depression (a dysfunctional reaction) is associated with unrealistic attributions. In sum, the two different indicators of attributional veridicality used by Försterling et al. (1998) in their Study 4 revealed opposing results with regard to attributional veridicality. The aggregated difference score between attributions and covariation information suggested unrealistic depressive pessimism and the intraindividual correlations suggested depressive realism. Försterling and Bühner (in press) have investigated whether this pattern of results is replicable while using an identical sample (Study 1) to the study just reported and while using a clinical sample (Study 2). In both studies we replicated both results: higher deviation scores in the direction of unrealistic depressive pessimism when the aggregated difference score (attribution minus perceived covariation information) was used and somewhat higher intra-individual correlations for depressed than for nondepressed individuals when intra-individual correlations were used as a dependent measure. Hence, we can summarize as follows: Concerning the difference score between attributions and covariation information, we found in all three studies that depressives make more internal, stable, and global attributions even when their perceived covariation information is taken into account. This means that depressives not only tend to assume, for instance, low consensus for their failures more so than do non-depressives, they also tend to draw more depressogenic conclusions from this (i.e., they make more internal attributions). However, with regard to the intra-individual correlations between attributions and perceived covariation information, we consistently found small (and only marginally significant) differences between depressives and non-depressives, pointing in the direction of a more realistic processing of covariation information by depressives. These findings seem to indicate that depressives have a consistent bias towards internal, stable, and global attributions but not a deficit in processing (covariation) information correctly. By contrast, our studies showed a (small but consistent) tendency for depressives to be even more sensitive to selfperceived covariation information than non-depressives (i.e., with regard to the intra-individual correlations). We will return in the conclusion section to this apparent discrepancy. The results of the studies that simultaneously manipulate or measure covariation information and assess attributions show that the integrative
248 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
attributional motivational model can be applied to the study of the phenomenon of depressive realism, and opens up new pathways for studying whether depressives or non-depressives are more realistic. We found that depressives’ attributions for failure are more internal, stable, and global in relation to experimentally induced (Försterling et al., 1998, Studies 1–3) and self-perceived (assessed) covariation information (Försterling et al., 1998, Study 4; Försterling & Bühner, in press, Studies 1 & 2) when compared to the causal ascriptions of non-depressives. However, the attributions of depressives are not differentially influenced by experimentally provided covariation (Försterling et al., 1998, Studies 1–3); by contrast, they are somewhat more sensitive to self-perceived covariation information when compared to the attributions of non-depressives. Meta-attributional accuracy of the mood groups. Recall that a major finding of the three studies reported above was that depressives also make more “pessimistic” attributions when covariation information is taken into account. One might argue, however, that although the attribution scale and the covariation scale both range from “1” to “10”, they might not be entirely comparable and that therefore our deduction that depressives are— overall—unrealistically pessimistic is premature. Försterling, Schuster and Morgenstern (2003) therefore used a different approach to assess attributional veridicality. They recruited heterosexual couples for their investigation and assessed—as in the studies reported above—depression scores and attributional style. In addition, participants were asked to make attributions for their partner in the depicted situations (i.e., self-partner attributions; e.g., suppose your partner would fail an important exam,…), and they had to indicate how their own behavioural outcomes would be explained by their partner (i.e., expected partner-self attributions; e.g., suppose you fail an important exam, how would your partner explain your failure?). The simultaneous assessment of the causal ascriptions of the depressives’ partner (self-partner attributions) and the attributions that the participant expects her or his partner to be making (expected partner-self attributions), in addition to attributions of the depressive person (self attributions), makes it possible to test assumptions about the veridicality of depressives’ and non-depressives’ (meta-) attributions (see Schuster, 2001, who introduced this idea). More specifically, we could test whether depressives have accurate, unrealistically pessimistic, or unrealistically optimistic assumptions about the attributions their partner would give by comparing the expected partner-self attributions to the attributions the partner actually made (self-partner attribution). We therefore computed a difference score by subtracting the expected partner-self attributions of a participant (across the eight success minus the eight failure scenarios) from the respective score of the attributions that the partner actually made (i.e., the self-partner attribution of the partner).
INTEGRATION OF ATTRIBUTION MODELS 249
When dividing our participants along the median of their depression score, we found (non-significantly) higher values for our depressive (M=–10.21) than for our non-depressive participants (M=–4.85); this difference reflects the fact that depressives tended to expect their partners to be making more depressogenic attributions than they in fact did (positive values of our attribution index reflect antidepressogenic attributions, hence a negative difference score between expected and actual attributions indicates that the actual attributions were more antidepressogenic than the expected attributions). In other words, these findings suggest that depressives are unrealistically pessimistic as to what they think their partners think about them, whereas non-depressives were comparatively accurate. When only comparing the least and the most depressed quartiles of our sample, this difference was more pronounced (M(depressed)=–16.58, M(nondepressed) =–1.46) and statistically significant. A second possibility for assessing the veridicality of participants’ attributions consists of a comparison of the intra-couple correlation between the eight success and the eight failure situations. High correlations reflect agreement and low correlations disagreement between couples as to how outcomes are attributed. A 2 (Depression Score: high vs low)×2 (Situation: positive vs negative) ANOVA with these intra-individual correlations as dependent variable revealed a marginally significant interaction. It reflected the fact that non-depressives were more accurate with regard to the partner’s attributions in the success condition, whereas depressives were more accurate for the failure conditions. When comparing only the lowest 25% and highest 25% depression groups, this interaction reached the traditional level of significance (see Figure 4). As an aside, a comparision of the attributions the mood groups made for own outcomes with those made for their partner’s outcomes revealed that depressives made clearly depressogenic attributions for themselves, but not so for their partner, suggesting that a depressogenic attributional style is not applied indiscriminately to all perspectives. Taken together, the studies that (a) simultaneously manipulate (or assess) covariation information and then assess, as dependent variable, causal attributions, and (b) compare self and partner attributions within the context of close heterosexual dyads, do not yet allow a definitive conclusion as to whether depressives make realistic or unrealistic attributions. The findings that depressives make more internal, stable, and global failure attributions in relation to induced (Försterling et al., 1998, Studies 1–3) and assessed (Försterling et al., 1998, Study 4; Försterling & Bühner, in press, Studies 1–2) covariation information, and the finding that they expect their partners to be making more internal, stable, and global attributions for themselves than they in fact do (Försterling et al., 2003), suggest that depressives have an unrealistically negative bias. Depressives
250 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 4. Mean intra-individual correlations between attributions expected from the partner and attributions actually made by the partner for depressed and nondepressed individuals separately for success and failure situations (Försterling et al., 2003).
seem to be drawing more pessimistic causal inferences from exactly the same information than do non-depressives. However, the fact that experimentally induced covariation information does not differentially influence depressives’ and non-depressives’ attributions, that depressives’ attributions are correlated more strongly with selfperceived covariation information, and that the attributions they expect from their partners are more closely correlated with the attributions their partners are actually making for them (at least for negative events), speak against the assumption that depressives are less realistic than nondepressives. These two apparently contradictory findings might best be summarised as follows: Depressives have an unrealistic negative bias operating across all kinds of different informational conditions. They seem to be adding in all informational combinations a pessimistic constant factor. However, depressives are not less sensitive to the available (covariation) information. Metaphorically speaking, they seem to be attentively observing the world with dark-coloured glasses.
INTEGRATION OF ATTRIBUTION MODELS 251
Hence, the application of our integrative attributional model of motivation to the question of depressive realism has not resulted in a definitive answer as to whether depressives are more or less realistic in their causal judgements than non-depressives, but it has shown that the question needs to be asked more differentially: One question concerns the (constant) depressogenic bias and the other one concerns differences between the two mood groups in information processing. We will return to this question in the conclusion section. NON-ATTRIBUTIONAL COGNITIVEDETERMINANTS OF DEPRESSION As indicated earlier, clinical depression constitutes a serious psychological impairment that is associated with considerable cognitive, motivational, and emotional deficits. In addition, the attributional analysis of depression maintains—and several studies have supported this idea—that depression is triggered by internal, stable, and global attributions. It follows from these hypotheses and empirical findings that therapy for depression should consist of convincing a depressed person to make fewer internal, stable and global attributions and to make more external, variable, and specific ones. It would be highly implausible from this perspective to suggest, as our integrative attributional model of motivation does, that—given a certain pattern of antecedent information—attributions to internal, stable, and global factors can be appropriate and lead to functional reactions. In fact, Taylor and Brown (1988) suggest that psychotherapy should not be concerned with the fostering of a realistic causal understanding of their clients but that it should instead teach them to maintain an unrealistically optimistic bias. Therefore, the question whether internal, stable, and global attributions necessarily lead to depression is not only theoretically interesting but also has important implications for applied issues in psychology. More relevant in the present context, however, is the question of whether attributions that are under certain conditions realistic (i.e., when failure occurs with low consensus, low distinctiveness, and high consistency) necessarily lead to a dysfunctional reaction (such as depression). Such a finding would constitute strong evidence against the central assumption of the presented model that realistic attributions lead to functional reactions. With regard to the applied implications, such a finding would suggest that therapy for depression should consist not of a change from unrealistic to realistic naive theories but to unrealistically positive naive theories. In four studies described in detail above (Försterling et al., 1998, Studies 1–4) a particularity concerning the correlation between explanatory style and depression, as assessed with the BDI, was found, which is relevant for
252 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 5. Scatter plot for BDI and ASQ scores (BDI=Beck Depression Inventory) (Försterling et al., 1998, Study 4).
this question. Figure 5 depicts the scatter plot of this correlation for Study 4. It can be seen that individuals who do not attribute failure to internal, stable, and global attributions (i.e., persons with an antidepressogenic attributional style) also typically have low depression scores. Or, in other words, individuals who make antidepressogenic attributions do not get depressed. With regard to persons with a depressogenic attributional style, Figure 5 reveals that there are about equal amounts of individuals with low and high values in depression. This finding is consistent with the previously articulated assumption that depressogenic attributions are a necessary condition for depression to occur, but that factors in addition to the depressogenic attributional style have to be present in order to trigger depression. But what might these additional factors be that determine whether depressogenic attributions lead to depression or not? Rational-Emotive Behaviour Therapy (REBT; Ellis, 1962; Ellis & MacLaren, 1998) suggests that the description of a certain causal relation (“my failure is due to lack of intelligence”) alone does not lead to an emotional reaction such as depression or sadness. In order to trigger an affective reaction, the stated causal relation needs to be evaluated (e.g., as good or bad). For Ellis, such evaluations can occur in two variants. One variant (which he labels “rational”) consists of wishes and preferences such as “I would like to
INTEGRATION OF ATTRIBUTION MODELS 253
possess high intelligence and it is unfortunate that my intelligence is low”). Such “rational” evaluations lead, according to Ellis, to appropriate (functional) emotional reactions such as sadness or disappointment when the wish is not fulfilled (e.g., one’s goal to possess high intelligence). The second variant of evaluation is labelled “irrational” and consists of demands and dogmas, such as “I must possess high ability, otherwise I am a worthless person”. Such “irrational” evaluations (or irrational beliefs) lead, according to Ellis, to depression when the respective goal is not attained. Försterling and Bühner (in press) investigated whether interindividual differences in the adherence to evaluative beliefs according to Ellis might explain the finding that only about half of the individuals with a depressogenic attributional style were depressed. More specifically, we hypothesised that depressed individuals with a depressogenic attributional style adhere more strongly to “irrational” evaluative cognitions than their nondepressed peers. To test this assumption, we assessed participants’ adherence to irrational beliefs in addition to their depression status and attributional style with the BDI and the ASQ (as in the studies described above). “Irrational” beliefs according to Ellis were assessed with an already existing instrument (the Irrational Attitudes Questionnaire [FIE; Fragebogen irrationaler Einstellungen], Klages, 1989) and with a six-item measure designed by Försterling and Bühner (in press). The six-item measure was created to avoid an overlap between the items assessing irrational beliefs and attributional style (such an overlap existed between the ASQ and the FIE). The new measure consisted of the following six items: (1) Sometimes, events happen that are not only uncomfortable but that are catastrophic and awful, (2) I must do the things that are important to me, perfectly, (3) I direly need the approval of others, (4) I cannot stand frustration, (5) after I have done something wrong, I frequently think that I am a failure, (6) there are some humans that are worthwhile and others that are worthless human beings. Participants had to indicate on scales ranging from 0 (“do not agree”) to 5 (“agree entirely”) to what degree they endorsed these beliefs. The correlation between the six-item-measure and the FIE were high in both studies (r=.77, Study 1, and r=.82 Study 2). As practically all results regarding the other variables obtained with these two measures were identical, we only report the findings using the six-item measure as this does not have the disadvantage of an item overlap. In addition to the correlation between attributional style (internal, stable, and global attributions) and depression (Study 1, r=.37; and Study 2, r=.58) we also found a significant correlation between depression and our measure of evaluative beliefs (r=.62, Study 1 and r=.53, Study 2). Moreover, both studies supported the hypothesis that depressive persons with a depressogenic attributional style more strongly adhered to irrational beliefs, according to Ellis, than non-depressives making internal, stable, and
254 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 6. Endorsement of (“irrational”) evaluative beliefs as a function of depression status (positive vs negative) (Försterling & Bühner, in press).
global attributions (see Figure 6). Hence, we found evidence for the assumption that attributions of failure to internal, stable, and global factors result in depression especially when “irrational” evaluative ideas are present. In addition, the findings are consistent with the assumption that individuals tend to be non-depressed when they make internal, stable, and global attributions in the absence of “irrational” evaluations. This finding has important theoretical and applied implications. With regard to our integrative model, it suggests that the advantages of making realistic attributions to internal, stable, and global factors are not necessarily accompanied by (dysfunctional) depressive reactions. By contrast, our results showed that many individuals make such attributions while remaining undepressed and that these persons are characterised by a lack of adherence to irrational cognitions according to Ellis. It follows— with regard to the applied implications—that therapeutic cognitive change for the treatment of depression can occur in different ways. One way consists of changing internal, stable, and global attributions to external, variable, and specific ones. A second possibility, however, would consist of an attempt to change evaluative cognitions, according to Ellis, while leaving internal, stable, and global attributions unchanged. This strategy would seem especially advisable when the “depressogenic” attributions are realistic. Of course, therapy could also consist of both a change of
INTEGRATION OF ATTRIBUTION MODELS 255
attributions and of evaluative cognitions, especially when the attributions are unrealistic (when unrealistic depressogenic attributions are made). CONCLUSIONS This chapter has presented a model integrating conceptions about the antecedents with those about the consequences of causal thoughts, a resulting definition of attributional veridicality, and recent empirical research designed to test the model in the contexts of achievement and depression. A new experimental paradigm for achievement situations was introduced in which overall performance is defined as the sum of the performances at subtasks that require different abilities. This paradigm revealed that realistic attributions lead to better overall performance than unrealistic attributions, and that unrealistic attributions of failure to non-ability factors have negative consequences for achievement results. Such a disadvantage of unrealistic attributions to high ability has not yet been voiced and/or experimentally demonstrated in any of the previous attributional models of motivation. With regard to the attributional analysis of depression, we have demonstrated that the integrative attributional model and the resulting definition of the veridicality of a causal attribution can be used to investigate whether depressives are more realistic than non-depressives. Two indicators of attributional veridicality were used, and each was assessed in two variants. The first indicator measured sensitivity to variations in covariation. This research (a) assessed whether depressives and non-depressives made different attributions depending on variations in covariation information and (b) calculated intra-individual correlations between perceived covariation information and attributions. The second indicator consisted of (a) deviation scores that related the absolute magnitude of the aggregated attribution score to actual and self-perceived covariation information, and (b) the attributions expected from the partner compared with the attributions the partner actually made. With regard to the first indicator (sensitivity to variations of covariation), we found that experimentally induced covariation information is processed similarly by depressives and non-depressives. In the studies that assessed (rather than experimentally manipulated) covariation information, we even found a (weak but consistent) trend indicating that causal attributions of depressives more closely reflect selfperceived covariation information than non-depressives’ causal inferences (i.e., higher intra-individual correlations). These findings seem to suggest that depressives are more realistic than non-depressives. Thus, these results are, at first glance, inconsistent with our model’s assumption that depressives should be more unrealistic than non-depressives. However,
256 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
these findings fit with results and theoretical interpretations of studies investigating the role of deprivation of control on the attribution process and with findings concerning depressive realism. For instance, Pittman and Pittman (1980) demonstrated that participants who made attributions following the experience of uncontrollability (a typical antecedent of depression) made attributions that were more in line with antecedent information than did individuals who were not deprived of control. And Alloy and Abramson (1988) reported that depressives performed better at a contingency detection task than non-depressives. With regard to the second indicator of veridicality (overall deviation scores), however, all our studies consistently revealed that given identical (experimentally provided or self-perceived) covariation information, depressives made more internal, stable, and global attributions for failure. For some of these studies it cannot be conclusively determined whether the mood group differences regarding this indicator reflect unrealistic pessimism or realism of depressives, or unrealistic optimism or realism of non-depressives. However, the study investigating (meta-)attributions of members of heterosexual couples used an objective criterion for “veridicality” (i.e., a comparison of the participants’ expectation of their partners’ attributions with the attributions partners actually made). The findings of this study suggested that depressives distort reality in a pessimistic fashion. Taken together, the findings regarding the second indicator (deviation scores) seem to suggest the existence of a negative bias among depressives, i.e., a “pessimistic” constant factor that is “added” in all covariation conditions. The finding of an (unrealistic) negative bias of depressives is compatible with our integrative attributional model and with Beck’s (1976) influential clinical theory of depression. Beck (1976) maintains that depressives have a tendency to see their own person, their situation, and their future in an unrealistically negative fashion (i.e., the cognitive triad). The divergent results of our two indicators of veridicality of depressives’ causal attributions mirror divergent findings and theoretical assumptions reported in the literature: the loss of control literature suggesting accurate information processing and Beck’s theory suggesting unrealistic pessimism. How can these findings be explained? The divergent findings would make sense when separating cognitive processes that trigger a depression from those that are instigated by the depressive episode. The unrealistic pessimistic bias that we find in our “second” veridicality measure (and which is central in Beck’s theory) might trigger and possibly maintain depression. Once triggered, however, the affective state of depression might induce the individual to examine information more carefully. In fact, the state of depression might serve the function of recalibrating one’s unrealistic appraisals, expectations and naive theories by inducing the individual to process information more analytically. Results of our first
INTEGRATION OF ATTRIBUTION MODELS 257
measure of veridicality, and the findings of the loss of control and depressive realism literature might reflect this mechanism. A test of this interpretation of our results, which can be summarised in the metaphor of depressives attentively observing the world with dark-coloured glasses, would however require longitudinal studies. On a metatheoretical level, I would like to point out that the integrative attributional model—in contrast to other attributional models of motivation (e.g., Abramson et al., 1978; Weiner, 1986)—is based on a relational concept: The “fit” or “relation” between an attribution and antecedent covariation information is viewed as a determinant of the functionality of a psychological reaction and not the attribution alone. This implies—on a methodological level—that our major independent or predictor variables are not merely attributions or attribution antecedents but indicators of relations between these two concepts such as correlations between attributions and covariation information. A further important finding concerns non-attributional cognitive correlates (possibly determinants) of depression. We have documented that the often reported correlation between attributional style and depression is not orthogonal: Among individuals with a depressogenic attributional style, we find about an equal amount of depressives and non-depressives. This suggests that internal, global, and stable attributions are not necessarily associated with dysfunctional reactions but that under certain circumstances, one can make such attributions without risk of becoming depressed. This finding is important for our model as it suggests that realistic attributions to internal, stable, and global causes should lead to functional reactions. The studies have shown that depressed individuals who make internal, stable, and global attributions adhere more strongly to evaluative “irrational” beliefs according to Ellis than non-depressed individuals with such an attributional style. Consistent with Ellis’ psychotherapy theory, individuals who make internal, stable, and global attributions for failure but do not, in addition, believe that they “must” succeed, that it is “awful” not to succeed, or that “individuals with low intelligence are worthless people”, are less depressed than individuals who endorse such irrational beliefs. The findings about the non-attributional determinants of depression are also relevant for cognitive psychotherapy of depression. They suggest that such therapeutic attempts can go two different, but not mutually exclusive, ways. One way implies changing internal, stable, and global failure attributions of depressed individuals to external, variable, and specific ones (i.e., attributional retraining). However, this way might be inappropriate when the “depressogenic” attribution is realistic from the perspective of the available covariation information (for instance, when a person does consistently fail at many different exams, whereas others succeed). For
258 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
these individuals, the second way to achieve cognitive change might be more appropriate: This way would imply not changing the (realistic) “depressogenic” attribution but modifying the evaluation of the event. Quite in line with the practice of Rational-Emotive Behaviour Therapy, according to Ellis, these individuals might benefit from being taught to accept the “pessimistic” attribution, to accept themselves still as worthwhile human beings, and not to consider the respective failure as catastrophic. REFERENCES Abramson, L.Y., Seligman, M.E.P., & Teasdale, J.D. (1978). Learned helplessnesss in humans. Journal of Abnormal Psychology, 87,49–74. Ackermann, R., & DeRubeis, R.J. (1991). Is depressive realism real? Clinical PsychologyReview, 11,365–384. Alloy, L.B., & Abramson, L.Y. (1988). Depressive realism: Four theoretical perspectives. In L. B.Alloy (Ed.), Cognitive processes in depression(pp. 223– 265). New York: Guilford Press. Beck, A.T. (1976). Cognitive therapy and the emotional disorders.New York: International Universities Press. Beck, A.T., Ward, C.H., Mendelson, M., Mock, J., & Erbaugh, J. (1961). An inventory for measuring depression. Archives of General Psychiatry, 4,561– 571. Brewin, C.R. (1985). Depression and causal attributions: What is their relation? PsychologicalBulletin, 98,297–309. Brewin, C.R., & Furnham, A. (1986). Attributional versus preattributional variables in self-esteem and depression: A comparison and test of learned helplessness theory. Journal ofPersonality and Social Psychology, 50,1013– 1020. Buss, D.M. (1995). Evolutionary psychology: A new paradigm for psychological science. Psychological Inquiry, 6,1–30. Campbell, J.D., & Fairey, P.J. (1985). Effects of self-esteem, hypothetical explanations, and verbalizations of expectancies on future performance. Journal of Personality and SocialPsychology, 48,1097–1111. Carey, S., & Gelman, R. (Eds.) (1991). The epigenesis of mind. Essays in biology and cognition.Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Colvin, C.R., & Block, J. (1994). Do positive illusions foster mental health? An examination of the Taylor and Brown formulation. Psychological Bulletin, 116,3–20. Cosmides, L. (1989). The logic of social exchange. Has natural selection shaped how humans reason? Cognition, 31,187–276. Dweck, C.S. (1999). Self-theories: Their role in motivation, personality, and development.Philadelphia, PA: Psychology Press. Dweck, C.S., Chiu, C., & Hong, Y. (1995). Implicit theories and their role in judgments and reactions: A world from two perspectives. Psychological Inquiry, 4,267–285.
INTEGRATION OF ATTRIBUTION MODELS 259
Ellis, A. (1962). Reason and emotion in psychotherapy.Secaucus, NJ: Citadel Press. Ellis, A., & MacLaren, C. (1998). Rational Emotive Behavior Therapy: A therapists’ guide.Atascudero, CA: Impact Publishers. Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7, 117–140. Försterling, F. (1985). Attributional retraining: A review. Psychological Bulletin, 98, 495–512. Försterling, F. (1986). Attributional conceptions in clinical psychology. American Psychologist,41,275–285. Försterling, F. (1988). Attribution theory in clinical psychology.Chichester, UK: John Wiley & Sons. Försterling, F. (1989). Models of covariation and causal attribution: How do they relate to the analogy of analysis of variance? Journal of Personality and Social Psychology, 57,615–625. Försterling, F. (1992). The Kelley Model as an analysis of variance analogy: How far can it be taken? Journal of Experimental Social Psychology, 28,475–490. Försterling, F. (1994). The functional value of realistic causal attributions. In W.Stroebe & M. Hewstone (Eds.), European review of social psychologyVol. 5 (pp. 151–179). Chichester, UK: Wiley. Försterling, F. (2001). Attribution: An introduction to theories, research and applications.Hove, UK: Psychology Press. Försterling, F., & Bühner, M. (in press). Attributional veridicality and evaluative beliefs: How do they contribute to depression? Journal of Social and Clinical Psychology. Försterling, F., Bühner, M., & Gall, S. (1998). Covariation information, attributions and depression: A test of depressive realism. Journal of Personality and Social Psychology, 75,1047–1061. Försterling, F., & Morgenstern, M. (2002). Accuracy of self-assessment and task performance: Does it pay to know the truth? Journal of Educational Psychology, 94,576–685. Fösterling, F., & Morgenstern, M. (2003). Naive theories about accuracy of self assessments andperformance.Manuscript submitted for publication. Försterling, F., & Rudolph, U. (1988). Situations, attributions and the evaluation of behavior. Journal of Personality and Social Psychology, 54,225–232. Försterling, F., Schuster, B., & Morgenstern, M. (2003). Perspectives of causal attributions: Astudy of couples and depression.Manuscript submitted for publication. Gigerenzer, G., & Hug, K. (1992). Domain-specific reasoning: Social contracts, cheating and perspective change. Cognition, 42,127–171. Glass, D.C, McKnight, J.D., & Valdimarsdottir, H. (1993). Depression, burnout, and perceptions of control in hospital nurses. Journal of Consulting and Clinical Psychology, 61,147–155. Gopnik, A., Meltzoff, A.N., & Kuhl, P.K. (1999). The scientist in the crib: Minds, brains andhow children learn.New York: Morrow. Hammen, C. (1999). Depression.Hove, UK: Psychology Press. Heider, F. (1958). The psychology of interpersonal relations.New York: Wiley.
260 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Kelley, H.H. (1967). Attribution theory in social psychology. In D Levine (Ed.), NebraskaSymposium on Motivation and Emotion Vol. 15, (pp. 192–238). Lincoln: University of Nebraska Press. Kelley, H.H. (1971). Attribution in social interaction.Morristown, NJ: General Learning Press. Kelley, H.H. (1973). The process of causal attributions. American Psychologist, 28, 107–128. Klages, U. (1989). Fragebogen irrationaler Einstellungen [Irrational Beliefs Questionnaire].Göttingen: Testzentrale. Kruglanski, A.W. (1980). Lay epistemology process and contents. Psychological Review, 87,70–87. Kuiper, N.A. (1978). Depression and causal attributions for success and failure. Journal ofPersonality and Social Psychology, 36,236–246. Kukla, A. (1972). Foundations of an attributional theory of performance. Psychological Review,79,454–470. Meyer, W.-U (1973). Leistungsmotiv und Ursachenerklärung von Erfolg und Mi erfolg[Achievement motive and causal explanations for success and failure].Stuttgart: Klett. Meyer, W.-U (1984). Das Konzept von der eigenen Begabung [The self concept of ability].Bern: Huber. Meyer, W.U. (2000). Gelernte Hilflosigkeit: Grundlagen und Anwendungen in Schule undUntericht.Bern: Hans Huber. Meyer, W.-U, & Försterling, F. (1993). Die Attributionstheorie. In M.Irle & D.Frey (Eds.), Theorien der Sozialpsychologie. Band L Kognitive Theorien(pp. 175– 214). Bern: Huber. Mill, J.S. (1872). A system of logic(8th edn). London: Longmans, Green, Reader & Dyer. Perry, R.P., Hechter, F.J., Menec, V.H., & Weinberg, L.E. (1993). Enhancing achievement motivation and performance in college students: An attributional retraining perspective. Research in Higher Education, 34,687–723. Peterson, C., & Seligman, M.E.P. (1984). Causal explanations as a risk factor for depression: Theory and evidence. Psychological Review, 91,347–374. Peterson, C., Semmel, A., von Baeyer, C., Abramson, L.Y., Metalski, G.I., & Seligman, M.E.P. (1982). The attributional style questionnaire. Cognitive Therapy and Research, 6,287–299. Pittman, T.S., & Pittman, N.L. (1980). Deprivation of control and the attribution process. Journal of Personality and Social Psychology, 39,377–389. Rescorla, R.A. (1988). Pavlovian conditioning: It’s not what you think it is. AmericanPsychologist, 43,151–160. Rizley, R. (1978). Depression and distortion in the attribution of causality. Journal of AbnormalPsychology, 87,32–48. Schuster, B. (2001). Rejection and victimization by peers: Social perception and social behavior mechanisms. In J.Jovonen & S.Graham (Eds.), Peer harassment in school(pp. 290–309). New York: Guilford. Schuster, B., Rudolph, U., & Försterling, F. (1998). Attributions or covariation information: What determines behavioral reaction decisions? Personality and Social Psychology Bulletin,24,838–854.
INTEGRATION OF ATTRIBUTION MODELS 261
Shanks, D.R., Medin, D.L., & Holyoak, K.J. (1996). Causal learning. The psychology oflearning and motivation, 34.San Diego, CA: Academic Press. Taylor, S., & Brown, J. (1988). Illusion and well-being: A social-psychological perspective on mental health. Psychological Bulletin, 103,193–210. Trope, Y. (1975). Seeking information about one’s ability as a determinant of choice among tasks. Journal of Personality and Social Psychology, 32,1004– 1013. Trope, Y., & Brickman, P. (1975). Difficulty and diagnosticity as determinants of choice among tasks. Journal of Personality and Social Psychology, 31,918–926. Trope, Y., & Liberman, A. (1996). Social hypothesis testing: Cognitive and motivational mechanisms. In. E.T.Higgins & A.W.Kruglanski (Eds.), Social Psychology: Handbook ofbasic principles(pp. 239–2370). New York: Guilford Press. Weiner, B. (1985). An attributional theory of emotion and motivation. Psychological Review,92,548–573. Weiner, B. (1986). An attributional theory of motivation and emotion.New York: Springer. Weiner, (1992). Human motivation. Metaphors, theories and research.Newbury Park, CA: Sage. Weiner, B. (1995). Judgements of responsibility: A foundation for a theory of social conduct.New York, London: The Guilford Press. Weiner, B., Frieze, I.H., Kukla, A., Reed, L., Rest, S., & Rosenbaum, R.M. (1971). Perceivingthe causes of success and failure.New York: General Learning Press. Wilkening, F., & Lamsfuß, S. (1993). (Miß-) Konzepte der naiven Physik im Entwicklungsverlauf. In W.Hell, K.Fiedler, & G.Gigerenzer (Eds.), Kognitive Täuschungen: FehlLeistungen und Mechanismen des Urteilens, Denkens und Erinnerns(pp. 271–290). Heidelberg: Spektrum.
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY, 2002, 13, 259–291
Morality and political orientations: An analysis oftheir relationship Nicholas Emler The London School of Economics, UK
Politics can be seen as a rule-governed contest between competing definitions of the morally desirable. This view gives rise to the following questions, examined in this chapter: How do individuals’ moral characteristics shape their choice of sides in this contest? How are the rules of this contest interpreted, and how and why do individuals differ in their interpretation of these rules? What determines the degree to which people become engaged in this contest? Answers derived from cognitive constructivism, for the last 30 years the dominant perspective in moral psychology, are contrasted with a socio-genetic interpretation. This latter approach treats moral perspectives as social identities that are socially constructed and communicated. The chapter reports on a research programme examining how moral reasoning communicates a political position. It also gives particular attention to the effects of participation in formal education upon the construction of political attitudes and identities.
© 2002 European Association of Experimental Social Psychology http://www.tandf.co.uk/journals/pp/02699931.html DOI: 10.1080/10463280240000082
MORALITY AND POLITICS 263
INTRODUCTION Politics is a moral contest, in two senses. First it is a competition between conflicting views of what is morally most desirable, what the priorities for a community or society should be and how they should be achieved. Second, it is a contest with rules about the manner in which it is to be conducted. It is also, however, a contest that occurs between collectives rather than individuals. To participate, individuals must align themselves temporarily or more permanently with one side or another in this contest. From the perspective of social psychology, the following questions emerge from this analysis: (1) how does the individual choose sides? (2) How does the individual interpret the rules of this contest? (3) To what extent does the individual participate at all? This is by no means an exhaustive list for a social psychology of politics, but it is a coherently interlinked set of questions about the moral character of political participation. It also provides the structure of this chapter. The first part considers the moral differences underlying politicalideological differences on a left-right dimension. It examines interpretations of these differences offered by alternative moral psychologies, giving attention principally to the cognitive constructivist perspective. The second part contrasts this with a socio-genetic interpretation of moral and political differences as socially constructed. This interpretation emphasises the expression and communication of political identities. The third part explores the case for inter-individual variations along a second dimension concerned with democratic principle and procedure, reflecting cognitive differences and shaped by participation in formal education. The fourth part returns to the social construction of political identities. It discusses variations in political engagement, including the motivational, cognitive, attitudinal, and behavioural elements that go to make up engagement, and the social processes that shape levels of engagement. MORALITY AND POLITICAL IDEOLOGY Most research into the psychological character and correlates of political orientation has assumed that different orientations can be ordered along an ideological “left-right” or radicalism-conservatism dimension.1 The implication is that the position an individual occupies on such a dimension is the basis for his or her behaviour as a political actor. This position shapes choices as to what issues, policies, agendas, or political figures to
Address correspondence to: Nicholas Emler, Department of Psychology, University of Surrey, Guildford, Surrey, GU2 7XH, UK, E-mail:
[email protected]
264 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
support or oppose, whether through passive acquiescence, or more active campaigning, demonstrating, organising collective action, joining parties or collective movements, or through the ballot box. There are several assumptions here and almost every one has its critics. Among these are arguments that social attitudes are no longer organised in terms of broad overarching ideological frameworks (Eagly & Chaiken, 1993), that left-right has ceased to be a meaningful representation of political differences (Inglehart, 1990) if indeed for a majority of adults it ever was (Converse, 1964), that in any event the left-right dimension is not a single dimension (Conover & Feldman, 1981; Kerlinger, 1984), that political action has more potent determinants than ideological commitments (Downes, 1957), and in particular that voting is seldom determined by ideological beliefs (Campbell, Converse, Miller, & Stokes, 1960). These criticisms do need to be taken seriously—and some more seriously than others—but they provide grounds for some qualification of the above analysis, not for its wholesale rejection. However, there is not the space to discuss them all in detail. Factor analytic studies of social attitudes from the 1930s through to the present have routinely produced the anticipated solution, namely a factor that can be identified with the left-right dimension (e.g., Bynner, Romney, & Emler, in press; Delli Carpini, & Keeter, 1996; Evans, Heath, & Lalljee, 1996; Eysenck, 1954; Fleishman, 1986; Heath, Evans, & Martin, 1993; Himmelweit, Humphreys, & Jaeger, 1981; Judd, Krosnick, & Milburn, 1981; Tygart, 1984; Wilson, 1973). This is strongly suggestive that people do organise their social attitudes in terms of a broad ideological left-right dimension, and that this remains true. This dimension has not been replaced by variations in terms of the kinds of post-materialist values described by Inglehart (1990; see e.g., Evans et al., 1996) or by new polarities organised around such issues as environmental concerns (Bynner et al., 2000). The most significant qualification to be added to this conclusion, however, is that the tendency for attitudes to be organised around a left-right dimension is itself variable; it is not equally true of all individuals. This issue forms the focus for the final part of this chapter. The remainder of this section examines explanations of ideological differences as moral differences. The three most influential traditions of moral psychology have been based respectively in psychoanalysis, behaviourism, and cognitive
1 In American research this dimension is generally identified as conservatismliberalism. But in some political cultures liberalism is identified with the political centre and in others with the political right. For these reasons, the simpler term “left-right” is preferred here.
MORALITY AND POLITICS 265
constructivism. Each has been the basis for a distinctive interpretation of the moral character of political views. Psychoanalysis took moral development to be the development of conscience, but was less interested in the consequences of a weakly formed conscience than in those of a conscience that was too strong. For Adorno, Frenkel-Brunswik, Levinson, and Sanford (1950) anti-Semitism and other kinds of racism were projections into social relations of an unduly rigid superego, generating punitive attitudes to all kinds of dissent and nonconformity, along with craven deference to symbols of authority. Because anti-Semitism and ethnocentrism turned out in their research to be correlated with political and economic conservatism, this suggested that such differences in moral perspective also predispose individuals towards particular ideological positions in the political sphere (cf. also Wilson, 1973). The classical psychoanalytic explanation for conscience development—in terms of castration anxiety and its resolution—has found little empirical support in the longer run, and Adorno et al.’s own developmental theory was never persuasively supported (Brown, 1965). But the implied explanation for politics attracted criticism mainly because it stigmatised the political right as morally immature and unenlightened. Eysenck (1954), Rokeach (1960), and others took the view that some of the same qualities of immaturity, rigidity, and intolerance also characterise the far left. Tetlock (1983a) aptly characterised this as the “ideologue hypothesis”: “… true believers, regardless of their specific cause, are more dogmatic, more intolerant of ambiguity and inconsistency, and more conceptually simple than moderates…” (p.119). Eysenck’s own psychological account of politics and social attitudes (Eysenck, 1954) was rooted in a behaviourist analysis of personality functioning. He accepted that social attitudes can be arrayed along a leftright dimension, but saw choice of position on this dimension as essentially an operant response. Just as for the Skinnerian morality reduces to doing habitually what is most consistently rewarded, so Eysenck proposed that individuals hold those political attitudes that prove the most rewarding to hold. However, Eysenck argued that social attitudes also vary along a second dimension, which he labelled tough- vs tender-minded; moreover this dimension corresponds to moral differences of another kind, namely differences in conscience strength. Eysenck interpreted conscience as conditioned anxiety, and proposed that tough-minded individuals were less likely to have formed these conditioned responses. Hence they would be more concerned with immediate gratification of aggressive and sexual impulses. He further anticipated that the supporters of different political parties would be distributed along this dimension from extremist to moderate, with communists and fascists together at the tough-minded end,
266 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
(British) liberal party supporters at the tender-minded end, and supporters of conservative and labour parties both in between. Eysenck’s moral psychology has received little empirical support (Emler & Reicher, 1995) but as an analysis of political ideology it also failed because the predicted association between tough-mindedness and far left politics was not substantiated; more specifically, communists did not get the same high scores on Eysenck’s tough-mindedness dimension as did fascists (Brown, 1965). The related argument advanced by Rokeach (1960), which linked political extremism to a close-minded cognitive style, suffered a similar fate: communists and supporters of other far left parties did not score more highly than moderates on measures of rigidity or dogmatism (Stacey, 1978). The third major tradition of moral psychology is represented by cognitive constructivism and in particular by the Piaget-inspired theorising and research of Lawrence Kohlberg (1984). An interpretation of political ideology from the perspective of this tradition, in contrast to those considered above, appears to be extensively supported by empirical evidence (Rest, Narvaez, Bebeau, & Thoma, 1999). So what is this interpretation, and is it the most appropriate one that can be made of the evidence available? Cognitive constructivism and political orientation A key observation underlying Kohlberg’s moral psychology is that most people endorse the same broad moral values—truth, the value of life, property rights, promise keeping, and so forth (Kohlberg, 1984). However, there are occasions when two or more of these values will be in conflict; realising one of them is incompatible with upholding another. The interesting differences among people are in how they reason about these values in such cases. Kohlberg’s method for studying this “moral” reasoning was to pose hypothetical dilemmas in which two or more values were in conflict and to question research participants about how these dilemmas should be resolved. Inspired by Piaget’s stage theory of cognitive development (Piaget, 1952; see also Flavell, 1963), Kohlberg (1984) argued that the manner in which an individual reasons about such moral questions has structural properties. That is to say, it forms a coherent, unified system of thinking about the nature of morality and about the basis of moral obligations towards others and the wider society. Kohlberg further proposed that this system of thinking undergoes qualitative change in the course of an individual’s development. He envisaged these changes as developmental stages; at each successive stage the system assumes a new and more cognitively complex form, replacing that used at the previous stage.
MORALITY AND POLITICS 267
Kohlberg interpreted his stages in a strict Piagetian sense as naturally occurring in an invariant, universal, and irreversible sequence. The developmental scale is also one of adequacy or adaptation (Kohlberg, 1971). Thus, higher stages are in a real sense better than lower stages; they are adaptive with respect to a wider range of moral problems— which is to say value-conflicts—and generate more adequate solutions to these problems. Kohlberg further supposed that moral stages are like Piagetian cognitive stages in setting an upper limit upon the individual’s capacity to understand; in the case of moral stages the limit is reflected in the individual’s ability to comprehend moral arguments supplied by others. Kohlberg also took a Piagetian view of developmental process. Individuals do not adopt patterns of thinking transmitted by parents or other “agents or socialisation”. Rather they progressively construct their own capacities to analyse moral conflicts, and do so spontaneously and independently. This notion of intrinsic tendencies towards internally consistent systems of thinking bears some resemblance to the concepts of balance (Heider, 1958) and psycho-logic (Abelson, 1968) in the field of attitudes. The difference is that Kohlberg believed cognitive balance and its associated logic take progressively more complex forms as reasoning develops. As to the number of stages through which moral reasoning may develop, Kohlberg thought that at least six were theoretically possible even if in practice most individuals were unlikely to progress so far. And in practice only five stages seemed to be identifiable in his research samples (Kohlberg, Levine, & Hewer, 1983). Moreover, with respect to political ideology the critical division has proved to be between the fourth and fifth. Kohlberg’s own terms for these stages were respectively “social systems and conscience” and “social contract/utility and individual rights”. Because he also saw these stages as either side of a more fundamental divide, between an approach to moral dilemmas in terms of shared social conventions and support for the established social order, and one in terms of universalisable principles, the distinction is referred to hereafter as social order versus principled reasoning. The cognitive constructivist explanation for political orientations is that they are the consequences of applying to the political domain moral perspectives developed initially to judge personal moral choices and interpersonal behaviour (Fishkin, Keniston, & MacKinnon, 1973; Rest, Cooper, Coder, Masanz, & Anderson, 1974). Social order reasoning leads an individual to adopt one kind of political posture, principled reasoning leads to the adoption of another. One implication is that development from the social order to the principled stage should produce a shift in political position: “at this specific point in moral judgment development there is a shift in political attitudes” (Rest, Thoma, & Edwards, 1997, p.5).
268 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Several studies have tested the prediction derived from this explanation that “knowledge of an individual’s competence in moral judgments should provide insight into his or her political choices and actions” (Thoma, 1993, p.359). Every test of this prediction drawing upon Kohlberg’s analysis of moral reasoning has shown the same thing: social order reasoning is associated with an inclination to favour the political right, while principled reasoning is associated more strongly with support for the political left. The same association has been found in several different political cultures (albeit all Western democracies), at different times, and using a variety of indicators of ideological orientation, including voting intentions, endorsement of parties and candidates, support for policies, issues, and causes, approval of political slogans, measures of social attitudes, and expressed political identities (e.g., Alker & Poppen, 1973; Barnett, Evens, & Rest, 1995; Candee, 1976; Emler, Remwick, & Malone, 1983; Fishkin et al., 1973; Fontana & Noel, 1973; Haan, Smith, & Block, 1968; Markoulis, 1989; Nassi, Abramonitz, & Youmans, 1983; O’Connor, 1974; Rest, 1975; Rest et al., 1974; Sampson, 1971; Sullivan & Quarter, 1972; Thoma, 1993). Despite the highly consistent pattern to the evidence, the cognitive constructivist interpretation of these findings is not without problems (for further discussion of these, see Barnett et al., 1995; Emler, PalmerCanton, & St. James, 1998). Perhaps the most serious is that if those on the political left are separated from those on the right by the different positions they occupy on a developmental scale then this should be reflected in other indicators of relative developmental level. Thus moral reasoning development is correlated with variations in intellectual level and with conditions associated with such variations, most obviously amount and level of participation in formal education (Colby & Kohlberg, 1987; Rest & Thoma, 1985). But the relevant variations in political orientation are not similarly correlated either with IQ scores or with level of education (Bynner et al., in press; Nie, Junn, & Stehlik-Barry, 1996). To summarise, then, the left-right polarity continues to be a reasonable representation of ideological differences in the political domain. In so far as people hold views on a range of policy issues these views tend to be organised around this polarity such that individuals can be placed in terms of their politics somewhere along this dimension. Attempts to explain the position occupied on this dimension based on the moral psychologies of, respectively, psychoanalysis and behaviourism were unsuccessful. A more recent interpretation, based on the cognitive constructivist view of morality, has generated an extensive body of evidence linking political positions to distinctive ways of reasoning about moral questions. But there are serious and perhaps fatal difficulties for the cognitive constructivist interpretation of these links. The next section considers an alternative interpretation.
MORALITY AND POLITICS 269
MORALITY AND POLITICS FROM ASOCIOGENETIC PERSPECTIVE The objective of this section is to sketch out a quite different interpretation of the association between moral reasoning and political orientation from those outlined above. This interpretation (Emler, 1983; Emler et al., 1998; Emler & Stace, 1999) concurs with the Kohlbergian position that supporters of left and right reason about moral questions in different ways and have different views about how conflicts between morally valued ends should be resolved. It also concurs with the proposition that ideological divisions in the political domain reflect differences in the way that basic values such as individualism and social equality are prioritised (cf. Kinder, 1998; Rokeach, 1979; Tetlock, 1983a). It differs from the cognitive constructivist position with respect to four distinctive claims that can be summarised briefly as follows. First, social order and principled moral reasoning are differentiated in terms of the content and organisation of the ideas they respectively embody but they are not consecutive stages on a developmental scale. The second claim follows from the first: these two forms of reasoning are equally available to supporters of left- and right-wing political positions; there is no asymmetry of access based on differences in developmental level. Third, political positions are also social roles and not just internal psychological states or dispositions. This draws attention to the functions of moral reasoning in communicating and performing a political role. Fourth, coherence in ideological positions is not generated by underlying cognitive structures; it is normatively defined and generated through social influence. For this reason this interpretation is described as a socio-genetic perspective (cf. Moscovici, 1976). This fourth proposal will be examined further in the final section. Here discussion focuses on the role of moral reasoning in the expression of political identity.2 Equality of access to social order andprincipled reasoning It is not proposed that everyone without qualification has the same degree of access to both social order and principled reasoning. On the contrary, there are likely to be prerequisites in terms of cognitive frameworks for the generation, interpretation, and comprehension of both kinds of argument. The point is rather that these prerequisites are not different, and not differentially available respectively to supporters of left and right. Conservatives, therefore, do not employ social order arguments because they have no access to principled moral arguments, but because the former more often, though not necessarily invariably, serve their social goals. Sparks and Durkin (1987) provide a persuasive demonstration of this point showing that occasionally principled arguments will better serve a
270 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Conservative’s goals and social order arguments will better serve those of their political opponents and under these circumstances each will readily accept the switch. Two earlier studies had tested the prediction that both kinds of moral argument are more or less equally available to people who differ ideologically (Emler, 1983; Emler et al., 1983). Participants in these studies, who defined themselves as left-wing, right-wing, or moderate, displayed spontaneous preferences for social order versus principled moral arguments in line with their politics; right-wingers showed a stronger preference for social order arguments, left-wingers for principled arguments. They were then asked to predict the kinds of argument that either a left-winger or a right-winger would prefer. If both forms of moral argument are equally accessible to supporters of both left and right, and if furthermore they all recognise the political positions conveyed by each form, then they should accurately predict the preferences of left and right irrespective of their own political leanings or preferred styles of moral reasoning. These predictions are at the same time inconsistent with view that people who spontaneously employ social arguments do so only because principled arguments are not accessible to them. The predictions were borne out and were subsequently replicated independently (Markoulis, 1989). Tetlock (1981) has made a similar point about cognitive complexity. Politicians and others may differ in their tendency to use cognitively complex versus simple arguments not because they are more or less able to do so, but by virtue of the different rhetorical goals served by complexity versus simplicity. The same user may therefore switch from complexity to simplicity and back again depending on the rhetorical needs of the moment. One such rhetorical need may be to communicate the particular political position one occupies. This brings us to the proposition that political positions are social roles and identities, communicated to others. Moral reasoning communicates a political position Part of the function served by the moral reasoning an individual chooses to employ is to express or communicate a particular political identity. Thus, if one side of the coin is the ability of, for example, right-wingers to predict that a left-winger will prefer and use principled moral arguments, the other side of the coin is the capacity to recognise that such arguments convey a leftwing identity. In other words, although alternative styles of moral
2
The reader interested in the detail of the argument and evidence against a stage developmental interpretation of the differences between social order and principled reasoning should consult Emler (1983) and Emler et al. (1998).
MORALITY AND POLITICS 271
reasoning are not the cause of political identities, they can be the instruments of their expression. Several studies bear upon this proposal. For example, Reicher and Emler (1984) found that a person described as selecting principled moral arguments was seen to differ from someone preferring social order arguments in political identity but not in cognitive qualities or moral virtue. However, this study employed a measure of moral reasoning, the Defining Issues Test (DIT; Rest et al., 1974), that does not operationalise social order and principled reasoning precisely as Kohlberg had done. It is possible that although the moral arguments in the DIT convey political identities, arguments that Kohlberg took to be definitive of the relevant forms of reasoning do not. We therefore repeated the Reicher and Emler study but using moral arguments taken from Kohlberg’s own coding manual (Emler & Stace, 1999). The participants in this study were presented with one of three different samples of moral argument from the Moral Judgment Interview coding manual (Colby & Kohlberg, 1987) and identified there as respectively examples of Stage 4A, Stage 4B, or Stage 5 reasoning. Sub-stage 4B is supposedly intermediate between 4A and 5; i.e., it is a mixture of social order and principled reasoning (see Kohlberg, 1984, pp.666–683, for a discussion of the A-B sub-stage distinction). Participants (n=211), drawn in approximately equal numbers from the United States and from Britain, were asked to evaluate a person supposedly using one of these forms of moral argument on a series of scales. The scales related to five kinds of variable, namely political, cognitive, moral, personal (e.g., warm-cold, masculine-feminine), and demographic attributes (e.g., age, social background, occupational status). Most relevant here are results relating to aggregate measures that were formed from scales in the first three categories. Factor analysis of the political scales supported a two-factor solution (eigenvalues 2.86 and 1.03, and accounting for respectively 41.5% and 16.8% of the variance). The two aggregated political scales formed on this basis were labelled “radical activism” and “progressive-left”. Factor analysis of the cognitive scales also indicated a two-factor solution (eigenvalues 2.80 and 1.09; variance accounted for 46.7% and 18.1%). The aggregated scales here were labelled “cognitive sophistication” and “intellectual maturity”. Only a single factor was identified underlying the moral scales (eigenvalue 2.60; variance accounted for 37.2%). The corresponding scale was labelled “moral virtue”. As can be seen from Table 1, the only differences in the impressions conveyed by the different forms of moral reasoning were with respect to the two political dimensions. In effect, someone using principled reasoning is seen as no more intellectually mature, cognitively sophisticated or morally virtuous than someone using social order reasoning. Principled
272 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
TABLE 1 Perceptions of traits of targets differing in moral reasoning (adapted from Emler & Stace, 1999)
Means on 7-point scale. Higher values reflect ratings closer to the intellectual maturity, cognitive sophistication, moral virtue, radical activism, and progressive left poles. Means in a row with different superscripts differ at p<.05.
reasoning does not, as one would expect if it is a higher stage in a developmental sequence, convey to others greater cognitive or moral maturity than social order reasoning. At the same time these two forms of reasoning do clearly communicate political differences. Persons using principled reasoning are seen by others as more radical and politically progressive (left-wing) than persons using social order reasoning; the latter are seen as more right-wing, conservative, and politically reactionary. Persons using principled reasoning are also seen as more politically active, more in favour of political change, and more hostile to established political authority than persons using social order reasoning. Four further studies (Emler et al., 1998) used a range of procedures to examine the political positions conveyed by different forms of moral argument. In the first two of these we included, in addition to social order and principled arguments, moral arguments explicitly critical of established authority. The reason for including these “anti-authority” arguments was to test an interpretation, offered by Barnett et al. (1995), of the apparent ability of conservatives to recognise that their opponents on the political left will prefer principled to social order moral arguments. Barnett et al. argue that right-wingers, who themselves use social order arguments, have no real understanding of principled arguments because these latter are indeed at a higher developmental stage and one they have not themselves reached. Conservatives, they suggest, primarily associate the political left with an anti-authority position; linking the left with principled reasoning is purely an artefact of the choices offered to participants in previous research.
MORALITY AND POLITICS 273
TABLE 2 Perceived relevance of moral arguments to people differing in political orientation (adapted from Emler et al., 1998)
Rated on a scale from 1=completely irrelevant to 5=highly relevant. Means in a row with different superscripts differ at p<.05.
We anticipated that distinctions may be made among different positions on the left and among the moral arguments expressive of each. More specifically, we predicted that principled arguments are seen to express moderate left positions while anti-authority arguments are more expressive of the extreme or hard left. We expected that the social order arguments would be those most closely associated with a right-wing position. In the relevance of each of these three kinds of moral arguments to supporters first study (n=50) participants were therefore asked to evaluate the of the right, the moderate left, and the extreme left respectively. As can be seen from Table 2, the predictions were substantially but not fully borne out. The social order arguments were rated as more relevant to a person on the political right than either principled or anti-authority arguments. Additionally, the arguments rated as most relevant to a person on the moderate left were the principled arguments. The unpredicted result was that anti-authority and principled moral arguments were seen as equally appealing to a person on the extreme left. In the second study participants (n=106) were asked to rate the politics of three imaginary candidates for election to political office and to do so on the basis of responses each had apparently made in selection interviews held by local political party officials. The responses in the case of each candidate, called respectively Jones, Brown, and Smith, were moral arguments of one of the three kinds examined in the previous study. That is, a candidate consistently gave replies based on either social order, principled, or antiauthority arguments. The ratings of the candidates’ politics, made on a 7-point scale where 1=very right-wing and 7=very left-wing, were ordered as predicted: Jones (M=2.40) who used social order arguments was seen as the most right-wing of the three; Smith (M=5.81) who used the anti-authority arguments was seen as the most left-wing; and Brown (M=4.17) using principled arguments as occupying an intermediate position, was actually close to the centre of the scale. However, participants’ own politics made a difference
274 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
TABLE 3 Perceptions of candidates’ politics as a function of own politics (adapted from Emler et al., 1998)
Ratings on a scale from 1=very right-wing to 7=very left-wing. Means in a row with different superscripts differ at p<.05.
here. Self-defined left-wingers perceived a significantly greater political difference between Jones (social order) and Brown (principled) than did self-defined right-wingers; the latter actually saw Brown to be slightly to the right of Jones (Table 3). The third study picks up this particular result. In a separate phase of the second study, participants were formed into groups to role-play political party selection committees and to rate the suitability of each candidate for their party. They role-played either Conservative or Labour party committees. The groups were formed on a random basis so that their individual members may or may not have played a role consistent with their own political sympathies. The relative suitability of Jones, Brown, and Smith for each party was precisely as predicted. Conservative party committees judged Jones as the most suitable candidate and Smith as the least, while Labour party committees saw Brown as the most suitable and Jones as the least (see Table 4). The third study was conducted to provide a further test of the inclination of left-wingers to perceive a greater difference in the political meanings of social order and principled reasoning than right-wingers. Here we simply presented character sketches of two individuals, Edward and Henry, which included their responses to two moral dilemmas. The responses attributed to one of these individuals, Henry, were all social order arguments, while those attributed to the other, Edward, were all principled arguments. Participants (n=60) evaluated the two characters on a number of scales, including one representing political identity (where 1=very right-wing, 7=very leftwing).3 Overall, Edward was rated as more left-wing than Henry but there was a significant effect of own politics on the ratings. The perceived difference was most marked among self-defined left-wingers (Ms respectively
MORALITY AND POLITICS 275
TABLE 4 Political party selection committee ratings of candidates’ suitability (adapted from Emler et al., 1998)
Ratings on 7-point scale; 1=highly unsuitable, 7=highly suitable. Means in a column with different superscripts differ at p<.05.
5.59 and 2.57; p<.05), less so among moderates (Ms 4.78 and 2.72; p<. 05), but non-significant among the self-defined right-wingers (Ms 4.42 and 3.17). At this point, therefore, there are some indications that social order moral reasoning can be more expressive of the political right than principled reasoning is of the political left. I shall return to this possibility in the next section. One further test of the communicative function of moral reasoning involved manipulating the salience of political identity (Emler et al., 1998, study 4). If different forms of moral reasoning do indeed convey different political identifications to others, then individuals should be more inclined to select the moral reasoning that is most expressive of their political views when political identity is made particularly salient. The salience of political versus personal identity was manipulated as follows. In one condition participants were asked first about their own political orientation and informed that we were interested in the views on a number of social and moral issues of people representing different political points of view. In the other condition we informed participants that we were interested in their personal opinions on social and moral issues. They were asked to indicate their own political position only at the end of the session. Political and other opinions were assessed on the basis of responses to Gold, Christie and Friedman’s (1976) New Left scales. These include three scales that are explicitly political in content, namely the New Left Philosophy, Radical Tactics, and Machiavellian Cynicism scales. Two other scales deal more directly with moral standards in personal relationships, the Traditional Moralism and Machiavellian Tactics scales. We varied the order of the measures across the two conditions. In the “political identity salient” condition participants completed the New Left scales first and a measure of moral reasoning second. In the “personal identity salient” condition this order was reversed. 3 In the original study (Emler et al., 1998) the scale runs from 1=very left-wing, to 7=very right-wing. It is reversed here for the sake of consistency.
276 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
TABLE 5 Effects of the salience of political versus personal identity upon correlations between moral reasoning and political versus interpersonal moral attitudes (adapted from Emler et al., 1998)
*p<.05, **p<.01, ***p<.001
Our predictions about the effects of manipulating identity salience were largely confirmed (see Table 5). When political identity was made salient, principled moral reasoning was more strongly positively correlated with political views (as reflected in responses to the New Left Philosophy, Radical Tactics, and Machiavellian Cynicism scales) than when personal identity was salient. On the other hand, in the “personal identity salient” condition, principled reasoning was more strongly negatively correlated with views about morals in personal relationships, namely the Traditional Moralism and Machiavellian Tactics scales. A corresponding pattern obtained for social order reasoning. Such reasoning was more strongly negatively correlated with the scales assessing political views when political identity was salient than when personal identity was salient. In contrast, when personal identity rather than political identity was salient, the positive correlations with the Traditional Moralism and Machiavellian Tactics were stronger. To determine whether political identity is in practice reliably communicated to others, we have conducted a series of studies using Kenny’s (1994) “round-robin” design for the assessment of interpersonal perceptions (Emler & St. James, 1999). In one of these studies 43 groups, each of four mutually acquainted individuals, were recruited to provide ratings of themselves and each other on a series of scales intended to provide assessments of various qualities including political orientation. Each quality was assessed with three 9-point scales. Each individual in the
MORALITY AND POLITICS 277
group rated him or herself and each of the other three group members on every scale. A second study replicated this procedure with 44 groups. If our participants have clearly articulated ideological identities then, according to the argument developed here, these should also be clearly apparent to their acquaintances. Three tests are therefore relevant. The first is simply one of self-consistency and is indexed by composite scale alphas. A high alpha for self-perceived ideological orientation reflects consistent self-ratings across the scales. A second test is degree of agreement between ideological orientation as self-defined and as perceived by others. For this test we averaged the ratings of each participant provided by the other three individuals in each group. The third test is whether observers agree with one another, whether there is consensus among them. For this we averaged the inter-observer agreements within each group for each target. The results of these three tests are summarised in Table 6.4 As these show, self-perceptions of ideological orientation not only exhibit a high degree of internal consistency but show significant levels of agreement with the perceptions of acquainted observers. Additionally, consensus or agreement among observers is at the same level as agreement between self and observers. To put these levels of self-other agreement and consensus in context, they are of the same order as values we have obtained for extroversion, which is one of the most perceivable of personal characteristics (e.g., Borkenau & Liebler, 1995). If the participants in these studies had consistent political identities, which they also communicated effectively to others, it is nonetheless likely that the coherence of such identities and the clarity with which they are expressed are variable qualities. In particular, as we shall see in the final section of the chapter, coherent political identities are more likely to be found within the highly educated population from which we have sampled, namely those engaged in a university-level education. To summarise, different forms of moral reasoning—specifically those identified by Kohlberg respectively as social order and principled reasoning —express political positions and are able to do so because their political meaning is understood by others, whatever their own position on the leftright dimension. The next section examines the possibility that develop mental differences may, nevertheless, underlie variations in other aspects of people’s attitudes towards politics, in particular their attitudes towards the rules of the political game. 4 The data reported here are not based on Kenny’s (1994) statistical procedures for partitioning variance in round-robin designs; these procedures are difficult both to apply and to interpret (Funder, 1999). However, we have analysed data from a parallel study using Kenny’s procedures, primarily to verify that variance in observer ratings of political identity was due to the target rather than the perceiver. The relevant effects were: perceiver 5%; target 47%.
278 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
TABLE 6 Communication of political identity
Values in the first column are Cronbach alphas. Those in second and third columns are mean correlations; all these correlations significant at p<.001.
COGNITIVE SOPHISTICATION AND THE RULES OFTHE POLITICAL GAME Notwithstanding the doubts outlined earlier about Eysenck’s two-factor model of political ideology, other evidence does indicate that a two- rather than a one-dimensional solution provides a better fit to data describing the covariation of socio-political attitudes. Heath et al. (1993), on the basis of opinion surveys of British adults, labelled the two dimensions respectively “socialist/laissez faire”, which reflects the traditional left-right divide in politics, and “libertarian/authoritarian”. A similar two-dimensional structure has been identified as the best fit for public opinion data derived from American adults (Fleishmann, 1988), and corresponding dimensions were also identified in a confirmatory factor analysis we conducted of the social attitudes of 2100 British 18–20-year-olds (Bynner et al., in press). The appropriateness of a two-factor solution is further supported by evidence for the discriminant validity of these factors. This includes the relative ability of position on each dimension to predict voting. Drawing on the same data set as Bynner et al. (2000), we were able to show that whereas the left-right dimension strongly predicts voting intentions, the second dimension does not (Banks et al., 1992). Similarly, Evans et al. (1996) showed that the Heath et al. (1993) left-right scale predicted party preferences more strongly than did the libertarian-authoritarian scale. Another indication of discriminant validity concerns the relation of the two dimensions to level of education. A dimension comparable to Heath et al.’s libertarian-authoritarian factor emerged from analysis of different kinds of data by Nie et al. (1996). This dimension, which they labelled “democratic enlightenment”, was defined by attitudes of tolerance to various kinds of non-conformists and minorities, and expressed as willingness to extend the rights of free speech to such groups. But of particular significance it was also defined by factual knowledge about aspects of the democratic process such as suffrage rules, the meaning of civil liberties, and the defining characteristics of a democracy as compared to a dictatorship. Position on this dimension was additionally, as the
MORALITY AND POLITICS 279
study’s authors anticipated, strongly related to education. In contrast, position on the left-right dimension is largely unrelated to level of education (Bynner et al., in press; Nie et al., 1996). The association between education and tolerance for people who differ in life-style, opinion, or race is one of the best documented and most reliable links in the public opinion literature. All the available evidence shows that, as Kinder (1998) puts it summing up American research findings, “more education is always associated with more tolerance” (p. 791; cf. also Sniderman, Brody, & Tetlock’s, 1991, findings regarding the effect of education upon support for “principled political tolerance”). Recent European studies likewise point in this direction (e.g., Bynner et al., in press; Pettigrew et al., 1998). But why does the association exist? Education and tolerance are potentially related in at least three distinguishable ways (Emler & Frazer, 1999). First, they may be related because both staying longer in education and being tolerant have a common cause; i.e., we are observing the effects of a third unmeasured variable. The obvious candidate for this causal role is intellectual ability, although there are others that have some plausibility including certain qualities of motivation and personality. A second and quite different possibility is that education has contingent or indirect social and psychological effects. Consider, for example, the consequences of educational attainment. People acquire social status and public esteem from their success in public examinations, their possession of certificates and degrees, and the prestige of the occupational positions to which these afford access. One might anticipate that an enlightened and tolerant attitude to minorities comes more readily to people who are already secure in a sense of their own value and that intolerance will be more characteristic of those who have no personal accomplishments of this kind on which to base their self-worth. This parallels the interpretation of ingroup favouritism advanced by Turner (1975); he suggested that self-esteem is derived from the perception that one’s own group is superior to some other, but also allowed that no such in-group bias is necessary if other, more personal sources of esteem are available. Similarly, Pettigrew et al. (1998) proposed that some of the effect of education on racial tolerance is mediated by education’s tendency to reduce feelings of group relative deprivation. And Devos, Deschamps, and Comby (1994) showed that the tendency to define the self in terms of group memberships was more pronounced among young people in low- than in high-status educational programmes. Two observations argue against this interpretation, however. First, education has little impact on self-esteem (West, Fish, & Stevens, 1980), and second, low self-esteem does not lead to greater in-group favouritism (Rubin & Hewstone, 1998). A third and more likely explanation is that there is an effect of education on tolerance or enlightenment mediated by one or more of the direct
280 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
consequences of having more or less education. The problem is that there are potentially several such consequences. Education develops cognitive skills and transmits information. It also conveys values. It exposes the individual to contacts with a particular range of other people. It entails subjection to certain forms of internal governance and power structure. It may provide experience in a variety of institutional roles and statuses. And so on. And each of these elements may have its particular direct effect on the individual’s political socialisation (cf. Albala-Bertrand, 1997). Nie et al. (1996) make a case for a very specific direct and long-term or enduring effect of education upon democratic enlightenment, namely the effect of years of formal education on cognitive skills. The skills they have in mind are those required to interpret, organise, and manipulate information in progressively more complex ways and according to increasingly broad and abstract categories, together with the ability to think analytically and critically, and to make moral judgements. Piagetian measures of operational thought provide one indicator of the development of such competencies and research evidence is clear that performance on these kinds of measures does improve as a direct effect of formal schooling (Rogoff, 1981). The case made by Nie et al. is consistent with several other indications. Among these are Schönbach, Gollwitzer, Stiepel, and Wagner’s (1980) evidence linking prejudice to cognitive ability variables; Doyle and Aboud’s (1995) demonstrations that childhood prejudice is inversely related to the development of logical operations; Rosenberg, Ward, and Chilteon’s (1987) identification of a link between cognitive development and endorsement of democratic concepts; and McClosky and Brill’s (1983) observations on the cognitive requirements for political tolerance. Moral reasoning scores have likewise been shown to correlate with such indicators of democratic enlightenment as support for human rights and civil liberties (Lonky, Reihman, & Serlin, 1981; Narvaez, Getz, Thoma, & Rest, 1999) and anti-racism (Davison, 1976). These same moral reasoning scores are also correlated with education. Indeed, education is the most potent predictor so far identified of level of moral reasoning (Colby & Kohlberg, 1987; Rest et al., 1999). In a sense there should be no surprise that measures of moral reasoning proved to be correlated with measures of democratic enlightenment or political tolerance. After all, the latter are in effect moral judgements about the appropriateness or otherwise of political procedures and the rights of participation in the political process of groups and individuals with differing views and values. The question is whether such moral judgements develop according to a stage-like pattern and on the basis of the kind of process proposed by cognitive constructivist theory. The potential contribution here of a cognitive constructivist perspective, such as Kohlberg proposed, is not so much in the specifics of his or any other stage theory as in the more general insight that social knowledge
MORALITY AND POLITICS 281
undergoes progressive change which brings with it changing moral judgements about, among other things, the legitimacy of democratic procedure. Piaget’s earlier research into children’s developing understanding of “the rules of the game” (Piaget, 1932) is a good illustration of this. As they become progressively more intellectually sophisticated, children are correspondingly increasingly able to argue that rules are more worthy of respect to the degree that they are decided through democratic procedures based on equality of participation. Adelson’s (1971) studies of changes in conceptions of law and government across adolescence also illustrate the process (see also: Berti, 1988; TorneyPurta, 1983); Adelson found, for example, a declining preference for authoritarian solutions to political conflicts over the course of adolescence. In accepting progressive change in socio-political knowledge and in moral judgements about political process, however, we are not also obliged to concede the cognitive constructivist view that these changes are selfconstructed. A case can be made instead for a process of construction in which political tolerance is an outcome of progressive socio-cognitive changes that are socially produced (Emler, 1987; Emler & Ohana, 1993), in other words a socio-genetic interpretation. To be more specific, justifications for political tolerance and insight into principles such as equality of representation in the political process are constructed though the social influences that individuals exercise upon one another as they discuss and negotiate their different viewpoints (cf. Doise & Mugny, 1984). Participation in formal education provides a particularly systematic, structured, and intensive framework for the operation of such influences. This is also the import of Doise, Staerkle, Clemence, and Savory’s (1998) conclusions from a study of the development of representations of human rights, including political rights, among Genevan youth. These young people’s representations become more sophisticated and elaborated with age, but this process was directly related to the extent and level of their participation in formal education. But education per se does not have these benign effects—pedagogical style, forms of internal organisation, and governance structures may all be relevant (Albala-Bertrand, 1997; Meloen & Farnen, 1996; Torney-Purta, 1983). The evidence reviewed by these authors indicates that it is not just cognitive skills in general terms but skills in critical analysis and judgement that are crucial. In particular, Torney-Purta (1983) and Meloen and Farnen (1996) both show that if time in school is taken up with patriotic ritual or the advocacy of nationalist agendas rather than with open discussion of opinions, the effects of more education upon tolerance can be negative. Finally, are the conclusions here about moral reasoning inconsistent with the claims presented earlier in the chapter? In particular can the same kind of moral reasoning—principled reasoning—be an expression both of a leftwing political position and of political tolerance, if ideological orientation
282 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
and political tolerance are distinct dimensions of social attitudes? A resolution of this apparent contradiction may lie in recognising that the form of moral reasoning Kohlberg (1984) and Rest et al. (1999) have identified as “principled”, and have sought to give operational definition in their various measurement procedures, may in practice contain distinct elements, some of which are expressive of a left-wing political position while others are expressive of political tolerance or democratic enlightenment. This is consistent with the findings from the Emler et al. (1998) studies 2 and 3 summarised earlier. We found there that when rightwing participants do not have to make a choice between social order and principled reasoning, they rate samples of principled reasoning as highly as examples of social order reasoning. We also found in the first study that perceived personal relevance of principled moral arguments was unrelated to own political position. For an educated right-winger the political tolerance conveyed by some elements of principled reasoning may be valued as highly as it is by a left-winger (cf. also Sniderman et al., 1991). However, whether existing definitions of principled moral reasoning do indeed confound two distinct kinds of moral argument awaits more direct and systematic test. POLITICAL IDENTITY: ALSO A MATTER OFMORE OR LESS If an ideological orientation or political identity reflects a choice of sides in a moral contest, the following question is still to be answered: what determines the direction of this choice? The political values of parents, it appears, are strongly related to the political identities adopted by their children (Niemi & Jennings, 1991). But this still takes us only so far. If parents have an influence, how is this influence exercised? Answers begin to emerge from responses to an issue raised most forcefully by Converse (1964). For Converse it seemed that questions about the determinants of political identity would have little relevance because most adults do not have political identities, at least not in the sense of stable, interrelated attitudes, coherently organised around an explicit ideological position. He came to this conclusion after examining successive waves of American survey data. These revealed that opinions expressed by respondents on one occasion were typically entirely unrelated to views expressed in answer to precisely the same survey questions on subsequent occasions. He reasoned that this reflected not rapid and frequent changes of attitude but the absence of clearly formed attitudes. Random responding occurred because the respondents concerned had in the first place no clear ideological position from which to generate consistent responses to particular policy issues.
MORALITY AND POLITICS 283
With respect to politics, however, it has emerged that people differ in all manner of ways, and not just in the stability of their opinions. So, people also differ in the number of political opinions they hold, or degree of “opinionation” (Krosnick & Milburn, 1990) and the degree to which these are related both to one another and to political or party identifications. But additionally, people differ in their interest in politics, their attentiveness to political matters, their awareness of political issues, their exposure to political information, their knowledge of basic political facts, of the policy positions of different political parties, of the identities of political representatives, their participation in political life whether through voting or other forms of activity or activism, the importance of politics to their self-concept, their sense of political efficacy, and so on. Despite the sheer number and variety of these differences, and almost endless proliferation of labels to describe them, there are grounds for regarding many if not all as facets of the same phenomenon (Nie et al., 1996): these are variations in degree of political engagement, and these differences are motivational, cognitive, and behavioural as well as attitudinal. That this has not yet been more widely apparent may reflect the tendency for researchers to focus upon some aspects of engagement rather than all of them simultaneously. It is proposed that six elements of engagement can usefully be distinguished. The first of these is interest, which is essentially motivational. It is usually operationalised as one or more questions about extent of interest in political matters. The second is attentiveness, understood as the extent to which individuals seek out or pay attention to political issues. Attentiveness may be regarded as a behavioural expression of engagement. Measures asking about the extent to which people talk about politics with others, and pay attention to news about politics by reading newspapers and by listening to or watching relevant broadcasts, constitute operationalisations of this element. The third element, knowledge, is the cognitive dimension of engagement. It is interpreted here as being close to the concept of political expertise (Fiske, Lau, & Smith, 1990). Potentially it is a complex element and could include factual knowledge about the political system, knowledge about political figures, knowledge of part, positions on policy issues, knowledge about political events, and so on. In practice, however, it seems that people tend to be generalists with respect to political knowledge (Delli, Karpini, & Keeter, 1996) and that measures of general factual knowledge are effective in capturing variance in political expertise (Fiske et al., 1990). Fourth is opinionation or the range of issues on which people have clear, confidently held, and stable opinions. Opinionation has most often been measured in terms of the number of items in a set for which responses other than the “don’t know” category are chosen (e.g., Krosnick & Milburn, 1990). It can, however, be assessed by asking respondents directly
284 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
whether they have formed opinions on a set of issues (Schuman & Presser, 1981). The fifth element is ideologically based consistency. This refers to the extent to which opinions held are consistently related to one another and organised around broader political categories and identifications such that they display ideological coherence or, in Converse’s (1964) terms, “constraint”. Finally, there are various forms of political action or participation, from voting in elections to demonstrating and protesting, through campaigning and organising political action, to joining and becoming active in political parties. Research on participation shows consistently that the most prevalent form of participation is voting and moreover that beyond this a majority of adults participate very little in political life (Banks et al., 1992; Delli Karpini, & Keeter, 1996; Kinder, 1998). At the same time, however, the various forms of political participation are positively interrelated. People who vote more often in elections are also more likely to take part in campaign work, to initiate contact with politicians, and to be active in community issues, for example (Nie et al., 1996). One argument for treating these different elements all as aspects of a single phenomenon, engagement, is that variation in one element is related to variations in others. Several studies provide partial confirmation of such covariations. Fiske et al. (1990), for example, found that knowledge, exposure to political information (attentiveness), and political activity were interrelated. Similarly Zaller (1990) found that measures of interest, knowledge, media exposure, and participation intercorrelated (see also Erber, Hodges, & Wilson, 1995). Nie et al. (1996), in a factor analysis of data from the US Citizen Participation Study (Verba, Schlozman, & Brady, 1995), showed that measures of political attentiveness (actually a combination of interest, media exposure, and political talk measures), frequency of electoral participation, other forms of political action, and various kinds of political knowledge, all loaded on a single factor, which they explicitly labelled as engagement. Finally, we have identified a similar factor in our analysis of data from a large sample of 16–20-year-olds (Bynner et al., in press). The factor was defined by interest in politics, intention to vote in parliamentary elections, communication (talking about politics, watching political broadcasts), and an attitude dimension labelled “political disaffection”. This can be regarded as another measure of interest. But it also suggests that low engagement may be more than indifference; it may also include an element of active aversion (cf. also White, Bruce, & Ritchie, 2000). A second argument for identifying these elements with participation is that they all relate in the same direction to education. More education is associated with more interest in politics, attention to politics, knowledge about politics, opinions about politics, ideological coherence, and participation (Bynner et al., 2000; Converse, 1972; Delli Karpini, & Keeter,
MORALITY AND POLITICS 285
1996; Fiske et al., 1990; Kinder, 1998; Krosnick & Milburn, 1990; Nie et al., 1996). At the same time, however, they are not necessarily all related to education for the same reasons. The possibility that different variables associated with extent of participation in education play distinct roles with respect to the different elements of engagement in politics emerges when one considers the determinants of political engagement. Two interpretations of these determinants can be distinguished. One treats engagement as unitary in the sense that interest, attentiveness, knowledge, and so forth are essentially alternative indicators of the same underlying quality. Determinants are then entirely external to the elements that make up or express engagement. This effectively is the view taken by Nie et al. (1996) who argue that education’s effects are mediated primarily by the consequences of amount of education for social position, and in particular for social network centrality in adult life. In their own research, network centrality was defined in terms of the number of political public figures and others professionally active in politics to whom a respondent was known. A second interpretation treats intercorrelations between the elements of engagement as reflecting a structure of causal relations. This interpretation is favoured by, among others, Delli Karpini and Keeter (1996). The focus of their interest is political knowledge and from this they work both backwards to its potential determinants and forwards to its consequences. Thus they interpret the association they find between extent of political knowledge on the one hand, and both interest in politics and attention to political matters, on the other, as reflecting the influence of interest and attention on the accumulation of knowledge. They interpret the findings that less politically knowledgeable adults “hold fewer, less stable and less consistent opinions” (Delli Karpini, & Keeter, 1996, p.265) and are less likely to participate politically as indicating the consequences of political knowledge. Whether this or indeed other causal structures linking the various elements of engagement can be confirmed remains to be seen. Consider the specific link from knowledge to opinionation. Numerous studies confirm the empirical association between these two aspects of engagement (see Krosnick & Milburn, 1990). But how can it be decided whether one is a causal influence on the other rather than both being expressions of the same latent variable? Evidence consistent with a causal interpretation of the correlation would be that the two elements relate to other variables in different ways. We have attempted to test this possibility in the following fashion (Earnshaw & Emler, 2000). Fifty students completed a range of measures including a political knowledge quiz, a measure of political opinionation, Caccioppo and Petty’s (1982) Need for Cognition scale, and Campbell, Trapnell, Heine, Katz, Lavallee, and Lehman’s (1996) scale to measure self-
286 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
concept clarity. We reasoned that because having opinions about political issues is a more assertive social posture than merely being knowledgeable about political affairs, people are more likely to take this step if they have a clear image of themselves in which they are confident. We therefore predicted that although opinionation and knowledge would be strongly correlated, self-concept clarity would relate to opinionation rather than knowledge, while the reverse would be the case for Need for Cognition. The opinionation measure in this study consisted of 15 statements about which respondents were asked not to give an opinion but to indicate the degree to which they had formed a clear opinion on the issue referred to in the statement. The results were as predicted. Opinionation and knowledge were strongly correlated (r=.62, p<.001). Additionally, although Need for Cognition also correlated with opinionation (r=.33, p<.02), it was more highly correlated with knowledge (r=.53, p<.001), and partialling out this association reduced the direct effect of need for cognition on opinionation to non-significance. In contrast, self-concept clarity was correlated with opinionation (r=.33, p<.02) but was uncorrelated with knowledge. This study by itself is hardly decisive. It shares with all correlational studies the problems such data pose for testing causal hypotheses. Nonetheless, an approach that treats the various elements of political engagement as related to one another in a causal structure has promise. Part of its promise lies in making sense of the role of other influences upon political engagement, notably those of education. Thus the fact that educational attainment and interest are associated (Bynner & Ashford, 1994) may reflect their shared links with ability. On the other hand, direct effects of education might be anticipated with respect to attentiveness in so far as formal education (a) develops the cognitive skills that underpin ability to attend to the spoken and the written word and (b) transmits norms for attentional priorities. Similarly, direct effects might be expected for political knowledge in so far as classroom instruction both conveys political facts and develops the skills to organise this information in memory. However, education’s effects on opinionation, ideological coherence, and political action could have more to do with the kind of social environment to which participation in formal education gives access. To appreciate this, consider again the association noted earlier between parents’ politics and their children’s political views. It is unlikely that parents directly transmit to their children an ideological belief system as a complete package. It is unlikely in part because, given the evidence of variability in levels of political engagement among the adult population, many parents will not have such a package to offer. It is more plausible that a collection of processes operate to create the observed association between parents’ political inclinations and those of their children. These may include shared location in the class structure and thus shared group interests. They may also include a geographic location common to parents and their children,
MORALITY AND POLITICS 287
exposing both to the same immediate social environment and its prevailing norms and value priorities. This is consistent with our finding (Abrams & Emler, 1992) that area of residence influences political partisanship over and above the effects of social class background. Additionally, parents, through various choices and opportunities, influence the kinds of schools their children attend and the amount of time they spend in education. Parents therefore directly and indirectly shape the influences on political outlook to which their children are subject. But they are also likely to shape the degree to which their children elaborate a political outlook at all or, in the terms used here, the degree to which they become politically engaged. Consider again the effects of education, which relate to both the direction and the degree of involvement. Formal education creates particular kinds of social contact patterns, contacts with a particular social and intellectual mix of other people (cf. Emler, 2000). Regarding the consequences of this for political commitments, we know little beyond the evidence provided by Newcomb’s studies of a cohort of females attending a small private university together in the 1930s (Alwin, Cohen, & Newcomb, 1991; Newcomb, 1943; for a rare recent exception, see Roker, 1993). It seems likely, however, that the consequences arise from the effects of social contact patterns on political talk. Indirect evidence for this comes from studies linking talk to political attitudes and identifications. I have previously reported evidence showing that awareness of one’s acquaintances’ political sympathies is associated with more ideologically organised attitudes (Emler, 1990). One interpretation of this finding is that high awareness of this kind reflects membership of a social world in which political sympathies are salient, explicit, and ideologically coherent enough to be communicated intelligibly among its members. A further interpretation is that the inhabitants of this social world collectively generate ideological consistency by regularly discussing their political views with one another. In line with this view, the awareness measure was highly correlated with a measure of political discussion. Further evidence for the role played by political talk comes from a study by Lalljee and Evans (1998). They found that talking politics with one’s acquaintances had some effect on opinion stability but rather clearer effects on inter-attitudinal consistency and upon stability of political party identification (see also Huckfeldt & Sprague, 1995). Palmer-Canton (1998) observed a similar effect of political talk on the internal consistency of political attitudes; people who reported talking more frequently about political issues with friends and acquaintances also expressed political attitudes that were more highly intercorrelated. To summarise, then, there are grounds for the view that coherent ideological positions emerge from the processes of social influence and social comparison that occur in face-to-face conversational interactions
288 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
among acquaintances. Judd and Krosnick (1989) may be right that thinking more often and more deeply about political questions supports the development of coherently organised political views, a judgement similar to Kohlberg’s view of the process that drives the development of moral reasoning. But these kinds of private and internal cognitive processes are more likely to occur if people also regularly experience demands to be accountable for their opinions (cf. Tetlock, 1983b). It is likely, moreover, that these processes of social influence give ideological positions their particular shape. Social and political attitudes do not covary because there is some inner logic that ties them together but because they are consensually defined as belonging together (Billig, 1976; Converse, 1964; Erickson, 1982). As Kinder (1998) observes, “ideological structure thus may reflect not only or even mainly the inner workings of psychologic, but rather processes of social diffusion” (p.797). But the extent to which an individual elaborates a political position will depend on the extent to which he or she is exposed to these processes of influence and diffusion. CONCLUSIONS For the last 30 years the dominant perspective on moral psychology and the development of social cognition has been cognitive constructivism. Its analysis of moral reasoning and social knowledge has taken these to be spontaneous and autonomous constructions, which owe their form to the inner logic of cognitive structures and not to any social influences. Its implications are that individuals orient to politics as a function of the level to which their abilities to understand, analyse, and reason about moral questions have developed. The most striking consequence of this is the conclusion that left and right as political orientations are tied to different points on a developmental scale. The evidence discussed in this chapter supports a different view of the relation of political orientations to cognitive development and moral reasoning. First, choice of political perspective is more likely to be driven by value preferences and the social influences from which these are derived than by cognitive level. Distinctive patterns of moral reasoning are associated with contrasting political positions because they effectively express and account for these value preferences. Moreover, the availability of different moral reasoning styles to supporters of both left and right is also consistent with their rhetorical function of communicating a political position. Second, however, the socio-genetic interpretation advocated here does not deny that individuals can differ in the cognitive complexity or sophistication that underlies their political views. It does anticipate that this complexity is socially constructed. This is particularly apparent with
MORALITY AND POLITICS 289
respect to democratic enlightenment, which entails complex insights and moral judgements. These cognitive attributes depend directly for their development upon the nature and extent of an individual’s participation in the social institution of formal education. Social construction is also apparent in the elaboration of an ideological posture. The moral judgements contained in political opinions appear to be heavily knowledgedependent, much more so than interpersonal moral judgements. But they are also subject to extensive calibration through face-to-face interaction while the knowledge base itself has social origins. And again the social relations accessed through participation in formal education play a major role. Future work might profitably focus upon the further unpacking of this role. REFERENCES Abelson, R.P. (1968). Psychological implication. In R.P.Abelson, E.Aronson, W.J.McGuire, T.M.Newcomb, M.J.Rosenberg, & P.H.Tannenbaum (Eds.), Theories of cognitiveconsistency: A source book.Chicago: Rand McNally. Abrams, D., & Emler, N. (1992). Self denial as a paradox of political and regional social identity: Findings from a study of 16- and 18-year-olds. European Journal of SocialPsychology, 22,279–295. Adelson, J. (1971). The political imagination of the young adolescent. Daedalus, 100,1013– 1050. Adorno, T.W., Frenkel-Brunswik, E., Levinson, D.J., & Sanford, R.N. (1950). Theauthoritarian personality.New York: Harper. Albala-Bertrand, L. (1997). What education for what citizenship? Preliminary results. EducationInnovation & Information, 90,2–8. Alker, H.A., & Poppen, P.J. (1973). Personality and ideology in university students. Journal ofPersonality, 41,652–671. Alwin, D.F., Cohen, R.L., & Newcomb, T.M. (1991). Aging, personality and social change:Attitude persistence and change over the lifespan.Madison: University of Wisconsin Press. Banks, M., Bates, I., Breakwell, G., Bynner, J., Emler, N., & Jamieson, L. et al (1992). Careersand identities.Milton Keynes, UK: Open University Press. Barnett, R., Evens, J., & Rest, J. (1995). Faking moral judgment on the Defining Issues Test. British Journal of Social Psychology, 34,267–278. Berti, A.E. (1988). The development of political understanding in children between 6–15 years old. Human Relations, 41,437–446. Billig, M. (1976). Social psychology and intergroup relations.London: Academic Press. Borkenau, P., & Liebler, A. (1995). Observable attributes as manifestations and cues of personality and intelligence. Journal of Personality, 63,1–25. Brown, R. (1965). Social psychology.New York: Free Press. Bynner, J., & Ashford, S. (1994). Politics and participation: Some antecedents of young people’s attitudes to the political system and political activity. European Journal of Social Psychology,24,223–236.
290 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Bynner, J., Romney, D., & Emler, N. (in press). An analysis of dimensions of identity in late adolescence. Journal of Youth Studies. Caccioppo, J.T., & Petty, R.E. (1982). The need for cognition. Journal of Personality andSocial Psychology, 42,116–131. Campbell, A., Converse, P.E., Miller, W.E., & Stokes, D.E. (1960). The American voter.New York: Wiley. Campbell, J.D., Trapnell, P.D., Heine, S.J., Katz, I.M., Lavallee, L.F., & Lehman, D.R. (1996). Self concept clarity: Measurement, personality correlates and cultural boundaries. Journal of Personality and Social Psychology, 70,141–156. Candee, D. (1976). Structure and choice in moral reasoning. Journal of Personality and SocialPsychology, 34,1293–1301. Colby, A., & Kohlberg, L. (1987). The measurement of moral judgment, Vol. 1, Theoreticalfoundations and research validation.New York: Cambridge University Press. Conover, P.J., & Feldman, S. (1981). The origins and meaning of liberal/ conservative self identifications. American Journal of Political Science, 25,617– 645. Converse, P.E. (1964). The nature of belief systems in mass publics. In D.E.Apter (Ed.), Ideology and discontent(pp. 206–261). New York: Free Press. Converse, P.E. (1972). Change in the American electorate. In A.Campbell, & P.E.Converse (Eds.), The human meaning of social change.New York: Russell Sage. Davidson, F.H. (1976). Ability to respect persons compared to ethnic prejudice in childhood. Journal of Personality Social Psychology, 34,1256–1267. Delli Karpini, M.X., & Keeter, S. (1996). What Americans know about politics and why itmatters.New Haven, CN: Yale University Press. Devos, T., Deschamps, J.-C, & Comby, L. (1994). Social insertions and representations of oneself and young people: Apprentices and high school students. European Journal ofPsychology of Education, 9,55–67. Doise, W., & Mugny, G. (1984). The social development of the intellect.Oxford: Pergamon. Doise, W., Staerkle, C, Clemence, A., & Savory, F. (1988). Human rights and Genevan youth: A developmental study of social representations. Swiss Journal of Psychology, 57,86–100. Downes, A. (1957). An economic theory of democracy.New York: Harper. Doyle, A.B., & Aboud, F.E. (1995). A longitudinal study of White children’s racial prejudice as a socio-cognitive development. Merrill Palmer Quarterly, 41,209– 228. Eagly, A., & Chaiken, S. (1993). The psychology of attitudes.Fort Worth, TX: Harcourt Brace Jovanovich. Earnshaw, O., & Emler, N. (2000). Psychological determinants of political knowledge andpolitical opinionation.Unpublished raw data, University of Oxford. Emler, N. (1983). Morality and politics: The ideological dimension in the study of moral development. In H.Weinreich-Haste, & D.Locke (Eds.), Morality in the making: Thought,action and social context(pp. 47–71). Chichester, UK: Wiley. Emler, N. (1987). Socio-moral development from the perspective of social representations. Journal for the Theory of Social Behaviour, 17,371–388.
MORALITY AND POLITICS 291
Emler, N. (1990). A social psychology of reputation. In W.Stroebe, & M.Hewstone (Eds.), European Review of Social Psychology(Vol. 1, pp. 171–193). Chichester, UK: Wiley. Emler, N. (2000). Social structures and individual lives: Effects of participation in the social institutions of family, education and work.,In J Bynner, & RK Silbereisen (Eds.) Adversityand challenge in the life course in England and the new Germany(pp. 62–84). London: Macmillan. Emler, N., & Frazer, E. (1999). Politics: The education effect. Oxford Review of Education, 25,251–274. Emler, N., & Ohana, J. (1993). Studying children’s social knowledge: Just old wine in new bottles? In G.Breakwell, & D.Canter (Eds.), Empirical approaches to social representations(pp. 63–89). Oxford: Oxford University Press. Emler, N., Palmer-Canton, E., & St. James, A. (1998). Politics, moral reasoning and the Defining Issues Test: A reply to Barnett et al. 1995. British Journal of Social Psychology, 37,457–476. Emler, N., & Reicher, S. (1995). Adolescence and delinquency: The collective management ofreputation.Oxford: Blackwell. Emler, N., Renwick, S., & Malone, B. (1983). The relationship between moral reasoning and political orientation. Journal of Personality and Social Psychology, 45,1073–1080. Emler, N., & Stace, K. (1999). What does principled versus conventional moral reasoning convey to others about the politics and psychology of the reasoner. European Journal ofSocial Psychology, 29,455–468. Emler, N., & St. James, A. (1999). The formation and communication of identity: Why it helps tohave friends and a social life.Paper presented at the 12th General Meeting of the European Association of Experimental Social Psychology, Oxford, 6–11July . Erber, M.W., Hodges, S.D., & Wilson, T.D. (1995). Attitude strength, attitude stability, and the effects of analysing reasons. In R.E.Petty, & J.A.Krosnick (Eds.), Attitude strength:Antecedents and consequences.Mahwah, NJ: Lawrence Erlbaum Associates Inc. Erickson, B.H. (1982). Networks, ideologies and belief systems. In P.V.Marsden, & N.Lin (Eds.), Social structure and network analysis.Beverley Hills, CA: Sage. Evans, G., Heath, A., & Lalljee, M. (1996). Measuring left-right and libertarianauthoritarian values in the British electorate. British Journal of Sociology, 47, 93–112. Eysenck, H.J. (1954). The psychology of politics.London: Routledge & Kegan Paul. Fishkin, J., Keniston, K., & MacKinnon, C. (1973). Moral reasoning and political ideology. Journal of Personality and Social Psychology, 27,109–119. Fiske, S., Lau, R., & Smith, R. (1990). On the varieties and utilities of political expertise. SocialCognition, 8,31–48. Flavell, J.H. (1963). The developmental psychology of Jean Piaget.Princeton, NJ: Van Norstrand. Fleishman, J.A. (1988). Attitude organisation in the general public: Evidence for a bidimensional structure. Social Forces, 67,159–184. Fontana, A.F., & Noel, B. (1973). Moral reasoning in the university. Journal of Personality andSocial Psychology, 27,109–119.
292 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Funder, D. (1999). Personality judgment: A realist approach to person perception.San Diego, CA: Academic Press. Gold, A.R., Christie, R., & Friedman, L.N. (1976). Fists and flowers: A social psychologicalinterpretation of student dissent.New York: Academic Press. Haan, N., Smith, B., & Block, J. (1968). Moral reasoning in young adults—political, social behaviour and family background correlates. Journal of Personality and Social Psychology,10,183–201. Heath, A., Evans, G., & Martin, J. (1993). The measurement of core beliefs and values: The development of balanced Socialist/Laissez Faire and Libertarian/ Authoritarian scales. British Journal of Political Science, 24,115–158. Heider, F. (1958). The psychology of interpersonal relations.New York: Wiley. Himmelweit, H.T., Humphreys, P., & Jaeger, M. (1981). How voters decide.Milton Keynes, UK: Open University Press. Huckfeldt, R., & Sprague, J. (1995). Citizens, politics and social communication: Information andinfluence in an election campaign.Cambridge: Cambridge University Press. Inglehart, R. (1990). Culture shift in advanced industrial society.Princeton, NJ: Princeton University Press. Judd, C.M., & Krosnick, J.A. (1989). The structural bases of consistency among political attitudes: Effects of political expertise and attitude importance. In A.R.Pratkanis, S.J. Breklery, & A.G.Greenwald (Eds.), Attitude structure and function.Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Judd, C.M., Krosnick, J.A., & Milburn, M.A. (1981). Political involvement and attitude structure in the general public. American Sociological Review, 46,660– 669. Kenny, D. (1994). Interpersonal perception: A social relations analysis.New York: Guilford. Kerlinger, F.N. (1984). Liberalism and conservatism: The nature and structure of social attitudes.Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Kinder, D. (1998). Opinion and action in the realm of politics. In G.T.Gilbert, S.T.Fiske, & G.Lindzey (Eds.), Handbook of social psychology(4th ed., Vol. 2). New York: McGraw Hill. Kohlberg, L. (1971). From is to ought. How to commit the naturalistic fallacy and get away with it in the study of moral development. In T.Mischel (Ed.), Cognitive development andepistemology.New York: Academic Press. Kohlberg, L. (1984). The psychology of moral development: Vol. 2. Essays on moral development.New York: Harper & Row. Kohlberg, L., Levine, C., & Hewer, A. (1983). Moral stages: A current formulation and aresponse to critics.Basel: Karger. Krosnick, J.A., & Milburn, M.A. (1990). Psychological determinants of political opinionation. Social Cognition, 8,49–72. Lalljee, M., & Evans, G. (1998). Political talk and the stability and consistency of political orientation. British Journal of Social Psychology, 37,203–212. Lonky, E., Reihman, J.M., & Serlin, R.C. (1981). Political values and moral judgment in adolescence. Youth & Society, 12,423–441. Markoulis, D. (1989). Political involvement and socio-moral reasoning: Testing Emler’s hypothesis. British Journal of Social Psychology, 28,203–212.
MORALITY AND POLITICS 293
McClosky, H., & Brill, A. (1983). Dimensions of tolerance: What Americans believe about civilliberties.New York: Russell Sage. Meloen, J., & Farnen, R. (1996). Authoritarianism and educational content/style: A crossnational survey of students from 30 countries.Paper presented at 19th Annual scientific meeting of the International Society of Political Psychology, Vancouver, July,. Moscovici, S. (1976). Social influence and social change.London: Academic Press. Narvaez, D., Getz, L, Thoma, S., & Rest, J.R. (1999). Individual moral judgement and cultural ideology. Developmental Psychology, 35,478–488. Nassi, A.J., Abramowitz, S.L, & Youmans, J.E. (1983). Moral development and politics a decade later: A replication and extension. Journal of Personality and Social Psychology, 45,1127–1135. Newcomb, T.M. (1943). Personality and social change: Attitude formation in a studentcommunity.New York: Dryden. Nie, N.H., Junn, J., & Stehlik-Barry, K. (1996). Education and democratic citizenship inAmerica.Chicago: Chicago University Press. Niemi, R., & Jennings, M.K. (1991). Issues and inheritance in the formation of party identification. American Journal of Political Science, 35,970–988. O’Connor, R.E. (1974). Political activism and moral reasoning: Political and apolitical students in Great Britain and France. British Journal of Political Science, 4,53–78. Palmer-Canton, E. (1998). Social processes in the development of political commitments.Unpublished D.Phil.Thesis, Oxford University. Pettigrew, T., Jackson, J.S., Ben Brika, J., Lemaine, G., Meertens, R.W., & Wagner, U. et al (1998). Outgroup prejudice in Western Europe. In W.Stroebe, & M.Hewstone (Eds.), European Review of Social Psychology (Vol 9, pp. 241– 273). Chichester, UK: Wiley. Piaget, J. (1932). The moral judgment of the child.London: Routledge. Piaget, J. (1952). The origins of intelligence in children.New York: International Universities Press. Reicher, S., & Emler, N. (1984). Moral orientation as a cue to political identity. PoliticalPsychology,5, 543–551. Rest, J. (1975). Longitudinal study of the Defining Issues Test of moral judgment: A strategy for analysing developmental change. Developmental Psychology, 11, 738–741. Rest, J., Cooper, D., Coder, R., Masanz, J., & Anderson, D. (1974). Judging the important issues in moral dilemmas—An objective measure of development. Developmentalpsychology, 10,491–501. Rest, J., Narvaez, D., Bebeau, M.J., & Thoma, S.J. (1999). Postconventional moral thinking: Aneo-Kohlbergian account.Mahwah, NJ: Lawrence Erlbaum Associates Inc. Rest, J., & Thoma, S. (1985). The relation of moral judgment development to formal education. Developmental Psychology, 21,709–714. Rest, J., Thoma, S., & Edwards, L. (1997). Designing and validating a measure of moral judgment: Stage preference and stage consistency approaches. Journal of EducationalPsychology, 89,5–28.
294 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Rogoff, B. (1981). Schooling and the development of cognitive skills. In H.C.Triandis, & A. Heron (Eds.), Handbook of cross cultural psychology, Vol 4. Developmental psychology.Boston: Allyn & Bacon. Rokeach, M. (1960). The open and closed mind.New York: Basic Books. Rokeach, M. (1979). The two value model of political ideology and British politics. In M Rokeach (Ed.), Understanding human values: Individual and social.New York: Free Press. Roker, D. (1993). Gaining the edge: Girls at a private school. In I.Bates, & G.Riseborough (Eds.), Youth and inequality.Milton Keynes, UK: Open University Press. Rosenberg, S., Ward, D., & Chilton, S. (1988). Political reasoning and cognition: A Piagetianview.Durham, NC: Duke University Press. Rubin, M., & Hewstone, M. (1998). Social identity theory’s self-esteem hypothesis: A review and some suggestions for clarification. Review of Personality and Social Psychology, 2,40– 62. Sampson, E.E. (1971). Social psychology and contemporary society.New York: Wiley. Schönbach, P., Gollwitzer, P., Stiepel, M., & Wagner, U. (1980). Education and intergroupattitudes.London: Academic Press. Schuman, H., & Presser, S. (1981). Questions and answers in attitude surveys: Experiments onquestion form, wording and content.New York: Academic Press. Sniderman, P.M., Brody, R.A., & Tetlock, P. (1991). Reasoning and choice: Explorations inpolitical psychology.Cambridge: Cambridge University Press. Sparks, P., & Durkin, K. (1987). Moral reasoning and political orientation: The context sensitivity of individual rights and democratic principles. Journal of Personality and SocialPsychology, 52,931–936. Stacey, B. (1978). Political socialisation in Western society.London: Arnold. Sullivan, E.V., & Quarter, J. (1972). Psychological correlates of certain postconventional moral types: A perspective on hybrid types. Journal of Personality, 40,149–161. Tetlock, P. (1981). Pre- to post-election shifts in rhetoric: Impression management or cognitive adjustment? Journal of Personality and Social Psychology, 41,207– 213. Tetlock, P.E. (1983a). Cognitive style and political ideology. Journal of Personality and SocialPsychology, 45,118–126. Tetlock, P.E. (1983b). Accountability and complexity of thought. Journal of Personality andSocial Psychology, 45,74–83. Thoma, S. (1993). The relationship between political preference and moral judgment development in late adolescence. Merrill-Palmer Quarterly, 39,359– 374. Torney-Purta, J. (1983). The development of views about the role of social institutions in redressing inequality and promoting human rights. In R.Leahy (Ed.), The child’sconstruction of social inequality.New York: Academic Press. Turner, J.C. (1975). Social comparison and social identity: Some prospects for intergroup behaviour. European Journal of Social Psychology, 5,5–34. Tygart, C.E. (1984). Political liberalism-conservatism among the clergy: The question of dimensionality. Human Relations, 37,10, 853–861.
MORALITY AND POLITICS 295
Verba, S., Schlozman, K.L., & Brady, H.E. (1995). Voice and equality: Civic voluntarism inAmerican politics.Cambridge, MA: Harvard University Press. West, C.K., Fish, J.A., & Stevens, R.J. (1980). General self-concept, self-concept of academic achievement and school achievement: Implications for ‘causes’ of self-concept. AmericanJournal of Education, 24,194–213. White, C., Bruce, S., & Ritchie, J. (2000). Young people’s politics: Political interest andengagement amongst 14- to 24-year olds.York, UK: Joseph Rowntree Foundation. Wilson, G.D. (1973). The psychology of conservatism.New York: Academic Press. Zaller, J. (1990). Political awareness, elite opinion leadership and the mass survey response. Social Cognition, 6,125–153.
EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY, 2002, 13, 293–323
Testing attitude—behaviour theories usingnon-experimental data: An examination of somehidden assumptions Stephen Sutton University of Cambridge, UK
This chapter presents a detailed causal analysis of the two leading theories of attitude—behaviour relations, the theory of reasoned action (TRA) and the theory of planned behaviour (TPB). It is noted that the direct path from perceived behavioural control to behaviour in the TPB is causally ambiguous. Focusing on the attitude—intention relationship, timeline and path diagrams are used to illustrate some of the hidden assumptions that underlie the way these theories are usually tested using standard multiple regression applied to nonexperimental data. Randomised experiments are recommended as the best solution to the serious problems arising from omitted causes. For the past quarter of a century, the theory of reasoned action (TRA; Ajzen & Fishbein, 1980; Fishbein & Ajzen, 1975) and its extension, the theory of planned behaviour (TPB; Ajzen, 1991; Ajzen & Fishbein, 2000) have been regarded as the leading theories of the attitude-behaviour relationship. These theories are usually “tested” in observational, nonexperimental studies using cross-sectional or prospective designs or a combination of the two. Numerous studies using this methodology have been published; see, for example, the recent meta-analysis of the TPB by Armitage and Conner (2001), and the quantitative reviews and metaanalyses of the TRA and the TPB summarised by Sutton (1998). This chapter examines some of the assumptions that underlie the way that the TRA and the TPB are usually tested. Many of these assumptions appear not to be widely appreciated by researchers in the field. They are © 2002 European Association of Experimental Social Psychology http://www.tandf.co.uk/journals/pp/02699931.html DOI: 10.1080/10463280240000019
SUTTON 297
implicit in the methods commonly used to investigate the theories, in particular in the standard multiple regression model applied to betweenindividuals data. (Note that structural equation modelling makes the same assumptions.) There are two reasons for examining these hidden assumptions. First, it would seem desirable as an end in itself that researchers are fully aware of the assumptions they are making. Second, the assumptions that are made have important methodological implications, for example with regard to whether a cross-sectional or a prospective design is more appropriate for investigating a particular causal relationship or, indeed, whether nonexperimental studies are appropriate at all. The present chapter treats the TRA and the TPB as causal models. It provides a detailed causal analysis (Heise, 1975) of these theories, focusing particularly on the attitude-intention relationship. The exposition is deliberately non-mathematical and makes minimal use of algebraic equations. Instead, diagrams (timeline and path diagrams) are used to illustrate the main points. THE TRA AND THE TPB AS CAUSAL MODELS Figure 1 shows a simplified version of the TRA. It is simplified in that it omits the beliefs (behavioural and normative) and their respective weights (outcome evaluations and motivations to comply) that are assumed to underlie the global constructs of attitude (A) and subjective norm (S). The TRA is a causal model. Attitude and subjective norm are assumed to influence intention (I), which in turn, under certain conditions (correspondence/compatibility; stability of intention; behaviour under volitional control), is assumed to influence behaviour (B). Figure 2(a) shows the TPB. Again, the belief components have been omitted for simplicity. All the paths in this model represent genuine causal effects with the exception of the direct path from perceived behavioural control to behaviour, which is causally ambiguous (Sutton, 2002). One rationale for this direct link is that perceived behavioural control can often be used as a substitute for actual control (Ajzen, 1991). It goes without
Address correspondence to: Stephen Sutton, University of Cambridge, Institute of Public Health, Robinson Way, Cambridge, CB2 2SR, UK, Email:
[email protected] I would like to thank the editors and four anonymous reviewers for valuable comments on earlier drafts of this chapter.
298 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 1. Path diagram showing a simplified version of the theory of reasoned action. (A=attitude towards the behaviour; S=subjective norm; I=intention; B=behaviour.)
saying that actual control directly influences behaviour or performance: a person cannot achieve a good mark in a maths exam if they do not have the ability; they cannot buy petrol if the garage is closed. But, it is argued, actual control is often difficult to measure and it is less interesting psychologically than perceived control. Perceived control can be used a proxy for actual control to the extent that people’s perceptions of control are accurate. According to this rationale, the direct link between perceived behavioural control and behaviour is not a causal path, and changing perceived behavioural control would not lead to behaviour change directly. (It could lead to behaviour change indirectly, of course, via a change in intention.) In order to change behaviour directly, it is necessary to change actual control. Figure 2 (b) shows a version of the TPB that includes actual control as well as perceived control. The path from actual control to behaviour represents a genuine causal effect. The correlation between perceived control and actual control, depicted by the double-arrowheaded path, represents the extent to which people accurately perceive the amount of actual control they have over the behaviour. This correlation may be large or small. The larger it is—in other words, the greater the accuracy of people’s perceptions—then the larger will be the direct predictive effect of perceived behavioural control on behaviour when estimated in the usual way, that is in the context of the model shown in Figure 2 (a). But this is a predictive effect of perceived behavioural control not a causal effect. Figure 2 (c) is identical to Figure 2 (b) except that it also includes a direct causal path from perceived behavioural control to behaviour. Ajzen (1991, p.6) suggested a rationale for this causal path: …holding intention constant, the effort expended to bring a course of behaviour to a successful conclusion is likely to increase with perceived behavioral control. For instance, even if two individuals have equally strong intentions to learn to ski, and both try to do so,
SUTTON 299
Figure 2. Path diagrams showing the theory of planned behaviour. (A=attitude toward the behaviour; S=subjective norm; PC=perceived behavioural control; AC=actual behavioural control; I=intention; B=behaviour.) Model (a) is a simplified version of the theory, omitting the beliefs that are assumed to influence A, S, and PC. The path from PC to B is represented as a dashed line to indicate that it is causally ambiguous (see text). Model (b) is a version of the theory that includes actual as well as perceived control. The correlation between PC and AC, depicted by the double-arrowheaded path, represents the extent to which people accurately perceive the amount of control they have over the behaviour; for legibility, the other correlations among the variables on the left-hand side are not shown. Model (c) is identical to model (b) except that it includes a direct causal path from PC to B, represented by a dotted line.
300 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
the person who is confident that he can master this activity is more likely to persevere than is the person who doubts his ability. Note that this effect is held to be mediated by “effort” and “perseverance”, neither of which are constructs in the theory. Thus the direct predictive effect of perceived behavioural control on behaviour shown in Figure 2 (a) can be thought of as comprising causal and non-causal components. When we investigate the theory in the usual way and estimate this path, we are conflating a genuine causal effect and what is merely a predictive effect. It is important to try to estimate the relative size of these two components. This would seem to require laboratory-based experimental studies in which actual control and perceived behavioural control are manipulated. A further complication concerns the interaction between perceived behavioural control and intention on behaviour that was postulated by Ajzen and Madden (1986). This interaction derives from an interaction between intention and actual control. In particular, intention is expected to have a stronger influence on behaviour the greater the degree of actual control the person has over the behaviour. Stated differently, the effect of actual control on behaviour will be larger the stronger the person’s intention to perform the behaviour. To give a simple example, a person will succeed in buying petrol from the garage only if they intend to do so and the garage is open. In the analysis that follows, we attempt to explain some of the assumptions that underlie the way that the TRA and the TPB are usually tested in nonexperimental studies. We use simplified versions of the models, focusing particularly on the attitude—intention relationship. In the next section, timeline diagrams are used to represent the relationship between attitude and intention within an individual over time. TIMELINE ANALYSIS OF THEATTITUDEINTENTION RELATIONSHIP Consider the effect of attitude and subjective norm on intention. This part of the TRA/TPB is usually tested for a given target population and target behaviour by conducting a cross-sectional study in which measures of A, S, and I are obtained in a suitable sample of individuals and the relationships betw een these variables are analysed using correlation and regression techniques. In particular, multiple linear regression is used to estimate the independent effects of A and S on I and the percentage of variance explained in I by these two predictors. The independent effects of A and S on I are commonly reported in terms of the beta weights (standardised partial regression coefficients) and their associated significance tests (t-tests of the hypothesis that each beta weight differs from zero). Alternatively,
SUTTON 301
the unique percentage of variance explained or the squared semi-partial or part correlation may be reported; note that the significance test is the same as that for the corresponding regression coefficient. We focus here on the regression weights because, as estimates of the structural (=causal) coefficients, they can be regarded as more fundamental. Note that this analysis relates differences between individuals in I to between individuals in S (controlling for A). For example, a beta weight differences between individuals in A (controlling for S) and to differences from this analysis of .4 for attitude can be interpreted as meaning that, holding S constant, a difference between individuals of one standard deviation unit on the attitude scale is associated with a difference between individuals of .4 standard deviation units on the intention scale. However, we usually wish to draw dynamic inferences from such crosssectional analyses (Berry, 1993). To state that attitude influences intention is usually understood to imply that if we could select an individual and intervene to increase their attitude, while holding all other variables constant, their intention would also increase. Thus, we wish to use a cross-sectional between-individuals analysis to draw inferences about processes that are assumed to occur within individuals over time. Drawing dynamic causal inferences from cross-sectional analyses requires a number of strong assumptions. In order to illustrate some of these assumptions, we consider a simplified version of the TRA/TPB in which attitude (A) is the only cause of intention (I). Let us assume that the following equation: I=.5A+.5, that is, that an increase of one unit in A withinindividual relationship between A and I is described by the produces an increase of .5 units in I. For the moment, we assume that this structural coefficient is unstandardised. Figure 3 shows timelines depicting the changes over time in A and I for three individuals (P1, P2, P3). A and I are shown as being measured on 7-point scales scored from 1 to 7, where the intervals are assumed to be equal and a high score means a more positive attitude or a stronger intention. The timelines in Figure 3 should be thought of as representing the “true scores” on attitude and intention for these three individuals. For simplicity, we assume perfect measurement. In other words, we assume that if A or I were measured for an individual at any time during the time period depicted in Figure 3, the observed score would equal the true score. [Note that the standard regression model assumes that the dependent variable (here, intention) is continuous and unbounded, that is, that it can take any numerical value. For present purposes, we will ignore problems that arise from the use of bounded scales.] Consider the period to the left of the bold vertical line, that is, the period cross-sectional study. For all three individuals, the cross-sectional relationbetween T0 and T1. This could represent, say, a 4-week period prior to a ship between A and I is described by the equation given above.
302 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 3. Timeline diagrams showing changes over time in attitude (A) and intention (I) for three individuals (P1, P2, P3).
For example, at the beginning of the period, P1 has a score of 2.5 on I, which equals .5 times their score on A (4) plus .5. Similarly, at the end of the period, P1’s score on I is 3, which equals .5 times A (5) plus .5. The discontinuous portions of the timelines for P1 and P2 are meant to indicate that these individuals had pre-existing attitudes and intentions. P1’s attitude is initially stable and so is their intention—remember that we are making the simplifying assumption that A is the sole cause of I, which means that I cannot change unless A changes. After a while, P1’s attitude increases (all changes in A and I are shown as occurring abruptly), and this is followed almost instantaneously by an increase in intention. The relationship between the change in I and the change in A reflects the structural coefficient of .5. An increase of 1 unit in P1’s attitude (from 4 to 5 on the 7-point scale) produces an increase of .5 units in their intention
SUTTON 303
(from 2.5 to 3). In causal analysis, it is assumed that cause always precedes effect by at least some minimal time period. Expositions of the TRA and the TPB are not explicit about the causal lag between attitude and intention (or between subjective norm and intention, or perceived behavioural control and intention). However, most applications of these models implicitly assume a very brief causal lag between a change in attitude and the corresponding change in intention; indeed, as we will see, this is one rationale for using a crosssectional rather than a prospective design to investigate the relationship between the two variables. Note that the increase in I does not produce any change in A (no reciprocal causation). During this same time period, P2 shows a decrease in attitude, which is immediately followed by a decrease in intention. The structural coefficient is again .5: a decrease of two units in attitude produces a decrease of one unit in intention. Note the symmetry of the relationship between A and I: a decrease of a given magnitude in A has the same effect on I (but in the opposite direction) as an increase of the same size. We have also assumed that the causal lag is the same in the two cases. Now consider P3. Initially, this person does not have an attitude or an intention towards performing the behaviour in question (they have never thought about performing it). But at some point in this period they form an attitude and this is immediately followed by formation of a corresponding intention. The same structural equation (I=.5A+.5) is assumed to apply here too. Now suppose that we conduct a cross-sectional study at time T1 in which we measure A and I perfectly (no measurement error) in a sample of individuals whose timelines are consistent with the patterns and characteristics of those shown in Figure 3. (For individuals who have never previously thought about performing the behaviour, completing a questionnaire may prompt the formation of an attitude and a corresponding intention. This measurement effect does not pose any problem so long as the same equation applies.) If all the assumptions referred to above hold for all individuals in the sample, then the crosssectional between-individuals relationship between A and I as measured at time T1 will accurately reflect the withinindividual relationship between A and I. The equation I=.5A+.5 will describe the cross-sectional betweenindividuals relationship between A and I as well as the within-individual relationship between these two variables. Furthermore, the relationship between differences between individuals in A and differences between individuals in I will reflect the structural coefficient of .5. For example, at time T1, P1 and P2 differ by 1 unit on A (5–4) and by .5 units on I (3–2.5). Let us examine some of the assumptions in more detail: (1) It is assumed that the same structural coefficient applies to every individual in the sample, i.e., that, for any individual in the sample, an
304 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Figure 4.Timeline diagrams showing changes over time in attitude (A) and intention (I) for four individuals (P4, P5, P6, P7).
increase in A of one unit would produce the same increase in I. This assumption is not inherent in the TRA or the TPB, which, in principle, allow different structural coefficients for different individuals from the same target population with respect to the same target behaviour. Rather, the assumption is required by the use of between-individuals cross-sectional regression analysis. Figure 4(a) Shows a case where two individuals have different coefficients. For person P4, the relationship between A and I is described by the equation I=.8A+.5, whereas for person P5 the relationship is described by the equation I=.5A+.5. A between-individuals crosssectional analysis at T1 yields a single coefficient (in this case, 1.1), which cannot provide an accurate estimate of the two withinindividual coefficients.
SUTTON 305
Within the framework of between-individuals cross-sectional regression analysis, structural coefficients can be allowed to vary for different subgroups of individuals by specifying interactions (Aiken & West, 1991; Cohen & Cohen, 1983). For example, an application of the TRA/TPB to the explanation of condom use intentions may allow men and women to have different coefficients for the attitude-intention relationship (cf. Sutton, McVey, & Glanz, 1999). However, within the standard regression model, it is still necessary to assume that all the men share the same coefficient and all the women share the same coefficient. If repeated measures of attitude and intention are obtained, either over time with respect to the same target behaviour or at one time point with respect to different target behaviours, then random-effects regression can be used to estimate different coefficients for different individuals (Hedeker, Flay, & Petraitis, 1996). In a random-effects regression analysis predicting I from A, a different coefficient for the effect of A is estimated for each individual, on the assumption that the coefficients vary randomly across individuals in the population and have a normal distribution. Thus, some individuals may show a strong positive relationship between A and I, others may show a weak positive relationship, and still others may show a near zero or even a weak negative relationship. The mean of these weights gives the “average” effect of A on I. The variance of these weights can be estimated, and the hypothesis that this variance is zero (i.e., that the same coefficient applies to everyone) can be tested directly. If there is significant individual variation in the regression coefficients, this variation can be modelled in terms of relevant individual-level characteristics. For example, it may be hypothesised that variation in the attitude coefficient is partly explained by variables such as gender (e.g., that women show a stronger relationship between A and I than men) or attitude strength. Random-effects regression has advantages over simply computing a correlation between attitude and intention for each individual in the sample or using standard regression to estimate a model for each individual, as has been done in a number of studies (e.g., Trafimow & Finlay, 1996). The paper by Hedeker and colleagues provides a very clear description of the method (although in their substantive example using the TRA, the measures of attitude and subjective norm departed widely from Ajzen and Fishbein’s 1980 recommendations). Randomeffects regression models can be estimated using PROC MIXED in SAS or a multilevel modelling program such as MLwiN (Goldstein et al., 1998). (2) It is further assumed that the structural equation (in this example, I=.5A +.5) is stable over time and has held true throughout the history of the
306 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
relationship between A and I for all individuals in the population, including the point at which the intention was formed. For some individuals (e.g., P3 in Figure 3), the history of the relationship may be brief; for others, it may extend back many years. It may be questioned whether this assumption need be so strict. Surely it is only necessary that the structural equation holds true for a relatively short period of time leading up to the cross-sectional study? If so, the cross-sectional analysis should still accurately reflect the recent history of the relationship between A and I. However, two problems arise from relaxing the assumption in this way. First, generalisability of the findings would be limited, and researchers would need regularly to repeat the same study in order to check that the same findings were obtained. Second, if the structural equation describing the relationship between A and I were believed or shown to be unstable over time, it would be necessary to try to explain why the structural relationship changes and why different equations hold at different time points. (3) It is assumed that the causal lag for the attitude-intention relationship is very short. Figure 4 (b) shows a relatively long causal lag in which the change in I produced by a change in A does not occur until after T1, that is, until after the cross-sectional study has been conducted. This creates a mismatch between the values of A and I measured at T1. P6’s score on A is being compared with the “wrong” score on I. So the equation derived from a between-individuals cross-sectional analysis will not accurately reflect the true relationship between A and I. Where the causal lag is relatively long, a cross-sectional design is inappropriate, even if all the other assumptions hold. Instead, a prospective design should be employed in which A is measured at one time point (T1, say) and I is measured at a later time point (T2), where, ideally, the length of the follow-up period is approximately equal to the expected length of the causal lag (Finkel, 1995). If the follow-up period is shorter than the causal lag, a change in A that occurs just prior to T1 will not have produced its effect on I by T2; thus, the effect of recent changes in A will be missed at the later time point. If the follow-up period is too long, a change in A that takes place during the follow-up period may produce a change in I within the same period. In both cases, the value of A at T1 and the value of I at T2 will be mismatched. However, if the causal lag is very brief, a prospective design is not appropriate and a cross-sectional design should be employed. Thus, although it is widely believed that prospective designs allow stronger inferences to be drawn, this is the case only where the causal lag is relatively long and the follow-up period is carefully chosen so that it approximately equals the expected causal lag. In the extreme case where A (and therefore I) is extremely stable over the follow-up
SUTTON 307
period (in the sense that individuals show little change over time relative to one another), it makes no difference whether a crosssectional or a prospective design is used; both should yield similar estimates of the effect of A on I. As mentioned above, applications of the TRA/TPB usually implicitly assume a very brief causal lag between all the variables in the model apart from behaviour; for example, between beliefs and attitude and between attitude and intention. However, there are some exceptions. For example, Liska (1984, p. 66) suggested that “…as information processing takes time, changes in attitudes may lag behind changes in beliefs, perhaps in some cases by months or even years…”, and Armitage and Conner (1999) employed a time lag of 3 months. Attitude-behaviour researchers should try to specify the causal lag for each of the relationships in the TRA/TPB or at least consider the implications of different causal lags. They should also consider the possibility that different relationships may have different causal lags. For example, it is conceivable, though admittedly not very plausible, that the attitude-intention and subjective normintention relationships have different causal lags. Furthermore, different individuals may have different causal lags (as well as different structural coefficients) for the same relationship. (4) Another key assumption is that there is no reciprocal causation. The causal flow is assumed to be one-way, in this case from attitude to intention. Changes in intention are assumed not to produce changes in attitude. (We also have to assume that changes in behaviour—not shown in the timeline diagrams—do not produce changes in attitude or intention.) Figure 4 (c) illustrates reciprocal causation between A and / for one individual (P7). The structural equation for the effect of A on I is I=.5A +.5, as in Figure 3; the structural equation for the reverse effect of I on A is A=4I+.5. The latter effect could be interpreted in terms of a tendency towards consistency, with attitude being “brought into line” with the new intention. For simplicity of presentation, we assume that the causal lags are the same for the two effects. Both coefficients are positive, creating a positive feedback loop. An increase of one unit in A produces an increase of .5 units in I. This in turn produces an increase of .2 units in A [=(.4)(.5)], which in turn leads to an increase of .1 units in I [=(.5)(.2)]. This in turn produces an increase of .04 units in A [=(.4) (.1)]. This process continues indefinitely with ever smaller effects (too small to be shown in the diagram). It can be shown that, given the structural coefficients that were (arbitrarily) selected for this example, A and I (will converge to the values of 5.25 and 2.625 respectively (Heise, 1975; Kenny, 1979). Assuming that these values are approximately achieved by T1, a cross-sectional analysis carried out on data obtained at this time point from a sample of individuals all of
308 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
whom share the same structural equations and assuming one-way causation from A to I would estimate the structural coefficient to be . 5. Thus, if the incorrect assumption of one-way causation is made, the crosssectional data would be interpreted as indicating that an increase in A of one unit produces an increase in I of .5 units. In fact, because of reciprocal causation, a one unit increase in A produces an increase of . 625 units in I (and A itself increases by a further .25 units as a consequence). Again, it might be thought that employing a prospective design would enable each of the two effects to be accurately estimated. Measures of attitude and intention would be obtained at two time points, T1 and T2. Attitude and intention at T2 would be regressed on attitude and intention at T1, in what is known as a crosslagged regression or a cross-lagged structural equation model. Interpretation focuses on the effect of attitude at T1 on intention at T2, and the effect of intention at T1 on attitude at T2. However, as noted in the preceding discussion of causal lags, the validity of this analysis would depend on choosing the correct follow-up period(s), taking into account the possible complication that the effects may have different causal lags— not implausible in this case. Moreover, if the causal lags were very brief, a prospective design would not be appropriate. An alternative to cross-lagged regression is cross-lagged panel correlation analysis, which has been used in several applications of the TRA/TPB (e.g., Armitage & Conner, 1999; Kahle & Beatty, 1987; Wittenbraker, Gibbs, & Kahle, 1983). This technique has attracted substantial criticism from methodologists over the years, most notably by Rogosa (1980), and can no longer be recommended. Even the two researchers who originally developed it have recently admitted that “it is certainly not the causal divining rod that some (including ourselves) once thought that it might be” and recommend its use only when “the goal is exploratory and the expectation is that there are few, if any, causal effects” (Campbell & Kenny, 1999, Chapter 9, p. 154). Standardised versus unstandardised coefficients Throughout the examples considered above, we used unstandardised coefficients, that is, coefficients expressed in terms of units on 7-point scales scored from 1 to 7. Most studies that use the TRA or the TPB report standardised coefficients or beta weights where relationships are expressed in terms of standard deviation units. The reason for preferring standardised coefficients is that the scales used to measure constructs like attitude and intention are arbitrary in the sense that a 7-point scale could be scored 1 to 7, 0 to 6, –3 to +3, or even 10 to 70. Although Ajzen and Fishbein (1980) present clear recommendations concerning how the TRA measures
SUTTON 309
should be scored, any scheme based on a linear transformation of the numbers 1 to 7 could be used in practice: correlations and standardised regression coefficients would be unaffected. (An important exception to this rule occurs when a measure is used as a component of a multiplicative composite, as in behavioural beliefs for example; see Evans, 1991, and French and Hankins, in press). The arbitrary nature of the scales creates a problem when a researcher wishes to compare two effects (e.g., the effects of attitude and subjective norm on intention). If attitude is scored 1 to 7 and subjective norm is scored 10 to 70 (to take an extreme example), it may appear that the effect of attitude is larger than the effect of subjective norm. The usual solution is to standardise the coefficients so that the effects are expressed in terms of standard deviation units. The researcher can then compare the effect of attitude (in terms of the effect on intention, in standard deviation units, of increasing attitude by one standard deviation unit) with the effect of subjective norm (in terms of the effect on intention, in standard deviation units, of increasing subjective norm by one standard deviation unit). However, it can be argued that standardised coefficients are no less arbitrary than unstandardised coefficients, particularly when one wishes to compare findings from different subsamples within the same study (e.g., men and women), to compare samples from different studies, or to combine findings across studies, as in meta-analysis. Standardised coefficients depend on the standard deviations of variables, which may differ arbitrarily from study to study or between subgroups within a single study in ways that are unrelated to the true structural relationships between variables. Even if the scales used to measure attitudes and intentions are arbitrary, if all studies (or even if most studies) used the same items, it might be preferable to use unstandardised coefficients. However, so long as authors report the standardised coefficients, the standard deviations of the measures, and how the measures were scored, it is possible to calculate the unstandardised coefficients and to rescale them so that they can be compared and combined across studies if desired. ALLOWING MULTIPLE CAUSES OF INTENTION In the preceding analysis using timeline diagrams, we made the simplifying assumption that attitude is the sole cause of intention. We now consider some of the assumptions that are necessary when this simplifying assumption is dropped. This section of the chapter was inspired by Clogg and Haritou (1997). The presentation in this chapter differs from that employed by Clogg and Haritou in a number of ways: it is less technical and uses path diagrams to make the argument more accessible; it adapts the general argument to the specific case of the TRA/TPB; and it considers
310 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
a number of potential solutions to the problem identified by Glogg and Haritou that were not examined by those authors. First, consider Model I in Figure 5. This shows an arrow from attitude towards the behaviour (A) to intention (I). This arrow represents the causal effect that we wish to estimate for a given target behaviour and target population. We now allow I to be directly influenced by other causes apart from A, although initially we do not specify what these additional causal factors are. The aggregate (weighted additive composite) of all other causes is represented by u, and the effect of this aggregate is represented by the small arrow pointing at I. u is often referred to as the error or disturbance term. The disturbance term can be thought of as comprising two components: a systematic component consisting of other important causes of I apart from A (e.g., subjective norm), some of which may be correlated with A; and a random component consisting of tens, hundreds, or even thousands of minor, independent, and unstable causes of I. The latter causes would be impossible to specify in practice. Although individually the effects of these “random shocks” are assumed to be small, in aggregate their effect may be quite large. Conceivably, the majority of the variance in intention could be explained by the random component, thus placing a severe limit on the proportion of variance that could be explained by the major causal factors. Suppose we have hypothesised that A is an important cause of I and that the causal lag is very short. In order to estimate the causal effect of A on I, we could measure A and I at the same time point, and then regress I on A using linear regression. If certain assumptions hold (including those discussed in the first section of this chapter), the standardised regression coefficient (beta weight) from this analysis can be interpreted as an unbiased estimate of the causal effect of attitude on intention. (Note that, in this simple bivariate case, the standardised regression coefficient equals the correlation coefficient.) Given that we are now allowing multiple causes of I, in order to be able to interpret the regression coefficient as an unbiased estimate of the causal effect of A on I, an additional assumption is required, namely that the correlation between A and the aggregate of all other causes of I is zero. In Figure 5, the correlation between A and u is represented by the dashed double-arrowheaded line. This assumption is necessary because the least-squares estimation procedure underlying linear regression forces the residuals (the differences between the observed and predicted values of the dependent variable) to correlate zero with the predictor variables. But is it reasonable to assume that this approximates the situation in the real world? The answer to this question depends on how the error term is constituted. If there were no systematic component in the error term so that it consisted only of the random component, the expected value of the correlation between A and the error term would be zero. Similarly, if the error term consisted only of
SUTTON 311
Figure 5. Path diagrams showing three models that could be estimated by regression analysis. (A=attitude; S=subjective norm; P=perceived behavioural control; I=intention; u, v, w are the disturbance terms for I; the continuous double-arrowheaded lines represent the [known] correlations among the predictor variables; the dashed double-arrow-headed lines represent the [unknown] correlations between the predictor variables and the disturbances.)
the random component and major causes of I each of which were uncorrelated with A, then the expected value of the correlation between A and the error term would again be zero. However, if the error term included major causes of I that were correlated with A, then the expected value of the correlation between A and the error term would not in general
312 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
equal zero. Unfortunately, it is not possible to distinguish between these three cases in practice. Thus the assumption that A correlates zero with the aggregate of all other causes of I is arbitrary and untestable. The true correlation could be zero. But it could also be .40, or •.70, or any other value between •1.0 and+1.0—we simply have no way of knowing. If the correlation is anything other than zero, our estimate of the causal effect of A on I will be “wrong” (biased) to an unknown extent. Now suppose we measure subjective norm (S) and include it in the regression model; in other words, we regress I on A and S (see Model II in Figure 5). In order to interpret the standardised partial regression coefficients from this analysis as unbiased estimates of the causal effects of A and S on I, we have to assume, among other things, that both A and S are correlated zero with the aggregate of all other causes of I; in other words, we now have to make two arbitrary and untestable assumptions. The aggregate variable in Model II is represented by v; it differs from u in Model I because we have removed S from the error term and made it one of the predictors. Note that, although Model II would be recognised by all attitude-behaviour researchers as the TRA (or, at least, part of it), Model I is not incompatible with the TRA. The difference between the two models is simply that, in Model I, S is unspecified and part of the error term, whereas, in Model II, it is explicitly identified as a variable that influences I. So if we choose to estimate Model I, we are not necessarily implying that we believe that subjective norm does not influence intention; rather we may have made a deliberate decision to leave subjective norm as part of the error term instead of specifying it as a predictor. If we expand the model by including a third predictor (P, which stands for perceived behavioural control), as in Model III, we have to make three arbitrary and untestable assumptions. Every time another predictor is added to the model, the number of such assumptions increases. The fact that drawing causal inferences from nonexperimental data rests on arbitrary and untestable assumptions involving unmeasured variables has very serious implications. It means that there is no basis for deciding which variables to include in a regression model and which to leave out, and that there is no basis for choosing between different models. In order to explain these implications, suppose that the true model is that represented by the path diagram in Figure 6(a). All relationships are assumed to be linear and additive, and all variables are assumed to be standardised. Two causes of I are specified (A and S);the correlation between them is a. v is a hypothetical variable representing the aggregate of all other causes of I apart from A and S; it includes both random and systematic components. The correlations between v and the two specified causes, A and S, are represented by e and f respectively. Note that this diagram is not incompatible with the TPB or with extensions to the TPB; perceived behavioural control and other major determinants of intention can be
SUTTON 313
Figure 6. Path diagrams showing three alternative representations of a true causal model. (A=attitude; S=subjective norm; I=intention; u and v are the disturbance terms for I;a, e,f, h are correlation coefficients; the other coefficients are path coefficients; the numbers in parentheses represent the three different examples discussed in the text.)
thought of as unspecified components of (the systematic part of) the aggregate variable v. Figure 6(b) is an alternative representation of the true model in which S has been absorbed into the error term—symbolised by u in order to
314 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
distinguish it from v in Figure 6(a). The correlation between A and u is represented by h. Figure 6(c) is yet another alternative representation of the true model. It shows the relationship between the two error terms u and v and thus how the diagram in Figure 6(b) can be obtained from the diagram in Figure 6(a). Specifically, u is a weighted additive composite of S and v: u = iS+jv. Note that u is completely determined by S and v. The causal effect of S on I in Figure 6(a) is decomposed in Figure 6(c) into an indirect effect consisting of a path from S to u and a path from u to I. By the rules of path analysis (Heise, 1975; Kenny, 1979), (1) Similarly, the causal effect of v on I in Figure 6(a) is decomposed in Figure 6(c) into an indirect effect comprising a path from v to u and a path from u to I: (2) The variance of u is given by the following equation: (3) Because u is standardised, as are all the other variables in the model, this expression equals 1. The correlation between A and u, represented by h in Figure 6(b), is given by the following equation: (4) which can be derived from Figure 6(c) using the rules of path analysis. Given numerical values for the coefficients in Figure 6(a), Equations 1–4 can be used to calculate h and g in Figure 6(b). The important point for present purposes is that Figure 6(a) and Figure 6(b) are two alternative representations of the same true model that differ in that in Figure 6(b)S is part of the error term, whereas in Figure 6 (a) it has been removed from the error term and specified as a cause of I. In order to limit the number of possible cases for discussion, we will assume that none of the coefficients a, b, c, d in Figure 6 equals zero and that a does not equal 1 or –1. Suppose that we obtain perfect measures of A, S, and I in the population (so that we can ignore the problems arising from measurement error and of estimating population values from sample statistics). Suppose that we start by regressing I on A alone. If the correlation h in Figure 6(b) happens to equal zero, we will obtain an unbiased estimate of the true causal effect b of A on I. This is the case even though we have omitted S, a true cause of I, from the regression model and even though including S in the regression model would increase the percentage of variance explained in I.
SUTTON 315
If the correlation h does not equal zero, the beta coefficient will be a biased estimate of the causal effect of A on I. In fact, the coefficient will equal b+gh [see Figure 6(b) ]; thus the amount of bias is equal to +gh. Of course, in practice there is no way of telling whether h is zero or not, or how close to zero it is. The assumption that h=0 is an assumption of necessity: unless we make this arbitrary assumption, we cannot interpret the beta coefficient from the regression of I on A as an unbiased estimate of the causal effect of A on I. Now suppose we regress I on A and S. If the correlations e and f in Figure 6(a)both happen to equal zero, the beta coefficients obtained from this regression analysis will be unbiased estimates of the true causal effects of A and S on I, i.e., of b and c in Figure 6(a). The coefficients shown in the first set of parentheses in the diagrams in Figure 6 provide an example. A and S are assumed to be correlated .60. The true causal effects of A and S on I are .40 and .20 respectively. Because the correlations e and I are both zero, regressing I on A and S yields unbiased estimates of these causal effects. Thirty percent of the variance in I is explained by these two predictors. This example shows that unbiased estimates of causal effects can be obtained even if the regression model does not explain a large proportion of variance in the criterion. If either or both of the correlations e and f do not equal zero, the beta coefficients obtained from regressing I on A and S will not in general be unbiased estimates of the true causal effects of A and S on I. [In fact, it can be shown that the coefficient for A will equal b+(d(e–af)/(1–a2)), that is, the amount of bias equals +(d(e–af)/(1–a2)); and that the coefficient for S will equal c+(d(f–ae)/(1–a2)) that is, the bias equals +(d(f–ae)/(1–a2)).] We have considered two different sets of assumptions: h=0, and e,f =0. It can be shown that these assumptions are not independent. Specifically, if h=0 then e and f cannot both equal zero. This means that if h=0 and we regress I on A alone, we will obtain an unbiased estimate of the effect of A on I, but if we also include S in the regression model, the estimates of the effects of A and S will be biased. This is known as included variable bias: including S, a true cause of I, in the regression model leads to biased estimates of the effects of A and S; excluding it yields an unbiased estimate of the effect of A. The reverse is also true: If e and f both equal zero, h cannot equal zero. Thus if e, f=0 and we regress I on A and S, we will obtain unbiased estimates of both coefficients, but if we omit S from the regression model, we will obtain a biased estimate of the effect of A. This is known as excluded variablebias: excluding S, a true cause of I, from the regression model yields a biased estimate. The assumptions we have been considering (h=0; e, f=0) are special cases. In general, we would expect neither set of assumptions to be satisfied, in which case both analyses (regressing I on A alone, and
316 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
regressing I on A and S) will yield biased estimates. The coefficients shown in the second set of parentheses in the diagrams in Figure 6 provide an example. Here e=.20,f=.30 [Figure 6(a) and Figure 6(c)], and h=.36 [Figure 6(b)]. The true causal effects of A and S on I are .40 and .20 respectively. Regressing I on A alone yields a biased estimate of the effect of A (.68 instead of .40). Regressing I on A and S also yields biased estimates (.39 for A and .42 for S). We would conclude from the latter analysis that the causal effects of A and S on I are approximately equal, whereas, in the true model, the causal effect of A is twice as large as the causal effect of S. Note that, in this example, bias sufficient to lead to an erroneous conclusion is generated by only medium-sized correlations between A and v and S and v [Figure 6(a)]. The regression model that includes both A and S has greater predictive power than the model that includes A alone (or, indeed, than the model that includes S alone). The proportions of variance in I explained by A alone, S alone, and A and S together are .46, .48, and .55 respectively. However, although it has better predictive power, the regression model that includes A and S yields biased estimates because the required assumptions (e,f =0) are not met. We now consider a more extreme example in which close to 100% of the variance is explained in I by a regression model that includes both A and S but where the estimates of the causal effects of these variables are seriously biased. Examine the coefficients in the third set of parentheses in the diagrams in Figure 6. In this example, e=.80, f =.50, and h=.51, so, as in the second example, neither set of assumptions holds. The true causal effects of A and S on I are .20 and .52 respectively. Regressing I on A alone yields a biased estimate of the effect of A (.65 instead of .20). Regressing I on A and S also yields biased estimates (.58 for A and .73 for S). This regression model explains 96% of the variance in I, which is 53 percentage points greater than is explained by A alone and 33 percentage points greater than is explained by S alone. This example is a stark illustration of the general point that the validity of a regression model depends on the validity of its assumptions, not on its predictive power. A common strategy among researchers who use the TRA/ TPB is to try to identify variables that add significantly and substantially to the proportion of variance explained in intention or behaviour. However, the arguments and examples presented above show that this emphasis on variance explained is misplaced. If the aim is to obtain unbiased estimates of causal effects, predictive power is not an appropriate criterion for preferring one regression model to another. Although it is often said that the decision concerning which variables to include in a regression model should be guided by theory, the above arguments demonstrate that theory provides no guidance at all. The TPB, for instance, postulates that A, S, and P (perceived behavioural control) are
SUTTON 317
important causes of I. A researcher investigating this theory would naturally measure A, S, P, and I and would include A, S, and P as predictors in the regression model. But, as we have shown, even if the theory is actually correct, this analysis may yield biased estimates of the effects of the three putative determinants of I, and, furthermore, leaving P (or P and S) out of the analysis may yield an unbiased estimate of A. Thus, in this context, theory does no more than identify the variables for which causal effects need to be estimated. It offers no help in deciding whether to include all of them or only a subset of them in a regression analysis. All the above arguments apply equally to prospective nonexperimental designs. For example, suppose we wish to estimate the causal effect of attitude (A) on behaviour (B). Even if we correctly specify the causal lag, we still have to assume that A correlates zero with the aggregate of all other causes of B, which, as we have argued, is an arbitrary and untestable assumption. Thus, the problem cannot be solved by employing a prospective design. It might be thought that if a number of studies were conducted using the TRA or the TPB and if the results were combined using meta-analysis, any biases would tend to cancel out. Unfortunately, this is not likely to be the case. For example, suppose again that the true model for a particular behaviour and target population is that shown in Figure 6 and that e and f do not both equal zero. Suppose that a number of studies using identical measures of A, S, and I are conducted on different samples drawn from the target population, and that I is regressed on A and S in each study. The findings will vary due to sampling variability, but the estimates of the causal effects of A and S on I will be biased on average across studies; hence, a meta-analysis will also yield biased estimates. We now consider four potential solutions to the problem of omitted variables before outlining the advantages of randomised experimental designs. 1. Mauro’s method The correlation between any two variables (r(1,2)) is constrained by their correlations with a third variable (r(1,3) and r(2,3)). Specifically, the mathematically possible upper and lower limits for r(1,2) are given by the following expression (see Cohen & Cohen, 1983, p.280):
(This constraint was used in constructing the second and third examples presented earlier to ensure that in each case the true model was mathematically possible.)
318 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Mauro (1990) used this constraint as the first step of a method for estimating the effects of omitted variables. Consider the simplest possible case in which we wish to estimate the effect of attitude (A) on intention (I). We measure A and I and compute the correlation between them, which is . 30 say. This is also the value of the beta weight. From this analysis, we may provisionally conclude that A has a substantial positive effect on I. However, we may have omitted a variable X that is an important cause of I. Mauro shows that for any value of the correlation between A and X, it is possible to determine (i) the range of values of the correlation between X and the criterion (I) that are mathematically possible (using the expression given above), and (ii) the size of this correlation that would be required for the omitted variable to reduce the effect attributed to A to nonsignificance or to reverse its sign. To apply Mauro’s method, we would first make an assumption about the size of the correlation between A and X (e.g., that it is positive and lies between .10 and .50). We would then calculate the range of values of the correlation between X and I that were mathematically possible given this assumption and given the observed correlation between A and I. In this simple example, the range of possible values is very wide (between •.92 and +.98). We may wish to rule out some of these values as implausible. For example, we might want to assume that the correlation takes only positive values. The final step is to examine the implications of particular values of this correlation. For example, suppose r(X, I)=.50 and r(X, A)=.10. Under these assumptions, it can be shown that, controlling for X, the beta weight for A would be reduced only slightly from .30 to .25. Now suppose r(X, I)=.50 and r(X, A)=.50. The beta weight for A would be further reduced to 0.07, which may be sufficient to change the initial conclusion that A has a substantial positive effect on I. If r(X, I) were as high as .80, then if r(X, A) =.50, the beta weight for A would equal •.13, which is of opposite sign to the original value. This is an example of suppression (Cohen & Cohen, 1983, pp.94–96). A value of .80 for r(X, I) may or may not be regarded as plausible. Mauro’s method is a useful technique for conducting sensitivity analyses (“what if” this correlation had this value?), but for the following reasons it cannot be regarded as providing a general solution to the problems raised in the second part of this chapter: (1) Although the method can be extended to the case of multiple predictors, it is limited to exploring the effects of a single omitted variable. (2) It requires assumptions to be made about the correlations between the omitted variable and the predictors in order to investigate the consequences for the correlation between the omitted variable and the criterion. (3) These assumptions may be difficult to make (although in some cases estimates of the relevant correlations may be available from other studies). (4) It abandons the aim of obtaining
SUTTON 319
Figure 7. Path diagram showing an instrumental variable model in which Z serves as an instrument for attitude (A) in its relationship with intention (I). a and b are path coefficients. u and v are the disturbance terms for A and I. The dashed doublearrowheaded line represents the (unknown) correlation between the disturbances.
accurate estimates of causal effects and replaces it with the weaker aim of determining whether or not a given predictor variable has a positive or negative causal effect on the criterion. (5) It makes the implicit assumption that a regression model in which the omitted variable were included would yield unbiased estimates; in other words, it assumes that the predictor variables in this model would correlate zero with the aggregate of all other causes of the criterion variable. 2. Instrumental variables Another potential solution is to use instrumental variables (Heise, 1975, pp. 160–185; Kenny, 1979, pp.93–95). Suppose we want to estimate the effect of A on I but we suspect that A is correlated with the disturbance of I. Figure 7 shows an additional variable Z, which: (i) has a direct influence on A; (ii) does not have a direct influence on I; and (iii) is not correlated with the disturbance of I (represented by v in this diagram). Z serves as an instrument for A in the A•I relationship. The effect of Z on A is represented by a, the effect of A on I by b. Since r(Z,I)=ab and r(Z,A)=a, an estimate of b can be obtained very simply from r(Z,I)/r(Z,A). More generally, estimates of effects in instrumental variable models can be obtained by means of two-stage least-squares (2SLS). Applied to this simple example, 2SLS would involve first regressing A on Z, saving the predicted values of A, and then regressing I on these predicted values. Because by assumption Z is uncorrelated with the disturbance of I, the predicted values of A will also have this property. Thus the variable consisting of the predicted values of A can be thought of as a “purified” version of A. Two-
320 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
stage least-squares is implemented in widely used statistical packages including SPSS and STATA. The estimates of causal effects obtained by this method are consistent, that is, they may be biased in small samples, but converge on the true value as the sample size increases. Alternatively, models can be estimated by maximum likelihood in specialist structural equation modelling packages; again, relatively large samples are recommended. As well as satisfying the conditions outlined above, an instrument should ideally have a high correlation with its target variable (A in this example); this increases the precision of the estimates. More than one instrument may be used in order to maximise the prediction of the target variable. The main problem with this method is the difficulty of identifying suitable instruments. Variables such as gender may be used, on the argument that variables whose values are fixed are less likely to be correlated with the disturbance term than those that change over time, but they may correlate only weakly with the target variable. Another possibility is to use a lagged version of the target variable. Thus, if A and I were measured at time T, an earlier measure of A, assessed at time T-1, would probably be highly correlated with the later measure and could be used as an nstrument if it could be assumed: (i) that it did not influence I at time T directly (but only via A at time T); and (ii) that it was not correlated with unmeasured causes of I at time T. 3. Clogg and Haritou’s method Clogg and Haritou (1997) propose an entirely different approach. Suppose again that we want to estimate the causal effect of A on I. Other variables in addition to A may influence I directly; for example, subjective norm (S), perceived behavioural control (P), anticipated regret, and moral norm. If we obtained measures of all these variables on the same sample, there would be 16 (i.e., 24) possible regression models that could be estimated in which I was the dependent variable and A was included as a predictor: A as the sole predictor; A and S as predictors; A and P as predictors; A, S, and P as predictors; and so on. In general, if there are k possible predictors in addition to A, there will be 2k different regression models and hence the same number of different estimates of the partial regression coefficient for A. Thus in our example we would have 16 different estimates of the causal effect of A on I. According to Clogg and Haritou, there is no basis for choosing one of these estimates in preference to any other, because we don’t know for which, if any, of the 16 regression models the required assumptions are satisfied, and the percentage of variance explained is not a relevant criterion for model selection. Their suggested solution is to compute the mean of the 16 coefficients and treat this as a point estimate of the true population effect, and to place a confidence interval around this
SUTTON 321
estimate to convey the degree of uncertainty attached to it. There are two sources of uncertainty: that due to sampling variability (reflected in the standard errors of the regression coefficients) and that due to the variability in the estimate of the A•I effect across the 16 different models (which can be measured by the variance of the estimates around the mean). They suggest computing an estimate of the standard deviation as the square root of the sum of the average squared standard error of the coefficients and the between-coefficients variance, and then using plus or minus two standard deviations to compute the 95% confidence limits around the mean coefficient. The result is an interval estimate that reflects both sampling variance and uncertainty in model selection. Clogg and Haritou’s method is simple enough to implement, but, as the number of predictors increases, the number of different regression models increases rapidly (with 10 predictors, there are 1024 different models), and it would require major changes in how the data from TRA/TPB studies were analysed and how the results were reported. 4. Abandon causal inferences A more radical response to the omitted variables problem is to abandon the attempt to draw causal inferences and to focus instead on prediction. This suggestion is not meant to be taken entirely seriously, but it provides an instructive contrast to causal analysis. Prediction may have huge practical value, particularly when the outcome is behaviour, rather than simply intention, and when the time interval allows interventions to be applied. For example, it would be extremely useful to know which criminals are most likely to re-offend when released from prison. Those considered to be at high risk could then be offered a special rehabilitation programme or released only under certain parole conditions. If the aim is prediction rather than explanation, the strong assumptions required for drawing causal inferences can be dropped. For example, the predictor variables do not have to be theoretical determinants of the target behaviour. Any predictor that works can be included in the regression model. Past behaviour, for instance, may be a particularly good predictor of future behaviour, even though its theoretical status as a determinant of future behaviour is contentious (Sutton, 1994). Similarly, it does not matter if relevant causal variables are omitted from the model. Furthermore, when the aim is prediction, emphasis on the overall predictive power of the model, as measured by the percentage of variance explained, is entirely appropriate.
322 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
EXPERIMENTAL DESIGNS The need to make the arbitrary and untestable assumptions discussed in the preceding section can in principle be avoided by the use of randomised experimental designs. For instance, suppose the aim is to investigate the causal effect of attitude on intention. Attitude towards a particular behaviour could be experimentally manipulated in a between-individuals design by randomly assigning participants to “high” or “low” attitude conditions. Assuming a very brief causal lag, intention would be measured within a few minutes of delivering the manipulation. Randomisation guarantees that all variables, apart from those that are influenced by the experimental manipulation, will correlate zero with the experimental manipulation—not necessarily within a single study (unless the sample size is extremely large), but on average in the long run, over a series of similar experiments. Thus, across a series of similar studies, use of the randomised experimental design will yield an unbiased estimate of the causal effect of the experimental manipulation on intention. According to the TRA/TPB, attitude towards the behaviour cannot be directly manipulated: changes in attitude are assumed to follow from changes in salient beliefs about the consequences of the target behaviour. Thus, an experimental study may attempt to manipulate attitude by presenting information about the possible consequences. This is expected to produce a change in attitude, and a measure of attitude may be included as a manipulation check. The effect of the manipulation on the manipulation check will also be estimated without bias over a series of similar experiments. Other variables may be measured in order to check that the experimental manipulation does not influence factors that it was not designed to influence (e.g., subjective norm). If, over a series of studies, the manipulation influences both the manipulation check and the main dependent variable (intention in this example) but does not influence other putative determinants of intention, then this could be interpreted as providing support for the theory. Of course, this pattern of findings does not show that the effect of the manipulation on intention is mediated entirely by attitude, as the TRA/ TPB would predict. Mediation analysis (Baron & Kenny, 1986) can be employed to test this hypothesis, although investigators should be aware that this procedure requires assumptions similar to those that have been considered in the present chapter (e.g., that causation is unidirectional, from attitude to intention, and that the disturbance terms for attitude and intention are uncorrelated). Furthermore, use of the randomised experimental design brings a new set of problems, for example the possibility of demand effects. It is not within the scope of the present chapter to discuss all the limitations and assumptions of randomised experiments. The main point
SUTTON 323
we wish to make here is that the randomised experimental design provides the best solution to the problems arising from omitted variables, and researchers in this field should make much more use of it. Indeed, it is remarkable how few studies to date have used this approach to test predictions from the TRA/TPB. The study by McCarty (1981) provides a rare example. Using messages about contraceptive use, he attempted to manipulate the attitudinal and normative components of the TRA orthogonally. Unfortunately, the manipulation was not effective; it had no differential effects on measures of attitude and subjective norm, and there was no significant effect on intention. Although this particular study was not successful in this respect, more studies of this kind are needed, using large samples, proper randomisation, and careful development and pretesting of messages. SUMMARY AND CONCLUSIONS This chapter has provided a detailed causal analysis of the TRA and the TPB. These theories were treated as causal models (although it was noted that the direct path from perceived behavioural control to behaviour in the TPB is causally ambiguous). They are usually tested in nonexperimental studies. Unfortunately, whether derived from cross-sectional or prospective designs, such data allow only weak conditional causal inferences to be drawn. A number of strong assumptions need to be made, including assumptions about the length of the causal interval, the stability of the structural equation over time, the invariance of the equation across individuals, and the direction of causation. Although it is widely believed that prospective designs allow stronger inferences to be drawn than crosssectional designs, this is not necessarily the case. If the causal lag is very short, a prospective design may not be appropriate. Attitude-behaviour researchers should carefully consider the degree of fit between the assumptions that underlie the TRA/TPB and those implied by the use of regression analysis applied to cross-sectional or prospective data. For example, they should try to specify the causal lag for each of the relationships in the TRA/TPB and select a research design accordingly. They should also consider how violations of the assumptions may affect the conclusions that are drawn. With nonexperimental data, it is also necessary to assume that each of the predictor variables is uncorrelated with the aggregate of omitted direct causes of the dependent variable (intention in our example). This assumption has serious implications for deciding which variables to include in a regression model and for choosing between different models. If the aim is to obtain unbiased estimates of causal effects, the percentage of variance explained is not an appropriate criterion for model selection. A regression model that explains a large amount of variance may yield seriously biased
324 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
estimates. Conversely, a model that explains only a small amount of variance may nevertheless yield unbiased estimates. On the other hand, if the aim is prediction rather than explanation, an emphasis on predictive power is entirely appropriate. Attitude-behaviour researchers who use nonexperimental designs to estimate causal effects should therefore be extremely cautious in interpreting their findings. The same caution should be applied to the interpretation of meta-analyses that integrate the findings from nonexperimental studies. Several potential solutions to the problems arising from omitted causes were described. Mauro’s (1990) method for estimating the effects of omitted variables is a useful technique for exploring the consequences of different patterns of correlations among included and omitted variables. It requires additional assumptions to be made, but in some cases data that inform these assumptions may be available from other studies. The instrumental variables approach also requires extra assumptions. The main problem with this method is identifying variables that satisfy these assumptions. However, the method may be useful in some situations when suitable instruments can be found. The technique proposed by Clogg and Haritou (1997) does not solve the problems arising from omitted causes, but it does provide a way of quantifying the uncertainty due to model selection. Although each of these approaches has its uses, the best general solution to the problems arising from omitted causes is to employ randomised experimental designs wherever possible. It is hoped that this chapter will encourage attitude-behaviour researchers to focus their efforts and resources on testing the TRA and the TPB in experimental studies. REFERENCES Aiken, L.S., & West, S.G. (1991). Multiple regression: Testing and interpreting interactions.Newbury Park, CA: Sage. Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human DecisionProcesses, 50,179–211. Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behavior.Englewood Cliffs, NJ: Prentice-Hall. Ajzen, I., & Fishbein, M. (2000). Attitudes and the attitude-behavior relation: Reasoned and automatic processes. European Review of Social Psychology, 11, 1–33. Ajzen, I., & Madden, T.J. (1986). Prediction of goal-directed behavior: Attitudes, intention, and perceived behavioural control. Journal of Experimental Social Psychology, 22,453–474. Armitage, C.J., & Conner, M. (1999). The theory of planned behaviour: Assessment of predictive validity and ‘perceived control’. British Journal of Social Psychology, 38,35–54.
SUTTON 325
Armitage, C.J., & Conner, M. (2001). Efficacy of the theory of planned behaviour: A metaanalytic review. British Journal of Social Psychology, 40,471–499. Baron, R.M., & Kenny, D.A. (1986). The moderator-mediator distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal ofPersonality and Social Psychology, 51,1173–1182. Berry, W.D. (1993). Understanding regression assumptions.Newbury Park, CA: Sage Publications. Campbell, D.T., & Kenny, D.A. (1999). A primer on regression artifacts.New York: Guilford Press. Clogg, C.C., & Haritou, A. (1997). The regression method of causal inference and a dilemma confronting this method. In V.R.McKim & S.P.Turner (Eds.), Causality in crisis?Statistical methods and the search for causal knowledge in the social sciences(pp. 83–112). Notre Dame, IN: University of Notre Dame Press. Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation analysis for the behavioralsciences.Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Evans, M.G. (1991). The problem of analyzing multiplicative composites. AmericanPsychologist, 46,6–15. Finkel, S.E. (1995). Causal analysis with panel data.Thousand Oaks, CA: Sage. Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior: An introduction totheory and research.Reading, MA: Addison-Wesley. French, D.P., & Hankins, M. (in press). The expectancy-value muddle in the theory of planned behaviour-and some proposed solutions. British Journal of Health Psychology. Goldstein, H., Rasbash, J., Plewis, I., Draper, D., Browne, W., & Yang, M., et al. (1998). Auser’s guide to MLwiN.London, UK: Institute of Education. Hedeker, D., Flay, B.R., & Petraitis, J. (1996). Estimating individual influences of behavioural intentions: An application of random-effects modelling to the theory of reasoned action. Journal of Consulting and Clinical Psychology, 64, 109–120. Heise, D.R. (1975). Causal analysis.New York: Wiley. Kahle, L.R., & Beatty, S.E. (1987). Cognitive consequences of legislating postpurchase behavior: Growing up with the bottle bill. Journal of Applied Social Psychology, 17,828–843. Kenny, D.A. (1979). Correlation and causality.New York: Wiley. Liska, A.E. (1984). A critical examination of the causal structure of the Fishbein/ Ajzen attitude-behavior model. Social Psychology Quarterly, 47,61–74. Mauro, R. (1990). Understanding L.O.V.E. (Left Out Variables Error): A method for estimating the effects of omitted variables. Psychological Bulletin, 108,314– 329. McCarty, D. (1981). Changing contraceptive usage intentions: A test of the Fishbein model of intention. Journal of Applied Social Psychology, 11,192– 211. Rogosa, D.R. (1980). A critique of cross-lagged correlation. Psychological Bulletin, 88,245– 258. Sutton, S. (1998). Predicting and explaining intentions and behavior: How well are we doing?Journal of Applied Social Psychology, 28,1317–1338.
326 EUROPEAN REVIEW OF SOCIAL PSYCHOLOGY
Sutton, S. (2002). Using social cognition models to develop health behaviour interventions: Problems and assumptions. In D.Rutter & L.Quine (Eds.) Changing health behaviour:Intervention and research with social cognition models(pp. 193–208). Buckingham, UK: Open University Press. Sutton, S., McVey, D., & Glanz, A. (1999). A comparative test of the theory of reasoned action and the theory of planned behavior in the prediction of condom use intentions in a national sample of English young people. Health Psychology, 18,72–81. Sutton, S.R. (1994). The past predicts the future: Interpreting behaviour-behaviour relationships in social psychological models of health behaviour. In D.R.Rutter & L. Quine (Eds.) Social psychology and health: European perspectives(pp. 71– 88). Aldershot, UK: Avebury. Trafimow, D., & Finlay, K.A. (1996). The importance of subjective norms for a minority of people: Between-subjects and within-subjects analysis. Personality and Social PsychologyBulletin, 22,820–828. Wittenbraker, J., Gibbs, B.L., & Kahle, L.R. (1983). Seat belt attitudes, habits, and behaviors: An adaptive amendment to the Fishbein model. Journal of Applied Social Psychology, 13,406–421.
Author index
Abelson, R.P., 115, 266 Aberson, C.L., 3 Aboud, F.E., 280 Abramowitz, S.I., 267 Abrams, D., 3, 13, 19, 20, 26, 114, 159, 286 Abramson, L.Y., 222, 223, 224, 225, 226, 234, 235, 241, 256, 257 Ackermann, R., 234 Acock, A.C., 190 Adams, J.S., 185, 192 Adelson, J., 281 Adorno, T.W., 96, 124, 264 Aiken, L.S., 304 Ajzen, L, 295, 297, 299, 300, 305, 308 Albala-Bertrand, L., 280, 281 Alker, H.A., 267 Alloy, L.B., 234, 256 Allport, G.W., 4, 6, 10, 22, 76 Altemeyer, R.A., 116, 124 Alwin, D.F., 286 Anastasio, P.A., 49, 58 Anastasio, P.S., 9 Anderson, D., 267, 270 Anderson, M., 80 Araji, S.K., 192 Archer, D., 100 Arcuri, L., 37, 38, 43, 44 Armitage, C.J., 295, 307, 308 Aron, A., 10, 11 Aron, E.N., 10, 11
Aronson, E., 115 Ashford, S., 286 Ashmore, R.D., 76 Ashton, M.C., 80, 81 Athenstaedt, U., 203 Avioli, P.S., 188 Babcock, J.C., 204 Bachman, B.A., 9, 49, 58 Banaji, M.R., 110, 112, 114, 115, 116, 117, 119, 120, 123, 124, 129, 132 Banks, M., 277, 284 Banner, M.J., 11, 12, 14 Bar-Tal, Y., 21, 22, 27 Bargh, J.A., 6, 8, 45 Barnett, R., 267, 272 Barnett, R.C., 188 Baron, J., 204 Baron, R.M., 54, 322 Baruch, G.K., 188 Bass, B.M., 153, 155, 156, 162, 166, 179 Bates, I., 277, 284 Baumeister, R.F., 9, 10, 14, 25, 159 Beatty, S.E., 308 Bebeau, M.J., 265, 280, 281 Beck, A.T., 241, 256 Belsky, J., 201 Bem, D.J., 115 Bem, S.L., 115 Ben Brika, J., 279 327
328 AUTHOR INDEX
Berger, U., 4 Berheide, C.W., 200, 206 Berk, S.F., 188 Berkowitz, L., 65, 66 Berry, W.D., 301 Berscheid, E., 187, 192 Berti, A.E., 281 Bettencourt, B.A., 49 Biel, A., 155, 156, 157 Biernat, M., 79 Billig, M.G., xi, 2, 4, 51, 287 Blair, S.L., 190, 191, 194, 199 Blanchard, D., 188 Blaney, P.H., 65 Blanton, H., 112, 120, 122, 137 Blanz, M., 3, 24 Block, J., 233, 234, 267 Blood, R.O., 187, 204 Bodenhausen, G.V., 35, 37, 50, 65 Bogart, L.M., 76, 77, 79, 81, 87, 88, 89, 9697, 99, 100, 101, 104 Bonnell, D., 204 Boodoo, G., 79 Borden, R.J., 65 Borkenau, P., 277 Bouchard, T.J.Jr, 79 Bowen, R., 65 Boyd, C, 100 Boykin, A.W., 79 Brady, H.E., 284 Branscombe, N.R., 126, 146 Brauer, M., 92, 93, 94, 104 Breakwell, G., 277, 284 Brehm, J.W., 115, 121 Brennacher-Kröll, S., 185, 191, 199, 200 Brewer, M.B., 2, 4, 6, 8, 17, 22, 23, 24, 27, 40, 43, 44, 49, 55, 58, 59, 60, 155, 156, 157, 158, 159, 160, 161, 164 Brewin, C.R., 234, 243 Brickman, P., 223 Brill, A., 280 Brines, J., 188 Brockner, J., 193 Brody, N., 79 Brody, R.A., 279, 282 Brown, J., 233, 234, 251
Brown, J.D., 4, 9 Brown, R., 2, 3, 4, 6, 23, 24, 113, 114, 119, 126, 130, 134, 264, 265 Browne, W., 305 Bruce, S., 284 Bruner, J.S., 48 Brunschko, B., 185, 191, 199 Buhl, T., 4 Bühner, M., 240, 241, 242, 243, 245, 247, 248, 250, 251, 252, 253, 254 Bundy, R.F., 51 Bundy, R.P., xi, 2, 4 Burgess, D., 112, 115, 116, 121, 122, 123, 124, 129, 134, 135, 138, 140, 141 Burks, N., 202 Buss, D.M., 202, 221, 230 Buunk, B.P., 185, 192, 193, 200, 202, 206, 210 Bynner, J., 264, 277, 279, 284, 286 Cacioppo, J.T., 45, 100, 285 Cadinu, M.R., 2, 15, 16, 22, 23, 45 Cairns, E., 38 Callahan-Levy, C.M., 120, 135 Callan, V.J., 204 Campbell, A., 263 Campbell, D.T., 43, 44, 308 Campbell, J.D., 233, 285 Candee, D., 267 Cantor, N., 63 Carey, S., 220 Carnevale, P.J.D., 204 Carter, J.D., 80, 95, 100, 102 Cartwright, D., 155, 156, 162, 178 Carvallo, M.R., 112, 119, 122, 131, 132, 133, 134, 141 Ceci, S.J., 79 Chaiken, S., 263 Chassin, L., 192 Chemers, M., 153 Chen, E.S., 144 Chilton, S., 280 Chin-A-Fat, B.E.S., 189, 194, 195, 205 Chiu, C, 222, 224, 225, 227 Chodorow, N., 202 Christensen, A., 202
AUTHOR INDEX 329
Christie, R., 275 Clark, M.S., 46, 185 Claypool, H.M., 11, 12, 14 Clemence, A., 281 Clement, R.W., 16, 17, 20, 82, 96 Clogg, C.C., 308, 320, 323 Coats, S., 11, 12, 13, 14, 27 Coder, R., 267, 270 Cohen, J., 304, 317, 318 Cohen, P., 304, 317, 318 Cohen, R.L., 286 Colby, A.S., 268, 270, 280 Cole, D., 80 Coltrane, S., 188 Colvin, C.R., 233, 234 Comby, L., 279 Conklin, L., 25 Conner, M., 295, 307, 308 Conover, P.J., 263 Converse, P.E., 263, 282, 284, 287 Conway, M., 124 Cook, S.W., 49 Cooper, D., 267, 270 Cooper, J., 115 Cooper, K., 192 Copper, C, 16 Corkran, L., 193 Corneille, O., 82, 96 Cosmides, L., 221 Coverman, S., 187, 188 Cozby, P.C., 49 Crettenden, A., 156 Crisp, R.J., 35, 37, 38, 39, 42, 43, 44, 46, 48, 51, 58, 59, 60, 62, 63, 64 Croak, M., 49 Crocker, J., 112, 120, 122, 137, 141, 146 Cronbach, L.J., 77, 82 Cropanzano, R., 185 Crosby, F., 192, 195, 208 Cross, S.E., 202 Crouter, A.C., 188 Czaja, S.J., 189, 202, 209 Dalbert, C., 144, 145, 146 Das, J.P., 45 Daubman, K.A., 63
Davidson, F.H., 280 Davis, M.H., 25 Dawes, R.M., 155, 156, 160, 163, 165 De Cremer, D., 155, 156, 157, 160, 161, 162, 164, 165, 167, 168, 170, 170, 172, 173, 174, 177 De Dreu, C.K.W., 204, 210 Deaux, K., 195 Deci, E.L., 159, 179 Degoey, P., 155, 156, 158 Del Boca, F.K., 76 Delli Karpini, M.X., 264, 283, 284, 285 DeMaris, A., 190, 191, 192, 194 Demo, D.H., 190 DeRubeis, R.J., 234 Deschamps, J.C., 25, 37, 43, 44, 82, 96, 279 Deutsch, F.M., 187, 188 Deutsch, M., 200 DeVault, M.L., 206 Devos, T., 25, 66, 279 Diehl, M., 3, 25 Diekman, A.B., 77, 80, 81, 86, 95 Diener, C, 9 Diener, E., 9 Dijksterhuis, A., 35, 37 DiMatteo, M.R., 100 Doise, W., 37, 38, 43, 44, 82, 96, 281 Doosje, B., 114, 119, 130 Dovidio, J.F., 6, 7, 9, 16, 22, 44, 45, 46, 49, 58, 62, 63, 64 Downes, A., 263 Doyle, A.B., 280 Draine, S.C., 7 Draper, D., 305 Duan, C, 39 DuBois, W.E.B., 138, 140 Duck, J., 160, 162 Duenas, G., 81 Dunning, D., 26, 27 Dunton, B.C., 44 Durkin, K., 269 Dweck, C.S., 222, 224, 225, 227 Eagly, A.H., 77, 80, 81, 86, 93, 94, 95, 123, 202, 263
330 AUTHOR INDEX
Earley, P.C., 193 Earnshaw, O., 285 Eccles, J., 77, 79, 86, 102 Edwards, L., 267 Eisenbeiss, K., 26 Ellemers, N., 160 Ellis, A., 252 Elster, J., 116, 121, 122 Emans, B.J.M., 204 Emler, N., 264, 265, 267, 268, 269, 270, 271, 272, 273, 274, 276, 277, 279, 281, 282, 284, 285, 286 Engels, F., 116 England, P., 188 Ensari, N, 49, 52, 53, 54, 56, 57, 58, 59, 60,60 Epstude, K., 11, 12, 13, 14, 27 Erbaugh, J., 241 Erber, M.W., 284 Erickson, B.H., 287 Esses, V.M., 80, 81 Etkoff, N.L., 35, 37 Evans, G., 264, 277, 286 Evans, M.G., 308 Evans, N., 44, 58 Evens, J., 267, 272 Eysenck, H.J., 264, 265 Fairey, P.J., 233 Farnen, R., 281 Fazio, R.H., 7, 8, 44, 58, 115 Feeney, J.A., 204 Fein, S., 126 Feinstein, J.A., 100 Feldman, S., 263 Festinger, L., 115, 193, 220 Feys, J., 10 Fiedler, K., 20 Fienberg, S.E., 80 Finlay, K.A., 305 Fischer, G.W., 76, 91 Fish, J.A., 279 Fishbein, M, 295, 305, 308 Fishkin, J., 267 Fiske, S., 283, 284 Fiske, S.T., 35, 37, 49, 93, 112, 121, 122, 124, 127, 140, 142
Flament, C, xi, 2, 4, 51 Flavell, J.H., 266 Flay, B.R., 305 Fleishman, J.A., 264, 277 Foddy, M., 156 Fogarty, S.J., 46 Folger, R., 193 Folk, K., 190, 199 Folkman, S., 146 Fontana, A.F., 267 Forgas, J., 20 Försterling, F., 220, 221, 224, 227, 228, 232, 233, 234, 235, 236, 237, 239, 240, 241, 242, 243, 245, 247, 248, 249, 250, 251, 252, 253, 254 Foufas, C., 197 Fraley, B., 10 Frazer, E., 279 French, D.P., 308 French, J.R.P., 127, 155, 160, 168, 178 Frenkel-Brunswik, E., 96, 124, 264 Freudenthaler, H.H., 185, 190, 191, 192, 194, 195, 196, 197, 197, 199, 200 Frey, B.S., 157 Friedman, L.N., 275 Frieze, I.H., 221, 222, 235 Funder, D., 277 Funder, D.C., 77 Furnham, A., 110, 115, 243 Gaertner, L., 2, 6, 19, 22, 24, 25, 27 Gaertner, S.L., 9, 49, 58, 62, 63, 64 Gage, N.L., 77 Gagnon, D., 120, 135 Gall, S., 240, 241, 242, 243, 245, 247, 248, 250, 251, 252 Gardham, K., 3 Gardner, W., 59 Geerken, M., 187 Gelman, R., 220 George, G., 112, 120, 122, 137 Getz, L, 280 Gibbs, B.L., 308 Gigerenzer, G., 221 Glanz, A., 304 Glass, B., 39
AUTHOR INDEX 331
Glass, D.C., 234 Glick, P., 112, 121, 122, 124, 140, 142 Gold, A.R., 275 Goldstein, H., 305 Gollwitzer, P., 280 Goodstein, L.D., 49 Goodwin, S.A., 127 Gopnik, A., 220 Gottman, J.M., 202, 204 Gove, G.R., 187 Graetz, K., 22, 27 Gramsci, A., 116 Gramzow, R.H., 145 Gramzow, R.M., 19, 25 Grant, P.R., 126 Gray-Little, B., 202 Greenberg, J., 185 Greenstein, T.N., 188, 190, 191, 194, 199, 206 Greenwald, A.G., 7, 112, 115, 132 Grieve, P.G., 20 Gross, S.R., 65 Grote, N.K., 185 Grove, J., 193 Guera, P., 62, 63, 64 Guermandi, G., 117, 124, 126, 127, 135, 137 Guinote, A., 21, 93, 94, 104 Gurtman, M.B., 6, 7, 9, 22, 45, 46, 58 Haan, N., 267 Hackel, L., 201 Hafer, C.L., 144 Haines, E.L., 112, 117, 122, 126, 128, 129, 130 Hains, S.C., 160, 162 Halal, M., 46 Hall, J.A., 80, 95, 100, 102 Hamilton, D.L., 77, 90, 96 Hammen, C, 240 Hankins, M., 308 Hardin, G., 158 Haritou, A., 308, 320, 323 Harmon-Jones, E., 66 Harrington, H.J., 49 Harris, D., 80 Harvey, R.D., 146
Haslam, S.A., 3, 13, 16, 19, 20, 25, 26, 48, 153, 160, 177, 178 Hastedt, C, 19, 22, 25 Hastie, R., 81, 91, 99 Hatfield, E., 192 Hausmann, L.R.M., 77, 98, 99 Havel, V., 110, 116, 144 Hawkins, A.J., 190, 191, 194, 199 Hay, J., 192 Hayes, A.F., 26, 27 Healy, M., 3 Heath, A., 264, 277 Heavey, C.L., 202 Hechter, F.J., 232 Hedeker, D., 305 Heesink, J.A.M., 185, 189, 194, 199, 200, 201, 202, 203, 204, 206 Heider, F., 195, 220, 221, 222, 228, 235, 266 Heimgartner, A., 203 Heine, S.J., 285 Heise, D.R., 297, 312, 318 Hemphill, J.K., 156, 178 Henry, S., 11, 14, 27 Hepburn, C, 24 Heschgl, S., 203 Hetts, J.J., 112, 120, 122, 137 Hewer, A., 267 Hewstone, M., 3, 19, 23, 35, 37, 38, 39, 40, 42, 43, 44, 46, 48, 51, 58, 59, 60, 60, 62, 63, 64, 90, 92, 96, 114, 170, 279 Higgins, E.T., 39, 47 Himmelweit, H.T., 264 Hinkle, S., 3, 4, 6 Ho, H.K., 40 Hochschild, A., 206 Hodges, S.D., 284 Hoffman, C, 117, 123 Hofstede, G., 119, 142 Hogg, M., 156 Hogg, M.A., 3, 4, 12, 13, 14, 17, 19, 20, 22, 26, 48, 114, 159, 160, 162, 172, 175, 177, 178 Hollander, E.P., 153, 155, 162, 178 Holmes, J.G., 200 Holtz, R., 43, 44, 49, 55, 58, 60, 65 Holyoak, K.J., 220
332 AUTHOR INDEX
Homans, G.C., 192, 204 Hon, A., 7 Hong, Y., 222, 224, 225, 227 Hoorens, V., 10 Hörtner, G., 197 Horwitz, M., 2 Howard, J.W., 6 Huckfeldt, R., 286 Hug, K., 221 Hulbert, L., 156 Humphreys, P., 264 Hurst, N., 117, 123 Huston, T.L., 188, 201 Inglehart, R., 263 Insko, C.A., 2, 24 Isen, A.M., 46, 62, 63, 64 Ishii-Kuntz, M., 188 Islam, M.R., 40, 60 Ito, T., 55 Jacklin, C.N., 202 Jackman, M.R., 110, 124 Jackson, J.R., 44 Jackson, J.S., 279 Jacobson, N.S., 202, 204 Jaeger, M., 264 Jamieson, L., 277, 284 Janoff-Bulman, R., 146 Jarvis, W.B.G., 100 Jennings, M.K., 282 Jepson, S.F., 157 Jermier, J.M., 177 Jetten, J., 114, 119, 130 Jimenez, M.C, 112, 115, 117, 122, 123, 124, 137, 145 John, D., 185, 187, 188, 190, 191 Johnson, C, 16 Johnson, M.P., 190, 191, 194 Jost, J.T., 110, 112, 114, 115, 116, 117, 117, 119, 120, 121, 122, 123, 124, 124, 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 137, 138, 140, 141, 142, 143, 144, 145, 146 Jourard, S.M., 49
Judd, C.M., 8, 10, 40, 60, 76, 77, 78, 79, 81, 82, 82, 89, 90, 91, 92, 93, 94, 96, 98, 99, 100, 102, 104, 264, 287 Junn, J., 268, 277, 280, 283, 284, 285 Jussim, L.J., 76, 77, 79, 86, 102 Kahle, L.R., 308 Kahneman, D., 204 Kaiser, C.R., 145 Kalbflesch, P.J., 200, 206, 207 Kalleberg, A.L., 185 Kambara, T., 65 Kanfer, R., 193 Kaplan, E., 188 Kardes, F.R., 7, 8, 58 Karp, L., 46 Katz, I.M., 285 Kay, A., 112, 115, 117, 122, 123, 124, 137, 145 Keeter, S., 264, 283, 284, 285 Keiper, S., 77, 79, 86, 102 Kelley, H.H., 158, 192, 204, 220, 222, 224, 228, 230, 235 Keltner, D., 204 Keniston, K., 267 Kenny, D.A., 54, 77, 276, 277, 308, 312, 318, 322 Kerlinger, F.N., 263 Kerr, N.L., 156, 157 Kerr, S., 177 Kessler, T., 4 Kieselbach, T., 145 Kim, S.H., 207 Kinder, D., 268, 279, 284, 287 King, G.A., 39, 47 Kishon-Rabin, L., 21 Kivetz, Y., 117, 124, 126, 127, 135, 137 Klages, U., 253 Klinetob, N.A., 202 Kluegel, J.R., 110, 115, 144, 146 Kluwer, E.S., 185, 187, 189, 194, 195, 199, 200, 201, 202, 203, 204, 205, 206, 209, 210 Kohlberg, L., 265, 266, 267, 268, 270, 271, 280, 281
AUTHOR INDEX 333
Komorita, S.S., 155, 156, 157, 163, 171 Kramer, R.M., 156, 158, 161, 164 Kraus, S., 92 Krosnick, J.A., 264, 282, 283, 284, 285, 287 Krueger, J., 16, 17, 19, 20, 22, 26, 27, 82, 96 Kruglanski, A.W., 20, 220 Kuhl, P.K., 220 Kuiper, N.A., 233 Kukla, A., 221, 222, 227, 235 Kulesa, P., 77, 95 Kunda, Z., 77, 82 Laird, J.D., 46 Lalljee, M., 264, 277, 286 Lambert, A.J., 81 Lamsfuß S., 220 Lane, R.E., 110, 119, 121, 122, 144, 146 Lang, M., 201 LaTour, S., 193 Lau, R., 283, 284 Lavallee, L.F., 285 Layne, C, 202 Layte, R., 187, 188 Lazarus, R.S., 146 Leary, M.R., 159 Lee, J.Y., 40 Lee, Y., 76, 81 Lehman, D.R., 285 Lemaine, G., 279 Lennon, M.C., 190, 191, 199 Lerner, M.J., 110, 114, 115, 144, 146, 189 Levenson, R.W., 202 Levine, C, 267 Levine, J.M., 26, 155, 156 Levinson, D.J., 96, 124, 264 Leyens, J.-P., 12 Liberman, A., 220 Liebler, A., 277 Liebrand, W.B.G., 156 Lilli, W., 43 Lind, E.A., 160, 161, 170, 177, 178, 193
Linville, P.W., 76, 91 Lippitt, R., 155 Liska, A.E., 307 Liu, T.J., 115 Locksley, A., 24 Long, K.M., 3 Longmore, M.A., 190, 191, 192, 194 Lonky, E., 280 Lord, R.G., 172 Lorenzi-Coldi, F., 93, 95 Lowrance, R., 62, 63, 64 Luce, C, 25 Luce, R.D., 158 Lui, L., 156, 158, 164 Lukács, G., 116 Luschen, K., 190, 191 Lussier, J.B., 187, 188 Lynch, L., 39 Maass, A., 6, 8, 45, 119 Maccoby, E.E., 202 Mackie, D.M., xi, 2, 3, 4, 63, 66, 96 MacKinnon, C.A., 116, 267 MacLaren, C, 252 Macrae, C.N., 35, 37 Madden, T.J., 300 Madon, S., 77, 79, 86, 102 Madson, L., 202 Maher, K.J., 172 Major, B., 110, 112, 120, 128, 135, 144, 145, 184, 185, 187, 190, 191, 192, 193, 194, 195, 208 Malle, B.F., 100, 116 Malone, B., 267, 269 Manis, M., 79 Mann, J.A., 49 Manstead, A.S.R., 3 Marco, C.A., 48 Marcus-Newhall, A., 43, 44, 49, 55, 58, 60 Markoulis, D., 267, 270 Marshall, C.M., 190, 191, 194, 199 Marshall, R., 110 Marshall-Goodell, B.S., 45 Martin, J., 114, 264, 277 Marx, K., 116 Masanz, J., 267, 270
334 AUTHOR INDEX
Mauro, R., 317, 323 McCarty, D., 322 McCauley, C.R., 76, 77, 79, 90 McClosky, H., 280 McCreary, L., 188 McFarlin, D.B., 120, 135, 195 McGarty, C, 13, 16 McGhee, D.E., 132 McGraw, K.M., 110, 114, 115 McGuire, C.V., 117, 122 McGuire, W.J., 115, 117, 122 McHale, S.M., 188 McKnight, J.D., 234 McVey, D., 304 Mederer, H.J., 191, 192, 194, 199, 201, 206 Medin, D.L., 220 Meertens, R.W., 279 Meiners, K.M., 190, 191, 194, 199 Meloen, J., 281 Meltzoff, A.N., 220 Menaghan, E.G., 185 Mendelson, M., 241 Menec, V.H., 232 Messé, L.A., 120, 135 Messick, D.M., xi, 2, 3, 4, 155, 156, 157, 158, 159, 164 Metalski, G.I., 241 Meyer, G., 82, 96 Meyer, W.-U., 222, 227, 236 Migdal, M., 35, 42, 43, 46, 58 Mikula, G., 185, 187, 189, 190, 191, 192, 193, 194, 195, 196, 197, 197, 199, 200, 203, 208 Milburn, M.A., 264, 282, 283, 284, 285 Mill, J.S., 220 Miller, C.T., 145 Miller, D.T., 110, 114, 115, 119, 144, 146 Miller, J.B., 202 Miller, N., 40, 42, 43, 44, 46, 48, 49, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 60, 62, 65, 66 Miller, N.M., 35, 42, 43, 46, 48, 49, 62 Miller, W.E., 263 Milne, A.B., 35, 37
Mock, J., 241 Moghaddam, F.M., 114 Montada, L., 144, 146 Moreland, R.L., 26, 155, 156 Morgenstern, M., 236, 237, 239, 240, 248, 249, 250 Moscovici, S., 269 Moses, J.L., 174 Moskowitz, G.B., 7, 8, 9, 14, 22, 27, 58 Mosso, C, 117, 121, 124, 126, 127, 135, 137, 141 Mount, L., 124 Mugny, G., 281 Mullen, B., 4, 13, 16, 20, 35, 42, 43, 46, 58, 119, 130 Mummendey, A., 3, 4, 6, 22, 24, 27 Murray, S.L., 200 Murrell, A.J., 49 Nanda, P.C., 45 Narvaez, D., 265, 280, 281 Nassi, A.J., 267 Nauta, A., 205 Neisser, U., 79 Nelson, G., 10, 11 Nelson, T.E., 79 Neuberg, S.L., 49 Newcomb, T.M., 115, 286 Nie, N.H., 268, 277, 280, 283, 284, 285 Niedenthal, P.M., 63 Niemi, R., 282 Nisbett, R.E., 77, 82 Noel, B., 267 Noller, P., 204 Nosek, B.A., 112, 132 Nuttin, J.M., 10 Oakes, P.J., 2, 4, 12, 13, 14, 16, 17, 22, 26, 40, 48, 77, 104, 114, 126, 159, 160, 175 O’Brien, M., 100 O’Connor, R.E., 267 Ohana, J., 281 Ohbuchi, K., 65 O’Higgins, M., 110
AUTHOR INDEX 335
Olson, J., 144 Olson, M., 155, 157, 158, 163 Onorato, R.S., 13, 27 Operario, D., 127 Orbell, J.M., 156, 163, 165 Ortiz, V., 24 Otten, S., 3, 4, 6, 7, 8, 9, 11, 12, 13, 14, 16, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 45, 46, 58 Overbeck, J., 117, 124, 126, 127, 135, 137
Poppen, P.J., 267 Porter, C.A., 114, 119, 146 Powell, M.C., 7, 8, 58 Pratto, F., 100, 110, 116 Presser, S., 283 Priester, J.R., 138, 140, 141 Procter, E., 115 Pruitt, D.G., 189, 202, 204, 207, 209
Palmer-Canton, E., 267, 268, 269, 272, 273, 274, 276, 282, 286 Palumbo, P., 77, 79, 86, 102, 103 Pantaleo, G., 22, 27 Paolini, S., 46, 59, 60, 64 Parcel, T.L., 185 Park, B., 8, 10, 38, 76, 77, 78, 79, 81, 82, 82, 89, 90, 91, 92, 93, 98, 99, 100, 102, 104 Parks, C.D., 155, 156, 157, 163, 171 Paul, B.Y., 55 Peirce, R.S., 189, 202, 209 Pelham, B.W., 112, 115, 119, 120, 121, 122, 131, 132, 133, 134, 137, 141, 143, 144, 145 Pepels, J., 39, 40 Pepitone, A., 128 Peplau, L.A., 115 Perdue, C.W., 6, 7, 9, 22, 45, 46, 58 Perry, R.P., 232 Perry-Jenkins, M., 188, 190, 199 Peterson, C, 223, 241 Petraitis, J., 305 Petri, B., 190 Pettigrew, T., 279 Petty, R.E., 45, 100, 138, 140, 141, 285 Piaget, J., 266, 280 Pittman, J.F., 188 Pittman, N.L., 255 Pittman, T.S., 255 Pizzamiglio, M.T., 124 Pleck, J., 206 Plewis, L, 305 Pomare, M., 49
Rabbie, J.M., 2, 3 Raiffa, H., 158 Rasbash, J., 305 Raven, B.H., 127, 155, 160, 168, 178 Reaven, J., 192 Reed, L., 221, 222, 235 Rehm, J., 43 Reicher, S.D., 4, 12, 13, 14, 17, 22, 26, 48, 114, 159, 160, 175, 265, 270 Reihman, J.M., 280 Reinecker, V.M., 49 Renwick, S., 267, 269 Repetti, R.L., 48 Rescorla, R.A., 220 Rest, J., 265, 267, 268, 270, 272, 280, 281 Rest, S., 221, 222, 235 Reynolds, K.J., 3, 19, 20, 25, 26, 77, 104 Richards, Z., 46, 59, 60, 64 Ridgeway, C, 124 Ritchie, J., 284 Ritchie, R.J., 174 Ritov, L, 204 Rizley, R., 233 Robinson, D.R., 77, 98, 99 Robinson, J., 191, 199, 200 Robinson, R.J., 204 Rogers, P.L., 100 Rogoff, B., 280 Rogosa, D.R., 308 Rokeach, M., 264, 265, 268 Roker, D., 286 Roman, R.J., 7 Romero, V., 3
Quarter, J., 267 Quinn, D.M., 141, 146
336 AUTHOR INDEX
Romney, D., 264, 277, 279, 284 Ronis, D.L., 115 Rosenbaum, R.M., 221, 222, 235 Rosenberg, M., 170 Rosenberg, M.J., 115 Rosenberg, S., 280 Rosenfeld, R.A., 185 Rosenfield, D., 193 Rosenfield, S., 190, 191, 199 Rosenthal, R., 100 Ross, M., 16 Rosselli, F., 96 Rothbart, M., 2, 6, 15, 16, 22, 23, 38, 45 Rubin, J.M., 43 Rubin, J.Z., 207 Rubin, M., 3, 19, 23, 43, 48, 58, 90, 279 Rubin, Z., 115 Rubini, M., 117, 124, 126, 127, 135, 137, 170 Ruble, D.N., 201 Ruderman, A.J., 35, 37 Rudolph, U., 224, 235 Russell, M.L., 46 Rust, M., 62, 63, 64 Rust, M.C., 9, 49, 58 Rutte, C.G., 156, 158, 164, 166, 179 Ryan, C.S., 76, 77, 78, 79, 81, 82, 87, 88, 89, 90, 91, 92, 94, 96, 97, 98, 99, 100, 101, 102, 104 Ryan, R.M., 159, 179 Sagar, H.A., 76 Salovey, P., 76, 91 Sampson, E.E., 267 Samuelson, C.D., 156, 158, 164, 166 Sanbonmatsu, D.M., 7, 8, 58 Sanchez, L., 190, 191 Sanford, R.N., 96, 124, 264 Sato, K., 156 Savory, F., 281 Schadron, G., 12 Schaller, M., 6, 8, 45, 100 Schiller-Brandl, R., 200 Schlozman, K.L., 284 Schmader, T., 128, 145
Schmaus, G., 110 Schmitt, M., 144, 146 Schmitt, M.T., 146 Schofield, J.W., 76 Schönbach, P., 280 Schopler, J., 6 Schot, J.C., 3 Schroeder, D.A., 156 Schuman, H., 283 Schuster, B., 224, 248, 249, 250 Schuurman, M., 200, 206 Schwartz, J.L.K., 132 Schwarz, N., 78 Schweitzer, M., 204 Scott, J., 130 Sedikides, C, 19, 22, 25, 27 Seligman, M.E.P., 222, 223, 224, 225, 226, 235, 241, 257 Semmel, A., 241 Senter, M.S., 124 Serlin, R.C., 280 Servis, L.J., 187, 188 Shah, J., 20 Shalker, T., 46 Shanks, D.R., 220 Shaver, K.G., 195 Sheldon, O., 112, 115, 121, 122, 141, 143, 144, 145 Shelton, B.A., 185, 187, 188, 190, 191 Sherif, C.W., 96 Sherif, M., 2, 96 Sherman, J.W., 45 Sidanius, J., 100, 110, 116 Siero, F.W., 200, 206 Sigelman, J., 66 Sillars, A.L., 200, 206, 207 Simon, B., 19, 22, 25, 27 Skowronski, J.J., 27 Smith, A., 25, 77, 79, 86, 102 Smith, A.E., 102 Smith, B., 267 Smith, C, 4, 119, 130 Smith, D.A., 202 Smith, E.R., 11, 12, 13, 14, 27, 66, 110, 115, 144, 146 Smith, H.J., 170, 178, 185, 189 Smith, R., 283, 284 Smollan, D., 10
AUTHOR INDEX 337
Sniderman, P.M., 279, 282 Snyder, M., 155, 156, 157 Sparks, P., 269 Spears, R., 3, 114, 119, 130, 160 Spencer, S.J., 126 Spitze, G., 191, 199, 200 Sprague, J., 286 Sprecher, S., 185, 192 St James, A., 268, 269, 272, 273, 274, 276, 282 Staats, A.W., 45 Staats, C.K., 45 Stace, K., 268, 270, 271 Stacey, B., 265 Staerkle, C, 281 Stallworth, L.M., 100, 116 Stangor, C, 39, 77, 100, 104 Steele, C.M., 115 Steffen, V., 123 Steffen, V.J., 81, 94 Stehlik-Barry, K., 268, 277, 280, 283, 284, 285 Steil, J.M., 185, 187, 202 Stephenson, G., 110 Stevens, R.J., 279 Stewart, T., 93, 95 Stiepel, M., 280 Stitt, C.L., 77, 79, 90 Stokes, D.E., 263 Stroebe, W., 157 Stroessner, S.J., 45 Sullivan, B., 112, 115, 121, 122, 141, 143, 144, 145 Sullivan, E.V., 267 Suls, J., 48 Susskind, J., 96 Sutton, S., 295, 297, 304 Sutton, S.R., 321 Swann, W.B., Jr, 77 Sweeney, P.D., 195 Swim, J.K., 80, 86, 95 Szegda, M., 46
Tanzer, N.K., 190 Tassinary, L.G., 45 Taylor, D.M., 114 Taylor, S., 233, 234, 251 Taylor, S.E., 4, 9, 35, 37, 51, 60 Taylor, S.P., 65 Teasdale, J.D., 46, 222, 223, 224, 225, 226, 235, 257 Tenbrunsel, A., 158 Tetlock, P., 270, 279, 282 Tetlock, P.E., 265, 268, 287 Thibaut, J., 158, 193 Thibaut, J.W., 192, 193 Thoits, P.A., 195, 200, 202, 203, 204 Thoma, S., 267, 268, 280 Thoma, S.J., 265, 280, 281 Thompson, E.P., 20, 112, 116, 117, 121, 122, 135, 136, 141, 142, 146 Thompson, L., 184, 185, 187, 190, 191, 192, 193, 194, 195, 208 Tiraboschi, M., 119 Törnblom, K.Y., 189 Torney-Purta, J., 281 Trafimow, D., 305 Trapnell, P.D., 285 Traupmann, J., 192 Trolier, T.K., 77, 90, 96 Trope, Y., 220, 223 Tropp, L.R., 13 Tudor, M., 10, 11 Turner, J., 3, 19, 20, 25, 26 Turner, J.C., xi, 3, 4, 12, 13, 14, 15, 16, 17, 22, 24, 25, 26, 27, 48, 77, 90, 113, 114, 126, 134, 159, 160, 164, 175, 178, 279 Tversky, A., 204 Tygart, C.E., 264 Tyler, R.B., 6, 7, 9, 22, 44, 45, 46, 58 Tyler, T., 155, 156, 157, 170, 170, 178 Tyler, T.R., 110, 114, 115, 144, 155, 156, 158, 160, 161, 170, 177, 178, 185, 189, 193
Tabak, N., 21 Tajfel, H., xi, 2, 4, 15, 24, 26, 43, 51, 82, 90, 96, 113, 114, 159, 175 Tannenbaum, P.H., 115
Uleman, J.S., 7 United Nations, 185 Urada, D., 46, 51, 52, 54, 59, 60, 60, 62
338 AUTHOR INDEX
Urban, L.M., 35, 42, 43, 46, 48, 49, 62, 65, 66 Utne, M., 192 Valdimarsdottir, H., 234 Van de Kragt, A.J.C., 156, 163, 165 Van de Vliert, E., 185, 189, 194, 199, 200, 201, 202, 203, 204, 206 Van den Bos, K., 193 Van Dijk, E., 156 Van Eimeren, B., 43 van Knippenberg, A., 35, 37, 39, 40 van Knippenberg, D., 162, 170 Van Lange, P.A.M., 177 van Twuvyer, M., 35, 37, 39, 40 Van Vugt, M., 155, 156, 157, 159, 160, 161, 164, 165, 167, 168, 170, 170, 172, 173, 174, 177 Van Yperen, N.W., 185, 192, 193, 200, 202, 206 Vanbeselaere, N., 43 Vanman, E.J., 42, 48, 55, 65, 66 VanNoy, M., 102 Vender, J.P., 76, 81, 89, 99, 101 Verba, S., 284 Vermunt, R., 193 Vescio, T.K., 48 Visser, L., 3 von Baeyer, C, 241 Vorauer, J.D., 16 Wagener, J.J., 46 Wagner, U., 279, 280 Walker, L., 193 Walling, D., 11, 12, 13, 14, 27 Walster, E., 187, 192 Walster, G.W., 187, 192 Waltz, J., 204 Wann, D.L., 126 Ward, C.H., 241 Ward, D., 280 Ward, R.A., 191, 199 Webster, D.M., 20 Weinberg, L.E., 232 Weiner, B., 220, 221, 222, 223, 224, 226, 235, 257 Weiss, R., 110
Wentura, 7, 8, 14, 17, 18, 19, 22, 25, 27, 45, 46, 58 West, C, 206 West, C.K., 279 West, S.G., 304 Wetherell, M.S., 4, 12, 13, 14, 17, 22, 26, 48, 114, 159, 160, 175 White, C.V., 284 White, K.K., 155 Wicklund, R.A., 115, 121 Wiesenfeld, B.M., 193 Wilder, D.A., 49 Wilke, H., 156, 158, 164 Wilke, H.A.M., 155, 156, 158, 164, 166, 179, 193 Wilkening, F., 220 Wilkes, A.L., 43, 82, 96 Williams, C.J., 45 Willis, H., 23, 58, 90 Wilson, G.D., 264 Wilson, T.D., 284 Wissink, I., 189, 194, 195, 205 Wit, A.P., 156 Wittenbraker, J., 308 Wittenbrink, B., 8, 79, 93, 99, 102, 104 Wolfe, D.M., 187, 204 Wolff, E.N., 110 Wolsko, C, 79, 93, 99, 102, 104 Wood, J.V., 193 Worth, L.T., 63 Wright, S.C., 13, 114 Wyer, N.A., 45 Yamagishi, T., 155, 156, 158, 159 Yang, M., 305 Yohannes, H., 100 Youmans, J.E., 267 Yukl, G.A., 153, 155, 156, 162, 166 Yzerbyt, V.-Y., 12 Zaller, J., 284 Zander, A., 155, 156, 162, 178 Zeiger, J.S., 16 Zeiss, A., 192 Zemke, P.E., 156, 158, 164 Zimmerman, D.H., 206
Subject index
Absolute discrepancies 74, 85, 86–9, 97 Achievement behaviour 218, 220, 222– xi, 225, 230–30, 236–8, 257 Additive pattern 40–6, 49, 51, 52, 59, 60–3 Advantaged groups, systemjustification theory 119, 121, 124, 144–6, 147 Affect influence Attribution models 218, 221–25, 230, 232–3, 240–55 Crossed categorisation 34, 44–8, 49– 6, 58, 59–8 Family work division 185–7 Ingroup representation 4–10, 20, 21, 25–6, 27 System justification theory 127, 128, 132, 138–3 Anti-authority beliefs 272–72, 281 Attentiveness, political orientation 283, 284, 285, 286 Attitude—behaviour theories, testing 295–324 Attitude—intention relationship 295, 297, 300–308, 317, 321 Attribution/attributional models, integration218–55 Attribution of blame model 195–6, 197, 208 Behaviour
Achievement 218, 220, 222–1, 225, 230–30, 236–8, 257 Attitude—behaviour theories 295– 324 Control 297, 299–7, 311 Behaviourism, morality and politics 265, 268 Bias Change 96–8 Individuating information 55–8 Novel ingroups 6–9, 20, 26 Self 10, 25, 78 Blame, attribution 195–6, 197, 208 Categorisation Affective valence 46–8, 60 Category importance 48–9, 49, 50– 6, 56, 60–3 Contextual inclusiveness 49–1, 58– 59, 60, 64 Crossed 34–67 Differentiation processes 44, 46 Dual route model 46–67 Individuating information 49–49, 50, 55–9, 60 Ingroup favouritism 5–27 Inhibition 35–7, 39–40 Multiple systems 34–42, 52–6, 56 Negative emotions 65–7 Self 12, 13–21
339
340 SUBJECT INDEX
Causal analysis, attitude—behaviour theories 295–324 Childcare 200, 201, 208–9 Clogg and Haritou, omitted variables method 320 Coefficients, standardised/ unstandardised 308–6 Cognitive constructivism 261, 263, 264, 265–5, 280–8, 287 Cognitive dissonance theory 115, 117, 121–3, 137 Cognitive factors Attribution models 218, 221–20, 224–5, 229, 232, 251–55 Cognitive load 46, 48 Crossed categorisation 34, 43–5, 49– 1, 55–7, 58–60, 63–8 Family work division 191, 199, 208 Minimal ingroups xi, 5, 6, 10, 15, 20–1 Morality and politics 261–65, 270, 277–9, 283, 284, 286–6 Stereotypes 76–7, 90–1, 100–1 System justification theory 114, 115, 116 Cognitive processes 74 Conflict management, family work division 185–7, 199–211 Connectionist networks, ingroup favouritism 5, 11–15 Conscience 264–2, 267 Consistency, political orientation 283– xi,286 Contextual inclusiveness 49–1, 58–59, 60, 64 Control, behavioural 297, 299–7, 311 Coping strategies, system justification theory 145–8 Covariation principle 218, 220, 227– 55, 277 Crossed categorisation 34–67 Cultural differences, system justification theory 124–7, 127, 133, 134, 141–3, 144 Cultural stereotyping 74–6, 79, 80, 90– 3, 95, 100, 101, 102
Decategorisation 49–49, 50, 56, 60, 64 Decision-making Family work division 193, 210 Leadership style 170 Demand—response model 188 Depersonalisation 14, 26 Depressed entitlement effect 109, 112, 120, 135–40, 143 Depression Attribution models 218, 222–2, 225– 5, 232–3, 240–55 Family work division 199 System justification theory 141, 142, 143, 147 Differential needs model 160–2, 163– 6, 168, 170, 172, 173, 175–81 Disadvantaged groups, system justification theory 112, 114, 117– 20, 121, 124–33, 140–2, 144, 147 Dissonance theory 115, 117, 121–3, 137 Distributive justice framework 184, 187, 189–99, 208 Dominance pattern 41, 62 Dual-route model, crossed categorisation 34, 46–67 Economic status 144–6, 226 Education, political orientation 277, 277–7, 281, 284–4, 288 Egocentric social projection 16–17, 19, 20, 27 Ego justification 112, 117, 120–2 Empathy 101 Entitlement Depressed entitlement effect 109, 112, 120, 135–40, 143 Violation of 195–7 Equivalence pattern 42, 52, 62, 63, 64 Evaluative response, family work division 185–99 Fairness, family work division 184, 185– 99, 206, 208, 210 Familiarity 100 Family work, gender-related inequalities 184–211
SUBJECT INDEX 341
Favouritism see ingroups; outgroups Freeriding 153–6, 157–9, 160–4, 168, 173, 175, 178 Gender Depressed entitlement effect 112, 120, 135–40 Family work inequalities 184–211 Ingroup ambivalence 140, 141, 142– 4 Role ideology 188, 191, 200, 206–7 Stereotyping 74, 77, 84, 95, 100 Groups see also ingroups; outgroups Characteristics 74, 77–89 Commitment 172–7, 178 Justification 117, 120–2, 138 Leader endorsement 153–81 Stereotypes 76, 78–9, 90–1, 93, 95– 6, 99–100, 102–5 Voluntary cooperation 155–62, 164, 167–77, 179 Group-value model 177, 178 Hedonistic mechanisms, attributional models 218, 222–1, 224–3, 227 Helplessness, learned 222, 223–2, 225, 234, 235 Hierarchical acceptance (ordering) pattern 41, 42, 54, 60, 63 Hierarchical rejection (derogation) pattern 41, 42, 53, 54, 60, 62, 63 Housework 200–3, 208–9, 210 Identity see personal identity; social identity Ideology Dissonance 109, 113, 115, 117, 121– 3, 141–4, 147 Gender role 188, 191, 200, 206–7 Palliative function 109, 110–14, 124, 143–8 Political 263–5, 269, 276, 277, 282, 287 Stereotype accuracy 74, 91–5 Individual differences
Family work division 191, 199 Morality and politics 282–80, 287– 6 Stereotype accuracy 74, 99–101 System justification theory 115, 116 Individuating information 49–49, 50, 55–9, 60, 102 Inequality Family work division 184–211 Internalisation 119–1, 129–40 Rationalising 109–13, 117–20, 123, 124, 141–6, 147 Information processing Ingroup representation 5, 9, 20–1 Multiple category use 38–9 Negative stimuli 60 Stereotypes 76–7 Ingroup favouritism Crossed categorisation 34–67 Minimal ingroups xi– 27, 62 Morality and politics 279 Role of self 5–27 Social dominance theory 117, 140, 141 System justification theory 119–1, 123–5, 124–7, 132, 133–6, 147 Ingroups Ambivalence 138–1, 141, 142–4, 147 Crossed categorisation models 40–2 Novel xi, 6–9, 10, 15, 20, 21, 26 Representation 4–10, 11–12, 15–22 Stereotypes 78–9, 90–1, 93, 95, 96, 99–100 Instrumental perspective, leader endorsement 153, 156, 158–61, 162, 165, 167–70, 170, 178, 179 Instrumental variables 318–17, 323 Intention Attitude relationship 295, 297, 300– 308, 317, 321 Multiple causes argument 308–19 Interest, political orientation 283, 284, 285 Intergroup processes Crossed categorisation 34–6, 40–2, 49
342 SUBJECT INDEX
Leader endorsement 160, 161–3, 164, 172, 178 Minimal ingroups 2–6, 21, 23, 24– 5, 26 Social identity 114, 160, 161–3, 164 Stereotype accuracy 74, 90–5, 100– 1, 102 System justification theory 112, 113, 114 Justice Family work inequalities 184, 185– 99, 207–11 Just world beliefs 114–16, 117, 141 Leadership style 170, 171–4, 175, 178 Knowledge, political orientation 283, 284, 285, 286 Labour division, domestic 184–211 Leader Commitment 172–7 Differential needs model 160–2, 163–6, 168, 170, 172–5, 175–81 Emergence 156, 164–8 Endorsement 153–81 Influence 156, 167–77 Instrumental perspective 158–60, 160 Leadership style 170–5, 177–81 Personal attributes 172–7, 178 Relational perspective 159–1 Types 165–8, 177 Life course perspective 188 Machiavellian Cynicism scale 275–4 Machiavellian Tactics scale 275–4 Manipulation check 321–20 Marxist-feminist theories 116, 117 Mauro, omitted variables method 317– 16, 323 Memory Ingroup representation 5, 11–15 Multiple category use 38
System justification theory 127, 128– 30, 130, 147 Meritocracy 144–6 Meta-analyses Attitude—behaviour theories 295 Stereotype accuracy 80–1 Minimal group paradigm (MGP), ingroup favouritism 2, 4, 5, 12, 14– 17, 20–3, 26 Mood see also affect influence; depression Crossed categorisation 46–8, 49, 50– 6, 58, 59–8 Ingroup bias effect 20 Morality, and political orientation 261– 86 Moral psychology 264–2, 287 Moral reasoning 266–75, 280, 281–9, 287 Motivation, political orientation 283, 284 Motivation model, attributions 218– 54 Multiple regression analysis 295–4, 300–321, 323 Negative emotions 65–7 Neuroticism 141, 142, 143, 147 New Left Philosophy 275–4 Omitted variables 317–19, 323 Opinionation, political orientation 282– 80, 284, 285–3 Outgroups Favouritism 109, 112, 114, 116, 119–1, 123–36, 143, 147 Representation 4–6, 10, 22–3, 40, 49 Stereotypes 78–9, 90–xi, 93, 95, 96, 100 Parents, political orientation 282, 286–4 Path diagrams, attitude—behaviour theories 295, 297–5, 310, 312, 319 Personal identity Leader endorsement 153, 159–6, 168, 168, 170, 172–81
SUBJECT INDEX 343
Morality and politics 275–4, 282 Personalisation 49–49, 55 Pessimism 218, 247, 248–8, 256–5 Planned behaviour theory (PBT) 295– 324 Politics Leadership 155 Orientation and morality 261–86 Participation 284, 285, 286 Positive attitudes Crossed categorisation 46–8, 49, 51, 60–5 Ingroup representation 4–10, 20, 25– 6, 27 Positive ingroup default xi, 5, 6–9, 22, 45 Positive ingroup distinctiveness Minimal ingroup favouritism xi, 5– 6, 8, 10, 16–17, 19–24 Social identity theory 159, 162 System justification theory 112 Power Family relationships 187, 193, 206, 210 Leaders 168–72, 177–9 Stereotype accuracy 74, 91–5 System justification theory 117–20, 126–30 Principled reasoning 266–4, 268, 269– 72, 277, 281–9 Prosocial norms 156 Psychoanalysis, morality and politics 264–2, 268 Public good dilemmas 157, 158, 163–77 Public records 79–80 Racism 264, 279, 280 Radical tactics scale 275–4 Randomised experiments 295, 305, 317, 321–20, 323 Real groups Ingroup favouritism xi, 26, 62 Positive ingroup default 45 System justification theory 123–6 Reasoned action theory (RAT) 295– 324 Reasoning, moral 266–75
Recategorisation 50, 63, 64 Relational comparisons 192, 194 Relational perspective, leader endorsement 153, 156, 159–1, 162, 165, 167–72, 175–81 Relationships Attribution models 218, 241, 248– 8, 256 Family work inequalities 184–211 Morality and politics 275–4, 286–5 Relative deprivation theory 195–6, 208 Relative resource approach 187 Resource dilemmas 157, 158–60 Self Anchoring 5, 15–21, 22, 24, 25–6, 45 Categorisation 12, 13–21 Cognitive dissonance theory 115 Family work division 193, 202 Interest 155, 157, 158, 161, 163, 178–1 Minimal ingroups xi, 5–6, 9, 10–21, 22, 27 Personal identity 153, 159–6, 168, 168, 170, 172–81 Rating 77–81, 97 Scapegoating 124 Self-concept Attributional model 222–1, 227, 232–2, 243, 253, 256 Morality and politics 285–3 Self-esteem Ingroup favouritism 3, 9, 43, 117 Leader endorsement 170–3, 178 Morality and politics 279 System justification theory 112, 117, 120–2, 132, 141–4, 147 Self-fulfilling prophecies 103 Sexism 142–4 Signed discrepancies 74, 82–4, 86–9, 97 Social categorisation Differentiation processes 44, 46 Ingroup favouritism 5–27 Multiple systems 34–42, 52–6, 56
344 SUBJECT INDEX
Social cognition, ingroup processes 6, 21 Social comparisons Attribution theory 220 Family work division 192–3, 194, 196–9, 207–8 Social desirability 78–9 Social dilemmas Leader endorsement 153–81 Structural solutions 156, 158, 175 Social discrimination Crossed categorisation 34, 37, 62 Minimal ingroups 2–4, 5, 6, 10 Stereotypes 74 Social dominance theory 116–18, 117, 140, 141 Social exclusion pattern 41, 42, 60, 65 Social identity Leader endorsement 153, 159–3, 164–7, 168–81 Minimal ingroups xi–4, 12, 15, 23– 5, 26 Morality and politics 261, 270, 275– 4, 282–85 System justification theory 113–15, 117 Social ideology, 74, 91–5 Social inclusion pattern 41–2, 51, 58– 59, 60, 60, 62, 64 Social inequality, rationalising 109– 13, 117–20, 123, 124, 141–5, 147 Social order 267, 268, 269–75, 272– 72, 282 Social perception Attribution models 224–3 Decategorisation effect 49, 60 Multiple categories 35–40 Social/personal identity 160–2 Stereotypes 81–9, 90–1, 96, 100, 102 Socio-genetic approach, morality 261, 268–75, 281, 288 Spontaneous trait inference paradigm 7– 9 Status Morality and politics 279 Stereotype accuracy 74, 91–5, 100
System justification theory 117–21, 121, 123–6, 130–6, 138–2, 147 Status quo Family work division 184, 192, 202– 6, 207, 209–10 Leader endorsement 164 Rationalisation 109, 115, 117–20, 122–30, 144 Stereotypes Absolute discrepancies 74, 85, 86–9, 97 Accuracy 74–104 Change and development 96–9, 104 Consequences of using 101–4 Individual differences 74, 99–101 Intergroup differences 90–5 Leadership 172, 178 Negative 49, 60, 76 Self-fulfilling prophecies 103 Signed discrepancies 74, 82–4, 86–9, 97 Social identity theory 113–15 Social perception 35–7, 39, 45n. System justification theory 110–13, 116, 117, 123–30, 132, 147 Within-subject correlations 74, 85– 9, 97, 100 Stress 145–7 System justification theory Background/development 109–18 Derived hypotheses 117–23 Empirical studies 122–44 Palliative function 143–8 Stress and coping perspective 145–7 Theory of planned behaviour (TPB) 295–324 Theory of reasoned action (TRA) 295– 324 Timeline analysis 295, 300–308, 323 Tolerance, political 277–9 Traditional moralism scale 275–4 Trust 156, 159, 161, 165 Variables Instrumental 318–17, 323 Omitted 317–19, 323
SUBJECT INDEX 345
Prediction 320–19, 323 Within-subject correlations 74, 85–9, 97, 100 Work, domestic labour division 184– 211
