Social Cognition: Selected Works of Susan T. Fiske 9781138734333, 9781315187280

Citation preview

Social Cognition

In the World Library of Psychologists series, international experts present careerlong collections of what they judge to be their finest pieces—extracts from books, key articles, salient research findings, and their major practical theoretical contributions. Susan T. Fiske has an international reputation as an eminent scholar and pioneer in the field of social cognition. Throughout her distinguished career, she has investigated how people make sense of other people, using shortcuts that reveal prejudices and stereotypes. Her research in particular addresses how these biases are encouraged or discouraged by social relationships, such as cooperation, competition, and power. In 2013, she was elected to the National Academy of Sciences, and, in 2011, to the British Academy. She has also won several scientific honours, including the Guggenheim Fellowship, the APA Distinguished Scientific Contributions Award, the APS William James Fellow Award, as well as the European Federation of Psychologists’ Associations Wundt–James Award and honorary degrees in Belgium, the Netherlands, Spain, and Switzerland. This collection of selected publications illustrates the foundations of modern social cognition research and its development in the late twentieth and early twenty-first century. In a specially written introductory chapter, Fiske traces the key advances in social cognition throughout her career, and so this book will be invaluable reading for students and researchers in social cognition, person perception, and intergroup bias. Susan T. Fiske is Eugene Higgins Professor, Psychology and Public Affairs, at Princeton University.

World Library of Psychologists

The World Library of Psychologists series celebrates the important contributions to psychology made by leading experts in their individual fields of study. Each scholar has compiled a career-long collection of what they consider to be their finest pieces: extracts from books, journals, articles, major theoretical and practical contributions, and salient research findings. For the first time ever, the work of each contributor is presented in a single volume so readers can follow the themes and progress of their work and identify the contributions made to, and the development of, the fields themselves. Each book in the series features a specially written introduction by the contributor giving an overview of their career, contextualizing their selection within the development of the field, and showing how their thinking developed over time. Discovering the Social Mind Selected Works of Christopher D. Frith By Christopher D. Frith Towards a Deeper Understanding of Consciousness Selected Works of Max Velmans By Max Velmans Thinking Developmentally from Constructivism to Neuroconstructivism Selected Works of Annette Karmiloff-Smith By Annette Karmiloff-Smith Acquired Language Disorders in Adulthood and Childhood Selected Works of Elaine Funnell Edited by Nicola Pitchford, Andrew W. Ellis Exploring Working Memory Selected Works of Alan Baddeley By Alan Baddeley Social Cognition Selected Works of Susan T. Fiske By Susan T. Fiske

Social Cognition Selected Works of Susan T. Fiske

Susan T. Fiske

First published 2018 by Routledge 2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN and by Routledge 711 Third Avenue, New York, NY 10017 Routledge is an imprint of the Taylor & Francis Group, an informa business © 2018 Susan T. Fiske The right of Susan T. Fiske to be identified as author of this work has been asserted by her in accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988. 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. Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library Library of Congress Cataloging-in-Publication Data A catalog record for this title has been requested ISBN: 978-1-138-73433-3 ISBN: 978-1-315-18728-0 Typeset in Bembo by Florence Production Ltd, Stoodleigh, Devon, UK

Contents

Acknowledgements 1

Not your grandparents’ social cognition: A family letter about progress through crisis

vii

1

SUSAN T. FISKE

PART I

Cognitive misers: The origins of social cognition 2

Attention and weight in person perception: The impact of negative and extreme behavior (1980)

13 15

SUSAN T. FISKE

3

The continuum model: Ten years later (1999)

41

SUSAN T. FISKE, MONICA LIN, AND STEVEN L. NEUBERG

4

Social science research on trial: Use of sex stereotyping research in Price Waterhouse v. Hopkins (1991)

76

SUSAN T. FISKE, DONALD N. BERSOFF, EUGENE BORGIDA, KAY DEAUX, AND MADELINE E. HEILMAN

PART II

Second wave: Motivated tacticians’ thinking is for doing 5

Controlling other people: The impact of power on stereotyping (1993)

99 101

SUSAN T. FISKE

6

The Ambivalent Sexism Inventory: Differentiating hostile and benevolent sexism (1996) PETER GLICK AND SUSAN T. FISKE

116

vi Contents PART III

Twenty-first-century activated actors: Social brain and social mind 7

A model of (often mixed) stereotype content: Competence and warmth respectively follow from perceived status and competition (2002)

161

163

SUSAN T. FISKE, AMY J. C. CUDDY, PETER GLICK, AND JUN XU

8

Dehumanizing the lowest of the low: Neuroimaging responses to extreme out-groups (2006)

215

LASANA T. HARRIS AND SUSAN T. FISKE

PART IV

Inequality enablers: Social cognition and social relevance 9

A prescriptive intergenerational-tension ageism scale: Succession, identity, and consumption (SIC) (2013)

227 229

MICHAEL S. NORTH AND SUSAN T. FISKE

10 Nations’ income inequality predicts ambivalence in stereotype content: How societies mind the gap (2013)

246

FEDERICA DURANTE, SUSAN T. FISKE, NICOLAS KERVYN, AMY J. C. CUDDY, ADEBOWALE (DEBO) AKANDE, BOLANLE E. ADETOUN, MODUPE F. ADEWUYI, MAGDELINE M. TSERERE, ANANTHI AL RAMIAH, KHAIRUL ANWAR MASTOR, FIONA KATE BARLOW, GREGORY BONN, ROMIN W. TAFARODI, JANINE BOSAK, ED CAIRNS, CLAIRE DOHERTY, DORA CAPOZZA, ANJANA CHANDRAN, XENIA CHRYSSOCHOOU, TILEMACHOS IATRIDIS, JUAN MANUEL CONTRERAS, RUI COSTA-LOPES, ROBERTO GONZALEZ, JANET I. LEWIS, GERALD TUSHABE, JACQUES-PHILIPPE LEYENS, RENEE MAYORGA, NADIM N. ROUHANA, VANESSA SMITH CASTRO, ROLANDO PEREZ, ROSA RODRIGUEZ-BAILON, MIGUEL MOYA, ELENA MORALES MARENTE, MARISOL PALACIOS GALVEZ, CHRIS G. SIBLEY, FRANK ASBROCK, AND CHIARA C. STORARI

Index

269

Acknowledgements

I would like to thank Guilford Publications for permission to include the following: Fiske, S. T., Lin, M. H., & Neuberg, S. L. (1999). The continuum model: Ten years later. In S. Chaiken & Y. Trope (Eds.) Dual process theories in social psychology (pp. 231–54). New York: Guilford. I would like to thank John Wiley & Sons for permission to include the following: Durante, F., Fiske, S. T., Kervyn, N., Cuddy, A. J. C., Akande, A., Adetoun, B. E., Adewuyi, M. F., Tserere, M. M., Al Ramiah, A., Mastor, K. A., Barlow, F. K., Bonn, G., Tafarodi, R. W., Bosak, J., Cairns, E., Doherty, C., Capozza, D., Chandran, A., Chryssochoou, X., Iatridis, T., Contreras, J. M., Costa-Lopes, R., González, R., Lewis, J. I., Tushabe, G., Leyens, J.-Ph., Mayorga, R., Rouhana, N. N., Smith Castro, V., Perez, R., Rodríguez-Bailón, R., Moya, M., Morales Marente, E., Palacios Gálvez, M., Sibley, C. G., Asbrock, F., & Storari, C. C. (2013). Nations’ income inequality predicts ambivalence in stereotype content: How societies mind the gap. British Journal of Social Psychology, 52, 726–746. I would like to thank the American Psychological Association for permission to include the following: Fiske, S. T., Cuddy, A. J. C., Glick, P., & Xu, J. (2002). A model of (often mixed) stereotype content: Competence and warmth respectively follow from perceived status and competition. Journal of Personality and Social Psychology, 82, 878–902. Glick, P., & Fiske, S. T. (1996). The Ambivalent Sexism Inventory: Differentiating hostile and benevolent sexism. Journal of Personality and Social Psychology, 70, 491–512. Fiske, S. T. (1993). Controlling other people: The impact of power on stereotyping. American Psychologist, 48, 621–628.

viii Acknowledgements Fiske, S. T., Bersoff, D. N., Borgida, E., Deaux, K., & Heilman, M. E. (1991). Social science research on trial: Use of sex stereotyping research in Price Waterhouse v. Hopkins. American Psychologist, 46, 1049–1060. Fiske, S. T. (1980). Attention and weight in person perception: The impact of negative and extreme behavior. Journal of Personality and Social Psychology, 38, 889–906. North, M. S., & Fiske, S. T. (2013). A prescriptive intergenerational-tension ageism scale: Succession, identity, and consumption (SIC). Psychological Assessment, 25(3), 706–713. I would like to thank Sage Publishing for permission to include the following: Harris, L. T., & Fiske, S. T. (2006). Dehumanizing the lowest of the low: Neuroimaging responses to extreme out-groups. Psychological Science, 17, 847–853.

1

Not your grandparents’ social cognition A family letter about progress through crisis Susan T. Fiske

Widely acknowledged as the founders of psychological science, Wilhelm Wundt and William James might also qualify as ancestors to social cognition researchers. Among other themes, Wundt (1897) viewed humans as motivated, thinking subjects, and James (1913) viewed humans as pragmatic thinkers for action. In those broad senses, they are our intellectual forebears. Social cognition research has deep roots. William James summered in the cool mountains of Chocorua, New Hampshire, in a rambling brown house. So did my great-grandmother, Mary Hutchinson Page, except her house was rambling and white. Family legend and the National Historic Register claim that she and her friends ‘frequently walked up [the hill] to enjoy the sunsets, where they had lively discussions about suffragettes, religion, prohibition, and living simply’ (Bowditch, 2005). One of those friends was William James, with whom she discussed her spiritual experiences, according to family legend. My great-grandmother was also a serious suffragist, as was her daughter. This is all to say that my family connection to psychological research (and to social issues) goes back some generations. Family tradition inspired this essay’s framework, a letter to my social psychologist daughter, Lydia Fiske Emery, about the psychological concerns of her forebears and of her own generation. The essay illustrates an intellectual history of social cognition research, using my selected work, but with visits from Lydia’s forebears.

Crisis in replicability, relevance, theory, methods All sciences undergo both gradual and sudden change as a function of ongoing discovery and self-correction. Sudden changes are harder to absorb. Crises in the field challenge a fresh PhD to evaluate the state of the field and her potential role in it. Just as now, in 1978, we had a crisis in social psychology, and the issues overlapped the current ones (Fiske, 2017), including replicability, relevance, theory, and methods. These issues persist from the early crises of scientific psychology (GinerSorolla, 2012). In his own time, Lydia’s grandfather, my father Donald Fiske,

2

Susan T. Fiske

faced related issues of replicability, relevance, theory, and methods, during his career starting in the 1940s. His quantitatively rigorous graduate research failed to replicate a senior professor’s observations about body shape and personality. Moreover, extant personality measurement methods needed work, in order to be relevant to World War II’s need to select personnel scientifically. Theory seemed unfalsifiable, as in the Freud my father later handed me when I wanted to learn about his field (a curious choice, as it did not represent his work at all). Finally, he grew concerned, as scientific findings did not replicate across distinct methods. My father addressed all these issues as a methodologist who pioneered the use of multiple measures to show convergent and divergent validity. The multitrait, multi-method matrix (Campbell & Fiske, 1959) was the most-cited article in Psychological Bulletin’s first 100 years. Advancing scientific methods was his response to crisis in early twentieth-century American psychology. He also first discovered the precursors to the Big Five personality trait factors (D. W. Fiske, 1949), advancing substance as well as methods. During the 1970s crisis, we faced these recurring issues of replication, relevance, theory, and methods. When I was a fresh PhD, we heard rumors that some flashy results did not replicate; only the original lab could pull them off. Social psychology then was dominated by laboratory experiments on cognitive dissonance and consistency theories more generally. That work seemed to us young firebrands to be irrelevant to pressing social problems: war, civil rights, environment. Besides being irreproducible and irrelevant, the field seemed stuck on grand theory, not practical, down-to-earth, falsifiable midrange theory. Finally, dubious method artifacts (e.g. experimenter expectancy) called into question all our findings. Or so it seemed to skeptics of the time. Several responses advanced the field beyond crisis. Arguably, social cognition research was one such response (for others, see Taylor, 1998). Social cognitive approaches answered the crisis by offering precision and reliability, generating fine-grained falsifiable theories, borrowing rigorous methods from more micro areas such as cognitive science, and addressing relevant issues such as stereotyping.

Social cognition models as crisis response Social cognition approaches developed in stages. The main part of the essay illustrates each phase. Immodestly, but because of the venue, all these illustrations come from my lab over the years. Countless others’ work would serve as well or better (for review, see Fiske & Taylor, 1984, 2017). But, these are the examples I know best. The articles exemplify various ways the field has characterized social perceivers: cognitive misers, motivated tacticians, activated actors, and inequality enablers. The end of the essay samples some of our current work to illustrate our ongoing responses to Crisis 3.0.

Not your grandparents’ social cognition

3

Cognitive misers: The origins of social cognition research Mid-century social psychology had generated many motivational meta-theories (e.g. drive for consistency) and some rational-actor models (e.g. reasoned attitudes driven by self-interest motives). Social cognition approaches dispensed with motivation, pushing purely cognitive explanations to their limits. Along the way, those cognitive mechanisms dethroned that rational actor who makes reasonable decisions, instead revealing a litany of fallibility, due to the human thinker’s imperfections. This toppling of the old models was front and center, as the field first modeled the person perceiver as a cognitive miser—saving scarce online processing capacity by taking shortcuts, such as stereotypes and decisionmaking heuristics (Fiske & Taylor, 2017). To illustrate, one cognitive shortcut assumes a moderately positive default for other people, all else being equal. A new acquaintance who is typically nice and sufficiently capable requires few online resources to get to know; conventional expectations can apply. In contrast, someone negative (mean or stupid) requires more scrutiny, as does someone extreme in either direction (monstrous or saintly, an idiot or a genius). The key cognitive-miser prediction is that deviation from the (moderately positive) baseline is informative and triggers effort. Diagnostic information elicits both attention and weight in the overall impression (Fiske, 1980): Pairs of photographs depicted an individual’s degree of (un)friendliness and civic (dis)engagement. As predicted, perceivers’ looking time and algebraic weight in forming an impression each reflected both negativity and extremity. Cognitive misers did not bother much with people who fit the baseline, reserving effort for when it mattered (informative negativity or extremes). To explain when people do and do not go beyond shortcuts, such as defaults, a framework built on this and related work (Fiske & Neuberg, 1990; Fiske, Lin, & Neuberg, 1999). In the continuum model, perceivers start with a shortcut impression, categorizing the other person by salient cues, such as gender, race, age. People prioritize these category-based impressions, stopping there if fit is good enough for their everyday purposes (e.g. interacting with the cashier). If fit is not good (e.g. the cashier’s gender is ambiguous) or motivation is high (e.g. the cashier is the friend of a friend), the perceiver devotes more attention. Increased attention can move impressions from simple category-based stereotyping to subgrouping to fully individuated, attribute-based impressions. Cognitive misers show up in all kinds of places, from shipyards to investment banks. To illustrate one relevant implication, in gender discrimination lawsuits, the plaintiff might not prove malignant intent, but could show reckless use of categories in workplace evaluations. Managers who rely on category-based shortcuts are failing to individuate candidates for hiring and promotion. In providing expert testimony to this effect, I aimed to bring up-to-date science into the courtroom, but I had no idea that the testimony would reach all the way to the Supreme Court. Fortunately for me and for social cognition research, the justices found the argument to be so obvious (‘icing on the cake’) that it hardly required an expert witness (Fiske, Bersoff, Borgida, Deaux, & Heilman, 1991).

4

Susan T. Fiske

From the outset, the cognitive-miser model proved useful by (a) catalyzing new cognitive methods that might minimize experimenter bias and subjectivity (e.g. looking time), (b) generating falsifiable theory, (c) with mental process variables, mediators (e.g. attention) and (d) moderators (e.g. fit) that should specify conditions for obtaining one effect (shortcuts) or another (thoughtful judgments). As a result, the approach had real-world relevance. Nevertheless, the cognitive miser lacked motivation and context. Second wave: Motivated tacticians’ thinking is for doing Our field’s founders had considered social cognitive questions from a pragmatic, cognitive-miser viewpoint. For example, James (1890) described the ‘finite and practical self’, who must be ‘always unjust, always partial, always exclusive’ (pp. 959–60). He described the perceiver’s understanding as in the service of acting, which my review paraphrased as thinking is for doing (Fiske, 1992). Other early person-perception frameworks followed this pragmatic perspective (Asch, Bruner, Allport, Heider, Tajfel, Jones). In the 1970s and 80s, the cognitive-miser view initially neglected the pragmatics of thinking, with a narrow focus on context-free cognitive process and outcome. But, toward the century’s end, the new field of social cognition began to recognize that social understanding serves social interaction (Fiske, 1992). Three patterns of findings were emerging: Perceivers adopt good-enough strategies for everyday purposes; they are not inevitably as error-prone as cognitive misers might seem. Perceivers construct meaning through concepts such as traits, stereotypes, and stories, going beyond the information given, in Bruner’s felicitous terms. Finally, perceivers’ thinking strategies depend on their goals: People are motivated tacticians—sometimes cognitive misers, but sometimes deeper thinkers—depending on their purpose in making sense of another person. Strategies depend on varying motives, such as self-enhancement or social interdependence, to determine when to make more or less effort. Expert witnessing uncovered one illustrative motive neglected in stereotyping research: power. Peers perceiving peers had been the paradigm. But, many consequential decisions entail power-holders judging the less powerful (e.g. bosses; Fiske, 1993). Power should moderate stereotyping for several reasons. First, power-holders, because they control resources, need not individuate subordinates as much as vice versa. Attention focuses up the hierarchy. Second, power-holders may not want to overcome their stereotypes; powerful positions select for individuals focused on self more than others. Finally, power-holders are outnumbered by subordinates, so they cannot attend fully to each one. Subsequent experiments supported the power-as-control model (Dépret & Fiske, 1999; Goodwin, Gubin, Fiske, & Yzerbyt, 2000; Operario & Fiske, 2001; Stevens & Fiske, 2000): Power motives make perceivers vulnerable to stereotyping. Motivated social cognition is not limited to how people think, superficially versus carefully. Motives also flavor what people think. For example, inter-

Not your grandparents’ social cognition

5

dependence motivates people to think harder about another person, whom they need (Erber & Fiske, 1984), but may introduce bias, as when people hope for a longer-term relationship and observe through rose-colored glasses (Goodwin, Fiske, Rosen, & Rosenthal, 2002). A case in point: Men and women are intimately interdependent, but men have more societal status. How do perceivers negotiate this complexity? Ambivalent Sexism (Glick & Fiske, 1996) provides one answer. Hostile sexism targets women who refuse intimate interdependence with men (career women, lesbians, feminists), viewing them as cold (but perhaps capable). Subjectively benevolent sexism—patronizing prejudice—views women in more traditional roles (secretary, housewife) as warm but incompetent. Sexism takes distinct forms, endorses different stereotypes, depending on interaction context, and motives inform this analysis. Overall, the motivated-tactician view, in returning to earlier pragmatism (thinking is for doing), inspired new methods (e.g. scales such as the Ambivalent Sexism Inventory), encouraged replicability by specifying moderating motives (e.g. power), provided societal relevance (e.g. power’s role), and encouraged new theory (e.g. Ambivalent Sexism). Twenty-first-century activated actors: Social brain and social mind By the turn of this century, social cognition research had discovered hosts of motives to moderate thinking and doing. But hosts are unmanageable. A socialevolutionary approach, based on adaptation to group life, proposed a taxonomy to organize core social motives relevant to social interaction (Fiske, 2000). First is belonging, wanting human connection and identity; people need other people to survive and thrive. Getting along in the group cognitively requires shared reality (understanding) and predictable contingency (controlling). Getting along affectively depends on sufficient self-regard (self-enhancing) and sufficient otherregard (trusting). These motives capture those most-often invented by psychologists generally and social cognition researchers specifically. Equipped with social motives, social cognition research expanded further to behavior, viewing the social perceiver as activated actor. Two of the mostoften recruited mechanisms were cognition-behavior activation in the mind and a direct route to behavioral tendencies correlated with brain activation. Our work illustrates as follows. The Stereotype Content Model (Fiske, Cuddy, Glick, and Xu, 2002) reacted to social cognition’s narrow focus on process, arguing that content also is predictable. Although stereotyping processes (attention, inference, memory) may be general, not all stereotypes are the same, but their content is systematic and predictable. Two dimensions capture impressions of groups encountered in daily life: What are their intentions—warm (friendly, sincere) or not? And, are they competent (capable, agentic) or not? The warmth-by-competence space differentiates the stereotypically high– high in-group (e.g. the admirable middle class) and the low–low out-groups

6

Susan T. Fiske

(e.g. allegedly disgusting homeless). Mixed combinations can be high warmth but low competence (pitiable older people), or high competence but low warmth (enviable rich people). Ethnic stereotypes are accidents of history and circumstances of immigration: which portion of an ethnic group happened to arrive under what circumstances at a given time. The migrants’ perceived cooperative (or competitive) intent predicts warmth stereotypes for their whole ethnicity. Their circumstantial status predicts competence stereotypes for their ethnicity. Stereotype content’s downstream consequences include distinctive emotional prejudices: pride for the middle class (and other groups stereotyped as high on both dimensions), disgust for the homeless (low on both), pity for the elderly (high warmth, low competence), and envy for the rich (low warmth, high competence). These emotions in turn predict behavioral tendencies—active help toward warm groups; active harm toward cold ones; passive association with competent groups; passive neglect of incompetent ones. Correlational and experimental data support these predictions across cultures, over time, and across levels (nations, groups, subgroups; for a recent review, see Fiske, 2015b). What’s more, the methods cut across levels of analysis. The standard surveys ask what most [Americans] think, to circumvent social desirability of people having to admit to stereotypes. Interpersonal experiments, however, show that the structural variables (cooperation, status) in dyadic interactions reliably produce expectations of individual warmth and competence (Russell & Fiske, 2008). If the twin dimensions of warmth and competence capture so much of the variance in social cognition and affect about individuals and groups, then each quadrant should also evoke typical neural signatures. Neural patterns indeed do characterize specific quadrants. For example, photographs of homeless people and drug addicts elicit insula activation, consistent with disgust ratings of these groups. Also, alone among the quadrants, these lowest of the low out-groups fail to activate the medial prefrontal cortex, which is otherwise reliably implicated in social cognition. These neural patterns fit rating data that likewise suggest dehumanization of allegedly disgusting outgroups (Harris & Fiske, 2006). In another quadrant, envied outgroups, such as rich people and business people, provoke resentment. This quadrant distinctively predicts Schadenfreude (malicious glee at another’s misfortune). Operationally, facial electrodes document subtle smiling at another’s misfortune—just for envied outgroups, but not others (Cikara & Fiske, 2012). Envy also appears in high-status competitive outgroups, such as rival sports teams who each perceive the other as threatening (highly ranked). In one historic baseball rivalry, extreme Yankee and Boston Red Sox fans showed reward-area activation to their rival’s loss to a third team (Cikara, Botvinick, & Fiske, 2011). This response, consistent with Schadenfreude, correlated with self-reported harms to rival fans. Activated actors respond affectively—assessed through neural responses, facial-muscle activity, and reported emotional prejudice; affect in turn predicts behavior, better than cognition alone. This era, in our lab and many others,

Not your grandparents’ social cognition

7

benefited from methodological advances that went beyond words and cognition, narrowly defined, to include affect and behavior. Regarding the 1970s crisis (by then forgotten): Predicting behavior required reliable and reproducible results. Theory proved useful in this regard, and societal relevance emerged. Inequality enablers: Social cognition and social relevance Although research in social cognition—and social psychology generally— has long studied intergroup issues, only recently has inequality itself emerged as a central societal-relevance focus. Besides the earlier work on power in our lab and others (Guinote, 2017), the status dimension generated Envy Up, Scorn Down: How Status Divides Us (Fiske, 2010, 2011). Social class too became focal (Fiske, 2015a; Fiske & Markus, 2012; Moya & Fiske, 2017). The essential message emerged: Social cognitive processes (e.g. biases, shortcuts, inattention), especially in higher-status perceivers, enable inequality at interpersonal levels. In our lab, a timely example describes ageism as intergenerational resource competition (North & Fiske, 2012). Although the default old-age stereotype affords pity for the allegedly warm but incompetent elder, this applies only to cooperative elders, who step aside for the next generation. The self-serving elder sacrifices all sympathy—more so than a comparably self-serving middleaged or younger person (North & Fiske, 2013). Generational tensions center on elders ceding timely succession (jobs, wealth, political power), shared consumption (healthcare benefits, highways), and identity (music, styles, technology). Ageism as resource tension is widespread, given rapidly graying populations (North & Fiske, 2015). Moving from the interpersonal to the cross-national, ambivalent stereotypes in general serve a broad social function: Income inequality predicts the use of ambivalent stereotypes (Durante et al., 2013). That is, more unequal countries have a more complicated status system to explain. Some high-status groups seem to deserve their good fortune (the hard-working middle class), and some do not (inheritors of wealth); some low-status groups seem deserving of help (elders, children, disabled), and some do not (drug addicts, homeless people). Explaining the trajectories of different immigrant groups also draws on narratives that reflect each quadrant’s distinctive profile. More equal countries have essentially two stereotypic groups: citizens of all kinds, who deserve the social welfare state, seeming both warm and competent, plus interlopers who do not deserve social welfare, seeming neither warm nor competent (refugees, Roma, nomads). In contrast to more unequal countries, equal ones produce fewer ambivalent stereotypes. All kinds of stereotypes prioritize warmth (trustworthiness, friendliness), underscoring the role of other people’s perceived intent (cooperative or competitive) in interpersonal and intergroup relations. Intent also determines who deserves sympathy and blame. Intent is central to social cognition. People are

8

Susan T. Fiske

motivated to attribute intent (or not), particularly for harm. Intent magnifies harm. That is, intentional harms seem worse, even when they are not, all else held exactly equal (Ames & Fiske, 2013). A nurse’s intentional drug mix-up seems more harmful than if it were unintentional. People are so motivated to blame and punish intentional wrongdoers that they will choose blaming to the exclusion of other appealing activities (Ames & Fiske, 2015). Online, they will click envelope icons—with more speed, quantity, and persistence—to recruit signatures supporting their blame-and-punish judgment for intentional harmdoers. People’s motivation to magnify such intentional harms has implications for moral and legal judgments. To summarize, inequality enablers endorse stereotypes and other judgments that determine who is deserving and who is blameworthy. Illustrative targets here include older people, people of lower socioeconomic status, and harmdoers. Finally, social cognitive research has become fully relevant to current issues. Although, certainly, earlier work had been socially relevant as well, the groundswell of interest in social class, power, morality, and politics qualifies as studying a new image of social perceivers. Social cognition models, as crisis response: Summary The crisis in social psychology upon my own doctoral graduation raised issues of methods, replicability, theory, and relevance. During my career, social cognition research has illustrated four phases of response to these challenges. First, the cognitive-miser approach coped with methods bias by introducing measures less prone to experimenter or participant interference: looking time as attention, clicking speed as motivation. Next, the motivated-tactician approach addressed replicability issues by acknowledging moderator variables, primarily goals, and motivations in social context. Perceivers’ attention, inference, evaluation, and memory all depend on their purpose in gaining social information. Getting to know a person as a long-term collaborator will differ from encountering a grocery cashier. The third wave, perceivers as activated actors, tackled the translation of mental states into behavior, using theory to guide prediction. In our lab’s earlier work, the continuum model predicted whether behavior would be based more on social categorization or individuation, although we barely began to pursue behavior empirically. Our power-as-control theory implied behavior mainly as evaluating and selecting others. The same holds for ambivalent sexism theory. But, finally, the stereotype content model specified behavioral tendencies distinctive to each combination of stereotypic warmth and competence. The distinctive emotional prejudices predicted behavioral tendencies even better (Cuddy, Fiske, & Glick, 2007). Going beyond reported behavior, distinctive activations emerged in brain-imaging and muscle responses. Finally, social cognitive approaches have been applied increasingly to stereotypes, power, class, inequality, morality, and intent. Viewing perceivers as inequality enablers answers any remaining doubts about the field’s relevance.

Not your grandparents’ social cognition

9

Crisis 3.0: Twenty-first century solutions that might please our elders In the current crisis, Grandfather Fiske would have told us to mind our methods, and many psychological scientists are doing so in many ways, often summarized as open science. But, going back to the first crisis responses, using multiple methods can still counteract artifacts due to particular methods. Accordingly, wherever possible, our lab seeks converging evidence across techniques. For example, stereotype content data come from current surveys, historical sources, online interactions, in-person laboratory experiments, brain-imaging, and electromyography. Most recently, we have been collecting open-ended responses and applying natural language processing to analyze the content (Nicolas & Fiske, in progress). Converging results are reassuring. Replicability goes beyond one lab’s converging methods. Another lab tried to replicate our warmth–competence space using multi-dimensional scaling (Koch, Imhoff, Dotsch, Unkelbach, & Alves, 2016). They reproduced our competence dimension, but not the primary one, warmth. And they found a new one—progressive/conservative beliefs. In such a crisis, Grandmother Barbara Page Fiske (herself the granddaughter of Mary Hutchinson Page) would have told us to mind our manners, as her husband, my father, would have told me to keep minding my methods. With their voices in my head, we engaged in an adversarial collaboration. In the process, we have discovered moderating conditions: Our results (warmth and competence) emerge in more person-level intergroup contexts (neighborhood). Their results (agency/status/competence and beliefs) emerge in more distant intergroup contexts (nation). Another recurring crisis concern, relevance, reminds us to consider contexts beyond American internet samples. As noted, the stereotype content model replicates across cultures, with moderators by region. My cultural psychologist brother, Alan Page Fiske, approves. And my demographer husband, Douglas Massey, appreciates moderators such as national inequality, peace–conflict, and ethnic diversity. Finally, there’s nothing so practical as a good theory, Kurt Lewin claimed (1951, p. 12), in defending applicable theory or theory-driven application. Our family’s traditions, from my suffragist great-grandmother onward, would add that nothing’s as relevant, certainly. But, the family methodologists would add that a theory is only as useful as its operationalization, because theory disciplines the mind in specifying methods for testing it. Both conceptual and empirical care should also improve replicability.

Conclusions To be sure, social cognition does not encompass psychology—or even social psychology—but serves here as a case study in scientific response to crises of confidence in a field. Modern social cognition research serves as one among many renovations to what, in dire moments, seemed a collapsing field. Science

10

Susan T. Fiske

sometimes advances abruptly (Kuhn, 1970), and these crisis responses, as well as ongoing science, give us many reasons to be optimistic for future generations.

References Ames, D. L., & Fiske, S. T. (2013). Intentional harms are worse, even when they’re not. Psychological Science, 24(9), 1755–1762. Ames, D. L., & Fiske, S. T. (2015). Perceived intent motivates people to magnify observed harm. Proceedings of the National Academy of Sciences, 112(12), 3599–3605. Bowditch, J. (2005). Chocorua Lake Basin Historic District [Application]. National Register of Historic Places Registration Form. Continuation Sheet, Section 7, p. 7 re Heavenly Hill. United States Department of the Interior National Park Service. Retrieved June 2017 https://npgallery.nps.gov/GetAsset/c0bd0bc2–47cf-4ad7-b496cd266e8b05c7 Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological Bulletin, 56(2), 81–105. Cikara, M., Botvinick, M. M., and Fiske, S. T. (2011). Us versus them: Social identity shapes neural responses to intergroup competition and harm. Psychological Science, 22, 306–313. Cikara, M., & Fiske, S. T. (2012). Stereotypes and Schadenfreude: Behavioral and physiological markers of pleasure at others’ misfortunes. Social Psychological and Personality Science, 3, 63–71. Cuddy, A. J. C., Fiske, S. T., & Glick, P. (2007). The BIAS map: Behaviors from intergroup affect and stereotypes. Journal of Personality and Social Psychology, 92, 631–648. Dépret, E. F., & Fiske, S. T. (1999). Perceiving the powerful: Intriguing individuals versus threatening groups. Journal of Experimental Social Psychology, 35, 461–480. Durante, F., Fiske, S. T., Kervyn, N., Cuddy, A. J. C., Akande, A., Adetoun, B. E., et al. (2013). Nations’ income inequality predicts ambivalence in stereotype content: How societies mind the gap. British Journal of Social Psychology, 52, 726–746. Erber, R., & Fiske, S. T. (1984). Outcome dependency and attention to inconsistent information. Journal of Personality and Social Psychology, 47, 709–726. Fiske, D. W. (1949). Consistency of the factorial structures of personality ratings from different sources. Journal of Abnormal and Social Psychology, 44(3), 329–344. Fiske, S. T. (1980). Attention and weight in person perception: The impact of negative and extreme behavior. Journal of Personality and Social Psychology, 38, 889–906. Fiske, S. T. (1992). Thinking is for doing: Portraits of social cognition from daguerreotype to laserphoto. Journal of Personality and Social Psychology, 63, 877–889. Fiske, S. T. (1993). Controlling other people: The impact of power on stereotyping. American Psychologist, 48, 621–628. Fiske, S. T. (2000). Stereotyping, prejudice, and discrimination at the seam between the centuries: Evolution, culture, mind, and brain. European Journal of Social Psychology, 30, 299–322. Fiske, S. T. (2010). Envy up, scorn down: How comparison divides us. American Psychologist, 65, 698–706. Fiske, S. T. (2011). Envy up, scorn down: How status divides us. New York: Russell Sage Foundation. Fiske, S. T. (2015a). Grolar bears, social class, and policy relevance: Extraordinary agendas for the emerging 21st century. European Journal of Social Psychology, 45(5), 551–559.

Not your grandparents’ social cognition

11

Fiske, S. T. (2015b). Intergroup biases: A focus on stereotype content. Current Opinion in Behavioral Sciences, 3, 45–50. Fiske, S. T. (2017). Going in many right directions, all at once. Perspectives on Psychological Science, 12(4), 652–655. Fiske, S. T., Bersoff, D. N., Borgida, E., Deaux, K., & Heilman, M. E. (1991). Social science research on trial: The use of sex stereotyping research in Price Waterhouse v. Hopkins. American Psychologist, 46, 1049–1060. Fiske, S. T., Cuddy, A. J., Glick, P., & Xu, J. (2002). A model of (often mixed) stereotype content: Competence and warmth respectively follow from perceived status and competition. Journal of Personality and and Social Psychology, 82, 878–902. Fiske, S. T., Lin, M. H., & Neuberg, S. L. (1999). The Continuum Model: Ten years later. In S. Chaiken and Y. Trope (Eds.) Dual process theories in social psychology (pp. 231–254). New York: Guilford. Fiske, S. T., & Markus, H. R. (Eds.) (2012). Facing social class: How societal rank influences interaction. New York: Russell Sage Foundation. Fiske, S. T., & Neuberg, S. L. (1990). A continuum model of impression formation, from category-based to individuating processes: Influence 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: Academic Press. Fiske, S. T., & Taylor, S. E. (1984). Social cognition. New York: Random House. Fiske, S. T., & Taylor, S. E. (2017). Social cognition: From brains to culture (3rd ed.). London: Sage. Giner-Sorolla, R. (2012). Science or art? How aesthetic standards grease the way through the publication bottleneck but undermine science. Perspectives on Psychological Science, 7(6), 562–571. Glick, P., & Fiske, S. T. (1996). The Ambivalent Sexism Inventory: Differentiating hostile and benevolent sexism. Journal of Personality and Social Psychology, 70, 491–512. Goodwin, S. A., Fiske, S. T., Rosen, L. D., & Rosenthal, A. M. (2002). The eye of the beholder: Romantic goals and impression biases. Journal of Experimental Social Psychology, 38, 232–241. Goodwin, S. A., Gubin, A., Fiske, S. T., & Yzerbyt, V. (2000). Power can bias impression formation: Stereotyping subordinates by default and by design. Group Processes and Intergroup Relations, 3, 227–256. Guinote, A. (2017). How power affects people: Activating, wanting, and goal seeking. Annual Review of Psychology, 68, 353–381. Harris, L. T., & Fiske, S. T. (2006). Dehumanizing the lowest of the low: Neuroimaging responses to extreme outgroups. Psychological Science, 17, 847–853. James, W. (1890). The principles of psychology. Cambridge, MA: Harvard. Koch, A., Imhoff, R., Dotsch, R., Unkelbach, C., & Alves, H. (2016). The ABC of stereotypes about groups: Agency/socioeconomic success, conservative–progressive beliefs, and communion. Journal of Personality and Social Psychology, 110(5), 675–709. Kuhn, T. S. (1970). The structure of scientific revolutions (2nd ed.). Chicago: University of Chicago Press. Lewin, K. (1951). Field theory in social science: Selected theoretical papers (p. 12). New York: Harper & Row. Moya, M., & Fiske, S. T. (2017). The social psychology of the Great Recession and social class divides. Journal of Social Issues, 73(1), 1–15. Nicolas, G., & Fiske, S. T. (in progress). What people want to know about stereotype content. Unpublished data.

12

Susan T. Fiske

North, M. S., & Fiske, S. T. (2012). An inconvenienced youth? Ageism and its potential intergenerational roots. Psychological Bulletin, 138(5), 982–997. North, M. S., & Fiske, S. T. (2013). Act your (old) age: Prescriptive, ageist biases over succession, identity, and consumption. Personality and Social Psychology Bulletin, 39(6), 720–734. North, M. S., & Fiske, S. T. (2015). Modern attitudes toward older adults in the aging world: A cross-cultural meta-analysis. Psychological Bulletin, 141(5), 993–1021. Operario, D., & Fiske, S. T. (2001). Effects of trait dominance on powerholders’ judgments of subordinates. Social Cognition, 19, 161–180. Russell, A. M., & Fiske, S. T. (2008). It’s all relative: Social position and interpersonal perception. European Journal of Social Psychology, 38, 1193–1201. Stevens, L. E., & Fiske, S. T. (2000). Motivated impressions of a powerholder: Accuracy under task dependency and misperception under evaluative dependency. Personality and Social Psychology Bulletin, 26, 907–922. Taylor, S. E. (1998). The social being in social psychology. In D. T. Gilbert, S. T. Fiske, & G. Lindzey (Eds.), The handbook of social psychology (pp. 58–95). New York: McGraw-Hill. Wundt, W. (1897/2009). Outlines of psychology (1897). In B. F. Gentile & B. O. Miller, Foundations of psychological thought: A history of psychology (pp. 36–44). Thousand Oaks, CA: Sage.

Part I

Cognitive misers The origins of social cognition

2

Attention and weight in person perception The impact of negative and extreme behavior Susan T. Fiske

People constantly form impressions and make evaluations of others. Consider the problem of hiring a research assistant or selecting a roommate. How do we go about evaluating each applicant? Obviously, we cannot fully appreciate each complex individual. Human beings present rich stimuli to an observer, and cognitive economy necessitates shortcuts. Categorization, for example, allows perceivers to apply prior knowledge about certain types of people to the perception of new ones (e.g., Allport, 1954; Bruner, 1973; Cantor & Mischel, 1979; Taylor, 1980a; Wyer, 1974). This allows information to be stored efficiently; in the case of interviewing job applicants, an employer can remember that this one came from Big Ivy University and that one from Backwater College and further that each applicant had the various credentials implied by their respective alma maters. Another illustrative cognitive shortcut is the much-researched attribution of dispositional consistency (Heider, 1944, 1958; Jones & Davis, 1965; Jones & Nisbett, 1972; Kelley, 1967, 1972; Ross, 1977). Attributing stable traits to another person allows the perceiver to predict that future behavior will be much like the present. If a job applicant is relaxed and friendly in the interview, a prospective employer usually infers that those attributes will extend to behavior on the job. Driven by the metatheory of parsimonious social information processing, current work in social cognition focuses in detail on how, for example, an employer would organize memory for each applicant (see Hastie et al., 1980) and how the employer would make inferences about each (see Higgins, Herman, & Zanna, 1980; Nisbett & Ross, 1980). The work depicts people as cognitive misers who carefully conserve scarce mental resources (Taylor, 1980b). In studying memory processes and post hoc inferences, much of social cognition is profitably guided by the idea of cognitive economy. The mainstream focus omits a crucial pair of issues, however. Consider first the employment interview itself, rather than inferences or memory after the fact. The potential boss, as a cognitive miser, processes selectively during the interview as well as post hoc; the boss attends to a subset of each applicant’s attributes available at the time of the interaction. To which attributes will the boss attend? Social cognition neglects, but is not entirely silent on, the issue of attention.

16

Susan T. Fiske

Previous analyses indicate that principles of relative salience in objects also predict people’s salience. Thus people who are statistically rare (e.g., the physically handicapped), rare in context (e.g., solos or tokens in groups), or visually highlighted (e.g., literally in the spotlight) all have been shown to attract attention (see McArthur, 1980; Taylor & Fiske, 1978, for reviews). For people and for objects, task conditions (Cohen & Ebbesen, 1979) and outcome dependency (Berscheid, Graziano, Monson, & Dermer, 1976) have both been shown to influence the process of ongoing perception. In addition to these not uniquely social principles of attention, more dramatic social features of people also predict attention, and this article provides a model for predicting selective attention on that basis. Before elaborating on that model, however, a second neglected issue in social cognition deserves discussion. In focusing on person memory and on attributions, current social cognitive work has neglected affect, preference, and evaluation (see Fiske, 1980; Higgins, Kuiper, & Olson, 1980; Zajonc, 1980, for further discussion of this). Evaluation, of course, is central to hiring a research assistant, picking a roommate, or any number of other daily social judgments. How would a cognitive miser process evaluations? This research aims in part to answer that question. One branch of social psychology does examine such evaluations, under the banner of person perception or impression formation (see Schneider, Hastorf, & Ellsworth, 1979, for a recent review). Since the current research borrows from methods used in this area and embellishes them, they require brief discussion now. The dominant paradigm in person perception was established by the germinal work of Solomon Asch (1946). In his work, people were given a list of traits describing a hypothetical other and were told both to write a descriptive paragraph and to choose descriptive adjectives from a series of antonym pairs. Nowadays, people are still given lists of traits, the antonym pairs have metamorphized into anchored rating scales, but the descriptive paragraph has been abandoned and, with it, the focus on overall impressions. In the current literature, the major contrast to Asch is Norman Anderson (e.g., 1974b), whose main emphasis has been on the evaluative dimension and the process of combining cues. The Asch paradigm remains, but the object of study has changed. Although its sole focus on person evaluation is narrow, Anderson’s model provides a start on how people make social evaluations. In particular, a cognitive miser view predicts selective processing of evaluative information—that is, weight to a subset of the available cues, presumably those that are relatively useful. The current analysis proposes a model for such selectivity: Perceivers put both weight and attention into a person’s most informative attributes. Informativeness (specifically, weighting and attention) is examined here as a direct function of attribute properties. Of course, in addition to properties of the cues themselves, the context within which the impression occurs should influence attending to and weighting of information. For example, presentation order and consistency affect relative weights in the trait-list paradigm. Although

Attention and weight in person perception

17

attention has not been studied directly in context effects, it has been evoked as an explanatory concept in work on order and consistency. (Fiske, Reference note 1, gives a conceptual discussion of informativeness as a function of context; this article will not consider context effects further.) To return to attribute effects on attention and weight, the current discussion focuses on aspects of attributes that are independent of context—specifically, the evaluative scale value of an attribute can be broken down into valence (sign or direction) and extremity (polarization or intensity). Valence and extremity will be considered in turn, each with regard to attention and the weighting of attributes. Considerable evidence shows that negatively valenced information receives more weight than positive information (e.g., Anderson, 1974b; Hamilton & Huffman, 1971; Hodges, 1974; Wyer, 1974). Kanouse and Hanson (1972), in a review of negativity in evaluations, discussed several explanations of this, two of which are entertained here. First, a figure-ground hypothesis suggests that most outcomes are perceived as positive, so that negativity stands out. A separate explanation involves two assumptions: (a) Most life outcomes are evaluated above the scale midpoint, and (b) extremely bad outcomes are more frequent than extremely good outcomes, which are almost nonexistent (see Figure 2.1a and Parducci, 1968). If the psychological mode of the perceived distribution is positive, then it follows that the valence of scale values has been studied by contrasting rare stimuli (negative) with frequent stimuli (positive). Further, given a moderately positive psychological expected value, negative attributes are psychologically more extreme than positive attributes of a comparable scale value. Such a viewpoint would argue that positive information, if its scale value were extreme enough, also would carry added weight. Figure 2.1b depicts this phenomenon by extending the positive side to a degree at which positive information would also be rare and psychologically extreme. This perspective is at variance with the above explanation by Kanouse and Hanson that the tails of the psychological distribution are not symmetrical (i.e., that extremely bad outcomes are more frequent than extremely good outcomes). Extremely positive as well as extremely negative information both should carry extra weight. This principle expands on typically obtained negativity effects and further applies to extremity. Much research demonstrates that impressions are most influenced by their extreme terms (e.g., Podell & Podell, 1963; Rosenberg, Nelson, & Vivekanathan, 1968; Warr & Jackson, 1975; Wyer, 1974). Extremity can constitute either more or less of an attribute than is usual, so extremity is a deviation from the central tendency. In accord with the above hypothesis that most of the negative behavior studied is psychologically more extreme than is the comparable positive behavior, an overall extremity effect is suggested for weighting and attention. If one examines a truly symmetrical distribution of extremely negative and more extremely positive attributes, then the positive attributes should carry more weight than they have been found to. If weight denotes informativeness of a stimulus, then the relationship of informativeness (and, therefore, weight or attention) to a single attribute’s evaluation should be as shown in Figure 2.1c.

18

Susan T. Fiske

FREQUENCY

a.

–

midpoint EVALUATION

+

midpoint EVALUATION

+

midpoint EVALUATION

+

FREQUENCY

b.

–

ATTENTION AND WEIGHT

c.

–

Figure 2.1 a) Hypothetical distribution of life outcomes or person attributes. (Note that the scale midpoint is hypothesized to be to the left of the psychological mode; most people are perceived as slightly positive on most dimensions.) Source: Redrawn from ‘Negativity in evaluations’ by D. E. Kanouse & R. Hanson, Jr., in E. E. Jones et al. (Eds) (1972). Attribution: Perceiving the causes of behavior (pp. 47–62). Morristown, NJ: General Learning Press. Copyright 1972 by Silver Burdett Company. Reprinted by permission.

b) An alternative hypothetical distribution of person attributes c) Hypothesized relationship of attribute evaluation to attention or weighting in impression formation. (Note that positive attributes usually carry less weight than negative attributes equidistant from the scale midpoint. Theoretically, it would be possible to obtain positive attributes with a scale value extreme enough to carry as much weight as do negative attributes with a less extreme scale value. In practice, however, such extraordinarily extreme positive scale values may be difficult to obtain, and the stimuli used here do not tap the furthest reaches of the positive tail)

Attention and weight in person perception

19

Let us return briefly to the employment interview. If the interviewer discovers that a potential research assistant has a BA degree, that moderately positive attribute is uninformative because it is fairly common. If the applicant has a Masters degree, that extremely positive cue carries more information and more weight because it distinguishes the applicant from the rest. The same would be true of a moderately negative and unusual attribute, such as an incomplete BA degree. An extremely negative cue, such as possessing only a high school diploma, would especially distinguish that unfortunate applicant and would carry enormous weight in evaluating the person as a potential research assistant. Research in categorization more formally supports the relationship of extremity to informativeness. A certain amount of variation between people serves to distinguish them, and one would expect higher weights to be placed on the cue that was the basis for categorization in that setting. This result has occurred (Bruner & Perlmutter, 1957; Deschamps, 1973–1974; Tajfel, 1969; Taylor, Fiske, Close, Anderson, & Ruderman, Reference note 2). Modal attributes, or familiar ones, have been hypothesized to have less impact (Bruner & Perlmutter, 1957; Langer, Taylor, Fiske, and Chanowitz, 1976; Malpass & Kravitz, 1969; Taylor et al., Reference note 2). Since modal attributes do not differentiate people, it follows that non-modal information should carry a higher weight. In this study, non-modal evaluative information is hypothesized to carry higher weight, whether its valence is positive or negative. In sum, both extreme and negative attributes are held to be informative, since they distinguish people. This research tests a model of attention and weight to these most informative attributes of people. This test requires an operationalization of weight and of attention. Anderson’s information integration theory (Anderson, 1959, 1968, 1971, 1974b) has been successfully applied to a wide variety of stimuli, including evaluations of books, meals, attitudes, behaviors, criminal offenses, life events, and performance of naval officers. But its primary emphasis has been on detailed exploration of the paradigmatic impression-formation task. Anderson’s model essentially describes the organism as an analog computer of continuous cognitive algebra. The crux of the equation for the combining of information is that a response, R, to a person or an attitude object equals the weighted average of the evaluative scale values of its components.

R = ∑ wi si + C + E n

i =0

For present purposes, the most crucial aspects of this equation are the weights, wi, in which the effective or relative weight of a given attribute, i, is wi/Σwi. The denominator forces the relative weights to sum to one. The si = each attribute’s evaluative scale value, where s0 = the organism’s initial response within a task and w0 = the weight of this starting predisposition, C = an arbitrary zero, and E = error. This general linear model depends on independent measurements of the organism’s pre-existing impression (w0s0) and

20

Susan T. Fiske

of weight (wi) and scale values (si) for each component. Although a complete review of this model is beyond the scope of this article, it seems that a weighted averaging approach is the most flexible and widely applicable algebraic model for the combining of person attributes. (Critics of Anderson’s model include Chalmers, 1971; Fishbein & Ajzen, 1975; and Ostrom, 1977.) In person perception, it has been used most in the minimal social situation with minimal cues, although some studies have used stimuli other than trait lists (e.g., Anderson & Farkas, 1973; Brewer, 1968; Dreben, Fiske, & Hastie, 1979; Lampel & Anderson, 1968). Weighted averaging eventually may be established as a general process underlying judgments. In any case, the present research will assume that people do integrate information in this way when forming an evaluation. We will satisfy one preliminary condition for assessing the relative influence of person cues by using Anderson’s model of information integration to solve for weights. Since relative weights (by definition of the averaging model) sum to one, more weight to one cue always excludes weight to another. Weight reflects the amount of information obtained from a cue relative to the other cues present (Anderson, 1974a). Weight varies depending on the context of a particular impression. Scale value, in contrast, is relatively fixed. A cue’s valence (positive or negative) and its extremity (polarization) are constant across people for any one perceiver on any one type of task. To most simply identify the model’s parameters, one independently estimates either weights or scale values. The focus here is on relative weighting, so the scale values are provided a priori by one group of subjects, and then relative weighting is investigated in another study. Given a way to estimate relative weight, the remaining prerequisite is an operationalization of attention. To enable a direct measure of attention, stimulus materials were visual depictions of social information, and attention was operationalized as time spent looking at them. Written stimuli, such as trait adjectives, would not have provided as sensitive a measure of attention as looking time, since one would not expect much interesting variation in time looking at written materials. Once the verbal stimulus has been read, no information remains to increase looking time beyond reading time. Photographs are more suited to such a direct measure, since they provide visually complex cues to which one may attend differentially. Given the main stimulus materials depicting particular behaviors, a series of these slides provided the within-rater stimulus unit. Raters were tested individually, and they controlled the remote switch on the slide projector, thus providing the direct measure of attention as looking time. Given measures for attention and weight, the stimuli were constructed to vary on informativeness. Under the general hypothesis that informativeness produces differential impact, negative and extreme information should attract both attention and weight (see Figure 2.2 for a diagram of this model). Consequently, the main experiment’s design independently varies both the valence and the extremity of cues. To determine valence and extremity, a preliminary

Attention and weight in person perception

21

LOOKING TIME

EXTREMITY

INFORMATIVENESS

NEGATIVITY

WEIGHT

Figure 2.2 A diagrammatic representation of the theoretical model for the research design

study was conducted to individually scale the stimuli. Once scale values had been established, a second group of subjects made evaluations of the people depicted by combinations of the slides. Each stimulus person was shown behaving on two different dimensions, sociability and civic activism. The stimuli, then, mimicked the traditional trait-list approach with a “behavior list.” Since the stimuli were behavior patterns, perhaps related to scripts or life themes (cf. Abelson, 1976; Schank & Abelson, 1977), this study modestly expands the person perception paradigm (cf. Fiske & Cox, 1979). More centrally, the main design tests the relationship of relative weight to looking time, as a function of valence and extremity.

Scaling study Method Overview and procedure The first step was to establish stimulus materials that would consistently elicit standard scale values on a measure of likability. After informal pretesting, two behavioral dimensions were found, each of which could be varied across four levels of likability, the minimum number of levels necessary to independently vary valence and extremity. After photographing each stimulus person engaged in each of the eight behaviors, the slides were scaled by 30 paid undergraduate subjects instructed to consider each slide independently. In response to the question, “How generally likable is this person?”, subjects marked an X on a 200 mm line, the endpoints of which were labelled not at all and extremely. Further, looking times were recorded by allowing subjects to control the slide changes. So that looking time would not include rating time, the instructions stated: “Just look at each slide until you feel like you understand it, until you are ready to rate it, and then push the button again.”

22

Susan T. Fiske

Stimulus materials After three practice examples irrelevant to sociability and activism, each subject first rated photographs of the faces of 17 young white males. Each slide was paired with a caption reading “This is X,” where X was a single-syllable common male name. Each subject then rated eight behavior slides of each stimulus person in a randomized sequence, making 136 ratings in all. The eight behavior slides for each stimulus person consisted of four levels, each on two dimensions. The first dimension, sociability, involved varying degrees of interaction with friends and was constructed so that two slides were negative and two were positive. Of each valence, one was more extreme than the other, making four levels: −−, −, +, and ++. The sociability slides all showed the stimulus person with friends in a park. The second dimension showed varying levels of civic activism; the target in all those cases reacted to a petition against child pornography—an issue shown by pretesting to elicit complete consensus from undergraduates. Setting, other actors, and most props were held constant within dimension. (More detailed descriptions are presented in Table 2.1.) To minimize variability of interpretation, descriptive captions were paired with each slide; these also are presented in Table 2.1. Equipment The slides were rear-projected onto a 17 × 12 in. (43 × 29 cm) translucent screen 38 in. (97 cm) from the subject. The slide changer switch was connected to the projector through a concealed relay rack so that the button simultaneously changed the slide and registered the change on an Esterline-Angus event recorder hidden next door. No subject expressed any suspicion about the event recorder, the noise of which was masked by the slide projector fan. Results Scaling results were analyzed by a three-way within-subjects analysis of variance. The factors were stimulus person (17 levels), dimension (sociability/activism), and level (−−/−/+/++). Likability ratings and looking times were the dependent measures. Likability All effects in the analysis were significant, due to the large number of withinsubjects degrees of freedom. However, relative effect sizes are informative in such a situation, and eta was calculated as follows: η=

SSeffect SSeffect + SSerror

Attention and weight in person perception

23

Table 2.1 Captions, descriptions of photos, and likability ratings for stimulus materials Dimension Level

Sociability

−− Caption “He resentfully ignores his friends.” Photo This showed the target person sitting huddled over, grim-faced, with his back to a male and female couple. All three people were eating lunch on the grass, in a park. The couple sat with their backs to the camera, while the stimulus person sat in profile or slightly faced the camera. Rating

77.01

Activism

Mean rating

“He deliberately rejects the cause.” The setting showed a table outdoors against a building, with a bright yellow poster reading Against Children in Pornography, and a female whose face was always obscured. The target person shoved away the petition clipboard offered to him by the young woman. He stood turning his back to her with a disgusted look on his face 67.44 72.22

− Caption “He intentionally withdraws from friends.” Photo In the above setting and configuration of people, the target person was lying on his stomach reading a book, and his face was neutral. Rating 94.84

“He apathetically ignores the cause.”

Caption “He casually eats with friends.” Photo With the above three people in the same setting, the target sat facing the couple and was offering a pretzel to one of them, with a neutral expression on his face.

“He moderately supports the cause.” In the same setting, the target leaned over to sign the petition, with a neutral expression.

Rating

108.53

In the same setting, the target simply walked past the young woman, who was holding up the petition for him. His expression was neutral, and he looked straight ahead. 70.76 82.80

+

111.43

++ Caption “He celebrates happily with his friends.” Photo In the same configuration as the + slide, all three people were holding beer cans and colorful napkins, and a crepe-paper streamer was in the background. The target person was shown laughing Rating 126.50 Mean

102.44

109.98

“He generously helps the cause.” In the same setting, the target was seated at the table, smiling, and offering the petition for the young woman to sign.

121.76

124.13

92.12

97.28

24

Susan T. Fiske

The largest effect was for level, F(3, 87) = 52.31, p < .001, η = .80, and the linear contrast clearly reflected the mean pattern, F (1, 29) = 67.55, p < .001, explaining 97% of the variance in the level effect. The residual also was significant, F(2, 58), p < .01. The significance of the predicted contrast supports the validity of the obtained scale values as representing four distinct levels of evaluation: extreme negative (−−), moderate negative (−), moderate positive (+), and extreme positive (++). Means are presented in Table 2.1. In addition, the dimension effect was quite substantial, F (1, 29) 21.24, p < .001, η = .65, such that sociability slides as a whole were rated higher than activism slides, as can be seen in Table 2.1. The other effects in the analysis were smaller. A main effect for stimulus person emerged, F(16, 464) = 6.60, p < .001, η = .43, such that some were more likable than others (means ranged from 83.85 to 109.07, with 14 out of 17 being within 10 mm of the grand mean). An effect due to the stimulus person × dimension interaction, F(16,464) = 4.61, p < .001, η = .37, showed that the effect of stimulus person varied somewhat by dimension (or vice versa). The interaction of stimulus person with level, F(48, 1392) = 2.21, p < .001, η = .27, showed a parallel result for evaluative degree with respect to target person. The three-way interaction of stimulus person, dimension, and level, F(48, 1392) = 2.98, p < .001, η = .30, reflected the interdependence of these factors. Finally, a dimension × level interaction, F(3, 87) = 6.02, p < .001, η = .17, stemmed from small disparities between the effects of level on the two dimensions. Based on these results, and since the main study called for exactly 16 stimulus people, the generally least likable of the 17 stimulus people was discarded from the set on the grounds that he introduced irrelevant variability into the stimuli. This left 16 targets whose mean ratings were each within 12 mm of the grand mean, or between 86 and 109. The major effects in the likability analyses, the main effects due to level and to dimension, justified retaining the stimulus set. The smaller effects due to stimulus person and to the various interactions were not judged to be prohibitively large. Thus, this analysis created a set of 16 stimulus persons of roughly equal likability, each with a face slide, four sociability slides, and four activism slides. Scale values of the 16 sets of eight slides each were calculated in the following manner. A primary assumption was made that the responses to individual slides actually represented a combination of the scale value of the slide with an initial predisposition (see e.g., Anderson, 1974b). If this initial scale value, or chronic expectation on the subject’s part, is integrated with the scale value for the slide, the relative weights of these two components must be determined. The simplest assumption is that they carry equal weight.1 In addition, the scale value of the initial impression must be determined. Again, given no a priori evidence, the simplest assumption seems to be best. One such assumption would be that the initial impression is exactly at the scale midpoint. However, given that the psychological midpoint may differ from the mathematical one (e.g., Parducci, 1968), the assumption was made that the scale midpoint was the grand

Attention and weight in person perception

25

Table 2.2 Scale values adjusted for initial impression a Level Dimension

−−

−

+

++

Sociability Activism

56.73 37.60

92.40 44.23

125.58 119.78

155.71 146.24

a

These are calculated with the assumption that s0 = 97.28, w0 = .5, and wi = .5.

mean of all the ratings across subjects. This value, 97.28, then was weighted .5 and yielded the adjusted scale values shown in Table 2.2. It should be noted that the assumptions to this point are conservative: They make the simplest possible theoretical assumptions, equal weighting and neutral impression. Looking times Although not the primary focus of the scaling study, looking times at both slides and captions potentially illuminate the process of evaluation: how subjects assign scale values. In other words, looking times may show how subjects evaluate or comprehend the stimuli. Examining looking times may also indicate possible confusions in their later interpretation as weights. If it should appear that the time to assign a scale value to, say, an extreme slide is larger than to assign one to a moderate slide, this would obviate using looking times later as direct measures of weight. Slide looking times were analyzed by analysis of variance on dimension (sociability/activism) × level (−−/−/+/++). Unfortunately, the looking times did vary systematically by level, F(3, 60) = 3.13, p < .03. Yet, as can be seen from Table 2.3, the abnormally high looking time at the high positive activism slide accounts for this and for the significant interaction, F(3, 60) = 7.81, p < .001. A Newman–Keuls post hoc comparison of means lent support to this interpretation. The high positive activism slide differed from six out of the seven remaining means at p < .05, and from the seventh at p < .10. None of the other means differed significantly from each other. Why this particular type of slide should elicit exaggerated attention is unclear. However, the pattern of looking times did not vary systematically in any other way. The planned contrasts that reflect predictions based on negativity and Table 2.3 Slide looking times from scaling study Level Dimension

−−

−

+

++

Sociability Activism

5.72b 5.42b

6.10a, b 5.25b

5.67b 5.11b

5.54b 6.74a

Note. Means not sharing a common subscript differ at p < .05.

26

Susan T. Fiske

extremity, as outlined in the introduction, were nonsignificant for both dimensions (both Fs < l).2 Thus it appears that slide looking times were all essentially similar, with one aberrant and unexplainable exception. However, even this exception could not produce the pattern of results predicted for the second study. It seems, then, that comprehension times will not confound interpretations of relative weighting. As mentioned above and described in Table 2.1, descriptive captions were paired with each slide. Caption reading times also were analyzed by a twoway analysis of variance. The grand mean was 1.82 s, and there were no significant effects for dimension, level, or their interaction.

Weighting study Method Design Each subject rated all 16 stimulus people and all 16 pairings of sociability (4 levels) × activism (4 levels); counterbalancing was accomplished via a 16 × 16 Latin Square (Cochran & Cox, 1957). Thus one dimension of the Latin Square comprised the 16-cell design assessing the effects for sociability, activism, and their interaction. The other dimension of the Latin Square gave the main effect for stimulus person. Due to this method of counterbalancing, the interaction of stimulus person with the dimension effects could not be assessed. Within the Latin Square, presentation sequence (i.e., activism first or sociability first) was randomly counterbalanced, but confounded with rater. Half the subjects received one form of the counterbalanced social–active sequences, and the second half received the other form. In both levels of sequence, sociability preceded activism eight times, and vice versa. This design was completely filled by 32 raters (given the 16 × 16 Latin Square with each of the two counterbalanced social–active sequences). Three replications of 32 subjects each were carried out, resulting in 96 subjects for each sociability–activism combination. Analyzing the within-subjects sociability (−−/−/+/++) × activism (−−/−/+/++) effects revealed relative weighting based on the two variables of major interest. Since each dimension varied across four levels (−−/−/+/++), weight based on extremity (−−, ++ vs. −, +) and valence (−−,− vs. +, ++) could be evaluated separately by planned comparisons. Weights should have been directly comparable to looking time, as hypothesized, and this could be tested by examining the fit of each to the predicted pattern. Subjects Ninety-six Harvard-Racliffe undergraduates participated, drawn from the same paid volunteer pool as those for the scaling study. Subjects had been randomly selected to be in the scaling study or the weighting study. A few additional

Attention and weight in person perception

27

weighting study subjects were tested, but their data were eliminated and replaced in the design, due to equipment failures. Procedure Subjects were tested individually, and the instructions were identical to the scaling study, except that the task was stated as involving three slides per person (i.e., subjects saw a standard face slide and two behavior slides— each of the three paired with its caption, as described in the first study). Subjects began with the same three practice examples as in the first study, except that each example involved three slides, a face and two behaviors, to parallel the new stimulus materials. The example behaviors were again irrelevant to the sociability and activism slides. After any questions had been answered, the subjects began the ratings with a fourth set, which was for practice but not labelled as such, and for which the behaviors shown were low negatives on both sociability and activism. The stimulus person for this set was the discarded 17th person from the original scaling set. Since he was the least likable and was shown with the two least likable behaviors, this set formed a low-end anchor for stabilizing subjects’ responses (M = 47.05). In contrast, the three previous practice slides ranged over the rest of the response scale (M = 87.68, 116.77, 130.14). Anderson (1974c) suggested such a procedure for stabilizing usage of the response scale. After the practice examples, subjects made the 16 critical likability ratings; each rater saw each stimulus person only once, in one particular social–active combination within the 4 × 4 matrix. Stimulus materials, equipment, and dependent measures were identical to those of the scaling study. After the study, subjects were paid, and the study was explained in its theoretical context. Results The Latin Square allowed separate assessment of effects for the 16 levels of sociability–activism combinations and for stimulus person but not for the interaction of these with each other. These interactions were of no substantive interest and presumably of small magnitude compared with the major effects. Another preliminary note: The analysis of sociability–activism combinations was not fully crossed with sequence at the individual-subject level, since each subject saw eight stimulus sequences with sociability presented first and eight with activism presented first. To solve this confounding, each subject was yoked with an immediately adjacent subject, who saw the same personcombination pairs (i.e., was in the same row of the Latin Square), but who saw the complementary pattern of sociability–activism sequences. Likability ratings Likability ratings were analyzed by a 4 (sociability) × 4 (activism) × 2 (sequence: sociability or activism first) analysis of variance. To test the precise form of the

28

Susan T. Fiske

effects obtained, contrasts were calculated to assess the predicted model and variants of it. The derivation of these contrasts is described first, followed by testing the fit of the results to the predicted pattern. Model. The reader will remember that relative weights were predicted to depend on both negativity and extremity. To create predicted scores based on this model, several assumptions were made. These involved, first, the weight and scale value of the initial impression (w0s0). As in the scaling study, s0, the scale value of the initial impression, was taken to be the grand mean of the scale values. Its relative weight, w0, was taken to be one third of the total weight, or the average of the weights of the other two components, (w1 + w2)/2. Since there were three items of information, assigning the one unknown weight, w0, to be one third of the total seemed to be the simplest assumption. The si scale values of the components were taken to be the adjusted scale values obtained from the first study and presented in Table 2.2. The weights, wi, were a more complex matter and the crux of the present investigation. The first prediction, greater weight to negativity, took the form 1/1/−1/−1, respectively, for −−/−/+/++; these were the weights for the contrast between positive and negative valence. Adding an arbitrary constant gave the more familiar positive weights of Anderson’s (1974b) model: 3/3/1/1 (see Table 2.4, Model 3). The second prediction, greater weight to extremity, took the form 1/−1/−1/1 or 3/1/1/3 (see Table 2.4, Model 2a). However, these extremity weights assumed equal intervals for the scale values. Since the scale values clearly did not meet this assumption, the weights for extremity were adjusted for the obtained scale values’ distance from the grand mean. This adjustment was accomplished by subtracting the obtained scale value from the grand mean and taking its absolute value. It was then divided by 15 to bring the relative weights to a scale comparable to the negativity weights, rather than letting them stay 15 times larger. In other terms, the extremity weights were calculated as follows: wi = |si – s0| / 15. This process yielded the adjusted extremity weights presented in Table 2.4, Model 2b. Combining the negativity and extremity predictions by simple addition gives the weights used in the main model. A more formal statement of the combined prediction includes as its first term the extremity weights. The second term is superficially complex, but the bracketed part of it simply yields zero if the scale value is positive (above the grand mean), thereby omitting the negativity weight. If the scale value is negative, the bracketed part yields one, and, when multiplied by three, the term yields the predicted negativity weight (see Table 2.4, Model 4).

wi =

|si -s0| ⎡|si -s0|− (si − s0 ) ⎤ + 3⎢ ⎥. 15 ⎣ 2(si − s0 ) ⎦

Attention and weight in person perception 1111 2 3 4 5 6 7 8 9 1011 1 2 3111 4 5 6 7 8 9 20111 1 2 3 4 5 6 7 8 9 30111 1 2 3 4 35 6 7 8 9 40111 1 2 3 4 45111

29

Table 2.4 Scale values and weights for model predictions Dimension Sociability −

+

Activism

Parameter

−−

++

−−

−

+

++

Scale values (si ) Weights (wi ) Model 1: equal weights Model 2a: extremity Model 2b: adjusted extremity* Model 3: negativity Model 4: combined

56.73 92.40 125.58 155.71

37.60 44.23 119.78 146.24

1

1

1

1

1

1

1

1

3

1

1

3

3

1

1

3

2.70

.33

1.89

3.90

3.98

3.54

1.50

3.26

3

3

1

1

3

3

1

1

5.70

3.33

2.89

4.90

6.98

6.54

2.50

4.26

Note: Parameters for the original impression were assumed to be: s0 = 97.28, w0 = (w1 + w2)/2. * Extremity adjusted for unequal intervals of obtained scale values.

The weights for the sociability and activism dimensions were presumed to be equal, on the grounds that it was the simplest assumption and could be verified in the subsequent analyses. The complete model, then, takes this form:

⎡ 2 ⎤ ∑ wi R = ⎢ i=1 ⎥ ( s0 × w1s1 ) + w 2 s2 ⎢ 2 ⎥ ⎣ ⎦ where wi is defined above. The scores predicted by the model are presented in Table 2.5. Fit of the model. The data were analyzed by a 4 (sociability) × 4 (activism) × 2 (sociability or activism first) repeated measures analysis of variance on dyads. Since sequence showed no significant main effect or interactions, it is not discussed further. The obtained data were compared to predictions based on the model and to predictions made by the three major alternatives: equal-weight averaging, extremity only, negativity only, and extremity and negativity combined (see Table 2.4). Contrasts testing the four models’ effects were obtained by calculating the predicted row, column, and interaction effects for each model and then using them as planned contrasts. As can be seen from the combined model predictions in Table 2.5, the main effect contrasts were linear, adjusted for scaling intervals, predicted weighting, and original impression. A linear main effect prediction also held for the other three models. Thus the more crucial

30

Susan T. Fiske

Table 2.5 Predicted likability means and effects based on combined weights and initial impression Sociability Activism −− M Cell effect − M Cell effect + M Cell effect ++ M Cell effect M Main effect

−−

−

+

++

M Main effect

63.23 5.56

69.28 −.89

74.66 −3.97

91.64 −.72

74.70 −19.35

65.80 5.08

72.74 −.41

78.53 −3.08

93.74 −1.60

77.70 −16.37

83.06 −6.58

101.89 −.52

114.36 3.52

128.14 3.57

106.86 12.86

95.78 −4.07

114.19 1.80

124.37 3.52

133.30 −1.27

116.91 22.87

76.97 −17.06

89.52 −4.52

97.98 3.94

110.70 17.66

94.04 –

Note: The main effects are deviation scores obtained by subtracting the grand mean from row and column means, and the interaction (“cell”) effects are deviation scores obtained by subtracting out the grand mean and main effects from the cell means. These predicted main effects and cell effects were used as contrasts to test the combined model against the data in Table 2.6.

test of the four models arose in the interaction pattern, but the main effects were considered first, as is traditional. The analysis showed a main effect for sociability, F(3, 141) = 109.98, p < .001, and for activism, F(3, 141) = 90.22, p < .001. Obtained means are presented in Table 2.6. The planned contrasts testing each of the four models were practically identical, since the scale values of the components were the major determinants of the main effects and they remained constant across the different models. Thus it is not surprising that, for the activism main effect contrasts, all four patterns were significant at p < .001 and accounted for between 97% and 99% of the main effect variance. The actual F values were: combined model, F(1, 47) = 192.71; extremity only, F(1, 47) = 192.45; negativity only, F(1, 47) = 193.24; and equal-weight averaging, F(1, 47) = 192.26. The residuals all were nonsignificant, indicating no significant lack of fit to the model. Similar results held for the sociability main effect contrasts; all four adjusted linear patterns were significant at p < .001 and accounted for between 90% and 97% of the sociability main effect: combined, F(1, 47) = 188.04 (94%); extremity only, F(1, 47) = 178.24 (90%); negativity only, F(1, 47) = 204.76 (97%); equalweight averaging, F(1, 47) = 311.69 (97%). Unfortunately, the sociability residuals all were significant at p < .001 also: combined, F(2, 96) = 15.54; extremity only, F(2, 96) = 31.23; negativity only, F(2, 96) = 8.09, but these were small effects compared to the linear patterns. To summarize, all four variants

Attention and weight in person perception

31

Table 2.6 Obtained likability means Sociability Activism −− M − M + M ++ M M

−−

−

+

++

M

56.55

69.91

91.46

94.96

78.22

70.92

77.32

87.62

98.57

83.61

83.43

100.35

119.34

121.62

106.19

89.88

115.50

128.68

133.26

116.83

75.19

90.77

106.78

112.10

96.21

on the averaging model accounted for almost all of each main effect, with slightly better success on the activism dimension than on the sociability dimension. The averaging model, in general, fared impressively on the main effects. Since the four models all made similar predictions concerning main effects, a more crucial test of the combined model came from the sociability × activism interaction. Although the equal-weight averaging model predicted no interaction, the effect was significant, F(9, 423) = 3.00, p < .002. Of the remaining three models, the combined model best described the data. Again, the contrasts were effects predicted by the model and shown in Table 2.5. These interaction contrasts were significant, F(1, 47) = 5.72, p < .02, and accounted for 27% of the interaction. Lack of fit to the model (i.e., the residual) also was significant, F(8, 376) = 2.54, p < .05, although its effect was smaller than the predicted pattern. The extremity predictions fit the data somewhat better than the negativity predictions did, since the predicted extremity effect accounted for 19% of the interaction variance and was significant, F(1, 47) = 5.64, p < .02, whereas the negativity effect accounted for 13% of the variance and was only marginally significant, F(1, 47) = 3.02, p < .10. Both their residuals were significant: extremity, F(8,376), p < .05; negativity, F(8, 376), p < .001. To summarize, the combined model fit the obtained interaction better than did the equalweighted average model, the extremity model, and the negativity model. Thus, all contrasts derived from the combined model described the data extremely well. The pooled significance test for the combined model as a whole reflected this, F(1, 141) = 461.48, p < .001. The residual was clearly far smaller, although also significant, F(14, 573) = 3.42, p < .001. The model as a whole explained 90% of the pooled likability variance. Subjects appeared to preferentially weight both negative and extreme information over both positive and moderate information. And in 15 of 16 cells, the discrepancy of the model from the data was less than 10 mm on a 200 mm response scale. However, it is instructive to see where the model was particularly discrepant from the data, in order to refine the model in the future. The residuals greater

32

Susan T. Fiske

than 3 mm can be summarized by three patterns. First, the moderately positive sociability and moderately negative activism estimates were consistently too low by a modal amount of about 5 mm. Drastic underestimation of these two scale values would account for such an effect, and the unusual rating distributions of these two scales support this interpretation. The moderately negative activism scale value had the most skewed curve (−.688, indicating negative skew), and the moderately positive sociability value had the most non-normal kurtosis (−.925, indicating a platykurtic or flat shape). Although neither of these measures was significantly non-normal, this evidence suggests that these particular scale values may be peculiar. Although a somewhat non-normal distribution suggests the use of central tendency estimates more robust than the typically-used mean, such alternatives (medians, modes, and windsorized means that omit outliers) did not improve the fit of the model. The second striking residual pattern appears in the two ++/−− combinations, that is, the highly inconsistent cells. These two cells both were weighted more toward the activism dimension than the model would predict. Note, however, that this greater weighting of activism was not true elsewhere in the residuals. A third outstanding element of the residuals was the −−/−− cell, the most extremely negative combination. The observed responses were even lower than the model would predict, as if extremely negative consistency carried extra weight. There was not a symmetrical pattern for the ++/++, +/+, or −/− cells. In addition to these effects—exaggerating the −−/−− cell, weighting activism over sociability in the extremely inconsistent cells (−−/++), and the constant underestimation of the activism −/sociability + combinations—one large residual remained. There is no apparent explanation for the model’s underestimation of the sociability ++/activism + cell. In any case, the model assuming preferential weighting of extreme and negative information accounted for virtually all of both main effects and a quarter of the interaction effect. None of the effects that involved sequence (sociability or activism first) were significant. Stimulus person. The analysis of stimulus person, at the level of single subjects rather than dyads, was conducted as an analysis of variance on 16 stimulus persons × 2 sequences (A or B),3 in which stimulus person was a repeated measure. The only significant effect was for stimulus person, F(15, 1,410) = 2.13, p < .01. This was a small effect, η = .15; only one stimulus person’s mean deviated more than 8 mm from the grand mean. Thus it was not considered a serious threat to generalizing over stimulus persons. In addition, such a small main effect made an effect on the untestable stimulus person × combination interaction less plausible. Looking Time Given that a preferential weighting model fit the likability ratings, is there any more concrete support for greater attention to negative and to extreme information? What is the connection between theoretical weights and actual

Attention and weight in person perception

33

attention? The slide looking times were analyzed to provide a behavioral indicator of relative weighting. Separate analyses were conducted on the first and second slides, since reactions to the second slide depended on which slide had come before it, and reactions to the first slide would not. First slide. Looking times at the first slide were examined in a 4 (level) × 2 (dimension) within-dyads analysis of variance. The critical test of the differential weighting pattern lay in the planned comparisons on the level main effect, although the level main effect was nonsignificant, F(3, 141) = 1.22, p = .30. As in the likability analysis, the looking time negativity predictions for the −−/−/+/++ levels were 1, 1, −1, −1. The extremity predictions, adjusted for unequal intervals of the obtained scale values, were averaged over the sociability and the activism dimensions: .70, −.70, −.94, .94. The combined model contrast weights were 1.70, .30, −1.94, −.06. The equal-weight averaging model would predict equal looking times at all slides, if looking time reflected weight. Of all four models, the combined model best fits the data, F(1, 47) = 5.48, p < .025. None of the other models were significant: extremity, F(1, 47) = 2.18, p > .10; negativity, F(1, 47) = 2.17, p > .10. The combined model’s residual was also non-significant, F(2, 96) < 1. As shown by Figure 2.3, the means were exactly as predicted. The dimension main effect was also significant, F(1, 47) = 5.60, p < .02, such that sociability slides received longer looking times than activism slides, M = 4.59 and 4.24, respectively. This main effect should be interpreted cautiously, for two reasons. First, since the settings and props were constant within dimension, it is possible that this effect merely represented the greater visual complexity of the sociability slides over the activism slides in general. Second, there was no evidence in the likability analyses for greater weighting of the sociability dimension, so it seems unlikely to generalize. To summarize, the pattern of greater attention to negativity and to extremity emerged as predicted for looking times at the first slide, and subjects also looked longer at sociability slides in general. Second slide. As already mentioned, looking times at the second slide were predicted to depend both on the slide itself and on the previous slide. Given the extremity and negativity predictions for both slides, looking time at the second slide should be proportional to its relative weight in the total impression. For example, given equal-interval scale values and only negativity predictions, a negative slide following a positive slide would receive a weight of 3, the positive slide a weight of 1, and the initial impression a weight of 2. The relative weight of the negative slide, then, is 3/6 or .5. For three of the four models considered here, the general prediction—that second slide looking time is proportional to its relative weight—creates contrasts in an analysis of variance on first slide (−−/−/+/++) by second slide (−−/−/+/++), in which the dependent measure is second slide looking time. The fourth model, equalweight averaging, predicts no main effects or interactions in this analysis. The combined model’s predicted contrast for level of the second slide was nonsignificant, F(1, 44) = 1.27, p < .25, as were the overall main effect,

34

Susan T. Fiske 0.30 Predicted Slide One Slide Two

Looking Time (Seconds)

0.20

0.10

0

–0.10

–0.20

–0.30 ––

–

+

++

Level

Figure 2.3 Predicted looking times and obtained looking times for first and second slides. (Since no predictions were made about the grand means, the looking times are presented as main effects by subtracting out the grand means. To obtain actual means, add the grand mean to each—predicted M = 4.64, Slide 1 M = 4.41, Slide 2 M = 6.40. Also, the predicted effects should be multiplied first by a factor of 10; no predictions had been made about the absolute scale of the effects.)

F(3, 132) < 1, and the contrasts for the other two models (both Fs < 1). This apparently stemmed from substantial amounts of error due to a few subjects who neglected to change the second slide before doing their ratings. Thus the second slide looking time measure showed weak significance, although the magnitudes of the effects are identical to those for the first slide. Further, the mean pattern was exactly as predicted by the combined model (see Figure 2.3). The main effect and the predicted contrast for level of the first slide were both nonsignificant (Fs < 1). As Table 2.7 shows, that mean pattern reflected only an extremity effect, such that an extreme first slide caused somewhat longer looking time at the second, but this contrast was nonsignificant, as was the negativity contrast (both Fs < 1). The interaction contrast predicted by the combined model was significant, F(1, 46) = 13.89, p < .001, accounting for 30% of the variance on the significant interaction effect, F(9, 396) = 3.24, p < .001. Although smaller than the contrast, the residual effect was significant also, F(8, 350) = 2.43, p < .01.

Attention and weight in person perception

35

Table 2.7 Looking times at second slides as a function of first and second slide levels Level of first slide Level of second slide

−−

−

+

++

M

−− − + ++ M

6.46 5.44 7.08 7.07 6.51

6.48 6.56 5.79 6.29 6.28

5.69 7.21 5.62 6.56 6.27

7.94 6.53 6.07 5.62 6.54

6.64 6.44 6.12 6.38 6.40

Neither the extremity-only nor negativity−only contrasts approached significance (both Fs < 1). The relative weight of the second slide showed the expected pattern of the combined model, as measured by looking time. There were no effects for dimension or its interactions in this analysis. To summarize, how does the predicted pattern of relative weighting fare with respect to looking time? The combined model’s predictions, pooled over the three contrast patterns, reflected the data at a level of marginal significance, F(1, 134) = 3.11, p < .07, and accounted for a combined 36% of the effects. The combined residual variation was nonsignificant, F(14, 526) < 1. The model, then, accurately explained the looking times for the second slide as a function of relative attention to extremity and negativity. Stimulus person. The analysis on looking time at the 16 stimulus people showed a main effect for stimulus person F(15, 1335) = 3.82, p < .01. Only one person elicited looking times that deviated more than one sec from the grand mean. There was no main effect for sequence (A or B) or the sequence × person interaction (both Fs < 1).

General discussion The predicted preferential weighting for negative and for extreme cues was the major effect in the likability analyses. Based on the moderate positivity bias well-established in person perception work, it was hypothesized that perceivers would give relatively high weight to cues that deviated from the modal position (unusual or extreme cues) and to cues whose evaluations fell below the psychological midpoint (negative cues). Both of these predictions were clearly supported. A model was successfully fit to the data, based on prior estimates of scale values for the component cues, on estimates of the weight and scale value of the original impression, and on the predicted pattern of weights for the components. Across two behavioral dimensions, perceivers’ judgments of likability were influenced especially by extreme cues, whose evaluation was highly positive or highly negative, and additionally by negative cues. Previous

36

Susan T. Fiske

work modeling evaluations of people has avoided prediction of differential weighting, but the current study demonstrates the systematic impact of negativity and extremity in person perception. The predicted pattern of differential weighting, together with the averaging model, accounted for 90% of the variance in likability judgments. Fitting the model did, however, depend on assigning scale values to the cues and on assigning a weight and scale value to the initial impression. Given these assumptions necessary to identifying parameters of the model, one might ask if it has any reality other than a mathematical one. The data concerning looking time address this issue. As predicted, greater weighting of the most informative cues was mirrored by greater visual attention to the same cues. Again, the predicted pattern fit the data extremely well. Apparently, this is the first time that a direct behavioral measure of the integration weights has been obtained. Previously, the weights have been mathematically deduced from scaled responses by making various preliminary assumptions. The looking times provide strong behavioral support for differential weighting as a function of informativeness. However, one possible objection to the looking time data could be that they not only tap relative weight but also comprehension. Consider, for example, a visually complex slide that yields little information relevant to the judgment. Or for example, irrelevant but unexpected behavior, such as reading comic books in an office, might take longer to comprehend than a more expected behavior that also happened to be more relevant to the judgment. A particular stimulus could be difficult to understand but have no impact on likability. In that case, looking time would reflect comprehensibility rather than weighting. This is not a strong objection, however, since the predicted effects replicated over two stimulus dimensions and 16 stimulus people. Another alternative related to comprehension time would object that certain levels of valence or extremity might be intrinsically harder to understand than others. For example, negative or extreme information plausibly might be easier to understand than other information. This would result in less comprehension time for the extremely negative slide, for example. But since that works against the current hypotheses, it would not constitute an alternative explanation for the obtained results. The opposite alternative, that extreme or negative information is rare and therefore hard to comprehend, is in fact integral to the current hypotheses. Rarity or novelty is by definition highly informative, and informativeness is hypothesized to cause increased looking time. Perceivers make a more detailed search of psychologically more informative cues and spend relatively less time “taking in” modal or uninformative cues. As a strategy of selective effort, such a process makes sense. Perceivers must ration social attention or be overwhelmed. In this instance, people seem to allocate processing time in an adaptive manner. Rather than equally weighting whatever comes along, social perceivers apparently attend to the most unusual cues—the non-modal ones. Modal cues are by definition uninformative; since

Attention and weight in person perception

37

they apply to many people, they do not discriminate. In attending to extreme cues, perceivers are following an adaptive strategy of attention to highly informative cues. Similarly, perceivers are proceeding adaptively in attending to negative cues. For a variety of reasons, negative cues are more informative than positive ones. First, a chronic positivity bias in person perception ensures that most person cues are positive, both in self-reports, which would mirror impression management efforts (e.g., Jones, Sensenig, & Haley, 1974), and in descriptions and ratings of others. Thus, as Kanouse and Hanson (1972) point out, negative cues stand out by virtue of being rare. And along the lines of the argument presented for extremity effects, such negative cues actually are more informative because of their rarity. Attention to negativity also is literally adaptive in the sense that one survives better by avoiding negative contacts. In view of controversies over the “irrationality” of various cognitive processes, these apparently adaptive strategies appear to be unusual (see e.g., Fontaine, 1975; Kahneman & Tversky, 1973; Nisbett & Ross, 1980). As cognitive misers (Taylor, 1980b), people selectively process a subset of the information available, and in this case, social perceivers allocate attention in an apparently sensible fashion. Although not designed to test the rationality of the social perceiver, this research demonstrates people’s apparently reasonable strategy of attention to maximally informative cues.

Notes 1

2

3

One could argue that the unequal weighting predictions of the second study should apply here as well. However, equal weights were assumed at this stage for several reasons. First, differential weighting is the hypothetical process under investigation, and to assume it true prior to demonstrating it would be empirically premature. Second, since the extremity weights depend on the extremity of the obtained scale values, determining the scale values would be a circular process. Third, the simplest predicted unequal weights (i.e., weights of 3, 2, 1, 2 for −−, −, +, ++) would alter the obtained scale values such that the presumably moderate positive stimuli would have scale values far above the presumably extreme positive stimuli, an improbable outcome. Fourth equal weighting, a simpler process, seems to be more reasonable for a task such as the first study, in which subjects evaluate only one slide of each stimulus person, rather than three as in the second study. Finally, the simple equal-weighting process seems most plausible in the first study, in which subjects rather wearily viewed a series of 153 separate slides; in contrast, the second study presented less than a third as many (48) in 16 groups of three each. On the grounds that the simplest assumption is best, equal weighting is assumed for the first study. A more precise analysis, along the lines of the weights analyses to be presented for the second study, also produced nonsignificant results (Fs < 1). This involved adjusting the extremity predictions for obtained intervals and assuming values for w0 and s0 to be .5 and the grand mean of the scale values, respectively. Since sequence is not fully crossed at the level of single subjects, it is an uninterpretable between-subjects counterbalancing variable in this analysis, although it is a meaningful within-dyads measure of sequence effects (sociability or activism first) in the sociability × activism analysis.

38

Susan T. Fiske

Reference notes 1. Fiske, S. T. Attention and the weighting of behavior in person perception. Unpublished doctoral dissertation, Harvard University, 1978. (Available from the author, Department of Psychology, Princeton University, Princeton NJ . 08544.) 2. Taylor, S. E., Fiske, S. T., Close, M. M., Anderson, C. E., & Ruderman, A. G. Solo status as a psychological variable. Unpublished manuscript, 1978. (Available from the first author, Department of Psychology, University of California, 405 Hilgard Avenue, Los Angeles, Calif. 90024.)

References Abelson, R. P. Script processing in attitude formation and decision making. In J. S. Carroll & J. W. Payne (Eds.), Cognition and social behavior. Potomac, MD.: Erlbaum, 1976. Allport, G. W. The nature of prejudice. Garden City, N.Y.: Addison-Wesley, 1954. Anderson, N. H. Test of a model for opinion change. Journal of Abnormal and Social Psychology, 1959, 59, 371–381. Anderson, N. H. A simple model for information integration. In R. P. Abelson et al. (Eds.), Theories of cognitive consistency: A sourcebook. Chicago: Rand McNally, 1968. Anderson, N. H. Integration theory and attitude change. Psychological Review, 1971, 78, 171–206. Anderson, N. H. Cognitive algebra: Integration theory applied to social attribution. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 7). New York: Academic Press, 1974. (a) Anderson, N. H. Information integration: A brief survey. In D. H. Krantz, R. C. Atkinson, R. D. Luce, & P. Suppes (Eds.), Contemporary developments in mathematical psychology (Vol. 2). San Francisco: Freeman, 1974. (b) Anderson, N. H. Methods for studying information integration. La Jolla, Calif.: Center for Human Information Processing, 1974. (c) Anderson, N. H., & Farkas, A. J. New light on order effects in attitude change. Journal of Personality and Social Psychology, 1973, 28, 88–93. Asch, S. E. Forming impressions of personality. Journal of Abnormal and Social Psychology, 1946, 41, 258–290. Berscheid, E., Graziano, W., Monson, T., & Dermer, M. Outcome dependency: Attention, attribution, and attraction. Journal of Personality and Social Psychology, 1976, 34, 978–989. Brewer, M. B. Averaging versus summation in composite ratings of complex social stimuli. Journal of Personality and Social Psychology, 1968, 8, 20–26. Bruner, J. S. Beyond the information given. New York: Norton, 1973. Bruner, J. S., & Perlmutter, H. V. Compatriot and foreigner: A study of impression formation in three countries. Journal of Abnormal and Social Psychology, 1957, 55, 253–260. Cantor, N., & Mischel, W. Prototypes in person perception. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 12). New York: Academic Press, 1979. Chalmers, D. F. Repetition and order effects in attitude formation. Journal of Personality and Social Psychology, 1971, 17, 219–228. Cochran, W. G., & Cox, G. M. Experimental designs (2nd ed.). New York: Wiley, 1957.

Attention and weight in person perception

39

Cohen, C. E., & Ebbesen, E. B. Observational goals and schema activation: A theoretical framework for behavior perception. Journal of Experimental Social Psychology, 1979, 15, 305–329. Deschamps, J. C. [Attribution, categorization, and intergroup representation.] Bulletin de Psychologie, 1973–1974, 27, 710–721. Dreben, E. K., Fiske, S. T., & Hastie, R. Impression and recall order effects in behaviorbased impression formation. Journal of Personality and Social Psychology, 1979, 37, 1758–1768. Fishbein, M., & Ajzen, I. Belief, attitude, intention, and behavior: An introduction to theory and research. Reading, Mass.: Addison-Wesley, 1975. Fiske, S. T. Social cognition and affect. In J. Harvey (Ed.), Cognition, social behavior, and the environment. Hillsdale, N.J.: Erlbaum, 1980. Fiske, S. T., & Cox, M. G. Person concepts: The effect of target familiarity and descriptive purpose on the process of describing others. Journal of Personality, 1979, 47, 136–161. Fontaine, G. Causal attribution in simulated versus real situations: When are people logical, when are they not? Journal of Personality and Social Psychology, 1975, 32, 1021–1029. Hamilton, D. L., & Huffman, L. J. Generality of impression formation processes for evaluative and nonevaluative judgments. Journal of Personality and Social Psychology, 1971, 20, 200–207. Hastie, R., et al. (Eds.). Person memory: Cognitive basis of social perception. Hillsdale, N.J.: Erlbaum, 1980. Heider, F. Social perception and phenomenal causality. Psychological Review, 1944, 51, 358–374. Heider, F. The psychology of interpersonal relations. New York: Wiley, 1958. Higgins, E. T., Herman, C. P., & Zanna, M. P. (Eds.). Social cognition: Structure and processes underlying person memory and social judgment. Hillsdale, N.J.: Erlbaum, 1980. Higgins, E. T., Kuiper, N. A., & Olson, J. M. Social cognition: A need to get personal. In E. T. Higgins, C. P. Herman, & M. P. Zanna (Eds.), Social cognition: Structure and processes underlying person memory and social judgment. Hillsdale, N.J.: Erlbaum, 1980. Hodges, B. H. Effect of valence on relative weighting in impression formation. Journal of Personality and Social Psychology, 1974, 30, 378–381. Jones, E. E., & Davis, K. E. From acts to dispositions: The attribution process in person perception. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 2). New York: Academic Press, 1965. Jones, E. E., & Nisbett, R. The actor and the observer: Divergent perceptions of the causes of behavior. In E. E. Jones et al. (Eds.). Attribution: Perceiving the causes of behavior. Morristown, N.J.: General Learning Press, 1972. Jones, R. A., Sensenig, J., & Haley, J. V. Self-descriptions: Configurations of content and order effects. Journal of Personality and Social Psychology, 1974, 30, 36–45. Kahneman, D., & Tversky, A. On the psychology of prediction. Psychological Review, 1973, 80, 237–251. Kanouse, D. E., & Hanson, L. R. Negativity in evaluations. In E. E. Jones et al. (Eds.). Attribution: Perceiving the causes of behavior. Morristown, N.J.: General Learning Press, 1972. Kelley, H. H. Attribution theory in social psychology. In D. Levine (Ed.), Nebraska Symposium on Motivation (Vol. 15). Lincoln: University of Nebraska Press, 1967.

40

Susan T. Fiske

Kelley, H. H. Attribution in social interaction. In E. E. Jones et al. (Eds.), Attribution: Perceiving the causes of behavior. Morristown, N.J .: General Learning Press, 1972. Lampel, A. K., & Anderson, N. H. Combining visual and verbal information in an impression formation task. Journal of Personality and Social Psychology, 1968, 9, 1–6. Langer, E. J., Taylor, S. E., Fiske, S. T., & Chanowitz, B. Stigma, staring, and discomfort: A novel stimulus hypothesis. Journal of Experimental Social Psychology, 1976, 12, 451–463. Malpass, R., & Kravitz, L. Recognition for faces of own and other race. Journal of Personality and Social Psychology, 1969, 13, 330–334. McArthur, L. Z. What grabs you? The role of attention in impression formation and causal attribution. In E. T. Higgins, C. P. Herman, & M. P. Zanna (Eds.), Social cognition: Cognitive structure and processes underlying person memory and social judgment. Hillsdale, N.J.: Erlbaum, 1980. Nisbett, R. E., & Ross, L. Human inference: Strategies and shortcomings in social judgment. Englewood Cliffs, N.J.: Prentice-Hall, 1980. Ostrom, T. M. Between-theory and within-theory conflict in explaining context effects in impression formation. Journal of Experimental and Social Psychology, 1977, 13, 492–503. Parducci, A. The relativism of absolute judgments. Scientific American, 1968, 219(6), 84–90. Podell, H. A., & Podell, J. E. Quantitative connotation of a concept. Journal of Abnormal and Social Psychology, 1963, 67, 509–513. Rosenberg, S., Nelson, C., & Vivekanathan, P. S. A multidimensional approach to the structure of personality impressions. Journal of Personality and Social Psychology, 1968, 9, 283–294. Ross, L. The intuitive psychologist and his shortcomings: Distortions in the attribution process. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 10). New York; Academic Press, 1977. Schank, R. C., & Abelson, R. P. Scripts, plans, goals, and understanding. Hillsdale, N.J.: Erlbaum, 1977. Schneider, D. J., Hastorf, A. H., & Ellsworth, P. C. Person perception (2nd ed.) Reading, Mass.: Addison-Wesley, 1979. Tajfel, H. Cognitive aspects of prejudice. Journal of Social Issues, 1969, 4, 79–97. Taylor, S. E. A categorization approach to stereotyping. In D. L. Hamilton (Ed.), Cognitive processes in stereotyping and intergroup behavior. Hillsdale, N.J.: Erlbaum, 1980. (a) Taylor, S. E. The interface of cognitive and social psychology. In J. Harvey (Ed.), Cognition, social behavior, and the environment. Hillsdale, N.J.: Erlbaum, 1980. (b) Taylor, S. E., & Fiske, S. T. Salience, attention, and attribution: Top of the head phenomena. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 11). New York: Academic Press, 1978. Warr, P., & Jackson, P. The importance of extremity. Journal of Personality and Social Psychology, 1975, 32, 278–282. Wyer, R. S. Cognitive organization and change: An information processing approach. Potomac, Md.: Erlbaum, 1974. Zajonc, R. B. Feeling and thinking: Preferences need no inferences. American Psychologist, 1980, 35, 151–175.

3

The continuum model Ten years later Susan T. Fiske, Monica Lin, and Steven L. Neuberg

[People’s windows on their impressions of others] have this mark of their own that at each of them stands a figure with a pair of eyes, or at least with a fieldglass, which forms, again and again, for observation, a unique instrument, insuring to the person making use of it an impression distinct from every other. He and his neighbors are watching the same show, but one seeing more where the other sees less, one seeing black where the other sees white, one seeing big where the other sees small, one seeing coarse where the other sees fine. . . . [The windows] are, singly, or together, as nothing without the posted presence of the watcher. –HENRY JAMES, Preface, The Portrait of a Lady (1881/1983, pp. ix–x)

The continuum model was designed to describe the range of ways in which— the many “windows” from which—people form impressions of other people, while acknowledging that they all do share some fundamental processes; they all inhabit the same human “house.” This chapter summarizes the continuum model and assesses its ongoing viability in light of the research exploring impression formation since the model’s formal publication in 1990. The chapter works from the specific to the more general. It first reviews the model’s history and its specific stages. Then the chapter examines the model’s five core premises, highlighting some of the over 300 citations of it in the social-scientific literature since 1990.1 Finally, the chapter revisits some of the model’s theoretical meta-assumptions, clarifying where appropriate some misinterpretations of our positions.

The continuum model: History and specific stages Historical context The central statement of the model (Fiske & Neuberg, 1990) dates back hardly 10 years, yet its origins date back farther. Fiske (1982) proposed the notion of “schema-triggered affect” to account for the immediate evaluation and affect associated with spontaneous social categorization. A 1984 grant proposal became the basis for a chapter (Fiske & Pavelchak, 1986) that delineated the subsequent empirical support for schema-triggered affect, contrasting category-based

42

Susan T. Fiske et al.

and piecemeal-based responses. Over the next several years, Fiske and Neuberg detailed the full model, making explicit the sequence of stages proposed, the core premises, and the support for both in the extant literature; the literature review was updated soon after (Riley & Fiske, 1991). As its centerpiece, the model tackled a contradiction between the literature on impression formation and social cognition—a contradiction pitting elemental, algebraic approaches to impression formation against Gestalt, holistic, configural approaches (Asch, 1946). The elemental, piecemeal view of impression formation posited that people form evaluative impressions of others by computing a weighted average of the isolated evaluations of the targets’ features (Anderson, 1981). For instance, a person known to be intelligent and altruistic should be evaluated favorably, because intelligence and altruism are each positively viewed characteristics (for most perceivers). Indeed, Anderson’s information integration model—the standard bearer of the elemental approach —did quite well in predicting people’s evaluations of others. Researchers expressed two main concerns with this and similar models, however. First, many believed that the piecemeal processes articulated by such models are psychologically peculiar, and perhaps even impossible for people to perform. Second, the elemental models viewed the meaning of each characteristic as fixed, not influenced by the other characteristics possessed by the target. For instance, regardless of whether it is paired with “altruistic” or “cruel”, the evaluation and meaning of “intelligent” is presumed to remain the same. This assumption conflicts with Gestalt, configural approaches, which posit that a characteristic’s meaning can change in light of a target’s other characteristics—that “intelligent” may mean something different and be valued differently, depending on whether it coexists with “altruistic” or “cruel” (for a review of this controversy, see Leyens & Fiske, 1994). These critiques were taken quite seriously, and many espoused instead a modern, information-processing version of the Gestalt approach. As the socialcognitive literature developed, aspects of the Gestalt approach informed researchers’ understanding of social categorization processes—a view represented in schema, category, prototype, and stereotype models (for a review, see Fiske & Taylor, 1991, Ch. 6). Theorists proposed both a richer role for perceivers’ prior knowledge in organizing their thinking about newly encountered people, and a more configural approach to understanding how elements of this prior knowledge might interact to create a more holistic meaning. This approach, too, garnered significant empirical support. And so the debate between the two approaches raged, until it was declared unresolvable, given the theories and methods then available (Ostrom, 1977). Synthesizing the two approaches was the aim of the continuum model. Sequence of processes Combining the social categorization and elemental approaches, the continuum model proposes that people can use a range of impression formation pro-

The continuum model: Ten years later 43 Encounter target person

INITIAL CATEGORIZATION occurs immediately upon perceiving person

Is person of minimal interest or relevance?

No

YES Allocate ATTENTION to target attributes

If successful

CONFIRMATORY CATEGORIZATION occurs when available information is interpreted to be consistent or nondiagnostic with respect to current category If unsuccessful

If successful

RECATEGORIZATION occurs when a person is interpreted as categorizable but not with respect to current category; includes accessing new category, subcategory, exemplar, or self-concept If unsuccessful PIECEMEAL INTEGRATION attribute-by-attribute analysis of person, occurs when the target is interpreted as not easily categorizable

Category-based affect, cognitions, and behavioral tendencies

Piecemeal-based affect, cognitions, and behavioral tendencies

Possible public expression of response

Is further assessment of target required?

YES

STOP NO

Figure 3.1 The continuum model of impression formation. It shows the range from category-based to individuating impression formation processes, as a function of attention and interpretation. Informational and motivational conditions determine the attentional and interpretive processes that result in the various impression formation processes Note: Copyright 1986 by Susan T. Fiske.

44

Susan T. Fiske et al.

cesses (see Figure 3.l), and that the utilized processes depend on two primary factors: the available information and the perceiver’s motivation. We briefly outline this continuum of processes next. Initial categorization Upon encountering an individual, perceivers rapidly categorize that individual on the basis of salient features. These features can take the form of physical characteristics, such as skin color or body shape; they can be a configuration of behavior that quickly and easily cues a social category, as when a smiling student strolls into one’s office and is categorized as a friendly person; or they can be verbally transmitted category labels, as when a friend introduces a new acquaintance as an “electrical engineer.” Many features can potentially elicit social categories, which then serve to organize and constrain the meaning and usefulness of subsequently identified features or attributes. Which features organize the others—taking on the role of category—depends on the cognitive and social context (Fiske & Neuberg, 1990, pp. 9–12). The model proposes, however, that certain social categories—such as gender, ethnicity, and age—are “privileged,” in that they can be easily applied to most people one encounters. Indeed, subsequent research has demonstrated that these categories are central and available to perceivers automatically, within milliseconds (see Fiske, 1998, for a review). These social categories have two advantages over others: They are physically manifested and thus have temporal primacy (most of all in visual encounters), and they have important cultural meanings that are often relevant for people’s immediate interaction goals. In any case, once perceivers categorize the encountered individual, they automatically tend to feel, think, and behave toward that individual in the same way they tend to feel, think, and behave toward members of that social category more generally, as research has amply demonstrated (Fiske, 1998). Personal relevance The continuum model proposes that personal relevance determines whether a perceiver stays with the initial category-based impression (in the case of lowrelevance circumstances) or tries to move beyond it (in the case of more relevant circumstances). The later sections reviewing the model’s core premises and meta-assumptions will explore the role of motivation. Our inclusion of personal relevance assumes a rich array of motivations relevant to belonging, understanding, controlling, self-enhancing, and trusting. The term “personal relevance” in the model merely implies motivational relevance. Attention and Interpretation The central mediators of this model, attention and interpretation, determine whether (and how far) perceivers go beyond the initial category and its

The continuum model: Ten years later 45 immediate cognitive, affective, and behavioral associates. To individuate, perceivers must examine other perceived attributes of the target, and they cannot do this unless they devote additional attentional resources to the task. “Attention to attributes mediates the extent to which people use relative stereotypic or relatively more individuating processes” (Fiske & Neuberg, 1990, p. 6). And how perceivers interpret the information they heed then determines whether the initial categorization remains plausible, or whether new (re)categorizations are needed. Attention does not guarantee accuracy (however defined) or individuation, but it does permit processes more individuating than the split-second initial categorization. Some colleagues (to remain anonymous) have sometimes read the model as claiming that attention necessarily creates accuracy, but instead “the attention stage is especially important because it provides the raw material that heavily influences which of the alternative processes is utilized. . . . This single mediator is posited to transmit the effect of both informational and motivational influences on subsequent impression-formation processes” (Fiske & Neuberg, 1990, p. 6, emphasis added). Informational fit and motivational pressure determine the impression formation processes used; attention and interpretation enable information and motivation to have this influence. Confirmatory categorization Upon receiving additional information, perceivers attempt to preserve the initial categorization. Originally, Fiske and Neuberg (1990) reviewed evidence that confirmatory categorization was encouraged by certain information conditions —namely, a label plus category-consistent attributes; a label plus mixed attributes; or a strong label plus judgment-irrelevant, category-irrelevant attributes. In addition, Fiske and Neuberg speculated that certain motivators (such as selfesteem threat) might increase the likelihood that a perceiver would view target information as consistent with the initial categorization, whereas other motivators (such as task outcome dependency, fear of invalidity, and accountability to an audience with unknown attitudes) might reduce the likelihood that a perceiver would view target information as consistent with the initial categorization. Since the publication of the original chapter, new evidence suggests not only that successful confirmatory categorization requires perceivers to attend effortfully to stereotype consistencies, but that such efforts indeed can be triggered by threats to oneself or one’s ingroup (Fiske & Leyens, 1996) and by the need to justify one’s power position (Goodwin, Gubin, Fiske, & Yzerbyt, 1998). Recategorization Sometimes people’s interpretation of new target information reveals the initial categorization to be faulty in some way. When this happens, and when perceivers have sufficient motivation (and attentional resources), the perceivers recategorize: “Recategorization represents an attempt to find a different

46

Susan T. Fiske et al.

category that can be interpreted as adequately organizing the bulk of current information” (Fiske & Neuberg, 1990, p. 7). Subcategories, exemplars, selfreferences, and new superordinate categories all represent attempts to install a better-fitting category in place of the initial category. Some commentators (Oakes & Reynolds, 1997) have not noted that all these are categories too, and thus subject to the same interplay between prior knowledge and perceived information fit, within a given social (cultural and interactional) context. Originally, Fiske and Neuberg described information conditions that encourage recategorization: a weak initial category confronted by judgment-irrelevant, category-irrelevant attributes, or an initial category confronted with clearly inconsistent attributes. In addition, they speculated that motivations would determine the specificity of categories used. Piecemeal Integration The most individuating stage of the continuum model is described as integrating, attribute by attribute, each relevant piece of information into an overall assessment. The initial category, under this process, does not evaporate, but becomes just another attribute that contributes to the overall impression. Some particular research programs have obtained piecemeal-type results, but they tend to require participants to judge multitudes of targets, using standardized information that varies according to a limited number of criteria (e.g., Anderson, 1981; Fiske, 1980). Comparable everyday examples include screening large quantities of admissions or job applications on a very few specified dimensions, with the aim of assigning each applicant some summary rating. In retrospect, purely elemental, attribute-by-attribute processing does not strike us as capturing every type of individuating response. The original version of the model also suggested that, in theory, one can make piecemeal sense of a single person who is otherwise uncategorizable. In hindsight, we are convinced that this latter sort of process is rare: Rather than simply adding up (or averaging) people’s good and bad points, people demonstrably construct naive theories to account for contradictory and novel combinations of attributes (Asch & Zukier, 1984; Kunda, Miller, & Claire, 1990; Leyens & Yzerbyt, 1992). Public expression and further assessment The model closes with a perceiver’s implicit or explicit decision to express the cognitions, affect, and behavior associated with the impressions resulting from processes along the continuum. Of course, such expressions can be made at any point along the way, even if a “final” impression has yet to be formed. Indeed, the further-assessment feedback loop captures the tendency for people to continue the thoughtful categorization–recategorization process (using the most recently accepted category as the foundation) as new information about the target becomes available or as they decide actively to seek more information. From the perspective of the continuum model, impression formation is a

The continuum model: Ten years later 47 dynamic process that responds both to the motives of perceivers and to the information impinging upon them. Conclusion The stages of initial categorization, confirmatory categorization, recategorization, and piecemeal processing are mediated by attention and interpretation, both of which are influenced by information and motivation. The subsequent research literature suggests that initial categorization and recategorization occur most easily, but this conclusion awaits an overview of the model’s broader premises.

Evidence for the core premises of the model Five key premises underlie the continuum model. Contemporaneous support for the initial formulation of the model was strong (see Fiske & Neuberg, 1990), and its premises have been further supported by subsequent research on impression formation. The following review explores this subsequent research, but is not meant to be exhaustive; instead, it concentrates on relevant recent research addressing the model or its five premises: 1. Perceivers give priority to categorization over individuation. 2. Ease of information fit between category and attributes influences progress along the continuum. 3. Attention to attribute information mediates the use of various impression formation processes. 4. Motivational factors influence progress along the impression formation continuum, according to the social interdependence structure and the criteria set by the primary motivating agent. 5. Attention to and interpretation of attribute information mediate the motivational influences on impression formation. Premise 1: Perceivers give priority to categorizing processes The model’s first premise assumes that perceivers typically use category-based processes before they use attribute-oriented processes, and that if the categoryoriented processes work well enough, perceivers do not engage additional, more attribute-oriented processes. In the context of the model, the premise implies that current knowledge about the target is fitted to the contents associated with the perceiver’s category to the degree tolerated by the perceiver’s interaction purposes and context. These purposes are determined by situational and individual differences in motivations, but the basic point is that category-based processes have priority and will be used without additional information search, to the extent that the category is pragmatic in context (especially in providing guidelines for interaction). The priority of categorizing processes continues to receive research support.

48

Susan T. Fiske et al.

First, people frequently use social categories to understand others, especially visually prominent categories such as gender, race, and age (Biernat & Vescio, 1993; Gardner, MacIntyre, & Lalonde, 1995; Hewstone, Hantzi, & Johnston, 1991; McCann, Ostrom, Tyner, & Mitchell, 1985; Stangor, Lynch, Duan, & Glas, 1992; Verkuyten, Masson, & Elffers, 1995; Zarate & Sandoval, 1995; Zebrowitz, Montepare, & Lee, 1993; for a review, see Fiske 1998). Indeed, perceivers can apply such social categories to targets quite quickly, usually in just fractions of a second after encountering them (Banaji & Hardin, 1996; Blair & Banaji, 1996; Devine, 1989; Dovidio, Evans, & Tyler, 1986; Fazio, Jackson, Dunton, & Williams, 1995; Gaertner & McLaughlin, 1983; Klinger & Beall, 1992; Lepore & Brown, 1997; Perdue & Gurtman, 1990; Wittenbrink, Judd, & Park, 1997; Zarate & Sandoval, 1995; Zarate & Smith, 1990; for reviews, see Dovidio & Gaertner, 1993; Fiske, 1998). Moreover, once a perceiver categorizes a target, the category works quickly and efficiently, making immediately accessible its associated affective, cognitive, and behavioral responses without requiring the perceiver to engage in much effortful thought (Macrae, Bodenhausen, & Milne, 1995; Macrae, Milne, & Bodenhausen, 1994; Macrae, Stangor, & Milne, 1994). People categorize others in part because doing so provides a wealth of information at little cognitive cost. Beyond providing information useful for impression formation, activated social categories can potentially bias attribute-oriented processing by eliciting selective perception, interpretation, inference, and memory (Heilman, 1995). In one study, for instance, men primed to think of women in stereotypical terms were not only more likely to ask a female job applicant more sexist questions and to exhibit more sexualized behavior toward her; they were also slower to recognize nonsexist words in a lexical-decision task (Rudman & Borgida, 1995). Indeed, the biasing effects of social categories occur so effortlessly that it becomes difficult to ignore or disregard these effects, even under relatively ideal circumstances (Nelson, Acker, & Manis, 1996). Furthermore, active attempts to inhibit category-based thoughts before they interfere with subsequent judgments or behaviors may create a ‘rebound effect’, whereby the stereotypic thoughts reappear with an insistence even greater than if they had never been suppressed (Macrae, Bodenhausen, Milne, & Jetten, 1994). Of course, this is not to say that social categories always inhibit conceptual processing of category-inconsistent individuating information. For instance, perceivers can preferentially recall stereotype-inconsistent information when they have plenty of available attentional resources (Macrae, Hewstone, & Griffiths, 1993), and this preferential recall may enable perceivers to modify their otherwise category-based impressions (Stangor & Duan, 1991). When attentional resources are restricted, however, the biasing effects of active social categories almost always increase (e.g., Gilbert & Hixon, 1991; Harris-Kern & Perkins, 1995; Macrae et al., 1993), even when perceivers are motivated to form accurate impressions (Biesanz, Neuberg, Smith, Asher, & Judice, 2001; Pendry & Macrae, 1994). And because people often focus their attention on

The continuum model: Ten years later 49 category-consistent information, their stereotypic expectations are easily reinforced (Leyens & Yzerbyt, 1992; Moberg, 1995; Wojciszke, 1994). In sum, the propensity for category-based cognitive processing to prevail speaks to the powerful effects of categorization over individuation. Nevertheless, the priority perceivers give to categorizing processes does not preclude their ability to use more attribute-oriented processes. Indeed, with the continuum model, we have hoped both to convey our belief that people rely on various kinds of impression formation and to describe the set of processes we think encourage perceivers to use each. Premise 2: Ease of information fit between category and attributes determines the processes people use The continuum model posits that different informational conditions elicit various impression formation processes, depending on the ease with which perceivers can fit a target’s attributes to the presently available category. When perceivers interpret a target’s attributes as fitting the category, they respond to the target in ways reflecting the contents of that category. In contrast, when perceivers interpret a target’s attributes as incompatible with the existing category, they are likely to move farther toward the individuating end of the continuum—first, by attempting to recategorize the target in a way that better takes account of target features perceived as relevant; and then, but only if necessary, performing a more piecemeal, attribute-by-attribute analysis. Recent research specifies information configurations that typically invoke different interpretations of category–attribute fit, leading to the impression formation processes of category confirmation, recategorization, and piecemeal integration. The initial categorization process differs from the others in that it is an automatic, perceptual process; perceivers spontaneously categorize targets. However, when perceivers judge targets to be of sufficient motivational relevance, and subsequently encounter additional information about them, perceivers move to the more thoughtful portion of the continuum, beginning with the process labeled “confirmatory categorization.” Here, reflecting on information they have about the target, perceivers assess whether the initial categorization still represents the target well enough for current purposes, given the target’s prototypicality and the perceiver’s sense of variability in the category (Haslam, Oakes, McGarty, Turner, & Onorato, 1995; Lambert, 1995; Perry, 1994). If the additional target information fits well with the initially selected category, the target’s perceived membership in that social category becomes more salient and compelling, leading perceivers to respond in ways heavily aligned with the initial category (Hamilton, Sherman, & Ruvolo, 1990; Jackson, Hansen, Hansen, & Sullivan, 1993; Oakes, Turner, & Haslam, 1991). When perceivers successfully confirm an initial categorization, they avoid the increased cognitive effort needed to shift toward more individuating processes. Nonetheless, perceivers may feel that the target’s characteristics do not fit well with the initial categorization. When this happens, they may attempt

50

Susan T. Fiske et al.

to identify a different category, one better able to capture the whole of the target’s characteristics. Fiske and Neuberg (1990) labeled this process “recategorization,” and viewed it as occupying an intermediate position along the continuum of impression formation processes: Although recategorization still involves categorizing the target, it relies more heavily on additional target characteristics to do so. Perceivers can recategorize a target in several ways (see Fiske & Neuberg, 1990). They can access more differentiated subcategories, or subtypes, of the initial category (e.g., Eckes, 1994; for a review, see Fiske, 1998). For instance, a surprisingly passive man may be viewed as one of those “strong, silent types.” Subcategorizing is probably the preferred method of recategorization, as it enables the perceiver to retain information from the initial categorization, often with the apparently incompatible target information as justification (Kunda & Oleson, 1995). People may also recategorize using exemplars (“He reminds me of Colin Powell”) or self-reference (“He reminds me of the way I used to be”), or may even select a new category altogether, discarding those target characteristics seen as less important and generating a new category that effectively captures the remaining characteristics (Stangor et al., 1992). Regardless of the particular recategorization technique used, the perceivers’ affective, cognitive, and behavioral tendencies will be those reflecting the contents of the newly accepted (re)category. What factors lead perceivers to view new target information as incompatible with their previous categorizations? Obviously, the target characteristics themselves play a critical role: The more extremely inconsistent they are with the category, the more likely it is that people will attempt to recategorize or further individuate the target (Seta & Seta, 1993). In addition, certain social categories are less well developed and entrenched than others, and perceivers may be more likely to doubt the ability of such categories to account adequately for even minimally incompatible target characteristics. Furthermore, when a particularly credible source provides information that a target does not seem to fit the initial category, perceivers are more likely to presume the information to be accurate and reliable, and thus more likely to recategorize or further individuate the target (Macrae, Shepherd, & Milne, 1992). And certain motivations probably influence people’s thresholds for accepting new target information as incompatible with existing categorizations; a later section explores such motives. Finally, when perceivers cannot confirm the initial categorization or recategorize the target—and we think such instances are rare—they are likely to integrate the target information attribute by attribute in a piecemeal fashion (e.g., Kashima & Kerekes, 1994; Levine, Halberstadt, & Goldstone, 1996), if they possess sufficient time, attentional resources, and motivation. This is the most individuating of the impression formation processes, for the perceiver considers target-based attributes with minimal reference to a category label. In summary, interpretation of category–attribute fit determines use of the impression formation continuum: When perceivers view a target’s characteristics as fitting easily with the initially selected category, they form impressions

The continuum model: Ten years later 51 consistent with that category; when perceivers cannot fit a target’s characteristics with the initially selected category, they engage more individuating processes, such as recategorization or (in unusual cases) piecemeal integration. Premise 3: Attention to attribute information mediates use of the continuum The third premise proposes that attention mediates the various types of impression formation: Increased attention to target attributes is necessary for perceivers to engage the more thoughtful individuating processes of impression formation. Why is attention to attributes so important to the model? First, target characteristics suggest categories that may represent the target well. Second, once a category is in place, additional target characteristics become the critical data for examining interpretive fit (Premise 2). And third, when perceivers decide to go beyond the initial categorizations, these attributes become the relevant information base: When perceivers recategorize the target, they need do so in a way that accommodates the attributes, and, when perceivers use piecemeal integration processes, the attributes become the pieces of information to be integrated. This premise was supported originally by work indicating that attention to attributes, but not attention to the category, is correlated with individuation (Fiske, Neuberg, Beattie, & Milberg, 1987), as well as by early work on time pressure (Jamieson & Zanna, 1989; see Fiske & Neuberg, 1990, for a review). Recent research expands these points. People stereotype at the nadir of their circadian rhythm (Bodenhausen, 1990), when they have less capacity to attend. Time pressure, as noted, encourages stereotyping (e.g., Heaton & Kruglanski, 1991; Kaplan, Wanshula, & Zanna, 1993; Kruglanski & Webster, 1991; Pratto & Bargh, 1991). Arousal created by exercise can be distracting enough to encourage stereotyping (Kim & Baron, 1988). Noise that diminishes attentional capacity has similar effects (Kruglanski & Webster, 1991). Anxiety, which in part amounts to capacity-reducing “mental noise,” also interferes with individuation (Wilder & Shapiro, 1989a, 1989b). Increased attention mediates the individuating influence of informational inconsistency. People use more individuating processes when they discover that targets’ characteristics are inconsistent with the way they have been categorized (e.g., Fiske et al., 1987). For instance, perceivers pay more attention to a target who violates their expectations than to a target who confirms them (Hilton, Klein, & von Hippel, 1991); this increased attention potentially leads to more individuating evaluations of the target. If people fail to elaborate disconfirming information about an individual target, they may not remember it and use it (Sekaquaptewa & von Hippel, 1994). Stereotype dilution occurs only when perceivers find it impossible to construe the attribute information as stereotypeconsistent (De Dreu, Yzerbyt, & Leyens, 1995). Thus, only when attribute information is highly inconsistent and highly attended are more individuating processes likely to supersede category-based ones.

52

Susan T. Fiske et al.

In the end, use of category information remains rather high across various information conditions (Fiske & Neuberg, 1990; Fiske, 1998). The changes in attention to and use of attribute information, however, are what mediate changes in impression formation. By retaining and emphasizing the crucial roles of both category and attribute information, the model proposes that category information has a constant role, while attribute information has a variable role, mediated by attention allocation. Premise 4: Motivational influences on impression formation operate according to the interdependence structure and the motivating agent’s criteria Thus far, the chapter has discussed impression formation in terms of the informational influences of prevailing category-based processes, interpretations of information configurations that guide use of the continuum, and differential attention to attribute information that mediate the processes. The model’s fourth premise posits that motivations also influence impression formation. This premise is represented in Figure 3.1’s inclusion of “personal relevance” as determining whether perceivers go beyond their split-second initial categorizations. Specifically, the premise proposes that outcome dependency (the perceived interdependence structure within a given situation) is crucial. For example, when a person from an underrepresented group suddenly comes in as department head, the department members are likely to adopt a goal of accurately understanding the person who controls many of their outcomes. If the same person is an irrelevant peer or a subordinate, they may not bother to attempt accuracy. Outcome dependency motivates effortful impression formation, in an attempt to be accurate about the person controlling outcomes—the person whom the model describes as the motivating agent. Outcome dependency operates according to the particular motivating agent’s criteria for providing the perceiver’s desired outcomes. To continue the prior example, it matters whether a bigoted or an equity-oriented person is department head. With a bigot for a boss, people will stay with more category-based decisions about subordinates and peers than they will with an equity-oriented boss. Perceivers are impelled toward the categorizing or the individuating end of the continuum, precisely because of their motivations that arise from outcome dependency. More formally, perceiver motivations can be analyzed structurally by focusing on the interdependence structure within the situation. In particular, consider the structural features of outcome dependency: the motivating agent (who controls the outcomes) and the criteria set for attaining some desired outcome. The motivating agent—the target, a third party, or the perceiver’s self— determines the specific criteria for attaining a desired outcome (or avoiding a feared outcome). Together, the motivating agent and the criteria shape the perceiver’s goals during impression formation (Riley & Fiske, 1991).

The continuum model: Ten years later 53 Supporting research underscores how outcome dependency motivates changes toward individuation and toward categorization. Our own research indicates a robust effect, such that outcome dependency increases individuating attention—that is, attention to expectancy-disconfirming information— in the service of forming individual dispositional attributions about the target (Erber & Fiske, 1984; Neuberg & Fiske, 1987). If the perceiver is not initially biased, outcome dependency creates an accuracy goal that leads to open-minded information seeking (Neuberg, 1989). In another study, an outcome dependency decreased bias only among nonpartisan (initially unbiased) perceivers. Nonpartisan observers of a negotiation were more likely to notice compatible interests, were more accurate with their judgments, and were more evenhanded with their judgments than partisan (initially biased) negotiators (Thompson, 1995). These results imply that nonpartisan perceivers presumably have less at stake and so are less motivated to process information in terms of partisan categories. Another example of outcome dependency shows its implications for behavior. If a perceiver expects short-term involvement in an interaction with another, tendencies toward behavioral confirmation may encourage more category-based processing. In one instance where perceivers were motivated to form stable, predictable social impressions of their targets’ personality, “to check out [their] first impression,” perceivers came to believe that their targets held the stereotypic traits expected of them (Snyder & Haugen, 1994). Other interaction goals, however, encourage more individuating processes. For instance, when perceivers are motivated to form accurate impressions of their targets, to get their targets to like them, or to create an easily flowing conversation, they are more responsive to target attributes and more individuating in their impressions (Neuberg, 1989; Neuberg, Judice, Virdin, & Carrillo, 1993; Snyder & Haugen, 1994). Their outcomes, in the short term, depend on the target, for whom accuracy is doubtless a virtue. Outcome dependency is clearest under conditions of cooperation, when people’s outcomes are mutually contingent. Structured cooperative contact can decrease stereotyping, according to the contact hypothesis. Cooperation (mutual outcome dependency) facilitates individuation of outgroup members and reduces bias toward the group as a whole. In a study involving cooperative learning between a perceiver and a confederate posing as a former mental patient, perceivers expected traits that fit the stereotype of a former mental patient. However, those perceivers in the cooperative-learning sessions with the confederate subsequently judged the target more positively and adopted more positive views of the target’s social group as a result, presumably because they had focused on the person’s individual attributes (Desforges et al., 1991). Perceivers’ outcomes can also be contingent on third parties—persons who are outside the interaction but hold a stake in it. Being held accountable to a third party with unknown or unbiased views can make perceivers attend carefully, with the goal of accuracy (Tetlock, 1992; Tetlock & Lerner, 1999). Information and accountability can moderate stereotype-driven processes

54

Susan T. Fiske et al.

during social decision making, such that perceivers who are held accountable attend to stereotype-inconsistent attributes (Hattrup & Ford, 1995). Moreover, accountability can influence perceivers who initially categorize a target at a superordinate level (e.g., ‘woman’) to process further any available attribute information regarding the target and to form a more differentiated, subtyped impression (e.g., ‘businesswoman’) (Pendry & Macrae, 1996). Outcome dependency on a third party—that is, accountability—can undercut even the effects of mood on stereotyping processes. In the absence of such motivations, angry perceivers rendered more stereotypic judgments, compared to neutral and sad perceivers, who did not significantly differ (Bodenhausen, Sheppard, & Kramer, 1994). Again, in the absence of accountability, happy perceivers used stereotypic processing (Bodenhausen, Kramer, & Susser, 1994). Nevertheless, when informed that they would be accountable for their judgments, even mood-manipulated participants greatly reduced their stereotypic thinking. Despite the pervasiveness of happy people rendering stereotyped judgments, they can avoid the influences of stereotypes in forming judgments when accountability (third-party outcome dependency) offers a motivational incentive (see also Bodenhausen, Macrae, & Sherman, 1999, and Bless & Schwarz, 1999). Given the research presented, outcome dependency in various forms plays a part in attentional and interpretive processes by altering how the perceiver forms impressions. The particular interdependence structure, which includes the motivating agent and the criteria set by the agent, moves the perceiver’s impression formation goals along the continuum. In other words, motivations arising from the interdependence structure push perceivers toward individuation or keep them toward the continuum’s category-based end. Premise 5: Attention to and interpretation of attributes mediate motivational influences on impression formation The preceding section suggests that perceivers’ motivations stem from the interdependence structure, but exactly how do these motivations operate? From an analysis of the motivational implications of interdependence, as well as from the overall continuum model, the fifth premise proposes that motivational influences are mediated by attentional and interpretive responses to attributes. As with the impact of different information configurations on the continuum (Premise 3), attention is the necessary mediating mechanism, and interpretation determines the amount of perceived category–attribute fit (Premise 2). Again, available research supports this theoretical framework. Perceiver outcome dependency or outcome control can influence how information relevant to a target is heeded and interpreted. As noted already, social information may be variously attended and may be interpreted stereotypically, given differential outcome dependency. Motivated to enhance prediction and control over their situations, the outcome-dependent powerless attend to the powerful others who control their outcomes. Thus, powerless perceivers tend

The continuum model: Ten years later 55 to formulate complex, potentially nonstereotyped (but not necessarily accurate) impressions (Dépret & Fiske, 1999; Goodwin et al., 1998; Stevens & Fiske, 2000). On the other hand, the powerful, who are by definition less outcomedependent on their subordinates, grant less individuating attention to those with less power, making them more vulnerable to stereotype-based information processing (Goodwin et al., 1998; Operario & Fiske, 2001; Zemore & Fiske, 1998). The powerful need not attend to target others to control their own outcomes; often cannot attend to others because they are attentionally overloaded; and, if they are high in need for dominance, do not desire to attend to others (Fiske, 1993a). For people whose cooperative outcomes depend upon others, as noted, individuating attention rather than stereotyping attention appears especially likely (e.g., Erber & Fiske, 1984; Neuberg & Fiske, 1987). Are individuating processes equally likely when the interdependence structure is characterized by a negative contingency between outcomes—that is, by competition? One study investigated this likelihood by having participants expect to compete or not to compete with a fictitious other portrayed as competent or incompetent (Ruscher & Fiske, 1990). They then commented about the individual’s attributes, some of which were consistent with the expectations and others of which were inconsistent. As predicted, competitors increased their attention to inconsistencies, rendered more dispositional inferences about the inconsistencies, and developed more varied, idiosyncratic impressions. When individual competition enters the interdependence structure, perceivers’ goals are geared toward understanding the competitor’s attributes—presumably to gain better, more accurate knowledge of the competitor, in order to enhance their chances of success. Within competitive contexts, expectancies of the target and the nature of the information available both work to influence information processing. When perceivers form expectancies of an opponent’s competence, they tend to limit their attention to the opponent’s discrete attributes to aid them in matching the expectancy with the attributes, perhaps in an attempt to “size up the competition” more accurately (Miki, Tsuchiya, & Nishino, 1993). But when the target expectancy is less well formulated, because the available attribute information is initially nondiagnostic, competitors may interpret the attribute information as congruent with their stereotype-based beliefs. Whereas interpersonal (one-on-one) competition facilitates individuating impressions of opponents, intergroup (group-on-group) competition tends to promote stereotyping of opponents. In one study, when a person was joined by others who would compete with that person, the competitors as a group managed their limited attentional resources by assigning greater priority to individuating each other as teammates (Ruscher, Fiske, Miki, & Van Manen, 1991). Expectancy information was manipulated, and those who competed alone were then compared to those who competed in teams. Results showed that individuating processes, reflected in attention to and dispositional inferences about expectancy-incongruent attributes, were apparent in interpersonal but not intergroup competition, a result also supported by Dépret and Fiske

56

Susan T. Fiske et al.

(1999). This suggests that one-on-one competition makes the goal of obtaining accurate impressions of opponents particularly more salient. Indeed, a watered-down version of outcome dependency—simply expecting to interact with someone—can encourage accuracy-driven attention to a target’s individual attributes in the service of one’s goal. For example, perceivers exert greater effort making inferences about individuals with whom they expect to interact (Johnston, Hewstone, Pendry, & Frankish, 1994). Thus prospective interaction has sometimes been considered as a means to reduce inaccurate impressions. Although the prospect of interaction with a target can prevent inferential error by motivating perceivers with already biased impressions of the target to revise such impressions, it can also promote error by prompting perceivers to consume cognitive resources in planning their own behavior instead of changing their biases. Certain features of prospective interaction, such as role demands, goal familiarity, and partner novelty, determine whether perceivers become preoccupied with preparing their own behavior, which in turn determines whether the prospective interaction prevents or promotes inferential error (Osborne & Gilbert, 1992). Attention to the target (specifically, to the target’s attributes) and the perceiver’s motivational goals apparently play key roles in the correction of biased impressions. In summary, motivational influences, like informational influences, are also mediated by attentional and interpretive responses to attributes. Interdependence structure is a central source of motivations prompting either categorybased or individuating responses, depending on the particular motivating agents and their established criteria. Thus, the variations in perceived type of interdependence directly affect a perceiver’s attention to and interpretations of the information configuration. Conclusion The five core premises of the continuum model have aged gracefully. In particular, the evidence for each has expanded since the original literature review.

Revisiting some meta-assumptions Behind the explicit stages and more abstract core premises of the continuum model, there lurk some background perspectives that we now bring into the foreground. The crucial role of motivation is one, and the role of individual differences is closely related, reminding us of Henry James’s admonition at the beginning of this chapter—“the posted presence of the watcher.” What is more, the model has long assumed that expectations interplay with information; it is neither a fully social constructionist model nor a totally data-driven one. Finally, as an information-processing model, it can be implemented as serial or parallel processes, and this was made explicit in the original statement (Fiske & Neuberg, 1990). This section, then, addresses two issues of motives (situational and individual) and two issues of information (its interplay and its implementation).

The continuum model: Ten years later 57 Situational motives matter Perceivers tend toward category-based processes as default processes. But, depending on motivation (“personal relevance” in Figure 3.1), perceivers may end up anywhere on the impression formation continuum; they pay attention and form interpretations in accord with their motivations. As just noted, Fiske and Neuberg (1990) described motivations as stemming from a motivating agent—self, target, or a third party—who controls outcomes that matter to the individual. The content of the outcomes matters less than who controls them and what the criteria are for obtaining them. The direction and nature of the perceiver’s attention are determined by the motivating agent. For example, when the motivating agent is the target, who must cooperate with the perceiver for both to obtain a desired outcome, the perceiver attends to the target with a goal of accuracy (e.g., Neuberg & Fiske, 1987). Or the motivating agent may be a third party who desires a particular outcome, as when an employee is accountable to an audience (e.g., the boss) with known values (Tetlock, 1992; Tetlock & Lerner, 1999); this situation encourages attention and interpretation in accord with the motivating agent’s desired outcome. And if the motivating agent is the self, attention and interpretation operate in the service of one’s own values (Fiske & Von Hendy, 1992). The original formulation of the model did not specify the range of motivations likely to influence impressions. Instead, it used illustrations from thencurrent literature, organized in terms of the self, perceiver, and target. Since that time, Stevens and Fiske (1995) have organized social motives in terms of social adaptation. A truly social view of interpersonal motivation starts with the premise that people need other people to survive; thus people are adapted to live in social groups. People whose core motives facilitate group life are more likely to be successful, in this view. Thus people are psychologically adapted to the social group as the immediate survival environment (see Caporael, 1997; Caporael & Brewer, 1991). Reviewing the range of motivations identified by two dozen personality and social psychologists during this century fits well with this social survival perspective (for references, see Stevens & Fiske, 1995). Under various names, the most frequently identified motives include belonging, understanding, controlling, self-enhancing, and trusting; social-cognition researchers have investigated only some of these. The boundaries among these motives are not absolute but overlapping, and the list is not necessarily exhaustive or uniquely appropriate. Still, it emphasizes the social quality of the core motives, each of which is applicable here. Belonging The main motivational emphasis both previously and in our ongoing research has been on interdependence, which may be considered a major aspect of the core motive of belonging. People’s need to be part of social groups and dyadic relationships enhanced their ancient survival skills and continues to enhance

58

Susan T. Fiske et al.

their mental and physical health in modern times (e.g., Baumeister & Leary, 1995). The other person or people are the context for adaptation and survival (Caporael, 1997). In this context, it is not surprising, then, that people try to maintain their relationships and social belonging by trying to form accurate (or at least good enough) impressions of the others on whom they depend. In the model’s previous terms, the target is then the primary motivating agent —the one whose criteria must be met to obtain outcomes that matter (in this case, constructive interactions). In most cases, interdependence is facilitated by trying to get an accurate impression of the other, and a program of research has established that interdependence facilitates various specific individuating processes: (1) increased attention to stereotype-discrepant information; (2) increased dispositional inferences to it; and (3) more varied, idiosyncratic impressions (Dépret & Fiske, 1999; Erber & Fiske, 1984; Goodwin et al., 1998; Neuberg & Fiske, 1987; Pendry & Macrae, 1994; Ruscher & Fiske, 1990; Ruscher et al., 1991; Snodgrass, 1992; Stevens & Fiske, 2000; for a review, see Fiske & Dépret, 1996). The need to get along has been newly identified as a motive relevant to impression formation (Smith, Neuberg, Judice, & Biesanz, 1997; Snyder, 1992; Snyder & Haugen, 1994). For perceivers, the need to get along interferes with their attempting to confirm their stereotypes. For targets, the motive to get along encourages them to accommodate, and thus to confirm the perceivers’ stereotypes. Getting along encourages people to go along with their partners’ presumed attitudes, even if these are stereotypic (Chen, Shechter, & Chaiken, 1996; Ruscher, Hammer, & Hammer, 1996). Essentially, people who are forming impressions comply with salient norms. If the local norms appear to promote stereotypes, people’s impression formation processes and expressed impressions tend to be more stereotypic. The converse is true when individuating, egalitarian norms are salient (Blanchard, Crandall, Brigham, & Vaughn, 1994; Blanchard, Lilly, & Vaughn, 1991; Fiske & Von Hendy, 1992; Leippe & Eisenstadt, 1994; Pryor, Geidd, & Williams, 1995). Accountability to a third party also motivates more effortful impression formation processes (Pendry & Macrae, 1996; Tetlock, Skitka, & Boettger, 1989; for a review, see Tetlock, 1992). To the extent that the third-party audience’s views are known, perceivers conform their decision-making processes to those views. The third party’s imagined views also embody perceived norms, so accountability is another form of social contingency. Thus there is evidence that interdependence, getting along, norms, and accountability forms of belonging all influence how impressions are formed along the continuum from more category-based to more individuating impressions. Understanding In social contexts, interdependent players must operate under shared understanding of the environment, group norms, and each other. The need to understand the social world, and the physical world as interpreted by the social

The continuum model: Ten years later 59 world, is a core motive from infancy onward. This “cognitive drive” relies on the self as a motivating agent, as White (1959) so aptly described “effectance motivation” many years ago (see Stevens & Fiske, 1995, for more references). But all cognition is social (Ostrom, 1984; Resnick, Levine, & Teasley, 1991) and responds to shared social reality, so targets and third parties are motivating agents here as well. Explicit instructions to be accurate (Chen et al., 1996; Neuberg, 1989; Neuberg & Fiske, 1987) and situationally manipulated fear of invalidity (Freund, Kruglanski, & Shpitzajzen, 1985; Kruglanski & Freund, 1983; Kruglanski & Mayseless, 1988) cause people to form more effortful, potentially individuating impressions. (A later section discusses individual differences in needs for accurate understanding.) This relatively “cognitive” motive allows people to function better in groups by moving toward shared understandings. Controlling The need for control also has important cognitive elements, and in the context of the continuum model, it subsumes primarily those goals that relate to maintaining control over the self and the target. People who are deprived of control have been demonstrated to search for information in their social environment (e.g., Pittman & D’Agostino, 1989; for reviews, see Pittman, 1998; Pittman & Heller, 1987), even apart from the work on interdependence, just reviewed. Time pressure may be viewed as another form of control deprivation, and time pressure limits perceivers’ abilities to move down the continuum toward more individuating impressions (for reviews, see Fiske, 1993b; Kruglanski & Webster, 1996). Time pressure can increase discrimination (Freund et al., 1985; Jamieson & Zanna, 1989; Kruglanski & Freund, 1983), as can the pressure to implement a decision (Gollwitzer & Kinney, 1989). People who have social control strive to maintain it. Power can be defined as control of others’ outcomes (for a review, see Dépret & Fiske, 1993). As noted earlier, people can maintain power by relying on stereotypic information and ignoring counterstereotypic information; stereotypes perpetuate asymmetrical power relations, and inaccuracy is presumably less costly for the powerful than for the powerless (Fiske, 1993a; Fiske & Dépret, 1996; Goodwin et al., 1998), who can confirm the stereotypic outcomes they expect (Claire & Fiske, 1998; Copeland, 1994). Self-enhancing Motives to feel good or at least to improve the self carry a less cognitive, more affective flavor than the previous two (more cognitive) motives. The crucial phenomena appear to be threats to self-esteem. People whose collective high self-esteem is threatened are most likely to discriminate (Crocker & Luhtanen, 1990). Social identity theory (Tajfel & Turner, 1986) describes self-esteem maintenance as critical to intergroup perception; favoring the ingroup, which often entails at least relatively disadvantaging the outgroup (see Brewer & Brown,

60

Susan T. Fiske et al.

1998), is the basis for a positive identity. Under threat, the individuating effects of interdependence break down (Dépret & Fiske, 1999; Stevens & Fiske, 2000). And being insecure or anxious enhances stereotyping (Fiske, Morling, & Stevens, 1996; Stephan & Stephan, 1984; Wilder & Shapiro, 1989a, 1989b). Thus self-esteem maintenance and self-enhancement can require outgroup derogation. And, relatedly, Altemeyer’s (1981, 1988) modern scale of authoritarianism incorporates personal threat as predicting categorical responses. Self-enhancement motives can also, in contrast, discourage category-based responses. Theories of subtle racism—aversive racism (Gaertner & Dovidio, 1986), racial ambivalence (Katz & Hass, 1988), and modern racism (McConahay & Hough, 1976)—all presuppose that self-esteem can depend on an unprejudiced self-concept. The theories of modern sexism (Swim, Aikin, Hall, & Hunter, 1995), neosexism (Tougas, Brown, Beaton, & Joly, 1995), and ambivalent sexism (Glick & Fiske, 1996) make similar assumptions about sexism; they all focus on self-enhancement as underlying stereotyping and prejudice. People’s motivation to maintain self-esteem or to improve themselves (hence the term “self-enhancing,” which covers both) arguably makes them strive to be better members of their group. This can entail category-based responses that favor the ingroup and (relatively) derogate the outgroup. Especially under threat, these responses motivate category confirmation processes. However, if egalitarian values are important to the self, the self-improvement form of selfenhancement can undercut stereotyping. Trusting People adapt better to immediate social groups when they are at least initially trusting of ingroup members. The impression formation literature has long demonstrated a positivity bias, such that perceivers expect positive things from others, all else being equal (e.g., Matlin & Stang, 1978; Parducci, 1968; Sears, 1983). These kinds of motive remain to be investigated for their bearing on category-based versus individuated processes. But a basic motive to trust the ingroup would seem related to individuating ingroup members, even given the effort required. Conclusion A variety of situational motives direct impression formation toward more categorizing or more individuating processes. Although different frameworks can parse social motives, one framework identifies core motives in terms of individuals’ functioning adaptively in social groups, and describes the implications of these motives for impression formation. Individual differences in motivation matter People come to their social encounters with different chronic goals, many of which influence the way they form impressions of others. This section briefly

The continuum model: Ten years later 61 considers several such goals: personal need for structure, depression, need for cognition, fear of invalidity, and content-specific goals. Personal need for structure Some people, more than others, like their lives to be organized, simply structured, and predictable. These people value their routines and become uncomfortable or distressed when their sense of order is disrupted. Their cognitive activities often show a similar penchant for simple structure and order. For instance, people scoring high on the Personal Need for Structure Scale (Thompson, Naccarato, & Parker, 1989; Thompson, Naccarato, Parker, & Moskowitz 2001) view themselves, others, and nonsocial objects in relatively simple ways (Neuberg & Newsom, 1993); they are especially likely to spontaneously generate potentially categorizing trait inferences (Moskowitz, 1993); they are more likely to overgeneralize failure experiences into learned helplessness (Mikulincer, Yinon, & Kabili, 1991); and they are more likely to fulfill responsibilities in a prompt fashion, thus avoiding the discomfort associated with a lack of completion (Neuberg & Newsom, 1993; Roman, Moskowitz, Stein, & Eisenberg, 1995). Of more direct relevance to impression formation, people with strong desires for simple structure are more likely to assimilate new information to previously existing structures (Thompson, Roman, Moskowitz, Chaiken, & Bargh, 1994), to create simple stereotypes of new groups (Schaller, Boyd, Yohannes, & O’Brien, 1995), and to use simplifying social categories to understand others (Naccarato, 1988; Neuberg & Newsom, 1993). People who desire simple structure in their lives prefer the more category-based end of the impression formation continuum; they are perhaps aware that a thoughtful, individuating consideration of new target information carries with it the potential to prove one’s existing categorizations inadequate. In sum, people who are dispositionally high in the need for structure want to preserve their initial category-based impressions when possible. Mild depression and the desire for control Mildly depressed individuals are especially likely to use individuating processes, perhaps in an attempt to gain control over their lives (Edwards & Weary, 1993). Consistent with this interpretation, people who score highly on the Desire for Control Scale are especially likely to explore more deeply their social situations (Burger, 1992). In one study, for instance, such individuals were less likely to succumb to the simplifying correspondence bias and more likely to search their social situation for causes of another’s behavior (Burger & Hemans, 1988). Of course, people who are severely depressed tend to view the world as uncontrollable, and may thus see little reason to expend scarce cognitive resources in an effort to understand others in a more individuated way. Indeed, severely depressed people tend to rely more heavily on category-oriented processing (Marsh & Weary, 1994).

62

Susan T. Fiske et al.

Need for cognition and personal fear of invalidity Some people enjoy solving life’s puzzles, view thinking as fun, and appreciate discovering the strengths and weaknesses of their arguments. These people are high in the “need for cognition” (Cacioppo & Petty, 1982); they are less likely to use simplifying cognitive heuristics and more willing to expend extra efforts to assess fully their social circumstances (Cacioppo, Petty, Feinstein, & Jarvis, 1996). Although data are lacking, such individuals seem more likely to stray deeply into the individuating end of the impression formation continuum. We would expect the same from people high in the “fear of invalidity” (Freund et al., 1985; Thompson et al., 1989, 2001), for a different but related reason. These individuals worry about making poor judgments. As such, they often set for themselves high thresholds for accepting any decision; in the absence of great confidence, they are reluctant to commit to decisions. People high in the fear of invalidity should thus be slow to apply simple categorizations for targets they encounter. Instead, they are likely to continue gathering information about the target, in the hope of forming a particularly accurate impression. “Content” goals The chronic goals just noted might well be characterized as “process” goals, as they capture individual differences in the ways people like to think about their social worlds. Such goals work by directing perceivers’ attention generally toward or away from target attributes, and perceivers care little about the content of the information they focus on (except, perhaps, for its relevance to the information previously collected or to the impressions previously formed). Other goals, in contrast, direct perceivers’ attention toward specific kinds of information. People who are dispositionally anxious to please may seek information about others that is diagnostic of social approval; people who are dispositionally insecure in their self-views may seek information about others’ weaknesses, so that they can boost themselves in the process; people seeking to land a spouse will focus disproportionately on such characteristics as fidelity, agreeableness, intelligence, social status, physical attractiveness, and age. In short, people think about others so that they can reach their goals more efficiently, and different target characteristics serve different social goals. Within the framework of the continuum model, “content” goals will influence which categories perceivers use for initial categorization, which target attributes come into focus, and the willingness of a perceiver to settle on a particular impression. Conclusion Recent research focusing on chronic social and epistemic goals can be profitably applied to the continuum model. Having considered situational and chronic motives, we now turn to information—specifically, its interplay with expectations and its implementation in the formal model.

The continuum model: Ten years later 63 Expectations interplay with information The continuum model has been labeled by some as a “strong” social-constructivist model—one that “assumes that social perception creates social reality as much or more than it reflects social reality” (Jussim, 1991, 54; emphasis in original). Such a representation misses the essence of the model. Indeed, its original purpose was to begin explaining the interplay between features of the social perceiver on the one hand, and the information received from the target on the other. A target’s characteristics influence how the target is categorized and when (or if) these categories are abandoned in the search for a more individuating understanding. A perceiver’s expectations and goals influence which target characteristics come into focus and how they are interpreted. And the social context influences both what a target reveals and what a perceiver apprehends. Although we do indeed believe that perceivers sometimes see primarily what they want to see, we believe just as strongly that “there is a there out there” that strongly constrains social perception and impression formation processes. Our position was never one of advocating for a socialconstructivist versus a social-realist view of impression formation. Rather, our interest has always been in understanding the mutual, interactive influences of the perceiver and the perceived. Serial and parallel processes both fit the continuum model In a recent paper, Kunda and Thagard (1996) proposed a parallel-constraintsatisfaction theory of impression formation, in which social stereotypes and target traits and behaviors constrain one another’s meanings and together influence the impressions people form. They contrasted their theory with the continuum model and found the latter wanting, due primarily to its alleged serial nature and to the priority it ostensibly gives to social stereotype information over individuating information. Their critique rests on several misunderstandings of the continuum model. Indeed, Fiske and Neuberg (1990) considered (albeit briefly) the possibility that its proposals could be instantiated equally well by parallel or serial mechanisms. Space prevents addressing all relevant issues here, so this section will briefly focus on two. First, the 1990 model stated quite clearly (as we have done again, earlier in the present chapter) that many different features of a target can take on the organizing role of social category. Although certain features, such as gender, race, and age, may gain preeminence because of their visual accessibility (and thus their temporal precedence), this need not be the case. Thus, although the model does claim that perceivers categorize targets, and that once targets are categorized, other features of the targets will often not demonstrate a similar impact, it does not limit the features that may serve as categories to those evoking common social stereotypes (as Kunda and Thagard imply). For instance, a target can be categorized on the basis of a single behavior, and as a result, the behavior will gain an enhanced amount of influence over the ultimate impression formed. Kunda and Thagard’s claim that common social stereotypes need not

64

Susan T. Fiske et al.

have a greater impact on the perceiver’s impressions than other target features is compatible with the continuum model. Second, Kunda and Thagard (1996) criticize the serial nature of the impression formation processes located on our continuum, noting that a parallel model can account for the existing laboratory data at least as well. Several points apply: Information about others usually arrives (or is captured via our limitedcapacity attentional system) sequentially; rarely do people in the real world simultaneously apprehend all the many pieces of information that Kunda and Thagard include in their modeling simulations. The continuum model was designed to capture this kind of active, thoughtful impression formation, and so we are unapologetic about its roughly serial nature. Note further that when the parallel-constraint-satisfaction model is made to accept information serially (as in Kunda and Thagard’s simulation of the primacy effect), it begins to look much like the continuum model in the predictions it makes. We should also note that Kunda and Thagard’s (1996) critique of the serial nature of the continuum model rests on several unusual features of their own model. Their simulations presume that all information is available to the perceiver simultaneously, that each piece of information receives the identical amount of attention, and that perceivers do not form impressions until all the information is in. Such assumptions may approximate the conditions in sparse experimental paradigms, but they do not capture most impression formation in the real world. Indeed, their model would generate perceiver responses quite compatible with the predictions of the continuum model (1) if it received target features in the order in which they typically appear (e.g., physical features such as race and gender, followed by social behaviors, followed by inferences about the traits these behaviors represent); (2) if these features received different amounts of processing as a function of their frequency and recency of activation, the perceiver’s goals and attentional capacity, and so forth; and (3) if the simulation were to “read out” the cognitive and affective responses in a more continuous manner. In particular, the flow from initial categorization through recategorization would be easily discerned if one were to model these more realistic conditions (see Bodenhausen et al., 1999). Moreover, the continuum model is entirely compatible with an implementation in which categorization processes and more attribute-based processes are set in motion simultaneously, but the categorization processes are faster and can constitute a “stop rule” for the more attribute-based processes if categorization results seem adequate for present purposes. In this parallel implementation (anticipated by Fiske & Neuberg, 1990), the categories and other attributes begin to interplay after the initial categorization, with the categorical processes carrying less relative weight as the attribute-oriented processes carry more relative weight, but both proceed in parallel. In other respects, we have been necessarily brief here and certainly not as thorough as we would like. For instance, space does not allow us to explore the commonalities between the process of constraint satisfaction and the process

The continuum model: Ten years later 65 of information fit assessment, or to address how the experimental procedures and demands of many laboratory studies might artifactually increase the apparent fit of Kunda and Thagard’s data to their model. Finally, although they are not alone in this, note that Kunda and Thagard (1996) fail to appreciate the dynamic nature of the continuum model. As we conceive of it, the impression formation process not only continually cycles back to the confirmatory categorization stage (using in later iterations the most recently accepted categorization as its foundation) as each piece of new information is attended, but also always evokes perceiver impressions appropriate to its last accepted categorization. Such a continually flowing system seems quite amenable to a parallel-processing mechanism—albeit one nested in a roughly serial information-gathering system. In sum, the alleged incompatibilities between the parallel-constraintsatisfaction model and the continuum model are less real than they may appear. Indeed, an integration of the two models could capitalize on each model’s strengths. The Kunda and Thagard model provides a mechanism by which target features can constrain one another’s meanings; a means of capturing the flow from initial categorization to recategorization in a more obviously dynamic way; and a thoughtful consideration of how response modalities (e.g., inferences about traits versus behavioral expectations) might respond differently, given the identical target information. The continuum model posits the factors that influence which categories become activated; articulates how motivation and attention bring target information roughly serially into the system, and differentially focus the perceiver on some information at the expense of other information; and so on. Such an integration would move us beyond debate (serial processing vs. parallel processing) to a resolution (serial processing and parallel processing).

Comparisons with other dual-process models In this chapter, we have first described the historical context and reviewed the specific stages of the continuum model; have then described its five core premises and some of the new research support for them; and, finally, have examined some of its meta-assumptions about situational and individual differences in motives, along with its emphasis on both expectations and information, and its compatibility with both serial and parallel implementations. In conclusion, we wish to note the compatibility and contrast of the model with some other dual-process models reviewed in Chaiken and Trope (1999). The continuum model is superficially closest to Brewer’s model of impression formation (Brewer & Harasty, 1999), in that both propose alternative sequences or stages of processing, with more category-based processes dominant. The essential differences (Fiske, 1988) include (1) Brewer’s emphasis on distinct types of cognitive representation versus our emphasis on one type that is modified online; (2) her emphasis on distinct inference rules following each stage, versus our standard reliance on informational fit and motivational involvement at each

66

Susan T. Fiske et al.

stage; (3) our focus on attentional and interpretive processes as the single, unifying mediator; (4) Brewer’s use of a branching model, versus our continuum; and (5) our more explicit emphasis on motivations and goals. Nevertheless, the similarities remain, in that both models synthesize a variety of literatures in person perception and stereotyping. The continuum model also bears some resemblance to the heuristic– systematic model (Chen & Chaiken, 1999) and the elaboration likelihood model (Petty & Wegener, 1999) of attitudes. The similarities include the specification of more automatic, superficial processes versus more in-depth, analytic processes, with the joint role of informational and motivational factors in both. One of the major differences, of course, is that the continuum model focuses on the cues provided by people, not the cues provided by persuasive communications. Moreover, whereas categorization plays a major role in the continuum model, it is mostly absent in the heuristic–systematic model and the elaboration likelihood model. Like the two attitude models, the continuum model provides a role for multiple types of motivation, but its most frequently studied motivations have been those most relevant to interactions between people. The continuum model also has some similarities to another model of attitudes— namely, Fazio’s “motivation and opportunity as determinants” (MODE) model (Fazio & Towles-Schwen, 1999). Both models posit affective associations to attitude objects (in our case, people), which are quickly triggered upon each encounter. Although schema-triggered affect is a central theoretical feature of the continuum model (Fiske, 1982), research on this model has never pursued the processing dynamics of such affect, whereas the MODE research has pursued these implications in detail. In short, some common themes underlie the various dual-process models, but they fit different circumstances.

Notes 1. A 2017 update shows more than 3500 citations.

References Altemeyer, B. (1981). Right-wing authoritarianism. Winnipeg: University of Manitoba Press. Altemeyer, B. (1988). Enemies of freedom. San Francisco: Jossey-Bass. Anderson, N. H. (1981). Foundations of information integration theory. New York: Academic Press. Asch, S. E. (1946). Forming impressions of personality. Journal of Abnormal and Social Psychology, 41, 303–314. Asch, S. E., & Zukier, H. (1984). Thinking about persons. Journal of Personality and Social Psychology, 46, 1230–1240. Banaji, M. R., & Hardin, C. (1996). Automatic stereotyping. Psychological Science, 7, 136–141. 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.

The continuum model: Ten years later 67 Biernat, M., & Vescio, T. K. (1993). Categorization and stereotyping: Effects of group context on memory and social judgment. Journal of Experimental Social Psychology, 29, 166–202. Biesanz, J. C., Neuberg, S. L., Smith, D. M., Asher, T., & Judice, T. N. (2001). When accuracy-motivated perceivers fail: Limited attentional resources and the reemerging self-fulfilling prophecy. Personality and Social Psychology Bulletin, 27(5), 621–629. Blair, I. V., Banaji, M. R. (1996). Automatic and controlled processes in stereotype priming. Journal of Personality and Social Psychology, 70, 1126–1141. Blanchard, F. A., Crandall, C. S., Brigham, J. C., & Vaughn, L. A. (1994). Condemning and condoning racism: A social context approach to interracial settings. Journal of Applied Social Psychology, 79, 993–997. Blanchard, F. A., Lilly, T., & Vaughn, L. A. (1991). Reducing the expression of racial prejudice. Psychological Science, 2, 101–105. Bless, H., & Schwarz, N. (1999). Sufficient and necessary conditions in dual-process models: The case of mood and information processing. In S. Chaiken & Y. Trope (Eds.) Dual process theories in social psychology (pp. 423–440). New York: Guilford. Bodenhausen, G. V. (1990). Stereotypes as judgmental heuristics: Evidence of circadian variations in discrimination. Psychological Science, 1(5), 319–322. Bodenhausen, G. V., Kramer, G. P., & Susser, K. (1994). Happiness and stereotypic thinking in social judgment. Journal of Personality and Social Psychology, 66, 621–632. Bodenhausen, G. V., Macrae, C. N., & Sherman, J. W. (1999). On the dialectics of discrimination: Dual processes in social stereotyping. In S. Chaiken & Y. Trope (Eds.) Dual process theories in social psychology (pp. 271–291). New York: Guilford. Bodenhausen, G. V., Sheppard, L. A., & Kramer, G. P. (1994). Negative affect and social judgments: The differential impact of anger and sadness. European Journal of Social Psychology, 24, 45–62. 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, 4th ed., pp. 554–594). New York: McGraw-Hill. Brewer, M. B., & Harasty Feinstein, A. S. (1999). Dual processes in the cognitive representations of persons and social categories. In S. Chaiken & Y. Trope (Eds.) Dual process theories in social psychology (pp. 255–270). New York: Guilford. Burger, J. M. (1992). Desire for control: Personality, social, and clinical perspectives. New York: Plenum. Burger, J. M., & Hemans, L. T. (1988). Desire for control and the use of attribution processes. Journal of Personality, 56, 531–546. Cacioppo, J. T., & Petty, R. E. (1982). The need for cognition. Journal of Personality and Social Psychology, 42, 116–131. Cacioppo, J. T., Petty, R. E., Feinstein, J. A., & Jarvis, W. B. G. (1996). Dispositional differences in cognitive motivation: The life and times of individuals varying in need for cognition. Psychological Bulletin, 119, 197–253. Caporael, L. R. (1997). The evolution of truly social cognition: The core configurations model. Personality and Social Psychology Review, 1, 276–298. Caporael, L. R., & Brewer, M. B. (1991). Reviving evolutionary psychology: Biology meets society. Journal of Social Issues, 47, 187–195. Chaiken, S., & Trope, Y. (Eds.) (1999). Dual process theories in social psychology. New York: Guilford. Chen, S., & Chaiken, S. (1999). The heuristic-systematic model in its broader context. In S. Chaiken & Y. Trope (Eds.) Dual process theories in social psychology (pp. 73–96). New York: Guilford.

68

Susan T. Fiske et al.

Chen, S., Schechter, D., & Chaiken, S. (1996). Getting the truth or getting along: Accuracy- vs. impression-motivated heuristic and systematic processing. Journal of Personality and Social Psychology, 71, 262–275. Claire, T., & Fiske, S. T. (1998). A systemic view of behavioral confirmation: Counterpoint to the individualist view. In C. Sedikides, C. Insko, & J. Schopler (Eds.), Intergroup cognition and behavior (pp. 205–232). Mahwah, NJ: Erlbaum. Copeland, J. T. (1994). Prophecies of power: Motivational implications of social power for behavioral confirmation. Journal of Personality and Social Psychology, 67, 264–277. Crocker, J., & Luhtanen, R. (1990). Collective self-esteem and ingroup bias. Journal of Personality and Social Psychology, 58, 60–67. De Dreu, C. K. W., Yzerbyt, V. Y., & Leyens, J.-P. (1995). Dilution of stereotypebased cooperation in mixed-motive interdependence. Journal of Experimental Social Psychology, 31, 575–593. Dépret, E. F., & Fiske, S. T. (1993). Social cognition and power: Some cognitive consequences of social structure as a source of control deprivation. In G. Weary, F. Gleicher, & K. Marsh (Eds.), Control motivation and social cognition (pp. 176–202). New York: Springer-Verlag. Dépret, E. F., & Fiske, S. T. (1999). Perceiving the powerful: Intriguing individuals versus threatening groups. Journal of Experimental Social Psychology, 35, 461–480. Desforges, D. M., Lord, C. G., Ramsey, S. L., Mason, J. A., Van Leeuwen, M. D., West, S. L., & Lepper, M. R. (1991). Effects of structured cooperative contact on changing negative attitudes toward stigmatized social groups. Journal of Personality and Social Psychology, 60, 531–544. Devine, P. G. (1989). Stereotypes and prejudice: Their automatic and controlled components. Journal of Personality and Social Psychology, S6, 5–18. 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. (1993). Stereotypes and evaluative intergroup bias. In D. M. Mackie & D. L. Hamilton (Eds.), Affect, cognition, and stereotyping (pp. 167–193). San Diego, CA: Academic Press. Eckes, T. (1994). Explorations in gender cognition: Content and structure of female and male subtypes. Social Cognition, 12, 37–60. Edwards, J. A., & Weary, G. (1993). Depression and the impression-formation continuum: Piecemeal processing despite the availability of category information. Journal of Personality and Social Psychology, 64, 636–645. Erber, R., & Fiske, S. T. (1984). Outcome dependency and attention to inconsistent information. Journal of Personality and Social Psychology, 47, 709–726. Fazio, R. H., Jackson, J. R., Dunton, B. C., & Williams, B. J. (1995). Variability in automatic activation as an unobtrusive measure of racial attitudes: A bona fide pipeline? Journal of Personality and Social Psychology, 69, 1013–1027. Fazio, R. H., & Towles-Schwen, T. (1999). The MODE model of attitude-behavior processes. In S. Chaiken & Y. Trope (Eds.), Dual process theories in social psychology (pp. 97–116). New York: Guilford. Fiske, S. T. (1980). Attention and weight in person perception: The impact of negative and extreme behavior. Journal of Personality and Social Psychology, 38, 889–906. Fiske, S. T. (1982). Schema-triggered affect: Applications to social perception. In M. S. Clark & S. T. Fiske (Eds.), Affect and cognition: The 17th annual Carnegie Symposium on Cognition (pp. 55–78). Hillsdale, NJ: Erlbaum.

The continuum model: Ten years later 69 Fiske, S. T. (1988). Compare and contrast: Brewer’s dual-process model and Fiske et al.’s continuum model. In T. K. Srull & R. S. Wyer (Eds.), Advances in social cognition: Vol. 1. A dual model of impression formation (pp. 65–76). Hillsdale, NJ: Erlbaum. Fiske, S. T. (1993a). Controlling other people: The impact of power on stereotyping. American Psychologist, 48, 621–628. Fiske, S. T. (1993b). Social cognition and social perception. Annual Review of Psychology, 44, 155–194. Fiske, S. T. (1998). Stereotyping, prejudice, and discrimination. In D. T. Gilbert, S. T. Fiske, & G. Lindzey (Eds.), The handbook of social psychology (Vol. 2, 4th ed., pp. 367–411). New York: McGraw-Hill. Fiske, S. T., & Dépret, E. (1996). Control, interdependence, and power: Understanding social cognition in its social context. In W. Stroebe & M. Hewstone (Eds.), European review of social psychology (Vol. 7, pp. 31–61). New York: Wiley. Fiske, S. T., & Leyens, J.-P. (1996). Let social psychology be faddish, or at least heterogenous. In C. McGarty & S. A. Haslam (Eds.), Message of social psychology: Perspectives on mind in society (pp. 92–112). Oxford: Blackwell. Fiske, S. T., Morling, B. A., & Stevens, L. E. (1996). Controlling self and others: A theory of anxiety, mental control, and social control. Personality and Social Psychology Bulletin, 22, 115–123. 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: Academic Press. Fiske, S. T., Neuberg, S. L., Beattie, A. E., & Milberg, S. J. (1987). Category-based and attribute-based reactions to others: Some informational conditions of stereotyping and individuating processes. Journal of Experimental Social Psychology, 23, 399–427. Fiske, S. T., & Pavelchak, M.A. (1986). Category-based versus piecemeal-based affective responses: Developments in schema-triggered affect. In R. M. Sorrentino & E. T. Higgins (Eds.), Handbook of motivation and cognition: Foundations of social behavior (Vol. 1, pp. 167–203). New York: Guilford Press. Fiske, S. T., & Taylor, S. E. (1991). Social cognition (2nd ed.). New York: McGrawHill. Fiske, S. T., & Von Hendy, H. M. (1992). Personality feedback and situational norms can control stereotyping processes. Journal of Personality and Social Psychology, 62, 577–596. Freund, T., Kruglanski, A. W., & Shpitzajzen, A. (1985). The freezing and unfreezing of impression primacy: Effects of the need for structure and the fear of invalidity. Personality and Social Psychology Bulletin, 11, 479–487. Gaertner, S. L., & Dovidio, J. F. (1986). The aversive form of racism. In J. F. Dovidio & S. L. Gaertner (Eds.), Prejudice, discrimination, and racism (pp. 61–89). San Diego, CA: Academic Press. Gaertner, S. L., & McLaughlin, J. P. (1983). Racial stereotypes: Associations and ascriptions of positive and negative characteristics. Social Psychology Quarterly, 46, 23–30. Gardner, R. C., MacIntyre, P. D., & Lalonde, R. N. (1995). The effects of multiple social categories on stereotyping. Canadian Journal of Behavioural Science, 27, 466–483. Gilbert, D. T., & Hixon, J. G. (1991). The trouble of thinking: Activation and application of stereotypic beliefs. Journal of Personality and Social Psychology, 60, 509–517.

70

Susan T. Fiske et al.

Glick, P., & Fiske, S. T. (1996). The Ambivalent Sexism Inventory: Differentiating hostile and benevolent sexism. Journal of Personality and Social Psychology, 70, 491–512. Gollwitzer, P. M., & Kinney, R. F. (1989). Effects of deliberative and implemental mind-sets on illusion of control. Journal of Personality and Social Psychology, 56, 531–542. Goodwin, S. A., Gubin, A., Fiske, S. T., & Yzerbyt, V. (1998). Power can bias impression formation: How powerholders stereotype by default and by design. Unpublished manuscript, Boston College. Hamilton, D. L., Sherman, S. J., & Ruvolo, C. M. (1990). Stereotype-based expectancies: Effects on information processing and social behavior. Journal of Social Issues, 46, 35–60. Harris-Kern, M. J., & Perkins, R. (1995). Effects of distraction on interpersonal expectancy effects: A social interaction test of the cognitive busyness hypothesis. Social Cognition, 13, 163–182. Haslam, S. A., Oakes, P. J., McGarty, C., Turner, J. C., & Onorato, R. S. (1995). Contextual changes in the prototypicality of extreme and moderate outgroup members. European Journal of Social Psychology, 25, 509–530. Hattrup, K., & Ford, J. K. (1995). The roles of information characteristics and accountability in moderating stereotype-driven processes during social decision making. Organizational Behavior and Human Decision Processes, 63, 73–86. Heaton, A. W., & Kruglanski, A. W. (1991). Person perception by introverts and extraverts under time pressure: Effects of need for closure. Personality and Social Psychology Bulletin, 17, 161–165. Heilman, M. C. (1995). Sex stereotypes and their effects in the workplace: What we know and what we don’t know. Journal of Social Behavior and Personality, 10, 3–26. Hewstone, M., Hantzi, A., & Johnston, L. (1991). Social categorization and person memory: The pervasiveness of race as an organizing principle. European Journal of Social Psychology, 21, 517–528. Hilton, J. L., Klein, J. G., & von Hippel, W. (1991). Attention allocation and impression formation. Personality and Social Psychology Bulletin, 17, 548–559. Jackson, L. A., Hansen, C. H., Hansen, R. D., & Sullivan, L. A. (1993). The effects of stereotype consistency and consensus information on predictions of performance. Journal of Social Psychology, 133, 293–306. James, H. (1983). The portrait of a lady. New York: Bantam. (Original work published 1881.) Jamieson, D. W., & Zanna, M. P. (1989). Need for structure in attitude formation and expression. In A. R. Pratkanis, S. J. Breckler, & A. G. Greenwald (Eds.), Attitude structure and function (pp. 383–406). Hillsdale, NJ: Erlbaum. Johnston, L. C., Hewstone, M., Pendry, L., & Frankish, C. (1994). Cognitive models of stereotype change: Motivational and cognitive influences. European Journal of Social Psychology, 24, 237–265. Jussim, L. (1991). Social perception and social reality: A reflection-construction model. Psychological Review, 98, 54–73. Kaplan, M. F., Wanshula, L. T., & Zanna, M. P. (1993). Time pressure and information integration in social judgment: The effect of need for structure. In O. Svenson & A. J. Maule (Eds.), Time pressure and stress in human judgment and decision making (pp. 255–267). New York: Plenum Press. Kashima, Y., & Kerekes, A. R. Z. (1994). A distributed memory model of averaging phenomena in person impression formation. Journal of Experimental Social Psychology, 30, 407–455.

The continuum model: Ten years later 71 Katz, I., & Hass, R. G. (1988). Racial ambivalence and value conflict: Correlational and priming studies of dual cognitive structures. Journal of Personality and Social Psychology, 55, 893–905. Kim, H., & Baron, R. S. (1988). Exercise and the illusory correlation: Does arousal heighten stereotypic processing? Journal of Experimental Social Psychology, 24, 366–380. Klinger, M. R., & Beall, P. M. (1992). Conscious and unconscious effects of stereotype activation. Paper presented at the 64th Annual Meeting of the Midwestern Psychological Association, Chicago. Kruglanski, A. W., & Freund, T. (1983). The freezing and unfreezing of lay-inferences: Effects of impressional primacy, ethnic stereotyping, and numerical anchoring. Journal of Experimental Social Psychology, 19, 448–468. Kruglanski, A. W., & Mayseless, O. (1988). Contextual effects in hypothesis testing: The role of competing alternatives and epistemic motivations. Social Cognition, 6, 1–20. Kruglanski, A. W., & Webster, D. M. (1991). Group members’ reactions to opinion deviates and conformists at varying degrees of proximity to decision deadline and use of environmental noise. Journal of Personality and Social Psychology, 61, 212–225. Kruglanski, A. W., & Webster, D. M. (1996). Motivated closing of the mind: “Seizing” and “freezing.” Psychological Review, 103, 263–283. Kunda, Z., Miller, D. T., & Claire, T. (1990). Combining social concepts: The role of causal reasoning. Cognitive Science, 14, 551–577. Kunda, Z., & Oleson, K. C. (1995). Maintaining stereotypes in the face of disconfirmation: Constructing grounds for subtyping deviants. Journal of Personality and Social Psychology, 68, 565–579. Kunda, Z., & Thagard, P. (1996). Forming impressions from stereotypes, traits, and behaviors: A parallel-constraint-satisfaction theory. Psychological Review, 103, 284–308. Lambert, A. J. (1995). Stereotypes and social judgment: The consequences of group variability. Journal of Personality and Social Psychology, 68, 388–403. Leippe, M. R., & Eisenstadt, D. (1994). Generalization of dissonance reduction: Decreasing prejudice through induced compliance. Journal of Personality and Social Psychology, 67, 395–413. Lepore, L., & Brown, R. (1997). Category and stereotype activation: Is prejudice inevitable? Journal of Personality and Social Psychology, 72, 275–287. Levine, G. M., Halberstadt, J. B., & Goldstone, R. L. (1996). Reasoning and the weighting of attributes in attitude judgments. Journal of Personality and Social Psychology, 70, 230–240. Leyens, J.-P., & Fiske, S. T. (1994). Impression formation: From recitals to symphonie fantastique. In P. G. Devine, D. L. Hamilton, & T. M. Ostrom (Eds.), Social cognition: Impact on social psychology (pp. 39–75). San Diego, CA: Academic Press. Leyens, J.-P., & Yzerbyt, V. Y. (1992). The ingroup overexclusion effect: Impact of valence and confirmation on stereotypical information search. European Journal of Social Psychology, 22, 549–569. 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 Social Psychology, 69, 397–407. Macrae, C. N., Bodenhausen, G. V., Milne, A. B., & Jetten, J. (1994). Out of mind but back in sight: Stereotypes on the rebound. Journal of Personality and Social Psychology, 67, 808–817. Macrae, C. N., Hewstone, M., & Griffiths, R. J. (1993). Processing load and memory for stereotype-based information. European Journal of Social Psychology, 23, 77–87.

72

Susan T. Fiske et al.

Macrae, C. N., Milne, A. B., & Bodenhausen, G. V. (1994). Stereotypes as energysaving devices: A peek inside the cognitive toolbox. Journal of Personality and Social Psychology, 66, 37–47. Macrae, C. N., Shepherd, J. W., & Milne, A. B. (1992). The effects of source credibility on the dilution of stereotype-based judgments. Personality and Social Psychology Bulletin, 18, 765–775. Macrae, C. N., Stangor, C., & Milne, A. B. (1994). Activating social stereotypes: A functional analysis. Journal of Experimental Social Psychology, 30, 370–389. Marsh, K. L., & Weary, G. (1994). Severity of depression and responsiveness to attributional information. Journal of Social and Clinical Psychology, 13, 15–32. Matlin, M., & Stang, D. (1978). The Pollyanna principle. Cambridge, MA: Schenkman. McCann, C. D., Ostrom, T. M., Tyner, L. K., & Mitchell, M. L. (1985). Person perception in heterogeneous groups. Journal of Personality and Social Psychology, 49, 1449–1459. McConahay, J. B., & Hough, J. C., Jr. (1976). Symbolic racism. Journal of Social Issues, 32, 23–45. Miki, H., Tsuchiya, H., & Nishino, A. (1993). Influences of expectancy of opponents’ competence upon information processing of their discrete attributes. Perceptual and Motor Skills, 77, 987–993. Mikulincer, M., Yinon, A., and Kabili, D. (1991). Epistemic needs and learned helplessness. European Journal of Personality, 5, 249–258. Moberg, S. (1995). Impact of teachers’ dogmatism and pessimistic stereotype on the effect of EMR-class label on teachers’ judgments in Finland. Education and Training in Mental Retardation and Developmental Disabilities, 30, 141–150. Moskowitz, G. B. (1993). Individual differences in social categorization: The influence of personal need for structure on spontaneous trait inference. Journal of Personality and Social Psychology, 65, 132–142. Naccarato, M. E. (1988). The impact of need for structure on stereotyping and discrimination. Unpublished master’s thesis, University of Waterloo, Waterloo, Ontario, Canada. Nelson, T. E., Acker, M., & Manis, M. (1996). Irrepressible stereotypes. Journal of Experimental Social Psychology, 32, 13–38. Neuberg, S. L. (1989). The goal of forming accurate impressions during social interactions: Attenuating the impact of negative expectancies. Journal of Personality and Social Psychology, 56, 374–386. Neuberg, S. L., & Fiske, S. T. (1987). Motivational influences on impression formation: Outcome dependency, accuracy-driven attention, and individuating processes. Journal of Personality and Social Psychology, 53, 431–444. Neuberg, S. L., Judice, T. N., Virdin, L. M., & Carrillo, M. A. (1993). Perceiver selfpresentational goals as moderators of expectancy influences: Ingratiation and the disconfirmation of negative expectancies. Journal of Personality and Social Psychology, 64, 409–420. Neuberg, S. L., & Newsom, J. T. (1993). Personal need for structure: Individual differences in the desire for simpler structure. Journal of Personality and Social Psychology, 65, 113–131. Oakes, P. J., & Reynolds, K. J. (1997). Asking the accuracy question: Is measurement the answer? In R. Spears, P. Oakes, N. Ellemers, & S. A. Haslam (Eds.), The social psychology of stereotyping and group life (pp. 51–71). Oxford: Blackwell. Oakes, P. J., Turner, J. C., & Haslam, S. A. (1991). Perceiving people as group members: The role of fit in the salience of social categorizations. British Journal of Social Psychology, 30, 125–144.

The continuum model: Ten years later 73 Operario, D., & Fiske, S. T. (2001). Effects of trait dominance on powerholders’ judgments of subordinates. Social Cognition, 19, 161–180. Osborne, R. F., & Gilbert, D. T. (1992). The preoccupational hazards of social life. Journal of Personality and Social Psychology, 62, 219–228. Ostrom, T. M. (1977). Between-theory and within-theory conflict in explaining context effects in impression formation. Journal of Experimental Social Psychology, 13, 492–503. Ostrom, T. M. (1984). The sovereignty of social cognition. In R. S. Wyer, Jr., & T. K. Srull (Eds.), Handbook of social cognition (Vol. 1, pp. 1–38). Hillsdale, NJ: Erlbaum. Parducci, A. (1968). The relativism of absolute judgments. Scientific American, 219, 84–90. Pendry, L. F., & Macrae, C. N. (1994). Stereotypes and mental life: The case of the motivated but thwarted tactician. Journal of Experimental Social Psychology, 30, 303–325. Pendry, L. F., & Macrae, C. N. (1996). What the disinterested perceiver overlooks: Goal-directed social categorization. Personality and Social Psychology Bulletin, 22, 249–256. Perdue, C. W., & Gurtman, M. B. (1990). Evidence for the automaticity of ageism. Journal ol Experimental Social Psychology, 26, 199–216. Perry, E. (1994). A prototype matching approach to understanding the role of applicant gender and age in the evaluation of job applicants. Journal of Applied Social Psychology, 24, 1433–1473. Petty, R. E., & Wegener, D. T. (1999). The elaboration likelihood model: Current status and controversies. In S. Chaiken & Y. Trope (Eds.), Dual process theories in social psychology (pp. 41–72). New York: Guilford. Pittman, T. S. (1998). Motivation. In D. T. Gilbert, S. T. Fiske, & G. Lindzey (Eds.), The handbook of social psychology (Vol. 1, 4th ed., pp. 549–590). New York: McGrawHill. Pittman, T. S., & D’ Agostino, P. R. (1989). Motivation and cognition: Control deprivation and the nature of subsequent information processing. Journal of Experimental Social Psychology, 25, 465–480. Pittman, T. S., & Heller, J. F. (1987). Social motivation. Annual Review of Psychology, 38, 461–489. Pratto, F., & Bargh, J. A. (1991). Stereotyping based on apparently individuating information: Trait and global components of sex stereotypes under attention overload. Journal of Experimental Social Psychology, 27, 26–47. Pryor, J. B., Giedd, J. L., & Williams, K. B. (1995). A social psychological model for predicting sexual harassment. Journal of Social Issues, 51, 69–84. Resnick, L. B., Levine, J. M., & Teasley, S. D. (Eds.). (1991). Perspectives on socially shared cognition. Washington, DC: American Psychological Association. Riley, T., & Fiske, S. T. (1991). Interdependence and the social context of impression formation. Cahiers de Psychologie Cognitive, 11, 173–192. Roman, R. J., Moskowitz, G. B., Stein, M. I., & Eisenberg, R. F. (1995). Individual differences in experiment participation: Structure, autonomy, and the time of the semester. Journal of Personality, 63(1), 113–138. Rudman, L. A., & Borgida, E. (1995). The afterglow of construct accessibility: The behavioral consequences of priming men to view women as sexual objects. Journal of Experimental Social Psychology, 31, 493–517. Ruscher, J. B., & Fiske, S. T. (1990). Interpersonal competition can cause individuating processes. Journal of Personality and Social Psychology, 58, 832–843. Ruscher, J. B., Fiske, S. T., Miki, H., & Van Manen, S. (1991). Individuating processes in competition: Interpersonal versus intergroup. Personality and Social Psychology Bulletin, 17, 595–605.

74

Susan T. Fiske et al.

Ruscher, J. B., Hammer, E. Y., & Hammer, E. D. (1996). Forming shared impressions through conversation: An adaptation of the continuum model. Personality and Social Psychology Bulletin, 22, 705–720. Schaller, M., Boyd, C., Yohannes, J., & O’Brien, M. (1995). The prejudiced personality revisited: Personal need for structure and formation of erroneous group stereotypes. Journal of Personality and Social Psychology, 68, 544–555. Sears, D. O. (1983). The person-positivity bias. Journal of Personality and Social Psychology, 44, 233–240. Sekaquaptewa, D., & von Hippel, W. (1994, May). The role of prejudice in encoding and memory of stereotype-relevant behaviors. Paper presented at the 66th Annual Meeting of the Midwestern Psychological Association, Chicago. Seta, J. J., & Seta, C. E. (1993). Stereotypes and the generation of compensatory and noncompensarory expectancies of group members. Personality and Social Psychology Bulletin, 19, 722–731. Smith, D. M., Neuberg, S. L., Judice, T. N., & Biesanz, J. C. (1997). Target complicity in the confirmation and disconfirmation of erroneous perceiver expectations: Immediate and longer term implications. Journal of Personality and Social Psychology, 72, 1396–1412. Snodgrass, S. E. (1992). Further effects of role versus gender on interpersonal sensitivity. Journal of Personality and Social Psychology, 62, 154–158. Snyder, M. (1992). Motivational foundations of behavioral confirmation. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 25, pp. 67–114). San Diego, CA: Academic Press. Snyder, M., & Haugen, J. A. (1994). Why does behavioral confirmation occur?: A functional perspective on the role of the perceiver. Journal of Experimental Social Psychology, 30, 218–246. Stangor, C., & Duan, C. (1991). Effects of multiple task demands upon memory for information about social groups. Journal of Experimental Social Psychology, 27, 357–378. Stangor, C., Lynch, L., Duan, C., & Glas, B. (1992). Categorization of individuals on the basis of multiple social features. Journal of Personality and Social Psychology, 62, 207–218. Stephan, W. G., & Stephan, C. W. (1984). The role of ignorance in intergroup relations. In N. Miller & M. B. Brewer (Eds.), Groups in contact: The psychology of desegregation (pp. 229–250). San Diego, CA: Academic Press. Stevens, L. E., & Fiske, S. T. (1995). Motivation and cognition in social life: A social survival perspective. Social Cognition, 13, 189–214. Stevens, L. E., & Fiske, S. T. (2000). Motivated impressions of a powerholder: Accuracy under task dependency and misperception under evaluative dependency. Personality and Social Psychology Bulletin, 26, 907–922. Swim, J. K., Aikin, K. J., Hall, W. S., & Hunter, B. A. (1995). Sexism and racism: Old-fashioned and modern prejudices. Journal of Personality and Social Psychology, 68, 199–214. 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. Tetlock, P. E. (1992). The impact of accountability on judgment and choice: Toward a social contingency model. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 23, pp. 331–376). San Diego, CA: Academic Press.

The continuum model: Ten years later 75 Tetlock, P. E., & Lerner, J. S. (1999). The social contingency model: Identifying empirical and normative boundary conditions on the error-and-bias portrait of human nature. In S. Chaiken & Y. Trope (Eds.), Dual process theories in social psychology (pp. 571–585). New York: Guilford. Tetlock, P. E., Skitka, L., & Boettger, R. (1989). Social and cognitive strategies for coping with accountability: Conformity, complexity, and bolstering. Journal of Personality and Social Psychology, 57, 632–640. Thompson, L. (1995). “They saw a negotiation”: Partisanship and involvement. Journal of Personality and Social Psychology, 68, 839–853. Thompson, E. P., Roman, R. J., Moskowitz, G. B., Chaiken, S., & Bargh, J. A. (1994). Accuracy motivation attenuates covert priming effects: The systematic processing of social information. Journal of Personality and Social Psychology, 66, 474–489. Thompson, M. M., Naccarato, M. E., Parker, K. C. H., & Moskowitz, G. B. (2001). The Personal Need for Structure and the Personal Fear of Invalidity Measures: Historical perspectives, current applications, and future directions. In G. B. Moskowitz (Ed.), Cognitive social psychology: The Princeton symposium on the legacy and future of social cognition (pp. 19–39). Mahwah NJ: Erlbaum. Thompson, M. M., Naccarato, M. E., & Parker, K. E. (1989, June). Assessing cognitive need: The development of the Personal Need for Structure and Personal Fear of Invalidity Scales. Paper presented at the annual meeting of the Canadian Psychological Society, Halifax, Nova Scotia, Canada. Tougas, F., Brown, R., Beaton, A. M., & Joly, S. (1995). Neo-sexism: Plus ça change, plus c’est pareil. Personality and Social Psychology Bulletin, 21, 842–849. Verkuyten, M., Masson, K., & Elffers, H. (1995). Racial categorization and preference among older children in The Netherlands. European Journal of Social Psychology, 25, 637–656. White, R. W. (1959). Motivation reconsidered: The concepts of competence. Psychological Review, 66, 297–333. Wilder, D. A., & Shapiro, P. (1989a). Effects of anxiety on impression formation in a group context: An anxiety-assimilation hypothesis. Journal of Experimental Social Psychology, 25, 481–499. Wilder, D. A., & Shapiro, P. (1989b). The role of competition-induced anxiety in limiting the beneficial impact of positive behavior by an out-group member. Journal of Personality and Social Psychology, 56, 60–69. Wittenbrink, W., Judd, C. M., & Park, B. (1997). Evidence for racial prejudice at the implicit level and its relationship with questionnaire measures. Journal of Personality and Social Psychology, 72, 262–274. Wojciszke, B. (1994). Inferring interpersonal attitudes: Hypotheses and the informationgathering process. European Journal of Social Psychology, 24, 383–401. Zarate, M. A., & Sandoval, P. (1995). The effects of contextual cues on making occupational and gender categorizations. British Journal of Social Psychology, 34, 353–362. Zarate, M. A., & Smith, E. R. (1990). Person categorization and stereotyping. Social Cognition, 8, 161–185. Zebrowitz, L. A., Montepare, J. M., & Lee, H. (1993). They don’t all look alike: Individual impressions of other racial groups. Journal of Personality and Social Psychology, 65, 85–101. Zemore, S. E., & Fiske, S. T. (1998). Who’s controlling me? The role of power in the selfregulation of prejudice. Unpublished manuscript, University of Massachusetts at Amherst.

4

Social science research on trial Use of sex stereotyping research in Price Waterhouse v. Hopkins Susan T. Fiske, Donald N. Bersoff, Eugene Borgida, Kay Deaux, and Madeline E. Heilman

The first Supreme Court case to use psychological research on sex stereotyping was Price Waterhouse v. Hopkins. The case was decided in May 1989 and remanded to Judge Gerhard Gesell, who rendered his final decision in May 1990. In this case, a social psychological expert testified to the antecedent conditions, indicators, consequences, and remedies of stereotyping, on the basis of recent cognitive approaches to stereotyping. The testimony was cited in decisions reached in the trial and appeals courts, as well as in the Supreme Court’s review. The American Psychological Association filed an amicus curiae brief supporting the validity of the field of stereotyping and the general methods used by the expert. Such legal application provides further lessons for psychological research on stereotyping. “The firm of Price Waterhouse refused to make Ann Hopkins a partner. Gender-based stereotyping played a role in this decision” (Hopkins v. Price Waterhouse, 1990, p. 1). Thus concluded federal district court Judge Gerhard Gesell in his May 1990 opinion that followed the United States Supreme Court landmark decision in the same sex discrimination case (Price Waterhouse v. Hopkins, 1989).1 The case is unique because it represents the first use of psychological evidence about sex stereotyping by the Supreme Court. Previous sex discrimination litigation has used only statistical approaches. Moreover, social science evidence has played a significant role in race discrimination cases in the Supreme Court since Brown v. Board of Education (1954), but such evidence had not before been used in sex discrimination cases. Originally filed in 1984, Hopkins v. Price Waterhouse was the first suit to be brought under a new ruling that partnership decisions qualify for protection under Title VII of the 1964 Civil Rights Act (Hishon v. King & Spaulding, 1984). Hopkins v. Price Waterhouse was eventually heard by the Supreme Court, was decided by the Court in May 1989, and was remanded to Judge Gesell, who rendered his final decision in May 1990 to comport with the new rules developed by the nation’s highest court for such cases. The federal court of appeals affirmed Judge Gesell’s opinion in December 1990. Testimony about the psychology of stereotyping played a crucial role at each stage of the litigation as it made its way through the judicial

Social science research on trial 77 process. The American Psychological Association (APA) submitted to the Supreme Court an amicus curiae brief (which follows the original article). As we will discuss later, the brief had a significant impact on that Court’s final decision. The relevant psychological literature was heavily cited in Judge Gesell’s original decision (Hopkins v. Price Waterhouse, 1985), and the testimony about the psychology of stereotyping was cited at all levels of the appeal and review process, including the Supreme Court’s decision and the subsequent remand. Moreover, the APA brief clearly contributed to the Supreme Court’s opinion regarding the credibility of this area of research. Now that the case has been decided, it is appropriate to describe and evaluate both the role of psychology in this process and what the drafters of the amicus brief learned about our field as a result.

Factual background of the Hopkins case In 1982, it seemed that Ann Hopkins had established her credentials as a topnotch performer at Price Waterhouse (PW), one of the nation’s big-eight accounting firms: She had more billable hours than any other person proposed for partner that year, she had brought in business worth $25 million, her clients praised her, and her supporters recommended her as driven, hard-working, and exacting. She was the only woman of 88 candidates proposed that year; of 662 partners at PW, only 7 were women. Instead of being promoted for her accomplishments, her candidacy was put on hold, and she was not proposed for partner the following year. Hopkins alleged that she was denied partnership because of her gender. Price Waterhouse countered that she was not admitted because she had interpersonal skills problems. According to some evaluators, this “lady partner candidate” was “macho,” she “overcompensated for being a woman,” and she needed a “course at charm school.” A sympathetic colleague advised that she would improve her chances if she would “walk more femininely, talk more femininely, dress more femininely, wear make-up, have her hair styled, and wear jewelry” (Hopkins v. Price Waterhouse, 1985, p. 1117). Instead, Hopkins took the firm to court. In 1982, she filed a complaint in the federal district court of the District of Columbia, alleging a violation of Title VIl of the 1964 Civil Rights Act. She had a strong case in many respects, but her attorneys, Douglas Huron and James Heller, still needed to demonstrate that these stereotypic remarks might account for discriminatory decision making. Huron and Heller had heard about attorney Sarah Burns’s novel construction of a prior sex discrimination case, using current research on sex stereotyping, particularly cognitive categorization theories (e.g., Allport, 1954; Ashmore & Del Boca, 1981; Hamilton, 1979; Tajfel, 1972; Taylor, 1981; for a review of subsequent contributions, see Fiske & Taylor, 1991, chap. 4, 5). Although that case had been settled before trial, the deposition by social psychologist Susan T. Fiske was viewed by the plaintiff’s attorneys as contributing to a favorable outcome for their client. Huron and

78

Susan T. Fiske et al.

Heller asked Fiske to testify about the psychology of stereotyping, with particular reference to the facts of their case. When they described their evidence of stereotyping, Fiske agreed, because, in her opinion, the case so closely fit the literature on sex stereotyping. Moreover, it seemed an excellent test of the utility of social psychological theories and research. Fiske’s testimony in the original Hopkins v. Price Waterhouse case drew on both laboratory and field research to describe antecedent conditions that encourage stereotyping, indicators that reveal stereotyping, consequences of stereotyping for out-groups, and feasible remedies to prevent the intrusion of stereotyping into decision making. Specifically, she testified first that stereotyping is most likely to intrude when the target is an isolated, one- or few-ofa-kind individual in an otherwise homogeneous environment. The person’s solo or near-solo status makes the unusual category more likely to be a salient factor in decision making (e.g., Crocker & McGraw, 1982; Heilman, 1980; Kanter, 1977; McArthur & Post, 1977; Spangler, Gordon, & Pipkin, 1978; S. E. Taylor, 1981; Wolman & Frank, 1975; for recent reviews, see Fiske & Taylor, 1991, chap. 7; Mullen, 1991; Pettigrew & Martin, 1987). Ann Hopkins qualified as a near-solo in the organization. Stereotyping is also likely to intrude when members of a previously omitted group move into a job that is nontraditional for their group. This is certainly true for women who are senior managers and partners in the big-eight accounting firms. The maximum percentage of women partners in a big-eight firm as of 1989 was 5.6%; PW had the lowest with 2% (Berg, 1988). As of May 1990, PW had 27 female partners out of 900 total, or 3% (Lewin, 1990). Another antecedent condition concerns the perceived lack of fit between the person’s category and occupation (Heilman, 1983). The attributes desirable in a manager—aggressive, competitive, driven, tough, and masterly—are not attributes typically expected of women (e.g., Heilman, Block, Martell, & Simon, 1989; Schein, 1973, 1975). Women who behave in those managerial ways are often disliked (e.g., Brown & Geis, 1984; Costrich, Feinstein, Kidder, Marecek, & Pascale, 1975; Deaux & Lewis, 1983, 1984; Hagen & Kahn, 1975; Heilman et al., 1989; for a recent meta-analytic review, see Eagly, Makhijani, & Klonsky, 1992) and create dissatisfaction among their subordinates (Petty & Lee, 1975; Rousell, 1974). The testimony also addressed antecedents regarding the information environment. Stereotyping is most likely when evaluative criteria are ambiguous (for reviews, see Arvey, 1979; Kanter, 1977; Nieva & Gutek, 1980; as a general principle of schema use, see Fiske & Neuberg, 1990; Fiske & Taylor, 1991, chap. 4, 5; Markus & Zajonc, 1985; Nisbett & Ross, 1980). Moreover, when information about the individual is ambiguous, it is most open to interpretation (e.g., J.M. Darley & Gross, 1983; Heilman, 1984; Heilman, Martell, & Simon, 1988; Locksley, Borgida, Brekke, & Hepburn, 1980, Study 2; Pheterson, Kiesler,·& Goldberg, 1971; Rasinski, Crocker, & Hastie, 1985; for reviews, see Arvey, 1979; Nieva & Gutek, 1980; Tosi & Einbender, 1985). Stereotypes provide structure and meaning, and they shape perceptions most

Social science research on trial 79 when the data themselves are open to multiple interpretations, as suggested by research on the cognitive bases of stereotyping (see Fiske & Taylor, 1991, chap. 4, 5, for a review). For example, a “counting” decision—based on millions of business dollars—is relatively immune to stereotypic biases. However, subjective judgments of interpersonal skills and collegiality are quite vulnerable to stereotypic biases. Cognitive models of stereotyping attest to the effects of well-developed expectancies and stereotypes on the interpretation of ambiguous information. This is not to say that decision makers should not use subjective criteria; merely that one must be alert to the possibility of stereotyping in their application. The testimony also indicated that the symptoms or indicators of cognitive processes that give rise to stereotyping are straightforward: categorical responding, such as unnecessarily labeling and evaluating someone according to gender. In Ann Hopkin’s case, people commented that her behavior was evaluated differently “because it’s a woman doing it,” and they suggested that she “overcompensated for being a woman” (for reviews, see Allport, 1954; Brewer, 1988; Deaux & Kite, 1993; Fiske & Neuberg, 1990; Fiske & Taylor, 1991, chap. 4; Tajfel, 1969). Another symptom of category-based judgment is evaluating people’s credentials along dimensions narrowly relevant to their group’s stereotype. For example, the sex role appropriateness of Ann Hopkin’s social skills, instead of her business-generating abilities, became the primary dimension along which she was evaluated. Selective perception and interpretation are also indicators of stereotyping. Hopkin’s detractors saw her as an aggressive woman and therefore abrasive and difficult, but her supporters and her clients saw her simply as a determined gogetter (for a gender-relevant example, see Taylor, Fiske, Etcoff, & Ruderman, 1978; as a general process in stereotyping, see reviews by Fiske & Taylor, 1991, chap. 4; Hamilton, 1979, 1981; Higgins & Bargh, 1987; Markus & Zajonc, 1985; Nisbett & Ross, 1980, chap. 2, 8). Finally, stereotyping is also indicated by extreme, polarized evaluations based on limited evidence. For example, some acquaintances claimed that Hopkins was “universally disliked,” which was demonstrably not true (e.g., Kanter, 1977; Taylor, 1981; Wolman & Frank, 1975; for a review, see Mullen, 1991). The testimony dealt with other examples, but these examples illustrate the sense of the argument. The testimony also noted that the consequences of such stereotyping are obvious. Evaluations are based on category membership, not individual merit (for reviews, see Arvey & Campion, 1982; Heilman, 1983; Nieva & Gutek, 1980; Olian, Schwab, & Haberfeld, 1988; D. N. Ruble & Ruble, 1982; T. L. Ruble, Cohen, & Ruble, 1984; Terborg, 1977). Negative attributes are exaggerated (Kanter, 1977; Linville, 1982; Linville & Jones, 1980; Taylor, 1981; Taylor et al., 1978), and positive ones can be discounted (Deaux, 1976; Deaux & Emswiller, 1974; Feather & Simon, 1975; Feldman-Summers & Kiesler, 1974; Frieze, Fisher, Hanusa, McHugh, & Valle, 1978; Garland & Price, 1977; Hansen

80

Susan T. Fiske et al.

& O’Leary, 1983; Heilman & Guzzo, 1978; Heilman & Stopeck, 1985a, 1985b; Nicholls, 1975; Pazy, 1986; Pence, Pendleton, Dobbins, & Sgro, 1982). Furthermore, there are clear constraints on the permissible behavior of the stereotyped person (e.g., Bartol & Butterfield, 1976; Brown & Geis, 1984; Costrich et al., 1975; S. Darley, 1976; Jago & Vroom, 1982; Kristal, Sanders, Spence, & Helmreich, 1975; Wiley & Eskilson, 1982). Regarding possible remedies, Fiske asserted that stereotyping is controllable (Fiske, 1989a; Fiske & Neuberg, 1990), and people are commonly aware that it is inappropriate. Consequently, adequate information undermines stereotyping (e.g., Deaux & Lewis, 1984; Dipboye & Wiley, 1977; Fiske, Neuberg, Beattie, & Milberg, 1987; Heilman, 1984; Heilman & Martell, 1986; Heilman et al., 1988; Locksley et al., 1980; Pheterson et al., 1971; Renwick & Tosi, 1978; Rasinski et al., 1985; Swim, Borgida, Maruyama, & Myers, 1989). In the PW partnership process, the opinions of people with limited hearsay information were given equal weight with the opinions of people who had more intensive contact. Wider distribution of information about the candidate’s performance would have provided a better basis for decisions. More careful weighting of comments by the level of the person’s knowledge would have improved overall judgments about performance. At a minimum, PW had no policy prohibiting sex discrimination (age and health were noted as impermissible reasons for excluding someone from partnership, but race and sex were not even mentioned). If a policy exists, at least some people will be motivated to comply with it, and it helps to establish counterstereotypic norms. Consistent with this failure to establish organizational norms emphasizing fairness, overt expressions of prejudice were not discouraged at PW. One partner commented that he did not see why they kept proposing women as partners, when women were not even suited to being senior managers. It was significant that no one rebuked him. In effect, the organizational climate can encourage or at least allow stereotypic judgments to go unchecked. Alternatively, it can actively discourage such processes from influencing decision making. As recent research indicates, motivational incentives force people to pay closer attention to their own possible biases (e.g., Erber & Fiske, 1984; Howard-Pitney, Borgida, & Omoto, 1986; Kruglanski & Freund, 1983; Tetlock, 1983; for a review, see Fiske & Neuberg, 1990; Fiske & Taylor, 1991, chap. 5). For example, interdependence research would suggest putting people on teams of mixed compositions, as well as making bonuses for partners and managers contingent on their ability to recruit and nurture subordinates from underrepresented groups. Thus, Fiske argued, these are solvable problems, and PW had not responsibly monitored its own decision making.

Judge Gesell’s response to the psychological testimony The preceding is an organized outline of the testimony, approximately as it was presented on the stand. Judge Gerhard Gesell (the son of the late, noted developmental psychologist Arnold Gesell) was extremely interactive, directing

Social science research on trial 81 the sequence of the testimony, asking many questions, and at times demanding that the expert speak in plain English, rather than in psychological jargon. For example, Judge Gesell was properly impatient with the roundabout way an academic makes a point, and he was openly critical of overly theoretical discursions. The attorneys and the expert had planned a rambling walk through the field of stereotyping, working uphill to the final viewpoint, the way one might in an academic talk or in a journal article, but Judge Gesell preferred that rules of evidence control the testimony: THE WITNESS:

There are general stereotypes of what people particularly expect men to be like and typically expect women to be like. People typically expect women to be strong on the social dimensions. Women are generally expected to be more tender and understanding and concerned about other people, and soft. THE COURT: You say that people who have dealt with women expect that? People who have dealt with women in the business context expect that or are you talking about people out on the farm? THE WITNESS: Well, I would – THE COURT: I mean we have got to talk about people dealing with people in a business context. Does that lady have an opinion that she is going to offer in this case? MR. HURON: Yes, sir. THE COURT: Why doesn’t she give me her opinion? And then tell me what she bases it on. MR. HURON: Fine. We were trying to – THE COURT: And if we did that then I think I would have a better understanding of where you are getting. MR. HURON: Dr. Fiske, have you examined whether stereotyping was occurring at Price Waterhouse; I am talking about sex role stereotyping, at the time and in connection with Ann Hopkins’ proposal for partnership which began in August ‘82 until she was placed on hold in March of 1983? THE WITNESS: I have examined evidence related to that. MR. HURON: Have you formed an opinion as to whether or not stereotyping was occurring? THE WITNESS: Yes, I have. MR. HURON: And what is your opinion? THE WITNESS: I am confident that stereotyping played a role in the decision about Ann Hopkins. THE COURT: Well, now, what kind of role and how confident? Are you able to say that you are confident within a reasonable degree of certainty in your discipline? THE WITNESS: Yes, I would say so, given – THE COURT: All right. That is what I want to know. And then you said it played some part. What part? I don’t know how you would express it in

82

Susan T. Fiske et al.

your discipline percentagewise or how you would express it, but minor, major, middle? I don’t know what the terminology is. THE WITNESS: Well, in lay language I would say it played a major determining role. THE COURT: A major determining role, with reasonable certainty? THE WITNESS: Yes. THE COURT: All right. (Price Waterhouse v. Hopkins, 1989, pp. 543–545) Judge Gesell ultimately ruled that an “employer that treats [a] woman with [an] assertive personality in a different manner than if she had been a man is guilty of sex discrimination” (Hopkins v. Price Waterhouse, 1985, p. 1119). He described the firm’s decision-making process as “tainted by sexually biased evaluations” (p. 1120). With regard to Fiske’s testimony, he noted that “a far more subtle process [than the usual discriminatory intent] is involved when one who is in a distinct minority may be viewed differently by the majority because the individual deviates from an artificial standardized profile” (p. 1118). He went on to say that the firm’s “partnership evaluation system permitted negative comments tainted by stereotyping to defeat her candidacy, despite clear indications that the evaluations were tainted by discriminatory stereotyping” (p. 1118). In a footnote, he added: Common sense is confirmed by the literature on the problem of sex stereotyping which suggests that making evaluators aware of the risks of biased evaluations and inquiring as to whether the generalizations are supported by concrete incidents can be effective in eliminating or minimizing stereotyping. (p. 1120, note 15)

The Court of Appeals decision Price Waterhouse appealed Judge Gesell’s decision, and the written record of the trial was reviewed by a three-judge panel of the U.S. Court of Appeals for the District of Columbia. Price Waterhouse argued that the social psychology testimony was “sheer speculation” of “no evidentiary value” (Price Waterhouse v. Hopkins, 1987, p. 467). The majority on the appeals court disagreed, ruling that “partners at Price Waterhouse often evaluated female candidates in terms of their sex . . . the partnership selection process at Price Waterhouse was impermissibly infected by stereotypical attitudes towards female candidates” (Price Waterhouse v. Hopkins, 1987, p. 468). Specifically regarding the testimony, the appeals court stated that “convergent indicators of stereotyping” . . . taken together provided Dr. Fiske a sufficient basis from which to draw her conclusions that Hopkins was a victim of stereotyping. To the extent Price Waterhouse believes

Social science research on trial 83 Dr. Fiske lacked necessary information, the firm is in fact quarreling with her field of expertise and the methodology it employs. (p. 467) However, the sole dissenting appeals judge described Fiske as someone “purporting to be an expert” (Price Waterhouse v. Hopkins, 1987, p. 477) in the field and protested “the remarkable intuitions of Dr. Fiske” (p. 478). His dissent implied that the argument was about the accuracy of perceptions of Hopkins’s personality, not the stereotypic manner in which those perceptions were cast. For example, he argued: “To an expert of Dr. Fiske’s qualifications, it seems plain that no woman could be overbearing, arrogant, or abrasive: any observations to that effect would necessarily be discounted as the product of stereotyping” (p. 477). He noted that without information about the so-called truth of the matter, “Dr. Fiske’s expertise rose to the occasion. Her arts enabled her to detect sex stereotyping based largely on the ‘intensity of the negative reaction’” (p. 477).

Issues in the Supreme Court Because they lost at the appellate court level, PW asked the Supreme Court to review the case. Review by the nation’s highest court is discretionary. It agrees to hear only about 150 of the 3,000 cases it is asked to review each year. But because the appellate court decisions in Hopkins and in similar cases had been in conflict, the Supreme Court accepted this case for review. The central legal issues that made this case ripe for review concerned the proper allocation of the burden of proof between the employer and employee, as well as the proper standard of proof required to support allegations of discrimination in what are called mixed-motive cases, cases in which the refusal to hire or promote is based on both legitimate and illegitimate reasons. The burden- and standardof-proof issues are irrelevant to this article, but some discussion of the nature of mixed-motive cases is necessary to understand the context of social science evidence in this case. Essentially, the mixed-motive issue asks whether it is permissible to refuse to hire or promote someone on the basis of membership in a protected category (e.g., sex, race, age) if the person also has actual areas of significant incompetence. In effect, one might argue that if people have real performance problems, it does not matter if they lose a job or promotion partly because of their race or gender. One might argue that the ultimate outcome is the same in either case, whether the grounds were stereotyping or performance. However, this argument would imply that one can only identify discrimination when the person is otherwise perfect. Most employees are not perfect, having records that, if closely examined, could indicate flaws. Is it permissible then to discriminate against the vast majority of people who are genuinely flawed in some important way? This question is identified in discrimination cases as the mixed-motive issue; it formed the basis of PW’s petition to the Supreme Court.

84

Susan T. Fiske et al.

The mixed-motive issue has important practical consequences. Can one discriminate against someone who has genuine problems, or do the targets of such discrimination have to prove that they were so exemplary that they would have been guaranteed the position if discrimination were absent? The mixed-motive question asks whether all else has to be above reproach for a person to prove discrimination. Alternatively, one could argue that decision makers can never use discriminatory motives in personnel judgments, even in the case of someone with known shortcomings. In this case, the defense was that if Ann Hopkins had actual interpersonal skills problems, she could not also be a victim of discrimination. The Supreme Court thought this a sufficiently important point of employment discrimination law that it ought to be resolved.

The APA amicus curiae brief The APA decided to enter the case for two reasons. First, the crucial finding of discrimination in this case, as characterized by the Court of Appeals, was grounded on direct evidence that the employer’s selection process “was impermissibly infected by stereotypical attitudes toward female candidates” (Price Waterhouse v. Hopkins, 1987, p. 468). As a result, the parties were forced to focus some of their attention on the issue of sex stereotypes. However, they lacked the necessary social science background to present the issue of stereotyping in an empirically based, legally relevant manner. Second, PW consistently disparaged Fiske’s testimony by criticizing the methodology and the concepts she used in arriving at her expert opinion that PW discriminated against Hopkins on the basis of sex. APA informed the Court that the methodology and literature Fiske used were consonant with generally accepted research practice. APA was careful not to support Fiske personally as an expert witness or as a researcher. Thus, as a good amicus should, APA addressed an issue (the psychology of stereotyping, as relevant to this case) that the parties did not have adequate space, time, or expertise to discuss. Price Waterhouse had conceded that the partners’ negative comments “might conceivably be taken as indicating that stereotypical thinking was sometimes present ‘in the air’ at Price Waterhouse” (Brief for Petitioner PW, 1988, p. 48), but throughout their briefs they disparaged the psychology of stereotyping. They placed the term expert in quotation marks, in a belated effort to discredit the validity of research on stereotyping. In addition, they placed the term sex stereotyping within quotation marks, falsely implying that it is an unaccepted neologism, and they characterized as an amorphous proposition the appeals court finding that the employer discriminated against Ms. Hopkins because of “stereotypical attitudes” (p. 15). They claimed that the finding was derived from “intuitions·about unconscious sexism—discernible only through an ‘expert‘ judgment” (p. 17). In addition to labeling Fiske’s opinion as “gossamer evidence” and “intuitively divined” (p. 45), PW claimed that Fiske’s conclusions were faulty because she never met Hopkins and only reviewed the partners’ evaluations of her. Price Waterhouse accused the lower courts of basing

Social science research on trial 85 a finding of intentional discrimination on a “chain of intuitive hunches about ‘unconscious’ sexism” which “were, in turn, magically transformed into evidentiary ‘facts’ by a shift in the burden of persuasion” (p. 44). The APA’s intent in its brief was to disabuse the Court of the notion that sex stereotyping was not an identifiable and legally cognizable source of sex discrimination prohibited by Title VII, and to inform the court of the scientific validity of the methods and literature used in Fiske’s testimony. Moreover, the goal of the brief was to represent the research literature as understood by the researchers who provided the material. The implications of the assessment did favor a particular side of the case, and the amicus used the facts of the case to illustrate the relevance of the sex-stereotyping literature, when appropriate. If the implications of the literature had instead favored PW, in the judgment of the researchers, that perspective would have been presented. The APA’s amicus curiae brief was drafted by a panel of social and industrial/ organizational psychologists—Eugene Borgida, Kay Deaux, Susan Fiske, and Madeline Heilman—and the final product was compiled, integrated, and rewritten by then-APA-counsel Donald Bersoff. The brief argued for the validity of the field of stereotyping in general, and sex-stereotyping research in particular. The brief noted, first, that empirical research on sex stereotyping has been conducted over many decades and is generally accepted in the scientific community; second, that stereotyping can create discriminatory consequences for stereotyped groups; third, that some conditions that promote stereotyping were present in the firm’s work setting; and finally that the firm took no effective steps to reduce discriminatory stereotyping, although such methods were available.

Oral arguments and the Court’s decision On October 31, 1988, the Supreme Court heard oral arguments in Price Waterhouse v. Hopkins. Kathryn Oberly argued the case for PW, James Heller for Hopkins. The two lawyers had each presented about a minute of their 30minute prepared statements when they were interrupted and peppered with questions by the justices for the remainder of their allotted times. On May 1, 1989, the Court handed down its decision. The vote was six to three. Justice Brennan delivered the opinion, in which Justices Blackmun, Marshall, and Stevens joined; Justices White and O’Connor filed concurring opinions. Justice Kennedy filed a dissenting opinion, in which Justices Rehnquist and Scalia joined. The Court found in favor of Hopkins on the mixed-motive issue and in favor of PW on the standard-of-proof issue. That is, in mixed-motive cases, it is not permissible for employers to use discriminatory criteria, and they (not the plaintiff) must bear the burden of persuading the trier of fact that their decision would have been the same if no impermissible discrimination had taken place. However, the court also said that PW had been held to too high a standard of proof (i.e., clear and convincing evidence) and that Judge Gesell should review the facts to see whether it could win under

86

Susan T. Fiske et al.

a less stringent (preponderance of evidence) standard. In all, it was viewed primarily as a victory for Hopkins, although the case was returned to Judge Gesell so that he could decide whether PW was liable under the lower standard of proof. APA contributed to making explicit, for the first time in any Title VII case, that sex stereotyping was a form of sex discrimination.2 The Court’s plurality specifically criticized PW’s placement of the phrase sex stereotyping in quotation marks. The plurality said, “such conduct seems to us an insinuation that such stereotyping was not present in this case or that it lacks legal relevance. We reject both possibilities” (Price Waterhouse v. Hopkins, 1989, p. 1791). The plurality went on to endorse the importance of sex stereotyping: “In forbidding employers to discriminate against individuals because of their sex, Congress intended to strike at the entire spectrum of disparate treatment of men and women resulting from sex stereotypes” (p. 1791, quoting Los Angeles Dept. of Water & Power v. Manhart, 1978, p. 707). More specifically, APA had argued that sex-stereotypic prescriptive demands to be feminine simultaneous with jobspecific demands to be aggressive place women in a double-bind situation; this was apparently convincing to the Court in their framing of Hopkins’s dilemma as a “Catch 22.” The Court did not dismiss Fiske’s expert testimony on grounds cited by PW that she did not personally interview Hopkins. In characterizing this criticism as scientifically naive and irrelevant, APA stated that PW confused the work of research psychologists such as Fiske with that of clinical psychologists who use interviews and other assessment devices to diagnose a patient. The issue in Fiske’s testimony was the presence of discriminatory stereotyping at PW, not the mental status of Hopkins. The proper focus, therefore, was on the conduct of PW partners, reflected in the stereotypic phrasing and content of their evaluations. Fiske brought precisely that focus when she evaluated the conditions at PW that evoke stereotyping and the nature of the comments by its partners in light of the research literature on stereotyping. The plurality agreed and refused to countenance the PW derogation of Fiske’s testimony. The Court said: We are not inclined to accept petitioner’s belated and unsubstantiated characterization of Dr. Fiske’s testimony as “gossamer evidence” based only on her “intuitive hunches” and her detection of stereotyping as “intuitively divined.” Nor are we inclined to accept the dissent’s dismissive attitude toward Dr. Fiske’s field of study and toward her own professional integrity. (Price Waterhouse v. Hopkins, 1989, p. 1793) Instead, they said, Indeed, we are tempted to say that Dr. Fiske’s expert testimony was merely icing on Hopkins’ cake. It takes no special training to discern sex stereotyping in a description of an aggressive female employee as requiring “a course at charm school.” Nor . . . does it require expertise in psychology

Social science research on trial 87 to know that, if an employee’s flawed “interpersonal skills” can be corrected by a soft-hued suit or a new shade of lipstick, perhaps it is the employee’s sex and not her interpersonal skills that has drawn the criticism. (Price Waterhouse v. Hopkins, 1989, p. 1793) One can interpret this comment in various ways: as dismissive, saying that the social science testimony was all common sense; as merely taking the social psychological expertise for granted; or as suggesting that one does not necessarily require expert witnesses to identify stereotyping when the evidence is egregious. Finally, in response to the argument that Ann Hopkins might really be an obnoxious person, as described by some acquaintances, the Court agreed with Judge Gesell that: The reactions of at least some of the partners were reactions to her as a woman manager. Where an evaluation is based on a subjective assessment of a person’s strengths and weaknesses, it is simply not true that each evaluator will focus on, or even mention, the same weaknesses. Thus, even if we knew that Hopkins had “personality problems,” this would not tell us that the partners who cast their evaluations of Hopkins in sex-based terms would have criticized her as sharply (or criticized her at all) if she had been a man. It is not our job to review the evidence and decide that negative reactions to Hopkins were based on reality; our perception of Hopkins’ character is irrelevant. We sit not to determine whether Ms. Hopkins is nice, but to decide whether the partners reacted negatively to her personality because she is a woman. (Price Waterhouse v. Hopkins, 1989, pp. 1794–1795) One could not have asked for a better understanding of the psychology of stereotyping. The Supreme Court concluded, In the specific context of sex stereotyping, an employer who acts on the basis of a belief that a woman cannot be aggressive, or that she must not be, has acted on the basis of gender. . . . We are beyond the day when an employer could evaluate employees by assuming or insisting that they matched the stereotype associated with their group. . . . An employer who objects to aggressiveness in women but whose positions require this trait places women in an intolerable Catch 22: out of a job if they behave aggressively and out of a job if they don’t. Title VII lifts women out of this bind. (pp. 1790–1791)

The state of the field, as revealed by the amicus process Strengths It is important to give credit where credit is due. In considering material for the brief, we were impressed with the advances made and knowledge

88

Susan T. Fiske et al.

accumulated during the past decade or so. Indeed, we gave the most weight to that material. Social scientists (and especially psychologists) have contributed significantly to an understanding of the nature and operation of stereotypes. Areas in which increments in knowledge are most pronounced include (a) the nature of categorization processes, including the development of subtypes; (b) conditions that encourage stereotyping; and (c) the influence of stereotypes on performance evaluation and causal explanation (Deaux, 1985; Deaux & Kite, 1993). The work on stereotyping is an excellent example of basic research that can have practical impact, both in analyzing the current situation and in suggesting future courses of action. The cognitive categorization approaches to stereotyping are particularly well suited to courtroom use. That is, the cognitive approach by definition concentrates on people’s thoughts, not on their feelings, which means the type of evidence one examines is highly compatible with the kinds of evidence available in the courtroom. We do not have to undertake a speculative analysis of an individual’s authoritarian personality or of an entire culture’s stance toward diversity; the cognitive analysis is more straightforward. Moreover, laboratory research has some special virtues for training experts in this kind of endeavor. Beginning with the antecedents of stereotyping, laboratory researchers are well suited to analyze the situation, looking for factors that encourage or discourage stereotyping. Training in experimental methods teaches one to divide a social situation into its critical components. In conducting laboratory research, one learns to isolate and manipulate the significant situational factors. Moreover, the same compatibility holds in examining the indicators of stereotyping. First, psychological researchers, particularly those with a cognitive bent, are used to dealing with the written record, hard evidence of people’s judgmental processes. Second, evidence suggesting how people categorize another person and evidence pointing to the attributes considered important in decision making are indicators that are often apparent in the written record. When a person labels another person, the written record or reports of conversations provide evidence that is similar to the responses of subjects in a laboratory experiment. Turning to the consequences of stereotyping, there, too, laboratory researchers are well suited to identify sensitive measures and interpret the meaning of a pattern of response. Finally, in suggesting remedies, many laboratory researchers have built their research programs around finding ways to undercut stereotyping. Experimenters are trained to deepen their understanding of an effect by making it come and go through various interventions. Laboratory scientists often assume that the interventions may be effectively translated into the organizational mold. For example, Fiske’s research on interdependence and stereotyping has been heavily influenced by the kinds of structural feature that operate in organizations and might usefully generalize to those contexts. This type of laboratory research on basic processes then converges with field research on real-world effects of interventions.

Social science research on trial 89 Weaknesses More work is needed, for both theoretical and practical reasons. There is still a need to test more laboratory findings in field settings and to use such findings to revise current theories. In turn, the key dynamics in field settings need to be incorporated systematically into laboratory experiments. A critical issue for managers as well as attorneys is how to minimize the use of stereotypes. The psychological literature is strong in showing how pervasive stereotypes are, in demonstrating the myriad conditions that promote them, and in showing how resistant they can be to new information. Relatively few research programs explore the conditions that discourage stereotyping. Psychologists need to work further to discover when stereotypes do and when they do not operate. Such efforts might investigate deterrent strategies, the influence of observers or models on stereotyping, and the broader contexts that may emphasize or minimize stereotypes. These issues, all raised by the Hopkins case, have not been addressed adequately on an empirical basis. Specific questions of cognitive process are also raised by this case. For example, how are two or more stereotypic categories combined (e.g., manager and woman)? What are the circumstances under which people use alternative subtypes (e.g., aggressive career woman) and more general categories (e.g., women)? The case specifically raises the issue of whether Hopkins was considered an exemplar of “aggressive career woman” or simply as a deviant from both categories, “woman” and “manager.” The continuing debate on perceivers’ use of category prototypes versus exemplars is relevant here (see Fiske & Taylor, 1991, chap. 4, for a review). Perhaps the central issue raised by this case is the distinction between descriptive and prescriptive aspects of stereotypes. This issue is particularly relevant to gender stereotyping, but it arises in other contexts as well. Although it is routinely accepted that sex stereotypes have both descriptive and normative components (e.g., Terborg, 1977), research has been focused primarily on the descriptive component. It has been concentrated on the following four issues: the attributes that constitute sex stereotypes; the processes by which they are ascribed to men and women; the consequences of such ascriptions on expectations, attributions, and evaluations; and the conditions that regulate their occurrence. Underlying this research is a working assumption that if psychologists can determine how and when stereotypic attributes are used to characterize an individual man or woman, differential treatment can be averted. The Hopkins case forces a reexamination of that assumption. Hopkins was acknowledged as competent, committed, hard-working, and effective. The extent of her accomplishments was not the main point of contention, her talents were not denied (although they may have been underrated), and the attributes considered part of the male stereotyped “competency cluster” were accepted. In short, in terms of work-relevant attributes, she did not appear to suffer the ills of a traditional stereotyped description (e.g., passive, weak, indecisive), and yet she was a victim of sex stereotyping. Why? Despite her work-related

90

Susan T. Fiske et al.

competence (or perhaps because of it), she was seen as behaving in ways that are considered inappropriate for women. She met certain expectations, but not others, and she was expected to meet both. The prescriptive aspects of sex stereotypes (i.e., the behaviors deemed respectively suitable for men and women) are clearly important. Nevertheless, in reviewing the literature for the brief, we found surprisingly little attention focused on this issue. Although there is evidence demonstrating that behaving out of sex role has unfavorable consequences and that women and men who violate the “shoulds” of the sex-stereotypic prescription are found to be objectionable, few attempts have been made to systematically explore this larger phenomenon of prescriptive stereotyping. Even fewer attempts have been made to address the question of how to prevent or remedy it. The interventions may be of a different sort than those targeting descriptive aspects of stereotyping. Here the dynamics are likely to involve motivational and affective factors as well as cognitive ones. How complete does the database have to be? In her testimony, Fiske based her assertions regarding the role of stereotyping on a well-established and thriving literature. Her assessment was grounded in decades of research. Moreover, the use of such “social framework” testimony is likely to continue in various areas of psychology (Goodman & Croyle, 1989). How broad and deep does the database have to be for psychologists to render expert opinions? A crucial concern emerges about the adequacy of scientific databases as social science is increasingly used in the legal system. The number of APA sponsored briefs has increased, and their influence has been examined (Acker, 1990; Roesch, Golding, Hans, & Repucci, 1991). The increased use of social science data in law, whether to address legislative or adjudicative matters, has generated important conceptual frameworks and evaluative criteria based on peer review standards for evaluating the scientific adequacy of the database in question (e.g., Monahan & Walker, 1988). Issues about the quality of the database, which also pertain to the use of social science evidence in the Hopkins case, have long been debated with regard to research on eyewitness identification (Kassin, Ellsworth, & Smith, 1989; Loftus, 1983; McCloskey & Egeth, 1983). More recently, this issue has emerged as central to the debate regarding the role of APA’s amicus curiae brief in death penalty cases, most notably Lockhart v. McCree, 1986 (see Bersoff, 1987). Again, the adequacy of the scientific database was at the heart of the controversy. Elliott (1991) took issue with APA’s Lockhart amicus position that “that stability and convergence of the findings over three decades lends impressive support to their validity” (Bersoff, 1987, p. 68). Elliott was quite critical of the database and argued that the team of experts assembled at APA may have exaggerated the conclusiveness of the database in order to have an impact on legal policy. Ellsworth (1991) strongly rebutted Elliott’s criticisms of the research database

Social science research on trial 91 and argued that the database was adequate by generally accepted peer review standards, that convergent validity was established in this research domain, and that Elliott’s claim (i.e., scientific authority was overstated) was unfounded. Most important, Ellsworth argued that “to keep silent until our understanding is perfect is to keep silent forever” (p. 277). The team that developed the amicus position in the Hopkins case would strongly endorse Ellsworth’s perspective. The amicus experts’ favorable assessment of the sufficiency of the scientific database in the Hopkins case, as in Lockhart, was determined by the convergent validity of the scientific research literature. The interplay between research and testimony A critic might argue that researchers who have contributed to a particular literature should not testify as experts or contribute to amicus briefs because they have a stake in the validity of that literature, and they might overlook the rough spots or misrepresent the degree of consensus in the state of the research. These ethical issues have been a concern from the outset of the expert testimony, in the drafting of the amicus brief, and in the composition of this article. Regarding the testimony of involved experts, it is difficult to imagine who is more qualified to testify than someone whose primary activity is to conduct research on the topic at hand. Certainly, perennial expert witnesses who conduct little research are not more credible than occasional expert witnesses whose primary identity lies in science. The main motivations for scientists to be involved in such expert testimony are to get well-established research literatures into the relevant legal settings and to receive feedback about the utility of their work. As researchers apply their expertise to legal issues, further research and greater expertise is inspired. Such a cycle must commence with a welldeveloped research area before it can be exported, but the point is that intellectual involvement of experts ultimately strengthens both research and its application to law.3 There are, of course, several safeguards against deliberately or inadvertently misleading legal applications. First, there is the conscience of the individual researcher. This can only be known in the privacy of one’s own thoughts, but ultimately there must be some degree of trust in the honor of our colleagues. One’s reputation as a scientist is far more important than one’s winning any one case. Without peer credibility, a scientist cannot carry on the requisite professional activities that enable his or her continued career. Another potential criticism of such legal application comes not from misrepresentation of the literature, but in application to a particular setting. The role of expert testimony, whether psychological, medical, statistical, engineering, or whatever, is to discuss a research literature, not to define or address the legal issues involved. The trier of fact (judge or jury) makes the legal decision. The role of the testimony is to be helpful to the trier of fact by providing information about an area of human behavior that might not be familiar.

92

Susan T. Fiske et al.

Specifically, expert testimony (a) represents the state of knowledge in the relevant field and (b) applies it to the facts of the particular case. Certainly, there are issues of expert judgment and informed opinion at both stages of the process. But why is the psychology of stereotyping ultimately any different from other areas of admissible expertise? We would argue that there are no fundamental differences that prevent stereotyping researchers from qualifying as experts, representing the database, and applying their knowledge to a particular case. Within an adversarial system, another safeguard is the alternative view posed by the actual or potential opposing expert, by the judge’s neutral expert (a strategy sometimes used for statistical testimony), and by the actual or prospective opposing briefs. The whole premise of an adversarial system is that the most veridical evidence, scientific or otherwise, is the most convincing. Although adversarial safeguards on expert testimony and amicus briefs are not perfect, it is a premise of the legal system that distortion of the data will come out, one way or another. Turning more specifically to the selection of the amicus drafters, deliberate efforts were made to represent a cross-section of expertise and opinion. The APA Legal Affairs Office approved the final composition of the group. Fiske was included because of her work on stereotyping and her intimate knowledge of the case; Deaux was included for her expertise about stereotyping based on gender per se; Heilman was included because of her knowledge of gender stereotyping in work contexts; and Borgida was included for his knowledge of forensic psychology and because his published work has taken a devil’s advocate position on some issues in gender stereotyping. All four have published extensively in the relevant areas, have held editorial positions on journals, and have been competitively funded for their research in these areas. Thus, the institutional controls (APA) and the orientations of the individual researchers both worked to assure a balanced perspective on the literature. One might argue that the amicus drafters, once chosen, were then driven by politics. However, expertise about the effects of stereotyping is not necessarily less neutral than expertise about the effects of a bullet to the brain. Medical experts can have political or self-serving motivations just as easily; they would fail to qualify if it could be shown that they were primarily motivated to push a policy agenda (e.g., gun control) or to become famous (through involvement in a notorious case despite private misgivings). Yet, one might argue, the researchers representing the psychology of stereotyping are somehow less neutral because they may be potential victims of stereotypes themselves and so are not necessarily neutral. By this argument, female scientists studying gender stereotypes have reduced credibility. As this argument implies, it would be like asking someone who lives in a neighborhood with a high crime rate to serve as an expert on the effects of fatal bullet wounds. Clearly, if such a person had scientific expertise in the topic, as established by peer review and reputation, the person’s private life would not be relevant. Obviously, this hypothetical expert would not be called in to testify as an expert against a spouse’s murderer,

Social science research on trial 93 but neither is the stereotyping expert called in to testify in a personal grievance. Careful methods, convincing data, reasonable argument, and debate among peers are the bulwark against bias in science. Careful methods, convincing data, reasonable argument, and debate among peers all protect against bias by experts testifying and drafting amicus briefs. We are not arguing that the database is perfect or that as researchers we are indifferent to the topics we study or their policy implications. Nor are we arguing that expert opinions may not differ. We are arguing that psychologists’ considered and self-critical judgments as researchers were reflected in the use of psychological expertise at trial and in the amicus process in this case.

Conclusions It is an extraordinary opportunity to have psychological research, in an area so well-established and thriving, be confronted by some of the most prominent legal minds in the country. We think psychology came out rather well, on the whole. But we also think, as psychologists, that we have our work cut out for us. To enhance our credibility, we must continue to test our theories in a range of laboratory and field settings. We must attend to those factors in the real world that demand research attention: How can decision makers best guard against the incursions of stereotypic judgments and consequent discrimination? How can people control their stereotypic thinking? What situational and organizational factors influence stereotyping? What are appropriate remedies? Psychology is well equipped to take on these challenges. Meanwhile, the role of expert testimony and amicus briefs is to educate and to bring the pertinent research to the attention of the trier of fact, and then let the courts decide.

Notes 1

2 3

The position of the names of the parties often changes in a case citation as it goes through trial, appeal, and review process. In this case, Ann Hopkins was the original plaintiff in the federal district (trial) court and Price Waterhouse (PW) was the defendant. Hopkins won at trial; PW then appealed to the intermediate federal appellate court (an appellant’s name appears first at this level). After it again lost, PW petitioned the U.S. Supreme Court to review the case. As petitioner; PW’s name again appears first. Regardless of the sequence of the names of the parties in this article, the authors are referring to the same underlying case. For a review of the impact of other social science amicus curiae briefs, see Acker (1990). As concrete examples of the benefits to research, we know of one meta-analysis that cited Hopkins as demonstrating some relevant dimensions for analysis (Eagly, Makhijani, & Klonsky, 1992) and one archival study of solo status and performance evaluations apparently inspired by Hopkins (Sackett, DuBois, & Noe, 1991). Moreover, in our own work, one research program has been heavily influenced by Hopkins (Fiske, 1989a, 1989b), and one literature review has described the Hopkins case as marking “a rite of passage for research on gender stereotypes” (Deaux & Kite, 1993).

94

Susan T. Fiske et al.

References Acker, J. R. (1990). Social science in Supreme Court cases and briefs: The actual and potential contribution of social scientists as amici curiae. Law and Human Behavior, 14, 225–242. Allport, G. W. (1954). The nature of prejudice. Reading, MA: Addison-Wesley. Arvey, R. D. (1979). Unfair discrimination in the employment interview: Legal and psychological aspects. Psychological Bulletin, 86, 736—765. Arvey, R. D., & Campion, J. E. (1982). The employment interview: A summary and review of recent research. Personnel Psychology, 35, 281–322. Ashmore, R. D., & Del Boca, F. K. (1981). Conceptual approaches to stereotypes and stereotyping. In D. L. Hamilton (Ed.), Cognitive processes in stereotyping and intergroup behavior (pp. 1–36). Hillsdale, NJ: Erlbaum. Bartol, K. M., & Butterfield, D. A. (1976). Sex effects in evaluating leaders. Journal of Applied Psychology, 61, 446–454. Berg, E. N. (1988, July). The Big Eight: Still a male bastion. New York Times: Business Day, pp. 1, 37. Bersoff, D. N. ( 1987). Social science data and the Supreme Court. American Psychologist, 42, 52–68. Brewer M. B. (1988). A dual process model of impression formation. In T. K. Srull & R. S. Wyer; Jr. (Eds.), Advances in social cognition (Vol. 1, pp. 1–36). Hillsdale, NJ: Erlbaum. Brief for petitioner Price Waterhouse. (1988, January). Petition for a writ of certiorari to the United States Court of Appeals for the District of Columbia Circuit. Brown v. Board of Education, 347 U.S. 483 (1954). Brown, V., & Geis, F. L. (1984). Turning lead into gold: Evaluations of men and women leaders and the alchemy of social consensus. Journal of Personality and Social Psychology, 46, 811–824. Costrich, N., Feinstein, J., Kidder, L., Marecek, J., & Pascale, L. (1975). When stereotypes hurt: Three studies of penalties for sex-role reversals. Journal of Experimental Social Psychology, 11, 520–530. Crocker, J., & McGraw, K. M. (1982). What’s good for the goose is not good for the gander: Solo status as an obstacle to occupational achievement for males and females. American Behavioral Scientist, 27, 357–369. Darley, J. M., & Gross, P. H. (1983). A hypothesis-confirming bias in labeling effects. Journal of Personality and Social Psychology, 44, 20–33. Darley, S. (1976). Big-time careers for the little woman: A dual-role dilemma. Journal of Social Issues, 32, 85–98. Deaux, K. (1976). The behavior of women and men. Monterey, CA: Brooks/Cole. Deaux, K. (1985). Sex and gender. Annual Review of Psychology, 36, 49–81. Deaux, K., & Emswiller, T. (1974). Explanation of successful performance on sexlinked tasks: What is skill for the male is luck for the female. Journal of Personality and Social Psychology, 29, 80–85. Deaux, K., & Kite, M. E. (1993). Gender stereotypes. In F. Denmark & M. Paludi (Eds.), Handbook on the psychology of women (pp. 176–202). New York: Greenwood Press. Deaux, K., & Lewis, L. L. (1983). Components of gender stereotypes. Psychological Documents, 13, 25. Deaux, K., & Lewis, L. L. (1984). The structure of gender stereotypes: Interrelationships among components and gender label. Journal of Personality and Social Psychology, 46, 991–1004.

Social science research on trial 95 Dipboye, R. L., & Wiley, J. W. (1977). Reactions of college recruiters to interviewer sex and self-presentation style. Journal of Vocational Behavior, 10, 1–12. Eagly, A. H., Makhijani, M. G., & Klonsky, B. G. (1992). Gender and the evaluation of leaders: A meta-analysis. Psychological Bulletin, 111(1), 3–22. Elliott, R. (1991). Social science data and the APA: The Lockhart brief as a case in point. Law and Human Behavior, 15, 59–76. Ellsworth, P. C. (1991). To tell what we know, or wait for Godot? Law and Human Behavior, 15, 77–90. Erber, R., & Fiske, S. T. (1984). Outcome dependency and attention to inconsistent information. Journal of Personality and Social Psychology, 47, 709–726. Feather, N. T., & Simon, J. G. (1975). Reactions to male and female success and failure in sex-linked occupations: Impressions of personality, causal attributions, and perceived likelihood of different consequences. Journal of Personality and Social Psychology, 31, 20–31. Feldman-Summers, S., & Kiesler, S. B. (1974). Those who are number two try harder: The effect of sex on attributions of causality. Journal of Personality and Social Psychology, 30, 846–855. Fiske, S. T. (1989a). Examining the role of intent: Toward understanding its role in stereotyping and prejudice. In J. S. Uleman & J. A. Bargh (Eds.), Unintended thought (pp. 253–283). New York: Guilford. Fiske, S. T. (1989b, August). Interdependence and stereotyping: From the laboratory to the Supreme Court (and back). Invited address given at the 97th Annual Convention of the American Psychological Association, New Orleans. 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). San Diego, CA: Academic Press. Fiske, S. T., Neuberg, S. L., Beattie, A. E., & Milberg, S. J. (1987). Categorybased and attribute-based reactions to others: Some informational conditions of stereotyping and individuating processes. Journal of Experimental Social Psychology, 13, 399–427. Fiske, S. T., & Taylor, S. E. (1991). Social cognition (2nd ed.). New York: McGraw-Hill. Frieze, I. H., Fisher, J. R., Hanusa, B. H., McHugh, M. C., & Valle, V. A. (1978). Attributions of the causes of success and failure as internal and external barriers to achievement. In J. L. Sherman & F. L. Denmark (Eds.), The psychology of women: Future directions in research (pp. 519–552). New York: Psychological Dimensions. Garland, H., & Price, K. H. (1977). Attitudes toward women in management and attributions for their success and failure in a managerial position. Journal of Applied Psychology, 59, 705–711. Goodman, J., &. Croyle, R. T. (1989). Social framework testimony in employment discrimination cases. Behavioral Sciences and the Law, 7, 227–241. Hagen, R. L., & Kahn, A. (1975). Discrimination against competent women. Journal of Applied Social Psychology, 5, 362–376. Hamilton, D. L. (1979). A cognitive-attributional analysis of stereotyping. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 12, pp. 53–84). San Diego, CA: Academic Press. Hamilton, D. L. (1981). Stereotyping and intergroup behavior: Some thoughts on the cognitive approach. In D. L. Hamilton (Ed.), Cognitive processes in stereotyping and intergroup behavior (pp. 333–353). Hillsdale, NJ: Erlbaum.

96

Susan T. Fiske et al.

Hansen, R. D., & O’Leary, V. E. (1983). Actresses and actors: The effects of sex on causal attributions. Basic and Applied Social Psychology, 4, 209–230. Heilman, M. E. (1980). The impact of situational factors on personnel decisions concerning women: Varying the sex composition of the applicant pool. Organizational Behavior and Human Performance, 26, 386–395. Heilman, M. E. (1983). Sex bias in work settings: The lack of fit model. Research in Organizational Behavior, 5, 269–298. Heilman, M. E. (1984). Information as a deterrent against sex discrimination: The effects of applicant sex and information type on preliminary employment decisions. Organizational Behavior and Human Performance, 33, 174–186. Heilman, M. E., Block, C. J., Martell, R. F., & Simon, M. C. (1989). Has anything changed? Current characterizations of men, women, and managers. Journal of Applied Psychology, 74, 935–942. Heilman, M. E., & Guzzo, R. A. (1978). The perceived cause of work success as a mediator of sex discrimination in organizations. Organizational Behavior and Human Performance, 21, 346–357. Heilman, M. E., & Martell, R. F. (1986). Exposure to successful women: Antidote to sex discrimination in applicant screening decisions? Organizational Behavior and Human Decision Processes, 37, 376–390. Heilman, M. E., Martell, R., & Simon, M. (1988). The vagaries of bias: Conditions regulating the undervaluation, equivaluation, and overvaluation of female job applicants. Organizational Behavior and Human Performance, 41, 98–110. Heilman, M. E., and Stopeck, M. H. (1985a). Attractiveness and corporate success: Different causal attributions for males and females. Journal of Applied Psychology, 70, 379–388. Heilman, M. E., & Stopeck, M. H. (1985b). Being attractive, advantage or disadvantage? Performance-based evaluations and recommended personnel actions as a function of appearance, sex, and job type. Organizational Behavior and Human Decision Processes, 35, 202–215. Higgins, E. T., & Bargh, J. A. (1987). Social cognition and social perception. In M. R. Rosenzweig & L. W. Porter (Eds.), Annual review of psychology (Vol. 38, pp. 369–425). Palo Alto, CA: Annual Review. Hishon v. King and Spaulding, 467 U.S. 69 (1984). Hopkins v. Price Waterhouse, 618 F. Supp. 1109 (D. D.C. 1985). Hopkins v. Price Waterhouse, No. 84–3040, slip op. (D. D.C. May 14, 1990) (on remand). Howard-Pitney, B., Borgida, E., & Omoto, A. M. (1986). Personal involvement: An examination of processing differences. Social Cognition, 4, 39–57. Jago, A. G., & Vroom, V. H. (1982). Sex differences in the incidents and evaluations of participative leader behavior. Journal of Applied Psychology, 67, 776–783. Kanter, R. (1977). Men and women of the corporation. New York: Basic Books. Kassin, S. M., Ellsworth, P. C., & Smith, V. L. (1989). The “general acceptance” of psychological research on eyewitness testimony: A survey of the experts. American Psychologist, 44, 1089–1098. Kristal, J., Sanders, D., Spence, J. T., & Helmreich, R. (1975). Inferences about the femininity of competent women and their implications for likability. Sex Roles, 1, 33–41. Kruglanski, A. W., & Freund, T. (1983). The freezing and unfreezing of lay-inferences: Effects of impressional primacy, ethnic stereotyping, and numerical anchoring. Journal of Experimental Social Psychology, 19, 448–468.

Social science research on trial 97 Lewin, T. (1990, May 16). Partnership in firm awarded to victim of sex bias. New York Times, pp. Al, A20. Linville, P. W. (1982). The complexity–extremity effect and age-based stereotyping. Journal of Personality and Social Psychology, 42, 193–211. Linville, P. W., & Jones, E. E. (1980). Polarized appraisals of out-group members. Journal of Personality and Social Psychology, 38, 689–70 3. Locksley, A., Borgida, E., Brekke, N., & Hepburn, C. (1980). Sex stereotypes and social judgment. Journal of Personality and Social Psychology, 39, 821–831. Loftus, E. F. (1983). Silence is not golden. American Psychologist, 38, 564–572. Los Angeles Department of Water & Power v. Manhart, 434 U.S. 702, 707 n.13 (1978). Markus, H., & Zajonc, R. B. (1985). The cognitive perspective in social psychology. In G. Lindzey & E. Aronson (Eds.), Handbook of social psychology (3rd ed., Vol. 1, pp. 137–230). New York: Random House. McArthur, L., & Post, D. (1977). Figural emphasis and person perception. Journal of Experimental Social Psychology, 13, 520–535. McCloskey, M., & Egeth, H. E. (1983). Eyewitness identification: What can a psychologist tell a jury? American Psychologist, 38, 550–563. Monahan, J., & Walker, L. (1988). Social science research in law. American Psychologist, 43, 465–472. Mullen, B. (1991). Group composition, salience, and cognitive representations: The phenomenology of being in a group. Journal of Experimental Social Psychology, 27, 297–323. Nicholls, J. G. (1975). Causal attributions and other achievement-related cognitions: Effects of task outcome, attainment value and sex. Journal of Personality and Social Psychology, 31, 379–389. Nieva, Y. F., & Gutek, B. A. (1980). Sex effects on evaluation. Academy of Management Review, 5, 267–276. Nisbett, R., & Ross, L. (1980). Human inference: Strategies and shortcomings of social judgment. Englewood Cliffs, NJ: Prentice-Hall. Olian, J. D., Schwab, D. P., & Haberfeld, Y. (1988). The impact of applicant gender compared to qualifications on hiring recommendations. Organizational Behavior and Human Decision Processes, 41, 180–195. Pazy, A. (1986). The persistence of pro-male bias despite identical information regarding causes of success. Organizational Behavior and Human Decision Processes, 38, 366–377. Pence, E. C., Pendleton, W. C., Dobbins, G. H., & Sgro, J. A. (1982). Effects of causal explanations and sex variables on recommendations for corrective actions following employee failure. Organizational Behavior and Human Performance, 29, 227–240. Pettigrew, T. F., & Martin, J. (1987). Shaping the organizational context for Black American inclusion. Journal of Social Issues, 43, 41–78. Petty, M. M., & Lee, G. K. (1975). Moderating effects of sex of supervisor and subordinate on relationships between supervisor behavior and subordinate satisfaction. Journal of Applied Psychology, 60, 624–628. Pheterson, G. I., Kiesler, S., & Goldberg, P. A. (1971). Evaluation of the performance of women as a function of their sex, achievement, and personal history. Journal of Personality and Social Psychology, 19, 114–118. Price Waterhouse v. Hopkins, 825 F.2d 458 (D.C. Cir. 1987). Price Waterhouse v. Hopkins, 109 S. Ct. 177.S (1989).

98

Susan T. Fiske et al.

Rasinski, K. A., Crocker, J., & Hastie, R. (1985). Another look at sex stereotypes and social judgments: An analysis of the social perceiver’s use of subjective probabilities. Journal of Personality and Social Psychology, 49, 3 17–326. Renwick, P. A., & Tosi, H. (1978). The effects of sex, marital status, and educational background on selection decisions. Academy of Management Journal, 21, 93–103. Roesch, R., Golding, S. L., Hans, V. P., & Repucci, N. D. (1991). Social science and the courts: The role of amicus curiae briefs. Law and Human Behavior, 15, 1–12. Rousell, C. (1974). Relationship of sex of department head to department climate. Administrative Science Quarterly, 19, 211–220. Ruble, D. N., & Ruble, T. L. (1982). Sex stereotypes. In A. G. Miller (Ed.), In the eye of the beholder: Contemporary issues in stereotyping (pp. 188–252). New York: Praeger. Ruble, T. L., Cohen, R., & Ruble, D. N. (1984). Sex stereotypes: Occupational barriers for women. American Behavioral Scientist, 27, 339–356. Sackett, P. R., DuBois, C. L. Z., & Noe, A. W. (1991). Tokenism in performance evaluations: The effects of work group representation on male–female and White– Black differences in performance ratings. Journal of Applied Psychology, 76, 263–267. Schein, V. E. (1973). The relationship between sex-role stereotypes and requisite management characteristics. Journal of Applied Psychology, 57, 95–100. Schein, V. E. (1975). Relationships between sex-role stereotypes and requisite management characteristics among female managers. Journal of Applied Psychology, 60, 340–344. Spangler, E., Gordon, M. A., & Pipkin, R. M. (1978). Token women: An empirical test of the Kanter hypothesis. American Journal of Sociology, 84, 160–170. Swim, J., Borgida, E., Maruyama, G., & Myers, D. G. (1989). Joan McKay versus John McKay: Do gender stereotypes bias evaluations? Psychological Bulletin, 105, 409–429. Tajfel, H. (1969). Cognitive aspects of prejudice. Journal of Social Issues, 25, 79–97. Tajfel, H. (1972). La categorisation social [Social categorization]. In S. Moscovici (Ed.), Introduction à la psychologie sociale (Vol. 1, pp. 272–302). Paris: Larousse. Taylor, S. E. (1981). A categorization approach to stereotyping. In D. L. Hamilton (Ed.), Cognitive processes in stereotyping and intergroup behavior (pp. 88–114). Hillsdale, NJ: Erlbaum. Taylor, S. E., Fiske, S. T., Etcoff, N. L., & Ruderman, A. J. (1978). Categorical bases of person memory and stereotyping. Journal of Personality and Social Psychology, 16, 778–793. Terborg, J. R. (1977). Women in management: A research review. Journal of Applied Psychology, 62, 647–664. Tetlock, P. E. (1983). Accountability and complexity of thought. Journal of Personality and Social Psychology, 45, 74–83. Tosi, H. L., & Einbender, S. W. (1985). The effects of the type and amount of information in sex discrimination on research: A meta-analysis. Academy of Management Journal, 28, 712–723. Wiley, M. G., & Eskilson, A. (1982). Coping in the corporation: Sex role constraints. Journal of Applied Social Psychology, 12, 1–11. Wolman, C., & Frank, H. (1975). The solo woman in a professional peer group. American Journal of Orthopsychiatry, 45, 164–171.

Part II

Second wave Motivated tacticians’ thinking is for doing

5

Controlling other people The impact of power on stereotyping Susan T. Fiske

Issues of power and stereotyping haunt our history and our present as human beings. Without stereotypes, there would be less need to hate, exclude, exterminate. For good reasons, people object to being stereotyped, categorized, and attributed certain characteristics in common. People do not want to be stereotyped because it limits their freedom and constrains their outcomes, even their lives. In short, stereotypes exert control. Obviously, stereotypes exert control through prejudice and discrimination. Victims of stereotyping know this and rightly resist stereotypes for those reasons. I want to go beyond these fundamental truths and argue that stereotypes are controlling by their very nature and all too easily result from power, from asymmetries in control. My argument focuses on some relationships between stereotyping and controlling others. It begins by discussing how stereotypes result from and maintain one person’s control over another; it claims that stereotypes are intrinsically controlling of other people. The focus here is on how power encourages stereotyping, as well as how stereotyping maintains power. The argument also describes how powerful people can be discouraged from stereotyping by getting them to pay attention. Essentially this account rests on the motivating impetus of social structure on the individual. I suggest that social control operates through the direction and nature of attention. People in power stereotype in part because they do not need to pay attention, they cannot easily pay attention, and they may not be personally motivated to pay attention. To illustrate these relationships between stereotyping and control, consider two real-life examples that both pertain to gender stereotyping, although the principles apply to other forms of stereotyping as well. Both examples came from legal cases in which I served as an expert witness.1 These cases presented superficially different but fundamentally similar cases of stereotyping; both revealed the impact of power, controlling others. Afterward, I will define terms more closely, note relevant literature, and describe some of our relevant research.

Tales of two women Lois Robinson worked as a welder in a certain Jacksonville, Florida, shipyard. Jacksonville Shipyards Inc. (JSI) repaired Navy and commercial ships in dry

102

Susan T. Fiske

dock—tough, sometimes dangerous work. Women made up less than 5% of the JSI workforce and less than half of l% of the skilled craftworkers. Typically there were no or few women on any given shift, so a woman was likely to be the only woman in the crowd getting on the shipyard buses or punching out at the time clock. The JSI shipyard has been described as a boys’ club, a man’s world, with “Men Only” painted on one of the work trailers. (When someone complained, the sign was painted over, but in a cursory way.) It is perhaps best summarized as a locker-room atmosphere, with a lot of practical joking and teasing. For example, one worker put a flashlight in his pants to show how well-endowed horses are; another carved the handle of a tool to resemble a penis, waving it in the face of the women. There was open hostility to women on the part of a few men: “There’s nothing worse than having to work around women; women are only fit company for something that howls.” More often, there was simply a great deal of off-color joking (including one often-repeated joke about death by rape). Obscenity and profanity were routine. Prominent in the visual environment (“every craft, every shop”) were many calendars showing women in various states of undress and various sexually explicit poses. Comparable magazines were widely shared, and pinups were torn out and posted spontaneously. Decorating various public walls were graffiti, both words and cartoons, with explicit sexual content depicting women. There were no pictures or graffiti of naked men. The workers were not allowed to bring other magazines on the job, and they were not allowed to post other material that was not work-related. The few women workers were typically called by demeaning or sexually explicit names (honey, dear, baby, sugar, momma, pussy, cunt, etc.). They were constantly teased, touched, humiliated, sexually evaluated, and propositioned; the incidents occurred “every day all day” involving “all crafts,” according to depositions. Lois Robinson, the welder, complained about the magazines and calendars, but she was brushed off, all the way up to the top. And even at the highest levels, one manager pointed out that he had his own pinups. Another manager reminded her patriotically that pinups had brought us through World War II. Robinson eventually filed a lawsuit alleging sex discrimination due to sexual harassment in a hostile work environment; she won her case at the trial court level. JSI has appealed, and that is pending. What does this case have to do with stereotyping, control, and power? Why was it so important to the men at JSI to keep their magazines and calendars, and why were some men so openly hostile to the women? One answer lies in the social structure, specifically the dramatic power asymmetries between the men and women at JSI. It was a man’s world where men controlled the distinctly male atmosphere, the coin of social acceptance, and the tangible rewards. Women as a group were radically powerless: outnumbered, out of place, and on trial. Men thus controlled the work environment and shaped it to their own needs. Essentially, one cause of stereotyping at JSI was the men’s impunity;

Controlling other people 103 they did not need the women for any workplace rewards. The power structure at JSI contributed to the generally aversive and stereotypic work environment. I will describe the mechanisms in more detail shortly. Moving from the male workers in general to the specific upper-level managers who heard and rejected Robinson’s complaints, one can see the effects of managerial overload. Not only were these men part of the traditionally dominant group but they were also given specific institutional power, that is, control over many people’s outcomes in the workplace. As managers, by definition each was attending to many underlings. Under such conditions of attentional overload, it was easier to form a superficial, stereotyped-based judgment and dismiss the complaint of one underling, especially an outsider. Another source of stereotyping at JSI was the small number of men who not only harassed the women but also were openly hostile. In effect, some men were really “bad apples,” such as the one who claimed that women were unfit company at work. Not knowing anything more about these particular men, one can only speculate, but an individual problem seems likely. I speculate that one possible problem was an overriding personal dominance orientation. The second case was set against the boardrooms of a Big Eight accounting firm, Price Waterhouse (PW). One of the top managers brought in millions of dollars in accounts, worked more billable hours than anyone in that cohort, was well liked by clients, and was described as aggressive, hard-driving, and ambitious. But this exemplary manager was denied partnership because she was not feminine enough. Ann Hopkins was not accepted as a partner because of “interpersonal skills problems” that would be corrected, a supporter informed her, by walking, talking, and dressing more femininely. Although the setting was not exactly Jacksonville Shipyards, it did encourage stereotyping of women in several comparable ways (see Fiske, Bersoff, Borgida, Deaux, & Heilman, 1991). First, Hopkins was in a firm that had approximately 1% female partners (7 of 662), and she was the only woman out of 88 individuals proposed for partner that year; the few women managers stood out. Second, being a manager in a Big Eight firm is a stereotypically masculine job, calling for tough, aggressive behavior; consequently people think there is a lack of fit between being a woman and being a manager (Glick, Zion, & Nelson, 1988; Heilman, 1983). Third, stereotypes operate more freely on ambiguous criteria, such as judgments of interpersonal skills, than on unambiguous counting criteria, such as number of billable hours. PW failed to guard against bias in these subjective judgments, and there were considerable differences of opinion about how to interpret Hopkins’s hard-driving managerial behavior. Fourth, the partnership evaluations were based on ambiguous and scant information in many cases; hearsay and casual opinions were given substantial weight. Finally, the firm had no explicit policy against gender discrimination, although it did prohibit discrimination on the basis of age or health in partnership decisions. Ann Hopkins also filed a lawsuit alleging sex discrimination, which she won, even though PW appealed it up to the Supreme Court. The American Psychological Association filed an amicus brief that apparently was helpful to

104

Susan T. Fiske

the Supreme Court in deciding this case (Fiske et al., 1991; but see Barrett & Morris, 1993a, 1993b; Fiske, Bersoff, Borgida, Deaux, & Heilman, 1993a, 1993b; Goodman, 1993; Saks, 1993). How does an analysis in terms of control fit here? Just as in the shipyard, the men were in power at PW, and the women were outnumbered, out of place, and on trial. The men controlled an atmosphere that might best be characterized as an exclusive gentlemen’s club, in which women were guests who were expected to defer to the men’s customs. The men as a group did not particularly need the few women in order to obtain workplace rewards, so again, there was a fundamental issue of resource control. In addition, these busy partners were evaluating up to 88 partner candidates, added to a grueling workload in one of the world’s premier accounting firms, and perhaps overloaded conditions contributed to their lack of attention to their own decision processes. In addition to these features of social structure, there were a few bad apples who seemed to have personal problems, such as the partner who each year complained that women should not even be senior managers, let alone partners. Again, we see power asymmetries in the social structure, attentional overload, plus a few individuals with special problems. At this point, it is important to note that this analysis is not engaging in what our graduate students call “male bashing.” I am not saying that men in general are the specific culprits in stereotyping. In fact, the whole burden of this article is that it is a matter of social structure and a matter of individual personality dynamics that are likely to encourage stereotyping. Any group in the kind of social structure described here would be likely to stereotype other people. Any individual with the kind of personality dynamics described here would be likely to stereotype other people. So, I argue for a more general theoretical basis underlying these two specific examples, even though both of them happen to include gender.

Stereotyping and control Stereotyping operates in the service of control. Stereotyping is a category-based cognitive response to another person. Apart from prejudice (affect) and discrimination (behavior), stereotyping describes people’s beliefs (cognitions) about an individual based on group membership. Category-based or stereotypic responses contrast with fully individuated, attribute-by-attribute consideration of another person (Fiske & Neuberg, 1990). It is useful to discuss two aspects of stereotyping in the context of stereotyping and control: descriptive and prescriptive beliefs (e.g., Terborg, 1977). If the stereotype is descriptive, it tells how most people in the group supposedly behave, what they allegedly prefer, and where their competence supposedly lies. People may believe that women in general are good secretaries and teachers, but poor welders, managers, or scientists (e.g., Heilman, 1983; Ruble & Ruble, 1982). Descriptive stereotypes also claim that African Americans are good athletes but poor scholars, that Asian Americans and Jews are good scholars

Controlling other people 105 but poor athletes, and so on (Miller, 1982). In these assumptions, there lurks an implicit pressure to fit a certain image; other people’s expectations create the starting point for one’s commerce with them. The easiest course for a stereotyped person is to stay within the bounds of those expectations. But the person who is stereotyped may try to contradict the expectations. In either case, the descriptive stereotype constrains a person because it anchors the interaction, weighing it down and holding it back. Either way, the stereotype must be dealt with. A friend in college once said that she was tired of being everybody’s “Black experience.” Not that she wanted to change who she was, but she was tired of having that limited dimension dominate her interactions, for better or worse. In short, a descriptive stereotype is controlling simply because it exists as an anchor or starting point in the mind of one person dealing with another. Anyone in the culture, whether actively biased or not, potentially knows the contents of the stereotype (Steele, 1997), so it becomes an implicit anchor for everyone. Another form of stereotype, the prescriptive aspect, is even more explicitly controlling. It purportedly tells how certain groups should think, feel, and behave. So, for example, women should be nice, African Americans should be spontaneous, Asian Americans should be good at math, and Jews should be good with money. In one sense, these are flattering stereotypes, but they also demand that the individual either conform or disappoint the holder of the stereotype. The penalties can be swift and severe if one disappoints someone else’s prescriptive stereotype (e.g., Eagly, Makhijani, & Klonsky, 1992). Think of the male adolescent in an all-male group who fails to conform to stereotypically masculine prescriptions. Prescriptive stereotypes are limiting and constitute a form of social control. The descriptive aspect of stereotypes acts as an anchor, and the prescriptive aspect of stereotypes acts as a fence. In short, stereotypes control people, which is one reason they are so aversive. No one wants to be stereotyped. Stereotypes reinforce one group’s or individual’s power over another by limiting the options of the stereotyped group, so in this way stereotypes maintain power. People with power do not have to put up with them, but people without power are victims. Power is control, and stereotypes are one way to exert control, both social and personal. One might argue that subordinates also stereotype those in power, which the next section will counter-argue. But even if underlings do stereotype, their beliefs simply exert less control than do those of people in power. Copeland (1994), for example, found that powerful people were more able to create self-fulfilling prophecies than were powerless people. The controlling impact of stereotypes also explains why power maintains stereotypes. Elsewhere, Eric Dépret and I (Dépret & Fiske, 1993) have defined power as asymmetrical control over another person’s outcomes (for another review of definitions, see Ng, 1980). Power has traditionally been defined as the ability to influence at will (e.g., Dahl, 1957; Huston, 1983; Pruitt, 1976). However, one may have power without influence if the subordinates refuse to be influenced, despite the control of the powerful over their outcomes. Power

106

Susan T. Fiske

has also been defined as status (e.g., Hogg & Abrams, 1988). However, one may have power with or without status, as when low-status groups control resources important to high-status groups. Dépret and Fiske’s (1993) definition of power in terms of the intrinsic characteristics of human interdependence, that is, as asymmetrical outcome control, follows some theoretical perspectives of Thibaut and Kelley (1959). Because power is essentially control, people pay attention to those who have power. It is a simple principle: People pay attention to those who control their outcomes. In an effort to predict and possibly influence what is going to happen to them, people gather information about those with power. Consider the direction of attention in a large organization. Attention follows power. Attention is directed up the hierarchy. Secretaries know more about their bosses than vice versa; graduate students know more about their advisors than vice versa. Similar dynamics operate at convention social hours, as people cluster around those perceived to be powerful. Thus, the powerless are attentive to the powerful. By the same token, the powerful need not attend very much to those with less power, because less is at stake for the powerful with regard to their subordinates. Besides outcome control and its attendant motivations, the powerful have more demands on their attention than do the powerless. By nature of the hierarchy, the powerful have more people competing for their attention than do the powerless. If stereotypes are shortcuts, overburdened people are more likely to use them. The literature indeed indicates that stereotyping is more likely when people are distracted, when their cognitive capacity is limited (for a review, see Fiske, 1993). For example, when people are busy, they do not modify initial categories, all else being equal (Gilbert, 1989). External factors decrease people’s mental capacity for thinking carefully about others, and the attentional overload of power predictably decreases people’s capacity. Finally, particular individuals, objectively powerful or not, seek power and dominance over other people, which should influence how they perceive those others (e.g., Battistich, Assor, Messe, & Aronoff, 1985). Individuals who seek to control the fates of other people may or may not more frequently end up in positions of power. Regardless, their motivation to control other people may result in the use of stereotyping as one form of control. Elsewhere, Emery and I (Fiske & Emery, 1993) have argued that such attempts at social control may come from a precarious sense of mental control. Whatever the mechanism, there may be personality analogs of the hypothesized social power processes, with lack of individuated attention as the cause of stereotyping. Attention may be determined by asymmetrical outcome control, capacity overload, and personal motivation, all in ways linked to one person’s actual or desired power over another. Attention then determines who has detailed knowledge of whom and who stereotypes whom. The powerless are stereotyped because no one needs to, can, or wants to be detailed and accurate about them. The powerful are not so likely to be stereotyped because subordinates need to, can, and want to form detailed impressions of them. The powerless

Controlling other people 107 need to try to predict and possibly alter their own fates. They may have fewer competing demands on their attentional capacity. And to the extent that a low personal need for power happens to coincide with a low-power position, they may be less motivated to stereotype. The next section presents data bearing on each of these points. Before turning to the data, one still might argue that the powerful are victims of stereotypes too. But, first, as noted, if the powerless stereotype the powerful, it simply does not matter as much; it demonstrably does not limit their behavior as much (Copeland, 1994) nor, by definition, control their outcomes as much. It is more an irritation than a fundamental threat, except when subordinates are given the power to evaluate, vote on, or otherwise judge those in power. Then the powerless have been given some outcome control, and they are by definition slightly more powerful. The other instance of the powerful being stereotyped might be argued to operate when the powerful stereotype themselves or each other. One might argue that the JSI workers stereotyped each other as all liking pornography or that the PW partners stereotyped themselves as necessarily male, but it is arguable whether more harm was done to themselves or to the women they excluded on that basis. Finally, stereotypes of one’s ingroup are more flexible and variable than stereotypes of the outgroup (for a review, see Fiske & Taylor, 1991, chap. 4); hence, they are less controlling.

Data on power and stereotyping from the bottom up A body of previous work from my laboratory supports half of the power– attention equation. When people were interdependent, when they needed each other to achieve their goals, they paid attention. In this work, my colleagues and I have typically manipulated expectancies, which are positive or negative. Sometimes the expectancies were simple expectancies about competence, and sometimes they were stereotypes, such as ethnic stereotypes. We also then manipulated the degree of interdependence (the degree to which the two people depended on each other for some valued outcome), and the interdependence was either high or low. Then we presented subjects with mixed information about the target person on whom their outcomes depended. Some of the information fitted the stereotype or expectancy, and some of the information did not fit it and disputed it. This simulated real life, where one gets mixed information about another person. We have shown in a variety of contexts three consistent and replicable results: First, people pay attention to others who control their outcomes (Erber & Fiske, 1984; Neuberg & Fiske, 1987; Ruscher & Fiske, 1990; Ruscher, Fiske, Miki, & Van Manen, 1991). lnterdependence increases attention in particular to stereotype-inconsistent information. This is the information that potentially undermines the stereotype; it is the most useful and informative. This means that outcome-dependent people attend to the most informative clues they can find, as if they are trying to be as accurate as possible, given the high stakes. Second, we find that people then draw inferences from the information they

108

Susan T. Fiske

gather. They make dispositional comments about the inconsistency. They in effect construct personality profiles of the person on whom they depend, perhaps in an attempt to see the other person (and therefore their own fate) as predictable. If they know the other person’s individual personality, they think they know what the other person will do and can infer how it will affect themselves. Finally, interdependence increases the variability of impressions across people, so they end up with more idiosyncratic impressions, often less reflective of stereotypes and expectations. This pattern occurs regardless of whether the interdependence is positive, as in cooperating pairs, or negative, as in competing pairs of people. So we have some evidence that attention follows power, at least when people are equally dependent on each other. Recently, we extended these findings to situations of asymmetrical power (Dépret & Fiske, 1999). In these studies, undergraduates expected to complete a task under the control of others who could judge, reward, punish, or interfere with their performance. That is, the other people had power over the subjects. When those others were individuals from various outgroups (in this case, from college majors far outside of psychology), the subjects paid more attention to them as their power increased. In particular, the increase was in attention to the inconsistent information, which is the most informative because it potentially challenges the stereotype. Again, as the stakes increased, people were more careful. As in our previous work, we also found that dispositional inferences increased with the target’s power, as if people were trying to make the other person as predictable as possible. The powerful people became intriguing individuals.2 Another lesson we are learning from our research is that the power dynamics have to be experientially real and significant to the perceiver. We have found that when power captures attention, the power is not just a matter of expecting to meet the other person, not just expecting to be evaluated by the other, and not just expecting to discuss a joint project (Stevens & Fiske, 2000). People have to expect actually to work together on some project. Demonstrable outcomes have to be present; something like effort or money has to depend on the interaction. In this case, the powerless attend to the powerful person, inconsistency and all, and they discount the inconsistencies less than do people not dependent. When no concrete outcomes depend on the other person, but the person is instead simply to evaluate the subject, then subjects concentrate on the downside, focusing on negative information and trying to discount it. In summary, then, we have evidence that powerful people, defined as those who control concrete outcomes, capture attention, and that subordinates form more detailed and idiosyncratic impressions of them. In that sense, then, the powerful are not stereotyped.

Data on power and stereotyping from the top down But what of the other half of the equation? Do the powerful not pay attention to the powerless? Our theory argues that the powerful do not need to pay

Controlling other people 109 attention because nothing is riding on the other person; their fates do not depend on the other, so their attention should be more superficial. Moreover, according to one of the other mechanisms, the powerful oversee many subordinates, and this too should interfere with careful attention. Goodwin, Gubin, Fiske, and Yzerbyt (2000) have recently found that power does indeed decrease attention to others, in a setting designed to mimic personnel decision making; that is, undergraduates were given the power to evaluate high school students’ summer job applications. As the percentage of their power in the decision increased, their attention to the applicants actually decreased in a baseline condition. I will come back to this study shortly, but the point is that it provides initial evidence that the powerful may not pay enough attention to the powerless. This then mimics what went on at Price Waterhouse and Jacksonville Shipyards; the powerful managers simply had no need to attend to the relatively powerless women as unique individual subordinates. At this point, one might well object that this analysis simply does not apply to the male workers at Jacksonville Shipyards. After all, the women there were receiving quite a lot of attention, although of a certain kind. And one might even argue that the men had very specific needs that depended on the women. So, one might argue, the women were indeed powerful because the men were sexually interested in them. This argument is as old as Aristophanes and Lysistrata, but examine the situation a bit more carefully. The men were powerful in several respects related both to the work and social environment. The women had only a modicum of social power, to the extent they were in a position to resist or to cooperate enthusiastically with the men’s sexual advances. Moreover, the attention they received was of a stereotypical sort. Thus, the motivations behind the attention determine the kind of attention, and only certain kinds of increased attention will undercut stereotypes. In a series of studies, we have examined what happens when a powerful person has social or sexual goals that depend on the subordinate. For example, what happens if a male manager wants to date a woman employee? What happens if a female boss wants to be close friends with her female assistant? If the subordinate’s goals are work-related, namely recognition for work well done, then the supervisor’s social interest is often experienced as interfering and irrelevant. Rather than focusing on the quality of one’s work, the supervisor is focusing on whether one is appealing and available. It follows from the specifically social goals that the supervisor will not be attending as carefully to the subordinate’s task performance. In effect, the supervisor is paying attention but not the right kind of attention. In our laboratory, we have investigated whether such social goals help or hinder the subordinate’s work-related goals. Goodwin, Fiske, Rosen, & Rosenthal (2002) recruited undergraduate men for a dating study that consisted of two parts: a task-oriented interaction and the actual series of dates. In the task-oriented part, the men were in a position to supervise and evaluate their female subordinate. For half the men, the subordinate was the same woman they expected to date, and half the men expected to date somebody else. Hence, for half the men their romantic fates

110

Susan T. Fiske

depended on this woman, and for the other half they did not. The woman was a confederate whose videotaped task performance was either competent or incompetent, according to pretesting. Consistent with this, the men who expected only a task-oriented interaction were able to distinguish competent from incompetent performance. But the men who expected to date her did not. This kind of bias is not limited to men in power; we have recently replicated this finding with women supervising men they expected to date. In both cases, a romantic goal clouds one’s judgment, presumably because people are not paying the right kind of attention. So far, I have focused on social structure as a cause and cure for stereotyping. But individuals are also directly involved here. Both in Jacksonville Shipyard and at Price Waterhouse, some people were worse than others; some people made particularly egregious comments. What about these bad apples? Maybe the bad apples stereotype other people because of personal problems that parallel the social structural ones. Goodwin, Gubin, Fiske, & Yzerbyt (2000) hypothesized that a personal need for power and dominance would have similar stereotype-confirming effects on attention. And indeed, people with highly dominant personalities pay less attention to the very information that could undermine their stereotypes. Like people who are powerful because of the social structure, people with dominant personalities ignore stereotype-discrepant information, preferring the information that confirms their stereotypes. Dominant people attend to consistent, stereotype-confirming information and attend very little to the information that could undermine their stereotypes. Nondominant people attend equally to both kinds of information. Perhaps, then, there are some similarities between those with power thrust upon them and those who take it for themselves. Control again is central here in that dominance-oriented people are distinguished by their chronic attempts to control other people for their own ends.

How intentional and responsible are people? So far, the portrait of the powerful seems discouraging. These are, after all, the people who control what happens to the rest of us. It would appear that the powerful do not pay a lot of appropriate attention as they manipulate our fates. And yet, they can be held responsible for their inattention. Each of us, when we have power over someone else, can be held responsible for our attention or inattention to the other person. What kind of evidence would be relevant here? Responsibility, according to a legal analysis, depends on intent. People are responsible if the act is intentional. So, for example, discrimination is not illegal if it can be proved to be totally unintentional. Penalties for killing another person vary with the degree of intent involved, and so on. Responsibility depends on state of mind and intent in particular. As psychologists, we know something about intent. As far back as William James, intent was defined by two factors: choice and attention (Fiske, 1989). If people have alternatives, if there is more than one way to behave according

Controlling other people 111 to a reasonable person, then one condition is met for recognizing intent. For example, in many circumstances, doing something at gunpoint does not count as intentional. Doing the only thing one knows how to do is not strictly intentional. The second feature of intent that emerges, over the decades, among psychologists and lay people, is attention. If people keep the chosen alternative in mind, they can be said to intend the one they follow. So, for example, if one is dieting and thinks about the box of Belgian chocolates on the table, one finds them harder to resist than if one’s thoughts are focused elsewhere. This analysis of intent applies to stereotyping: People’s tendency to stereotype is intentional in that first they demonstrably have alternative ways of thinking about people, as members of a category or as unique individuals; everyone can do this (Fiske & Neuberg, 1990). Second, people can implement their alternative ways of thinking about other people according to how much attention they pay to those other people. Hence, we find again that attention is central in whether or not people stereotype. In particular, this suggests that people with power can overcome the tendency to stereotype the powerless by the very processes of attention that have been discussed so far.

What can be done? The powerful can be influenced to be more careful by strategies consistent with the theory proposed here. The powerful are by definition not motivated by outcomes that depend directly on their subordinates. Hence, the powerful are more likely to be influenced by their own higher-ups, by their own peers, or by their self-concept. In that sense, the motivations of the powerful are independent of their relationship with the person being judged; the motivations of the powerful are more autonomous in that sense. In this view, then, the powerful are motivated by what they perceive to be acceptable, according to the norms, and by their own self-concepts. There is some evidence for this. In the Goodwin et al. (2000) study in which undergraduates made job decisions about high school students, power decreased decision makers’ attention to the applicants, as hypothesized, in a baseline condition. But we also predicted that shared norms concerning humanitarian and egalitarian values might remind people of their better selves, in effect. That is, if we could make accessible the decision makers’ sense of responsibility, we might be able to get them to pay attention. So, some people completed a scale of shared humanitarian–egalitarian values; when we primed a sense of responsibility, their attention increased dramatically. Notice that we were not telling them what to do but only reminding them of their own values. The powerful can also be motivated by their own self-concepts as fair-minded and careful people. Fiske and Von Hendy (1992) predicted and found that people had the capacity to think about other people in either categorical or individuating ways and that they could use either strategy depending on which aspect of their self-concept was salient to them at the time. In effect, we were bringing out different sides of people by the feedback we gave them. We found

112

Susan T. Fiske

that people could be influenced by bogus feedback about their supposed special abilities to treat other people categorically or individually, if they were the kind of people who strictly followed their own proclivities. Similarly, people can be influenced by bogus information about the appropriate norms in the situation, if they are the kind of people who strictly use other people’s standards as a guide. This last set of studies focuses on values, self-concepts, and norms as motivators for powerful people. People, including those with power, can also be influenced by their own compunctions (Devine, Monteith, Zuwerink, & Elliot, 1991), public accountability (e.g., Tetlock, Skitka, & Boettger, 1989), fear of invalidity (e.g., Kruglanski, 1990), instructions to be accurate (Neuberg, 1989), and the like (see Fiske, 1993, for a review). In my opinion, they relate to the shipyard and the accounting firm in that an organization can make certain values salient, can encourage the constructive sides of people’s self-concepts, can promote norms of fairness, and so on. Conversely, an organization can ignore these issues and let the powerful take the easy way out, not bothering to pay much attention to the powerless. Our main program of research, then, has been showing that social structure affects attention, and if people pay more attention, at least some of them are less likely to stereotype. Moreover, organizations can encourage individuating attention by the structures they create.

Conclusion The type of illustrative data presented here deserves some comment. These are laboratory experiments, and one might wonder about the laboratory as a way to study power, stereotyping, sexual harassment, and so forth. Does the laboratory trivialize such real world issues? These laboratory studies were designed as simulations, as microcosms. Our strategy is to create a miniature world in which we can analyze and isolate the features of the social setting that we think are important. We transport these isolated features to the lab, and we show that these are sufficient conditions to produce a phenomenon in which we are interested. It allows us to see the fine-grained mechanisms that link power, control, and stereotyping. And it allows us to demonstrate the sufficient conditions for stereotypic processes (such as lack of attention to stereotype-disconfirming information). Obviously, we depend heavily on feedback from field studies as well, and we also draw a lot of our ideas from the world outside the lab. Similarly, one might wonder about our focus on cognitive mechanisms and attention in particular. First, attention here specifically means time looking at written materials. Attention also is an indicator of weight in a judgment (Fiske, 1980). Second, we do have other dependent variables, but attention has been the focus in this article for simplicity’s sake. But also it is a strategic choice in our research to focus so much on attention as a main dependent variable, because it is the beginning of the process, without which nothing else can occur.

Controlling other people 113 It has a central role in whether people stereotype or not, although there are no guarantees that getting somebody to pay attention to somebody else will undercut stereotypes. But at any rate, it is a first step, and with some people it is effective some of the time. If we can begin to find ways to capture people’s attention, perhaps we can undermine the control of stereotypes. Jacksonville Shipyards, Price Waterhouse, and our laboratory data have all linked power to stereotyping as mediated through the amount and nature of attention. People in power stereotype subordinates because they do not need to pay attention to them and because it may not be easy to do so. Moreover, sometimes their own personal dispositions may be oriented toward power, and this may compound the lack of individuating attention. Power affects stereotyping through attention (or a lack of the right kind of attention), and stereotyping controls those who are stereotyped. Our organizational structures and incentives can ameliorate this problem or make it worse.

Notes 1 2

The descriptions of the cases (but not their relevance to stereotypes, power, and control) also appear in Fiske and Stevens (1993). A more detailed description of the second case appeared in Fiske, Bersoff, Borgida, Deaux, and Heilman (1991). This process is entirely altered when the interaction is intergroup rather than interpersonal. When the powerful others are perceived to be a homogeneous outgroup, power creates a sense of threat—in effect an outgroup conspiracy. The powerful outgroup is instead viewed more stereotypically, as people expect the ingroup to be treated in an unfair and discriminatory fashion. Similarly, Ruscher, Fiske, Miki, and Van Manen (1991) found that intergroup competition increased stereotyping processes relative to interpersonal competition. Some of these group level phenomena explain the threatened reactions of some of the JSI workers and PW partners, to the extent that they viewed the women as a group threatening the men as a group.

References Barrett, G. V., & Morris, S. B. (1993a). The American Psychological Association’s amicus curiae brief in Price Waterhouse v. Hopkins: The value of science versus the values of the law. Law and Human Behavior, 17, 201–216. Barrett, G. V., & Morris, S. B. (1993b). Sex stereotyping in Price Waterhouse v. Hopkins. American Psychologist, 48, 54–55. Battistich, V., Assor, A., Messe, L.A., & Aronoff, J. (1985). Personality and person perception. In P. Shaver (Ed.), Review of personality and social psychology (Vol. 6, pp. 185–208). Beverly Hills, CA: Sage. Copeland, J. T. (1994). Prophecies of power: Motivational implications of social power for behavioral confirmation. Journal of Personality and Social Psychology, 67, 265–277. Dahl, R. A. (1957). The concept of power. Behavioural Science, 2, 201–215. Dépret, E. F., & Fiske, S. T. (1999). Perceiving the powerful: Intriguing individuals versus threatening groups. Journal of Experimental Social Psychology, 35, 461–480. Dépret, E. F., & Fiske, S. T. (1993). Social cognition and power: Some cognitive consequences of social structure as a source of control deprivation. In G. Weary,

114

Susan T. Fiske

F. Gleicher, & K. Marsh (Eds.), Control motivation and social cognition (pp. 196–202). New York: Springer-Verlag. Devine, P. G., Monteith, M. J., Zuwerink, J. R., & Elliot, A. (1991). Prejudice with and without compunction. Journal of Personality and Social Psychology, 60, 817–830. Eagly, A. H., Makhijani, M. G., & Klonsky, B. G. (1992). Gender and the evaluation of leaders: A meta-analysis. Psychological Bulletin, 111, 3–22. Erber, R., & Fiske, S. T. (1984). Outcome dependency and attention to inconsistent information. Journal of Personality and Social Psychology, 47, 709–726. Fiske, S. T. (1980). Attention and weight in person perception: The impact of negative and extreme behavior. Journal of Personality and Social Psychology, 38, 889–906. Fiske, S. T. (1989). Examining the role of intent: Toward understanding its role in stereotyping and prejudice. In J. S. Uleman & J. A. Bargh (Eds.), Unintended thought: The limits of awareness, intention, and control (pp. 253–283). New York: Guilford Press. Fiske, S. T. (1993). Social cognition and social perception. In M. R. Rosenzweig & L. W. Porter (Eds.), Annual review of psychology. Palo Alto, CA: Annual Reviews. Fiske, S. T., Bersoff, D. N., Borgida, E., Deaux, K., & Heilman, M. E. (1991). Social science research on trial: The use of sex stereotyping research in Price Waterhouse v. Hopkins. American Psychologist, 46, 1049–1060. Fiske, S. T., Bersoff, D. N., Borgida, E., Deaux, K., & Heilman, M. E. (1993a). A brief rejoinder: Accuracy and objectivity on behalf of the APA. American Psychologist, 48, 55–56. Fiske, S. T., Bersoff, D. N., Borgida, E., Deaux, K., & Heilman, M. E. (1993b). What constitutes a scientific review? A majority retort to Barrett and Morris on gender stereotyping. Law and Human Behavior, 17, 217–233. Fiske, S. T., & Emery, E. J. (1993). Lost mental control and exaggerated social control: Social-cognitive and psychoanalytic speculations. In D. M. Wegner & J. W. Pennebaker (Eds.), Control motivation and social cognition (pp. 171–199). New York: Springer-Verlag. Fiske, S. T., & Neuberg, S. L. (1990). A continuum model of impression formation, from category-based to individuating processes: Influence of information and motivation on attention and interpretation. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 23, pp. 1–74). San Diego, CA: Academic Press. Fiske, S. T., & Stevens, L. E. (1993). What’s so special about sex? Gender stereotyping and discrimination. In S. Oskamp & M. Costanzo (Eds.), Gender issues in contemporary society: Applied social psychology annual (Vol. 6, pp. 173–196). Newbury Park, CA: Sage. Fiske, S. T., & Taylor, S. E. (1991). Social cognition (2nd ed.). New York: McGrawHill. Fiske, S. T., & Von Hendy, H. M. (1992). Personality feedback and situational norms can control stereotyping processes. Journal of Personality and Social Psychology, 62, 577–596. Gilbert, D. T. (1989). Thinking lightly about others: Automatic components of the social inference process. In J. S. Uleman & J. A. Bargh (Eds.), Unintended thought: The limits of awareness, intention, and control (pp. 189–211). New York: Guilford Press. Glick, P., Zion, C., & Nelson, C. (1988). What mediates sex discrimination in hiring decisions? Journal of Personality and Social Psychology, 55, 178–186. Goodman, J. (1993). Evaluating psychological expertise on questions of social fact: The case of Price Waterhouse v. Hopkins. Law and Human Behavior, 17, 249–256.

Controlling other people 115 Goodwin, S. A., Fiske, S. T., Rosen, L. D., & Rosenthal, A. M. (2002). The eye of the beholder: Romantic goals and impression biases. Journal of Experimental Social Psychology, 38, 232–241. Goodwin, S. A., Gubin, A., Fiske, S. T., & Yzerbyt, V. (2000). Power can bias impression formation: Stereotyping subordinates by default and by design. Group Processes and Intergroup Relations, 3, 227–256. Heilman, M. E. (1983). Sex bias in work settings: The lack-of-fit model. Research in Organizational Behavior, 5, 269–298. Hogg, M. A., & Abrams, D. (1988). Social identifications: A social psychology of intergroup relations and group processes. New York: Routledge, Chapman & Hall. Huston, T. L. (1983). Power. In H. H. Kelley, E. Berscheid, A. Christensen, J. H. Harvey, T. L. Huston, G. Levinger, E. McClintock, L. A. Peplau, & D. R. Peterson (Eds.), Close relationships (pp. 169–219). New York: Freeman. Kruglanski, A. W. (1990). Motivations for judging and knowing: Implications for causal attribution. In E. T. Higgins & R. M. Sorrentino (Eds.), Handbook of motivation and cognition: Foundations of social behavior (Vol. 2, pp. 333–368). New York: Guilford Press. Miller, A. G. (Ed.). (1982). In the eye of the beholder: Contemporary issues in stereotyping. New York: Praeger. Neuberg, S. L. (1989). The goal of forming accurate impressions during social interactions: Attenuating the impact of negative expectancies. Journal of Personality and Social Psychology, 56, 374–386. Neuberg, S. L., & Fiske, S. T. (1987). Motivational influences on impression formation: Outcome dependency, accuracy-driven attention, and individuating processes. Journal of Personality and Social Psychology, 53, 431–444. Ng, S. H. (1980). The social psychology of power. San Diego, CA: Academic Press. Pruitt, D. G. (1976). Power and bargaining. In B. Seidenberg & A. Snadowsky (Eds.), Social psychology: An introduction. New York: Free Press. Ruble, D. N., & Ruble, T. L. (1982). Sex stereotypes. In A. G. Miller (Ed.), In the eye of the beholder: Contemporary issues in stereotyping (pp. 188–252). New York: Praeger. Ruscher, J. B., & Fiske, S. T. (1990). Interpersonal competition can cause individuating processes. Journal of Personality and Social Psychology, 58, 832–843. Ruscher, J. B., Fiske, S. T., Miki, H., & Van Manen, S. (1991). Individuating processes in competition: Interpersonal versus intergroup. Personality and Social Psychology Bulletin, 17, 595–605. Saks, M. J. (1993). Improving APA’s science translation amicus brief. Law and Human Behavior, 17, 235–248. Steele, C. (1997). A threat in the air: How stereotypes shape intellectual identity and performance. American Psychologist, 52(6), 613–629. Stevens, L. E., & Fiske, S. T. (2000). Motivated impressions of a powerholder: Accuracy under task dependency and misperception under evaluative dependency. Personality and Social Psychology Bulletin, 26, 907–922. Terborg, J. R. (1977). Women in management: A research review. Journal of Applied Psychology, 62, 647–664. Tetlock, P. E., Skitka, L., & Boettger, R. (1989). Social and cognitive strategies for coping with accountability: Conformity, complexity, and bolstering. Journal of Personality and Social Psychology, 57, 632–640. Thibaut, J. W., & Kelley, H. H. (19 59). The social psychology of groups. New York: Wiley.

6

The Ambivalent Sexism Inventory Differentiating hostile and benevolent sexism Peter Glick and Susan T. Fiske

Is sexism a form of prejudice? Although the question might appear absurd, consider Allport’s (1954) influential definition of ethnic prejudice. Prejudice, Allport wrote, “is an antipathy based upon a faulty and inflexible generalization” (p. 9). The existence of prejudice is commonly indexed by measures of antipathy, such as social distance (e.g., Crosby, Bromley, & Saxe, 1980) and negative stereotypes (e.g., Sigall & Page, 1971). Relationships between men and women, however, do not easily fit the mold of ethnic prejudice, at the very least because no other two groups have been as intimately connected (S. T. Fiske & Stevens, 1993). Furthermore, cultural images of women from ancient to modern times are not uniformly negative; women have been revered as well as reviled (Eagly & Mladinic, 1993; Guttentag & Secord, 1983; Tavris & Wade, 1984). Sexism is indeed a prejudice, but in this article we argue that it is, and probably always has been, a special case of prejudice marked by a deep ambivalence, rather than a uniform antipathy, toward women. Our goals are to: (a) reveal the multidimensional nature of sexism, (b) offer a theoretical and empirical analysis of the sources and nature of men’s ambivalence toward women, (c) compare our conception of ambivalent sexism with other theories of ambivalence (including ambivalent racism), and (d) provide a validated measure of ambivalent sexism.

Hostile and benevolent sexism Sexism has typically been conceptualized as a reflection of hostility toward women. This view neglects a significant aspect of sexism: the subjectively positive feelings toward women that often go hand in hand with sexist antipathy. We view sexism as a multidimensional construct that encompasses two sets of sexist attitudes: hostile and benevolent sexism. Hostile sexism needs little explanation; by it we mean those aspects of sexism that fit Allport’s (1954) classic definition of prejudice. We define benevolent sexism1 as a set of interrelated attitudes toward women that are sexist in terms of viewing women stereotypically and in restricted roles but that are subjectively positive in feeling tone (for the perceiver) and also tend to elicit behaviors typically categorized as

The Ambivalent Sexism Inventory 117 prosocial (e.g., helping) or intimacy-seeking (e.g., self-disclosure). We do not consider benevolent sexism a good thing, for, despite the positive feelings it may indicate for the perceiver, its underpinnings lie in traditional stereotyping and masculine dominance (e.g., the man as the provider and woman as his dependent), and its consequences are often damaging. Benevolent sexism is not necessarily experienced as benevolent by the recipient. For example, a man’s comment to a female coworker on how “cute” she looks, however well-intentioned, may undermine her feelings of being taken seriously as a professional. Nevertheless, the subjectively positive nature of the perceiver’s feelings, the prosocial behaviors, and the attempts to achieve intimacy that benevolent sexism generates do not fit standard notions of prejudice. Evidence for benevolent sexism can be gleaned from a variety of research areas. Research on helping behavior shows that female targets are more likely to elicit help than male targets are (see Eagly & Crowley, 1986 for a metaanalysis of this effect). Both men and women are more likely to seek intimacy with female than with male strangers, as indexed by interpersonal distance (Riess & Salzer, 1981), touching (Major, Schmidlin, & Williams, 1990), and selfdisclosure (Cozby, 1973; Morton, 1978).2 Even the commonly accepted notion among social scientists that stereotypes of women are more negative than those of men has been called into question by Eagly and her colleagues (Eagly & Mladinic, 1993; Eagly, Mladinic, & Otto, 1991), who have found evidence for more positive stereotypes of women than men, on certain dimensions. To balance the picture, however, it is important to note the prevalence of hostile sexism. In nearly all cultures and time periods for which information is available, women have been restricted to social roles with less status than those of men (Tavris & Wade, 1984). In our own society, there is evidence that women still face discrimination in gaining employment (Fitzgerald & Betz, 1983; Glick, 1991) and sexual harassment on the job (Gutek, 1985) and are perceived less favorably than men when enacting leadership roles in a masculine manner or domain (Eagly, Makhijani, & Klonsky, 1992). Even though stereotypes of women contain many positive traits, the positive traits relate to social– emotional, not agentic dimensions, so women are portrayed as being nice but incompetent at many important tasks (e.g., analytical thinking). Finally, there is ample evidence that sexual violence toward women is disturbingly frequent (Unger & Crawford, 1992).

Sources of hostile and benevolent sexism We propose that hostile and benevolent sexism have their roots in biological and social conditions that are common to human groups. Although “anthropologists do not totally agree on whether male dominance characterizes all human cultures” (Stockard & Johnson, 1992, p. 89), they do agree that patriarchy (men possessing structural control of economic, legal, and political institutions) is prevalent across cultures. Furthermore, “virtually all anthropologists doubt the existence of matriarchies at any phase of cultural evolution”

118

Peter Glick and Susan T. Fiske

(Harris, 1991, p. 10). The bias toward patriarchy is probably due to several factors related to the biology of sexual reproduction: sexual dimorphism (men’s greater size and strength may be one factor that typically allowed men to dominate preindustrial societies; Harris, 1991), the tendency for men to have a stronger social dominance orientation than women (as a result of sexual selection; Pratto, Sidanius, & Stallworth, 1993), and gender-based role divisions in which women perform the bulk of domestic duties (the mother carrying the fetus to term and providing nourishment for infants precipitated the traditional division of labor that restricted women to domestic roles; Stockard & Johnson, 1992). Although hostility between groups that differ in physical appearance is an all-too-human condition, the biology of sex creates a situation that is uniquely different from other in-group–out-group distinctions. As Guttentag and Secord (1983) pointed out, sexual reproduction lends women “dyadic power” (power that stems from dependencies in 2-person relationships) in that it compels men to rely on women as bearers of children and, generally, for the satisfaction of sexual needs. In addition, men may seek to fulfill needs for psychological intimacy with women (Berscheid, Snyder, & Omoto, 1989; Derlega, Winstead, Wong, & Hunter; 1985), perhaps because such needs are not as easily met with other men, who typically are competitors for status and resources (Harris, 1991). Cross-cultural and historical evidence gathered by Guttentag and Secord (1983) shows that, within patriarchal societies, women’s dyadic power is reflected in a particular form of social ideology: protective attitudes toward women, a reverence for the role of women as wives and mothers, and an idealization of women as romantic love objects. These are precisely the attitudes we define as characterizing benevolent sexism. The degree of hostile as compared with benevolent sexism may vary widely among societies (from those in which women are treated as chattel to those dominated by an ideology of chivalry), depending on factors such as sex ratios (Guttentag & Secord, 1983). However, the balance of power between the sexes is typically complex, reflecting the coexistence of male structural power and female dyadic power (see Harris, 1991; Stockard & Johnson, 1992). Furthermore, even though benevolent sexism suggests a subjectively positive view of women, it shares common assumptions with hostile sexist beliefs: that women inhabit restricted domestic roles and are the “weaker” sex. Indeed, both hostile and benevolent sexism serve to justify men’s structural power. Hostile sexist beliefs in women’s incompetence at agentic tasks characterize women as unfit to wield power over economic, legal, and political institutions, whereas benevolent sexism provides a comfortable rationalization for confining women to domestic roles. Similar ideologies (e.g., the “White man’s burden”) have been used in the past to justify colonialism and slavery (see Tajfel, 1969). Like hostile and benevolent sexism, these ideologies combine notions of the exploited group’s lack of competence to exercise structural power with selfserving “benevolent” justifications (“We must bear the burden of taking care of them”) that allow members of the dominant group to view their actions as

The Ambivalent Sexism Inventory 119 not being exploitative. Thus, benevolent sexism may be used to compensate for, or legitimate, hostile sexism (“I am not exploiting women; I love, protect, and provide for them”). Although the ideology of “the White man’s burden” seems archaic, the “man’s burden” as protector and provider still provides a positive image for men that subtly reinforces notions of dominance over women (Nadler & Morrow, 1959).3 The above analysis suggests that both hostile and benevolent sexism revolve around issues of social power, gender identity, and sexuality. We propose that HS and BS are composed of three shared components: paternalism, gender differentiation, and heterosexuality. Each component reflects a set of beliefs in which ambivalence toward women is inherent (i.e., each construct has a hostile and a “benevolent” aspect) and which serves to justify or explain the underlying social and biological conditions that characterize relationships between the sexes. Together, these three components form the core of our theory. Paternalism In common discourse, paternalism and sexism are often used synonymously, yet the former term, surprisingly, is not indexed in PsycLit, despite many references to the latter. Paternalism literally means relating to others “in the manner of a father dealing with his children” (Random House College Dictionary, 1973). This definition meshes well with the view that sexism is a form of ambivalence, for it includes connotations of both domination (dominative paternalism) as well as affection and protection (protective paternalism). Advocates of dominative paternalism justify patriarchy by viewing women as not being fully competent adults, legitimizing the need for a superordinate male figure. Yet protective paternalism may coexist with its dominative counterpart because men are dyadically dependent on women (because of heterosexual reproduction) as wives, mothers, and romantic objects; thus, women are to be loved, cherished, and protected (their “weaknesses” require that men fulfill the protector-andprovider role). Research on power in heterosexual romantic relationships confirms that dominative paternalism is the norm (see Brehm, 1992, chapter 9; Peplau, 1983). In its most extreme form, the traditional marriage (see Peplau, 1983), both partners agree that the husband should wield greater authority, to which the wife should defer. Protective paternalism is evident in the traditional male gender role of provider and protector of the home, with the wife dependent on the husband to maintain her economic and social status (Peplau, 1983; Tavris & Wade, 1984). Gender differentiation All cultures use physical differences between the sexes as a basis for making social distinctions, which are manifested as notions about gender identity (Harris, 1991; Stockard & Johnson, 1992). Developmentally, gender is one of the earliest and

120

Peter Glick and Susan T. Fiske

strongest forms of group identity to be internalized (Maccoby, 1988), and people are more likely to categorize others on the basis of gender than on the basis of race, age, or role (A. P. Fiske, Haslam, & Fiske, 1991; Stangor, Lynch, Duan, & Glass, 1992). Social identity theory (Tajfel, 1981) suggests that the tendency to differentiate between groups will be strong when social status is bound up with group membership, helping to create social ideologies that justify the status differences. Like dominative paternalism, competitive gender differentiation presents a social justification for male structural power: Only men are perceived as having the traits necessary to govern important social institutions. This creates downward comparisons, in which women serve, in Virginia Woolf’s (1929/1981) words, as “looking-glasses possessing the magic and delicious power of reflecting the figure of a man at twice its natural size” (p. 35), allowing individual men to enhance their self-esteem by association with a male social identity (Tajfel, 1981). Alongside the competitive drive to differentiate, however, the dyadic dependency of men on women (as romantic objects, as wives and mothers) fosters notions that women have many positive traits (Eagly, 1987; Eagly & Mladinic, 1993; Peplau, 1983) that complement those of men (complementary gender differentiation). Just as the traditional division of labor between the sexes creates complementary roles (men working outside the home, women within), the traits associated with these roles (and hence with each sex) are viewed as complementary. The favorable traits ascribed to women compensate for what men stereotypically lack (e.g., sensitivity to others’ feelings). Hence a man may speak of his “better half”; for the benevolent sexist, the woman completes the man. Heterosexuality Virginia Woolf (1929/1981) hazarded her own answer about the reasons for polarized images of women in literature: “the astonishing extremes of her beauty and horror; her alternations between heavenly goodness and hellish depravity” are as “a lover would see her as his love rose or sank, was prosperous or unhappy” (p. 83). Heterosexuality is, undoubtedly, one of the most powerful sources of men’s ambivalence toward women. Heterosexual romantic relationships are ranked by men (and women) as one of the top sources of happiness in life (see Berscheid & Peplau, 1983; Brehm, 1992), and these relationships are typically nominated as the most psychologically close and intimate relationships men have (Berscheid et al., 1989). Men’s sexual motivation toward women may be linked with a genuine desire for psychological closeness (heterosexual intimacy). Although, at their best, heterosexual relationships are the source of euphoric and intimate feelings (Hatfield, 1988), romantic relationships between men and women also pose the greatest threat of violence toward women (Unger & Crawford, 1992). Men’s dyadic dependency on women creates an unusual situation in which members of a more powerful group are dependent on members of a subordinate group. Sex is popularly viewed as a resource for which women act as the gatekeepers (Zillmann & Weaver, 1989).

The Ambivalent Sexism Inventory 121 This creates a vulnerability that men may resent, which is reflected in the frequency with which women are portrayed in literature as manipulative “temptresses,” such as Delilah, who can “emasculate” men. The belief that women use their sexual allure to gain dominance over men (who would, in vulgar parlance, be called “pussy-whipped”) is a belief that is associated with hostility toward women (Check, Malamuth, Elias, & Barton, 1985). As Bargh and Raymond (1995) and Pryor, Giedd, and Williams (1995) demonstrated, for some men sexual attraction toward women may be inseparable from a desire to dominate them (heterosexual hostility).

The nature of sexist ambivalence We have suggested that sexist ambivalence stems from simultaneously holding two sets of related sexist beliefs: hostile and benevolent sexism. We label this ambivalent sexism because we believe that these two constructs subjectively entail opposite evaluative feeling tones toward women (a claim for which we offer supportive data in the studies that follow). However, the present conception of ambivalence proposes that hostile and benevolent sexism may be positively correlated, whereas other ambivalence theorists have assumed (and have found) that beliefs associated with ambivalence are typically conflicting (and therefore negatively correlated) or, at best, are unrelated (Cacioppo & Bernston, 1994; Thompson, Zanna, & Griffin, 1995). This raises the question: If the two sets of beliefs are positively correlated, can they be called “ambivalent”? We characterize them as ambivalent because, even if the beliefs about women that generate hostile and benevolent sexism are positively related, they have opposing evaluative implications, fulfilling the literal meaning of ambivalence (“both valences”). Eagly and Chaiken (1993) and Thompson et al. (1995) suggested that many different forms of ambivalence are possible because of the multidimensional nature of attitudes. For example, a man may hold two beliefs about women that he views as entirely consistent with each other (e.g., “Women are incompetent at work” and “Women must be protected”), yet these beliefs could yield opposing evaluations. Thus, a measure that focuses on beliefs about women, as ours does, could show a positive correlation between beliefs that are, nevertheless, diagnostic of opposing valences toward women. Another reason why sexist individuals may, in our terms, be ambivalent toward women without experiencing any sense of confusion, conflict, or tension about these attitudes is that sexist ambivalence may generally take the form of dividing women into favored in-groups—consisting of women (e.g., homemakers) who embrace traditional roles that fulfilled paternalistic, gender identified, and sexual motives of traditional men—versus disliked out-groups— consisting of women (e.g., feminists) who challenge or threaten these needs and desires. Many researchers (e.g., Deaux, Winton, Crowley, & Lewis, 1985; Taylor, 1981) have argued that women are typically classified in terms of such subtypes. Ambivalent sexism may be most evident in polarized views of these different types (e.g., the notion of women as “saints” or “sluts”). It is worth

122

Peter Glick and Susan T. Fiske

noting that it is precisely this form of ambivalence (polarized reactions toward different target individuals who arouse either the positive or the negative aspect of ambivalent attitudes) that is typically demonstrated in research on racial ambivalence (e.g., Rogers & Prentice-Dunn, 1981). By differentiating women in this manner, men could maintain a sense of attitudinal consistency (“I hate some women but love others”), even though they are quite ambivalent toward women as a whole. Furthermore, this differentiation into subtypes may help ambivalent sexists justify their attitudes as not prejudicial toward women overall because it is only certain types of women whom they dislike. This “unconflicted” form of ambivalence is not mutually exclusive with the possibility that particular female targets might arouse consciously conflicting feelings in ambivalent sexists. Women who simultaneously fit into a desired subtype on one dimension but fit into a hated subtype on another may arouse a conflicted form of ambivalence. Imagine, for instance, a sexist man’s attitudes toward a daughter who is a radical feminist. Or consider sexist men’s attitudes toward sexy women. We have argued above that such women may arouse conflicting feelings among sexist men, who find them sexually attractive but potentially dangerous as “temptresses” who can use their allure to dominate men. Thus, ambivalence may be evident in both an unconflicted form, in which different subtypes of women elicit either extremely positive or extremely negative reactions, as well as a conflicted form, in which particular female targets activate both hostile and benevolent motives.

Ambivalent sexism and ambivalent racism Because recent theories of racism emphasize ambivalence, it is important to consider how these theories compare with our approach. Gaertner and Dovidio’s (1986) theory of aversive racism postulates that White hostility toward Blacks is well learned and automatic. A desire to be egalitarian conflicts with these feelings, leading aversive racists to bend over backward to demonstrate their egalitarianism but to exhibit hostility whenever the target or situation provides attributional ambiguity. A related approach is the construct of symbolic racism (Kinder & Sears, 1981; McConahay, 1986). Advocates of this theory view racism as emerging covertly in policy-related attitudes (e.g., opposition to affirmative action) for reasons similar to Gaertner and Dovidio’s (the individual can attribute the attitude to non-racial motives). Readers interested in how this approach may extend to sexism should refer to Swim, Aikin, Hall, and Hunter’s (1995) Modern Sexism scale and to Tougas, Brown, Beaton, and Joly’s (1995) Neo-Sexism scale, which examine gender-related policy attitudes. Our ambivalent sexism approach shares some similarities with these theories. We believe that sexist attitudes also become automatic and that sexist ambivalence polarizes responses to different members of the target group. Unlike the two racism theories, however, we propose that sexist men have genuinely positive feelings, as well as hostile attitudes, toward women and that the desire to project and protect an egalitarian image is much less relevant to explaining

The Ambivalent Sexism Inventory 123 sexist behavior (assuming, as Fiske & Stevens, 1993, argued, that people are not as worried about appearing sexist as they are about appearing racist). Rather than viewing ambivalent sexism as a conflict between more recent egalitarian beliefs toward women and continuing hostility, we have traced the roots of the “positive” side of sexist ambivalence to the dyadic power that women have always held, by virtue of the intimate ties between the sexes that sexual reproduction creates. Our conception bears more similarity (and debt) to Katz and his colleagues’ (Katz & Hass, 1988; Katz, Wackenhut, & Hass, 1986) analysis of racial ambivalence, for Katz et al. presumed that there are genuine pro-Black feelings among Whites. This view has been disputed by others (e.g., Gaertner & Dovidio, 1986) who have argued that “sympathy for the underdog” is not a truly pro-Black attitude. Indeed, there may be an element of paternalism in what Katz et al. termed pro-Black beliefs, as we believe there is in benevolent sexist views. Whether or not truly positive subjective feelings toward Blacks are common among Whites, we think that such feelings toward women are common among men. We do not label the “positive” side of this ambivalence “pro-women” attitudes for, as we have detailed above, we believe that these attitudes are themselves sexist. Furthermore, Katz et al., like other racism theorists, viewed the desire to protect an egalitarian self-image as the source of the positive side of ambivalence, which we do not, and Katz et al. stressed that pro- and anti-out-group beliefs are in conflict, which we do not. Thus, these aspects of their theory are dissimilar to our own.

Measuring ambivalent sexism Although we have discussed the wider social conditions that foster ambivalent sexism, our research takes an individual-differences approach to measuring ambivalent sexism because, as Snyder and Ickes (1985) argued, it allows a methodological entrée into the investigation of the social psychological processes of interest. Furthermore, there is likely to be a great deal of variability in the degree to which hostile and benevolent sexism characterize individuals. Many individual-differences measures of sexism currently exist. Most follow in the footsteps of Spence and Helmreich’s (1972; Spence, Helmreich, & Stapp, 1973) Attitudes Toward Women Scale (AWS). Such scales (e.g., Beere, King, Beere, & King’s, 1984, Sex-Role Egalitarianism Scale) are more precisely labeled as measures of sex role traditionalism versus egalitarianism or, as the AWS is subtitled, as measures of “Attitudes Toward the Rights and Roles of Women” (Spence & Helmreich, 1972). Because egalitarian social beliefs have become widely embraced by Americans in recent years, researchers have now turned to assessing more subtle aspects of sexism (see Swim et al.’s [1995] Modern Sexism scale and Tougas et al.’s [1995] Neo-Sexism scale). Rather than directly assessing egalitarianism and traditionalism, these scales focus on current genderrelated political issues (particularly on the denial of continuing discrimination against women, which is thought to reflect underlying hostility).

124

Peter Glick and Susan T. Fiske

None of the measures just listed, however, distinguishes between sexism’s hostile and benevolent components, which was our primary goal in developing the measure presented here, the Ambivalent Sexism Inventory (ASI). The ASI was developed to tap the three subcomponents hypothesized to make up hostile and benevolent sexism: paternalism (dominative and protective), gender differentiation (competitive and complementary), and heterosexuality (hostile and intimate). We constructed our measure with the following goals in mind: (a) it must assess both hostile and benevolent sexism and (b) for pragmatic reasons, the final measure should be easily administered, able to be completed quickly, and simple to score (these concerns dictated a short self-report measure).

Methods In this article we present data from six studies conducted to develop and validate the ASI. For conceptual simplicity and comparative purposes, the results are grouped according to the theoretical and empirical issues addressed rather than study by study (although the source of the data is clearly labeled in each case). Participants and procedures The six studies reported here involved 2,250 individuals. Recruitment procedures varied among studies. Participants were assured of the anonymity of responses, either by returning surveys anonymously through campus mail (Study 1) or in a sealed container with other completed surveys (Studies 2–6; with the exception of Study 3, a large participant pool prescreening study for which responses were confidential but not anonymous). When a cash lottery was used as incentive (Studies 1, 4, and 5), participants filled out lottery entry forms that were separate from their survey responses. Sample 1 Study 1 included three subsamples of roughly similar size used for the initial development of the ASI: 833 students (353 men and 480 women) at three different colleges (University of Massachusetts at Amherst, Amherst College, Lawrence University) completed a 140-item survey of “attitudes toward men and women and their relationships in contemporary society” (the initial pool of items from which the ASI was derived). All participants took part in a cash lottery (prizes of $100, $60, and $40) as an incentive. Although the three subsamples were similar in terms of average age (ranging from 19.54 to 20.69 years), ethnic composition (76%–86% of respondents on each campus were White), and education (virtually all respondents were full-time undergraduate students), the three campuses differ in terms of geographical location (two in the East, one in the Midwest) and in selectivity and size (two are small, selective, private undergraduate liberal arts colleges, whereas the other is a less selective, large, state-funded institution).

The Ambivalent Sexism Inventory 125 Sample 2 The second sample consisted of 171 University of Massachusetts undergraduates in introductory psychology courses, who were recruited for a survey of “men and women and their relationships in contemporary society.” Although the age and ethnicity of participants were not recorded, the sample appeared to be similar to the students in Sample 1. The participants (77 men and 94 women) earned extra credit points for their classes. A series of questionnaires was administered in same-sex groups ranging from 4 to 35 participants. In all sessions, the researcher was of the same sex as the participants. Participants completed a short version of the ASI and a number of other measures of sexism (e.g., the AWS; see below). Answers to the surveys were filled out on computerized optical scan sheets on which respondents were instructed not to include their name or any other identifying information (other than their sex). Sample 3 In Study 3, all University of Massachusetts undergraduates enrolled in introductory psychology classes were recruited for a general prescreening that would make them eligible for subsequent studies. A total of 937 participants (396 men and 541 women) received extra credit points in their classes. The participants were similar in age to those in Studies 1 and 2; 81% were White (with Asian being the next largest category at 6%). These individuals completed a short version of the ASI embedded among a battery of scales (measures from a variety of researchers who wished to use the prescreen for selection purposes), including personality inventories (e.g., California Personality Inventory Dominance scale) and attitude measures (e.g., Modern Racism scale). Samples 4 and 5 Studies 4 and 5 tapped nonstudent samples. In Study 4, 72 men and 72 women were recruited in public areas (malls, restaurants, laundromats) in a variety of locations in Massachusetts (including urban neighborhoods of Boston as well as more rural settings in western Massachusetts). An effort was made to avoid areas populated by undergraduate college students (e.g., Amherst, Cambridge). In Study 5, 36 men and 76 women were recruited. About half of these were obtained in the same geographical area and in similar fashion to Study 4, whereas the other half were recruited in the Midwest by a number of Lawrence University student volunteers, each of whom recruited about 2 or 3 adult nonstudents for the survey (many of these were parents of students living in Wisconsin, Minnesota, and Illinois). Participants were asked to complete a survey “about men and women and their relationships in contemporary society.” Participants’ ages ranged from 18 to 77 years (with a median of 34), were typically White (83%), and held a variety of occupations (including blue-collar jobs, such as plumber, truck driver, mechanic; pink collar jobs, such as receptionist, clerk, secretary; white collar jobs, such as editor, electrical engineer, architect,

126

Peter Glick and Susan T. Fiske

lawyer, physician; and some homemakers). Although these samples by no means represent a random cross-section of Americans, they were quite different from and much more diverse than the samples described earlier. Lotteries (with prizes of $50, $25, and $10) were used as an incentive for participation. Participants completed a short version of the ASI and a questionnaire concerning their attitudes toward and the traits they ascribe to women (or men, in Study 4 only, depending on random condition assignment). Sample 6 Study 6 included a small University of Massachusetts sample of 44 men and 41 women, similar in characteristics to the students in Study 2 and recruited and run in the same manner. This study replicated methods used in Studies 4 and 5 (with nonstudent samples) for comparative purposes. The ASI The questionnaire used in Study 1 consisted of 140 statements with which respondents were asked to indicate agreement or disagreement on a scale that ranged from 0 (disagree strongly) to 5 (agree strongly) with no midpoint (respondents were forced to agree or disagree at least slightly with each item). In all subsequent studies reported here, 22–32 items selected from the initial pool were used as a short version of the ASI. In some instances, items were reworded to yield the reverse meaning to control for acquiescence bias. Overall, we generated the initial items to represent the conceptual categories derived from our theoretical analysis of hostile and benevolent sexism. Hostile sexism Hostile sexism items tapped the categories of dominative paternalism (e.g., “The world would be a better place if women supported men more and criticized them less”), competitive gender differentiation (e.g., “A wife should not be significantly more successful in her career than her husband”), and heterosexual hostility (e.g., “There are many women who get a kick out of teasing men by seeming sexually available and then refusing male advances”). Each subcategory was represented by approximately 15 items. Benevolent sexism Benevolent sexism items tapped the categories of protective paternalism (e.g., “Every woman should have a man to whom she can turn for help in times of trouble”), complementary gender differentiation (e.g., “Many women have a quality of purity that few men possess”), and heterosexual intimacy (e.g., “People are not truly happy in life unless they are romantically involved with a member of the other sex”). Each subcategory was represented by approximately 15 items.

The Ambivalent Sexism Inventory 127 Although we do not believe that egalitarianism is part of the subjectively positive feelings men have toward women, because of its importance to modern racism and modern sexism theories, we included a number of such items. Nine items were adapted from Katz and Hass’s (1988) pro-Black scale by converting the target group to women (e.g., “Women do not have the same employment opportunities that men do”). In addition, we generated several similar items expressing a recognition of continuing discrimination against women (e.g., “Popular culture is very sexist”). Finally, six obviously correct or incorrect statements were included among the initial pool of 140 items as validity items to check whether respondents were responding with due care (e.g., “Few secretarial jobs are held by women”).

Results Data from the six studies documenting the factor structure, convergent and discriminant validity, and predictive validity of the ASI scales are presented below. For all analyses, sex differences between participants were examined and are reported in those cases in which significant differences were found. Factor analyses Exploratory factor analysis with Sample 1 Study 1 was aimed at winnowing the initial 140-item pool down to a short set of items that tapped hostile and benevolent sexism. We excluded from further analysis items with extreme means (i.e., not much room for variation), based on arbitrary cutoffs of 1 or less and 4 or more (on our 0–5 scale). The 6 validity items and an additional 22 items were excluded on this basis.4 The remaining 112 survey items were factor analyzed with principalcomponents analysis in SPSSX (Version 4.0, 1993) with a varimax rotation. Separate analyses of respondents from the three colleges involved in Study 1 yielded impressively consistent results, as did separate analyses of male and female respondents (pooled across the different college samples).5 The consistency of results, across colleges and between the sexes, enabled us to perform a more powerful analysis involving all respondents (resulting in a cases-to-item ratio of 7.46:1, which is within the range generally deemed acceptable for factor analysis). We conducted a principal-components analysis on all of the 112 acceptable items using SPSSX with a varimax rotation. The two strongest factors corresponded closely to our theoretical constructs: a Hostile Sexism (HS) factor (Factor 1, with an eigenvalue of 25.64, accounting for 23% of the variance) and two Benevolent Sexism (BS) factors (Factor 2: protective paternalism and heterosexual intimacy, eigenvalue of 6.30, accounting for 6% of the variance; Factor 3: complementary gender differentiation, eigenvalue = 3.45, accounting for 3% of the variance). Although BS emerged in two separate factors in this

128

Peter Glick and Susan T. Fiske

analysis, the BS items converged into one factor in separate analyses for male and female respondents (reported in note 5). There were 21 additional factors with eigenvalues greater than 1, one of which (Factor 4) is of particular interest. Factor 4 (eigenvalue = 2. 72, accounting for 2% of the variance) consisted mainly of the items adapted from Katz and Hass (1988). All eight items adapted from their pro-Black scale loaded .3 or better on this factor. The three other items with high loadings were similar in tone to the Katz and Hass items; these items expressed the view that discrimination against women is a serious problem in our society. We labeled this factor recognition of discrimination. Disagreement with the items on this factor would (in our opinion) correspond closely to the gender equivalent of symbolic or modern racism (Kinder & Sears, 1981; McConahay, 1986). Indeed, the strongest component of Swim et al.’s (1995) notion of modern sexism is denial of continued discrimination against women. That these items cluster on a separate factor from our hostile and benevolent sexism items provides preliminary evidence of discriminant validity for our constructs. The remaining 20 minor factors with eigenvalues greater than 1 are not reported, either because they consisted of too few items for meaningful interpretation or were of little theoretical interest (e.g., Factor 5, which accounted for 2% of the variance, measured heterosexual arousability independently of sexism with items such as “Watching physically attractive members of the other sex is exhilarating”). Factors 6–24 were generally uninterpretable; each accounted for less than 2% of the variance. We used our initial exploratory analyses as a sieve to aid selection of a smaller set of items, both to yield a short scale and so that analytic techniques of greater sophistication and power could be performed. To avoid redundancy, we use these subsequent analyses to detail the nature of the two subscales. In the Appendix we report the final 22-item ASI scale (with HS and BS subscales) constructed on the basis of these analyses. Selection of restricted item set We narrowed the ASI to 22 items (11 HS and 11 BS) on the basis of: (a) the items’ tendency to load consistently highly on the HS and BS factors that emerged in the separate factor analyses for men and women (reported above), (b) maintaining diversity in the various aspects of sexism apparently tapped by the items, and (c) consistent performance by the items in subsequent studies (after a “cut” to 16 items each, the scales were reduced to their 11-item forms based on their performance in Studies 1–4). Exploratory factor analyses with the 22-item ASI suggested one HS factor and three BS factors (corresponding to the three predicted BS subfactors). Rather than reporting the exploratory analyses in detail, we present structural models of the ASI based on confirmatory factor analysis conducted with LISREL 8.0 (Jöreskog & Sörbom, 1993). The preferred structural model was guided by our theoretical analysis and the exploratory principal-components analyses in SPSSX.

The Ambivalent Sexism Inventory 129 Confirmatory factor analysis The advantages of using confirmatory factor analytic procedures with LISREL are threefold: (a) it allows the construction of specific structural models based on theoretical models, (b) models can be rigorously tested and compared through assessment of their goodness of fit to the observed data, and (c) models can include second-order factors. These features are particularly important because our theoretical model proposes that BS is a distinct component of sexism and that it has three subfactors (paternalism, gender differentiation, and sexuality). Further exploratory analyses based solely on the restricted 22-item set suggested that our proposed subfactors were represented in the data for the BS items but not for the HS items (which proved to be strongly unidimensional; separate analysis of the HS items revealed only a single factor with an eigenvalue greater than 1, accounting for 50% of the variance). Because there was no apparent empirical basis for distinguishing the theoretical HS subfactors, we did not attempt to include HS subfactors in the model. The 11 BS items, however, did split into three subfactors in exploratory analyses involving only the restricted item set, leading us to test a theoretical model with second-order factors. If BS is a separate (though possibly related) entity from HS, then a one-factor model should account for the data significantly less well than a two-factor model, which, in turn, may be improved by adding the three BS subfactors. We tested each of these models: a one-factor model (all items assigned to a single sexism factor); a two-factor model (each item assigned to load on either an HS or a BS factor, with an assigned loading of 0 on the other factor); and a full model with two second-order factors—HS and BS—and three BS subfactors: complementary gender differentiation, heterosexual intimacy, and protective paternalism. The full model was highly restrictive (and hence a rigorous test of the proposed structure): (a) all HS items were assigned a loading of 0 on BS and vice versa, and (b) for all BS items, loadings were estimated on one subfactor only (each item was assigned a loading of 0 on the other two subfactors). If the observed data did not fit these many restrictions, the model would fail to have a good fit to the data. The full model is depicted in Figure 6.1.

Hostile Sexism

Benevolent Sexism

Protective Paternalism

Complementary Gender Differentiation

Heterosexual Intimacy

Figure 6.1 Preferred (full) model for confirmatory factor analysis of the Ambivalent Sexism Inventory

130

Peter Glick and Susan T. Fiske

Factor structure and model comparisons across five studies We present five independent replications of the confirmatory factor analysis (Study 6 was excluded from the factor analyses because of low sample size). These replications include three undergraduate samples (Studies 1–3) and two nonstudent samples (Studies 4–5). Two of the samples were relatively large (more than 800 respondents in each), and three were relatively small (fewer than 150 people in each). Because all of the items on the initial scale were worded such that agreement indicated a higher level of sexism, a number of items were reworded for Studies 2–6 to control for measurement biases. Acquiescence bias could inflate (or could even completely account for) any positive correlation that emerged between the HS and BS factors without item reversals (e.g., Green, Goldman, & Salovey, 1993). To control for this possibility, 8 of the 22 items (4 HS and 4 BS items) were reversed in Studies 2 and 3. For Studies 4–6, the 2 reversed items that had performed least well (1 HS and 1 BS item) were restored to their original wording. Thus, for Studies 2 and 3, 36% of the items were reversed, and for Studies 4–6, 27% of the items remained reversed. Goodness of fit. We used Jöreskog and Sörbom’s (1993) Goodness of Fit Index (GFI) and Adjusted Goodness of Fit Index (AGFI) to assess the fit of the data to the models. These measures, as opposed to chi-square, are standardized and can be interpreted easily.6 Both statistics are on a 0-to-1 scale, with 1 indicating a perfect fit (i.e., that the observed correlation matrix can be perfectly reproduced by the model’s estimates). Jöreskog and Sörbom (1993) consider AGFI estimates of .85 and above to indicate a good fit. In addition to these measures, we used chi-square to test differences between the fit of the alternative models through the likelihood ratio or chi-square difference test (Bollen, 1989). Fit measures for the three models are reported in Table 6.1 for the five samples. As is indicated in Table 6.1, in all five studies the two-factor model fit the data significantly better than the one-factor model, suggesting that BS is indeed a separate entity from HS. Furthermore, the full model represented a significant improvement over the two-factor model, indicating that BS can best be described as being composed of three subfactors. The full model performed well across all five studies and yielded acceptable GFIs. Although the fit is weaker in three of the five samples (Studies 2, 4, and 5), the reader should note that these are the three small samples and that item loadings (reported below) remain high for these studies. Given the complexity of the model being tested, we consider these results strong evidence for the validity of this structural model. Factor correlations. Even though BS is distinct from HS, the two factors are strongly and positively correlated (see Table 6.2 for factor correlations).7 Acquiescence bias cannot explain this result. The reversal of a significant number of items in Studies 2–5 did not consistently reduce the correlation between the HS and BS factors: Studies 2, 3, and 5 yielded fairly strong correlations between HS and BS that were, on average, only slightly lower than the .71 correlation in the first study. Thus, the correlation between HS and BS cannot be attributed to acquiescence bias. The two nonstudent samples (Studies 4 and 5),

The Ambivalent Sexism Inventory 131 Table 6.1 Goodness of fit (GFI) of full and restricted models across five samples Study Fit index

1

2

3

4

5

One-factor model GFI AGFI x2 (209)

.77 .80 .72 .76 1,862.75** 460.47**

.79 .69 .75 .62 2,017.19** 554.24**

.72 .66 441.45**

Two-factor model (no subfactors) GFI AGFI Decrease in x2 (vs. one factor)a

.89 .83 .87 .80 892.82** 106.61**

.83 .82 .80 .78 810.64** 161.30**

.79 .74 69.35**

.92 .86 .90 .82 413.56** 103.86**

.80 .75 20.40**

Full model GFI AGFI Decrease in x2 (vs. one factor)b

.94 .93 416.68**

N

811

.86 .83 31.56** 171

937

144

112

Note: AGA = Adjusted Goodness of Fit Index. a = distributed as χ2 with 1 df. b = distributed as χ2 with 2 df. ** p < .01.

Table 6.2 Correlations among ASI factors across five samples Study ASI factor

1

Subordinate ASI factors: HS and BS Loadings of subfactors on BS factor: – protective paternalism – complementary gender differentiation – heterosexual intimacy N

811

2

3

4

5

.71

.74

.62

.37

.58

.93

.98

.92

.98

.93

.81 .79

.75 .76

.77 .79

.72 .75

.92 .72

171

937

144

112

Note: ASI = Ambivalent Sexism Inventory; HS = Hostile Sexism; BS = Benevolent Sexism.

Table 6.3 Factor loadings for ASI items across five studies Study Scale item

1

2

3

4

5

Hostile Sexism Women exaggerate problems at work Women are too easily offended Most women interpret innocent remarks as sexist When women lose fairly, they claim discriminationb Women seek special favors under guise of equality Feminists are making reasonable demandsa Feminists not seeking more power than mena Women seek power by gaining control over men Few women tease men sexuallya Once a man commits, she puts him on a tight leash Women fail to appreciate all men do for them

.71 .76 .74

.70 .81 .69

.71 .66 .61

.80 .69 .55

.73 .66 .70

.74

.49

.31

.77

.66

.68

.74

.70

.59

.71

.75 .73 .67

.60 .50 .64

.50 .47 .72

.49 .56 .70

.42 .64 .69

.60 .73

.51 .65

.37 .65

.25 .81

.46 .77

.68

.69

.66

.64

.58

.68 .69

.58 .43

.66 .28

.58 .66

.62 .49

.69 .62

.54 .48

.73 .35

.67 .33

.69 .47

.69 .82

.74 .80

.75 .82

.77 .78

.56 .61

.72

.69

.71

.67

.71

.67

.57

.69

.64

.55

.69 .79

.70 .75

.51 .84

.63 .66

.55 .71

.67

.50

.37

.36

.44

Benevolent Sexism Protective Paternalism – A good woman should be set on a pedestal – Women should be cherished and protected by menb – Men should sacrifice to provide for women – In a disaster, women need not be rescued firsta Complementary Gender Differentiation – Women have a superior moral sensibility – Women have a quality of purity few men possess – Women have a more refined sense of culture, taste Heterosexual Intimacy – Every man ought to have a woman he adores – Men are complete without womena – Despite accomplishment, men are incomplete without women – People are also happy without heterosexual romancea N

811

171

937

144

112

Note: ASI =Ambivalent Sexism Inventory. a indicates items reverse-worded (and reverse-scored) for Studies 2–6 and on the final scale. b Indicates items for which reversed wording (and reversed scoring) was used in Studies 2 and 3 but which were returned to their original wording for the final version of the scale and for Studies 4–6.

The Ambivalent Sexism Inventory 133 however, yielded lower correlations between HS and BS. Analyses that shed further light on this difference are reported below. The reader should keep in mind that the magnitude of the correlation between the HS and BS factors does not affect the fit of the models we tested, because this parameter is not restricted. Factor loadings of individual items. Comparisons across the studies reveal that although the reversed items typically did not fare as well as the original items, their loadings were generally acceptable, and the factor structure was consistent across studies. Because item reversals did not have any dramatic effects on the correlations among the factors, recall that the two reversed items that performed least well were restored to their original wording beginning with Study 4. Individual items (with a “gist” wording) and factor loadings across the studies are reported in Table 6.3; all reversed items were reverse scored. The reader should keep in mind that, in this confirmatory factor analysis (which fit the observed data rather well), items were allowed to load only on one factor (or subfactor); all other loadings were fixed at 0 (and therefore are not reported). Consistency of factor structure for men and women. A formal test of the consistency of factor structure between men and women was possible within Studies 1 and 3, each of which had sufficient sample sizes to perform subgroup analyses that tested the fit of the data to the model simultaneously but separately for male and female respondents. The GAs for these analyses were .92 for Study 1 and .91 for Study 2, which indicates that for each of these samples the factor structure was similar for male and female respondents. Properties of the raw score ASI scales across six studies Reliability Researchers interested in using the ASI will most likely rely on raw score averages when using the scale. Reliability analyses of a total ASI score (average of all items) and average scores for the two major subscales of the ASI yielded acceptable alpha coefficients across all six studies (see Table 6.4). The BS scale consistently yielded lower alpha coefficients; this is not surprising given the Table 6.4 ASI scale reliabilities (alpha reliability coefficients) across six samples Study ASI scale

1

ASI Hostile sexism Benevolent sexism N

2

.92 .92 .85 811

3

.88 .87 .75 171

Note: ASI = Ambivalent Sexism Inventory.

4

.83 .80 .77 937

5

.83 .87 .78 144

6

.87 .91 .73 112

.90 .89 .83 85

134

Peter Glick and Susan T. Fiske

multidimensional nature of this scale. The BS subscales are composed of too few items (three for complementary gender differentiation, four each for protective paternalism and heterosexual intimacy) to yield acceptable reliability scores when analyzed separately. Relationship between HS and BS scales The correlations between the raw score versions of the HS and BS scales were consistently lower than the correlation between the HS and BS factors estimated in the LISREL analyses reported above (most likely because the LISREL analyses control for error variance that otherwise attenuates the correlation). Although the two studies (1 and 3; both with undergraduates) with sufficient sample size yielded similar factor solutions for men and women, the correlations between the raw scores for the HS and BS scales (see Table 6.5) suggest that they differ for men and women in the nonstudent samples (Studies 4 and 5). Although the correlation for men is significantly lower than the correlation for women in one student sample (Study 3, z = 2.47, p < .01), the differences between men and women are dramatic only for the nonstudent samples: HS and BS were positively correlated only for women in the nonstudent samples, not for men. In both cases, the differences between the correlations for women and men were significant (Study 4: z = 3.78, p < .01; Study 5: z = 4.10, p < .01). Sex differences in mean scores We expected mean differences between men and women on the ASI scales. Indeed, if men were not found to be more sexist than women, the validity of the scale would be in doubt. We examined sex differences in average raw scores for the overall ASI (all items), HS, and BS in each of the six studies. We also performed a 2 (sex of respondent) × 2 (ASI subscale: HS and BS) analysis of variance (ANOVA) within each study, using the HS and BS scores as a repeated measures variable. The main effect for sex of respondent in each analysis Table 6.5 Correlations between hostile and benevolent sexism scales across six samples by sex of respondent Study Sex of respondent

1

Men: r N

344

Women: r N

467

** p < .01.

2

.55**

3 .45**

77 .56**

.31** 396

.57** 94

4 −.12 72

.45** 541

5 −.15 36

.48** 72

6 .53 44

.61** 76

.56** 41

The Ambivalent Sexism Inventory 135 is equivalent to a univariate test for sex differences in the overall ASI means (because the overall ASI score is the average of the HS and BS scores). Means for these analyses are reported in Table 6.6. In all cases, the sex-of-respondent main effect was significant (all Fs > 4.82, p < .05), such that men scored higher on the ASI than did women. Although men tended to score higher than women on both subscales of the ASI, the differences were more extreme for HS than for BS. Indeed, the interaction terms for all analyses were significant (all Fs > 3.79, p < .05). Pairwise comparisons (within each study) of men’s and women’s scores on the HS and BS scales (based on Tukey’s test for post hoc comparisons) revealed that men scored higher than women on HS and on BS in every study (all ps < .05) with the exception of Study 6, for which men scored higher on HS only. The significant interaction terms, however, indicate that the difference between the HS scores of men and women were significantly larger than the differences between the BS scores. It is not

Table 6.6 ASI Scale means for men and women across six samples Study ASI scale

1

2

3

4

5

6

ASI – M – SD

2.96 .88

2.53 .61

2.46 .61

2.46 .62

2.52 .63

2.45 .75

Hostile Sexism – M – SD

3.05 1.04

2.49 .74

2.38 .78

2.63 .95

2.72 .97

2.54 .86

Benevolent Sexism – M – SD

2.87 .97

2.58 .69

2.53 .74

2.31 .92

2.33 .95

2.36 .85

Male respondents

N

344

77

396

72

36

44

Female respondents ASI – M – SD

2.41 .82

1.85 .76

1.97 .72

1.82 .87

1.78 .89

2.07 .84

Hostile Sexism – M – SD

2.38 .95

1.49 .88

1.73 .84

1.67 1.03

1.66 1.05

1.87 .98

Benevolent Sexism – M – SD

2.43 .96

2.21 .83

2.20 .84

1.98 1.01

1.90 .94

2.27 .92

N

467

94

541

72

Note: Each scale ranged from 0 (disagree strongly) to 5 (agree strongly).

76

41

136

Peter Glick and Susan T. Fiske

surprising that women would be more likely to reject sexist attitudes that are hostile toward women than those that carry potential benefits, as BS does (i.e., protected status and favorable stereotypes).8 Convergent and discriminant validity of the ASI Relationship of HS and BS to recognition of discrimination We suggested earlier that the recognition of discrimination (RD) factor (based on a gendered translation of items from Katz and Hass’s [1988] pro-Black scale), which emerged in Study 1 as a separate factor in exploratory analyses, provides evidence of discriminant validity for the ASI. The reader may wonder, however, given the multifaceted nature of the BS scale, whether the RD factor is indeed part of BS (given that Katz and Hass designed these items to express sympathy with the plight of the out-group). We formed a reliable RD scale (α = .77) by averaging the 11 items that loaded .3 or better on the RD factor (items included in Study 1 only). A regression analysis in which HS and BS (raw score versions) were entered as predictors of RD revealed that RD is more strongly (and negatively) related to HS. Once HS was partialed out, RD was weakly (but positively) related to BS. The standardized regression coefficients for the combined regression were -.52 for HS (t = −12.76, p < .001) and .25 for BS (t = 6.24, p < .01). This result is consistent with a sexism equivalent of the claim that the denial of discrimination against the minority group masks an underlying hostility (e.g., McConahay, 1986; see Swim et al., 1995). Furthermore, the different directions of the relationships of RD to HS and RD to BS provide additional evidence of the importance of distinguishing between these two aspects of sexism. The subjectively positive feelings associated with BS may lead people high in BS to be more sympathetic toward women and, as a result, to be somewhat more aware of the obstacles women face (although perhaps unaware of the sexism inherent in their own views). The ASI and social desirability Participants in Study 2 (N = 161) completed Paulhus’s (1988) Balanced Inventory of Desirable Responding (BIDR), which separately measures impression management and self-deception, the two dimensions Paulhus found to underlie other measures of socially desirable responding. We administered the BIDR using the 0 (strongly disagree)-to-6 (strongly agree) scale recommended by Paulhus. With an issue as sensitive in the current cultural climate as relationships between men and women, particularly on a college campus noted for political correctness, it would be surprising if the ASI were completely unrelated to socially desirable response tendencies. Although the ASI scales were not significantly related to the self-deception scale of the BIDR, the correlations of the impression management (IM) scale of the BIDR with the ASI scales

The Ambivalent Sexism Inventory 137 Table 6.7 Socially desirable responding and the Ambivalent Sexism Inventory (ASI) BIDR scale ASI scale

Self-deception

Impression management

ASI HS BS HS, controlling for BS BS, controlling for HS

−0.10 −.04 .04 −.07 .07

−3.10* −.29* −2.60* −.18* −.13*

Note: N = 171; BIDR = Balanced Inventory of Desirable Responding; HS = hostile sexism; BS = benevolent sexism. * p < .05.

were significant but not large (see Table 6.7). Partial correlations, in which each of the ASI subscales (HS and BS) were controlled, while the other was correlated to the social desirability scales, show that both scales are independently correlated with IM. Despite the statistical significance of the relationships between the ASI scale and IM, the magnitude of the correlations indicates that the two are not redundant. No particular ASI items can be tagged as the “culprits” for this correlation; in no case does any individual item on the ASI correlate more strongly than −.27 with IM, but virtually all of the items tend toward a weak relationship with IM (6 items in the −.20s, 12 items in the −10s). Thus, the overall relationship between the ASI and IM reflects an aggregation of many weak relationships. The ASI and other measures of sexism Four additional measures of sexism and hostility toward women were included in Study 2: The AWS (Spence & Helmreich, 1972); two new scales developed by Swim et al. (1995)—the Modern Sexism scale (which measures political attitudes thought to be related to a more subtle, “modern” sexism) and the Old-Fashioned Sexism scale (which measures more overtly traditional sexist beliefs about women, e.g., “Women are generally not as smart as men”); and the Rape Myth Acceptance Scale (Burt, 1980). Given that sexism has typically been conceptualized as hostile attitudes toward women or women’s rights, we expected the HS scale to correlate well with other measures of sexism, showing convergent validity for this subscale. Because other researchers have neglected the “benevolent” side of sexism, we made the following prediction: Any correlation between the BS scale and the other measures of sexism would be wholly accounted for by its relationship to HS. We constructed three different types of correlation between the ASI scales and other measures of sexism (see Table 6.8): first-order correlations, partial correlations in which IM was controlled, and partial correlations for each of the ASI subscales (HS and BS) in which

138

Peter Glick and Susan T. Fiske

Table 6.8 Relationship of the Ambivalent Sexism Inventory (ASI) and its subscales to other measures of sexism Other sexism measures ASI scale

AWS

Old-fashioned Modern sexism sexism

RMA

ASI HS BS Controlling for impression management – ASI – HS – BS HS, controlling for BS BS, controlling for HS

.63** .68** .40**

.42** .48** .24**

.57** .65** .33*

.54** .61** .32**

.61** .67** .38** .60** .04

.38** .44** .19** .43** -.03

.62** .70** .36** .60** -.06

.54** .61** .31** .55** -.02

Notes: N = 171; AWS = Attitudes Toward Women scale; RMA = Rape Myth Acceptance Scale; HS = Hostile Sexism; BS = Benevolent Sexism. ** p < .01.

each subscale was partialed out from the other (to obtain “pure” measures of HS and BS). The overall ASI score correlated well with other measures of sexism. Furthermore, the relationship of the ASI scales to these measures of sexism and hostility toward women were unaffected when IM was partialed out (giving at least some evidence that the relationship between the ASI and IM does not affect the ASI’s relationship to other measures). The relationship between the ASI and other sexism measures, however, is wholly attributable to the HS subscale, which was as strongly related to other measures of sexism as the overall ASI score. In contrast, the BS scale had consistently lower first-order correlations to the other measures. These correlations completely disappeared once HS was partialed out, indicating that benevolent sexism is not directly tapped by the other sexism scales.9 The ASI and modern racism Although we had no control over which measures (other than the ASI) were included in Study 3 (the prescreen sample, N = 937), one measure that happened to be included had clear relevance to the ASI: the Modern Racism scale (McConahay, 1986). A positive relationship has frequently been found between prejudice against different out-groups; those who are prejudiced against one group are likely to be prejudiced against others (Adorno, FrenkelBrunswik, Levinson, & Sanford, 1950; Allport, 1954; Altemeyer, 1988; Esses, Haddock, & Zanna, 1993). According to McConahay (1986), the Modern

The Ambivalent Sexism Inventory 139 Table 6.9 Relationship of the Ambivalent Sexism Inventory (ASI) to modern racism Modern racism ASI scale

Men

Women

ASI HS BS HS, controlling for BS BS, controlling for HS

.38** .44** .16** .42** .01

.51** .47** .40** .35** .24**

Note: N = 937. HS = Hostile Sexism; BS = Benevolent Sexism. **p < .01

Racism scale taps symbolic attitudes toward Blacks that are related to racial hostility. Therefore, we expected a positive correlation between the Modern Racism scale and HS, but of a magnitude less than the relationship of HS and other sexism scales. We had no reason to expect a correlation between the Modern Racism scale and BS once HS was controlled for (given that BS does not measure hostility). Because the results differed between men and women, the correlations are broken down by respondent sex in Table 6.9. As expected, HS and the Modern Racism scale correlated moderately well. Hostile sexists are also likely to be “modern racists.” Although Modern Racism and BS did not correlate for men, they were weakly but significantly correlated for women. Predictive validity of the ASI We have argued that the ASI taps two aspects of sexism that, though related, have different evaluative valences. In short, the ASI is intended to capture the ambivalent sentiments expressed in the oft-heard lament of men about women: “Can’t live with them, can’t live without them.” We have also claimed that HS and BS underlie ambivalent images of women, with HS being related to negative images of women and BS being predictive of positive stereotypes about women. HS, then, should be correlated with a negative general attitude toward women and negative stereotypes about women as a group. BS, on the other hand, should be correlated with a positive general attitude toward women and the acceptance of positive stereotypes about women. The total ASI score may remain unrelated to any of these measures because it is composed of two subscales with opposing relationships to these criteria. However, the overall ASI score should correlate with indices of men’s ambivalence toward women. We consider these predictions an “acid test” of the ASI as a scale that should help to further distinguish it from other measures currently in use, such as the AWS. As Eagly and Mladinic (1989) noted, the AWS is often misinterpreted by researchers as measuring attitudes toward women rather than attitudes

140

Peter Glick and Susan T. Fiske

toward equal rights for women, which was Spence and Helmreich’s (1972) explicit intention. Eagly and Mladinic (1989) did in fact show that the AWS is unrelated to attitudes toward women. As they noted, a person who is firmly against equal rights for women because he or she embraces traditional beliefs about gender roles may indeed have a very positive attitude toward women and positive, although stereotypic, images of women. This observation is entirely consistent with our own notions of benevolent sexism, which we see as embracing a traditional set of sexist beliefs that are associated with positive feelings about (and positive trait ascriptions to) women. We have adopted, to the extent practically possible, the methods Eagly et al. recommended for measuring the favorability or unfavorability of attitudes toward and stereotypes about women and men (Eagly & Mladinic, 1989; Eagly et al., 1991). We used these methods in Studies 4–6, two nonstudent samples and one sample of undergraduates. For these studies, in addition to the ASI, respondents were asked to rate their general attitudes toward and stereotypes about women (or men, Study 4 only, depending on random assignment).10 Unfortunately, many respondents in these studies failed to complete the entire survey after filling out the ASI, which was at the beginning of the survey (33 incomplete surveys in Study 4, 36 in Study 5, and 14 in Study 6). As a result, sample sizes are relatively low in these studies, but we compensate by offering three replications.11 The attitude measure was a 5-item semantic differential scale used by Eagly et al. (1991) to measure attitudes toward specific social groups (e.g., men, women). The 7-point semantic differential scales included the following bipolar pairs: good–bad, positive–negative, valuable–useless, pleasant–unpleasant, nice–awful. All items were subsequently coded in the positive direction. The scale was highly reliable (αs = .92, .89, and .79 in Studies 4–6, respectively). We measured stereotyping of the group by having participants indicate the percentage of group members who possessed each of 32 traits. These traits, also used by Eagly and Mladinic (1989), originally come from Spence, Helmreich, and Holahan’s (1979) Extended Personal Attributes Questionnaire. The traits are grouped into four sets of eight: masculine–positive (e.g., independent, self-confident), masculine– negative (e.g., arrogant, hostile), feminine–positive (e.g., helpful, gentle), and feminine–negative (e.g., whiny, spineless). All four sets formed reliable scales (alphas ranged from .76 to .91 for each scale across the three studies). Ambivalent attitudes toward women Does the ASI measure ambivalent attitudes toward women? We used the semantic differential items aimed at measuring overall evaluations of women and the positive and negative trait ratings to answer this question. Because the former measure was based on bipolar scales, it cannot directly measure ambivalence. Because the subscales of the ASI ought to tap different poles of ambivalence, however, the HS and BS scales should show opposite correlations to overall attitudes (semantic differential scales). For the latter measures (traits

The Ambivalent Sexism Inventory 141 ascribed to men and women), participants rated positive and negative traits separately, allowing the construction of a direct index of ambivalence with the ambivalence formula suggested by Thompson et al. (1995). This formula, Ambivalence = (Positive + Negative)/2 - Positive - Negative|, indexes the simultaneous intensity of the positive and negative ratings. We constructed two such ambivalence scores—one based on all the traits rated (both stereotypically masculine and feminine) and one based only on the stereotypically feminine traits. The latter score was constructed with the notion that ambivalence toward women may emerge most clearly in traditional stereotypes of them. Men’s ambivalent attitudes toward women. Correlations of the overall ASI score, HS (controlling for BS), and BS (controlling for HS) with the average of the overall attitude measure and the two ambivalence measures are reported in Table 6.10 for men in Studies 4–6.12 For men, the results were very much as predicted: Although the overall ASI score did not predict general attitudes toward women (as expected, because these were bipolar ratings that do not index ambivalence), for men in the nonstudent samples the two subscales had opposite relationships to overall attitude: HS was significantly negatively related to favorable attitudes toward women and BS was significantly positively related to favorable attitudes toward women. Neither correlation was significant for men in the student sample. In general, the more nonstudent men expressed positive semantic differential attitudes toward women, the more benevolent and less

Table 6.10 The Ambivalent Sexism Inventory (ASI) as a predictor of men’s ambivalent attitudes toward women ASI scale

General attitude

Ambivalence (all traits)

Ambivalence (feminine traits)

.21 .24 .03

.35* .38* .08

.63** .73** .27

.57** .70** .15

.41** .37* .00

.49** .48** .00

Nonstudent men, Study 4 (N = 31) ASI HSa BSb

.07 −.37* .29*

Nonstudent men, Study 5 (N = 25) ASI HSa BSb

−.09 .46* .33*

Undergraduate men, Study 6 (N = 36) ASI HSa BSb

−.03 −.12 .08

Note: HS = Hostile Sexism; BS = Benevolent Sexism. a Correlations in this row are partial correlations controlling for Benevolent Sexism scores. b Correlations in this row are partial correlations controlling for Hostile Sexism scores. * p < .05; ** p < .01.

142

Peter Glick and Susan T. Fiske

hostile sexism they expressed. The opposing correlations of the overall attitude ratings with the ASI subscales imply that, for nonstudent men at least, the overall ASI score is a measure of ambivalence. This is directly confirmed by the relationship between total ASI scores and the ambivalence measures. With only one exception across the three studies (including the student sample), both ambivalence measures were positively and significantly correlated with the overall ASI score that men achieved. Interestingly, the HS scale alone also was significantly correlated to the ambivalence measures in most cases. This curious finding is discussed in light of data presented below. Women’s ambivalent attitudes toward women. The results for women (reported in Table 6.11) are less consistent. Although in Study 5 (the second nonstudent sample) the HS and BS subscales parallel the findings with male respondents (showing opposing correlations to overall attitudes about women), the other two studies did not show this effect. In Study 4, correlations of overall attitude with the total ASI score and with both subscales tended to be negative for female respondents, with the overall ASI score significantly negatively correlated with the attitude measure. The correlations of the ASI and BS scale scores with overall attitude toward women were both significantly different from the corresponding correlations for male respondents (both zs > 2.13, p < .05). Study 6 showed no significant relationships between the ASI scores and the attitude measure. The ambivalence measures were more consistent,

Table 6.11 The Ambivalent Sexism Inventory (ASI) as a predictor of women’s ambivalent attitudes toward women ASI scale

General attitude

Ambivalence (all traits)

Ambivalence (feminine traits)

.31 .45* −.05

.41* .53** −.08

.39** .35** −.08

.43** .32** −.01

.48** .17 .29*

.52** .11 .39*

Nonstudent men, Study 4 (N = 22) ASI HSa BSb

−.51** −.12 −.33

Nonstudent men, Study 5 (N = 52) ASI HSa BSb

−.05 −.25* .23*

Undergraduate men, Study 6 (N = 35) ASI HSa BSb

−.15 −.12 −.04

Note: HS = Hostile Sexism; BS = Benevolent Sexism. a Correlations in this row are partial correlations controlling for Benevolent Sexism scores. b Correlations in this row are partial correlations controlling for Hostile Sexism scores. * p < .05; ** p < .01.

The Ambivalent Sexism Inventory 143 Table 6.12 The Ambivalent Sexism Inventory (ASI) as a predictor of men’s stereotypes about women Feminine traits ASI scale

Positive

Masculine traits Negative

Positive

Negative

.20 −.09 .38*

.36* .46** .01

Nonstudent men, Study 4 (N = 31) ASI HSa BSb

.19 −.20 .44*

.49** .52** .14

Nonstudent men, Study 5 (N = 25) ASI HSa BSb

−.04 −.12 .17

.46** .63** .02

.31 .02 .51**

.54** .61** .20

Undergraduate men, Study 6 (N = 36) ASI HSa BSb

.39** .27* .20

.55** .51** .02

−.15 .01 −.20

.24* .33* −.13

Note: HS = Hostile Sexism; BS = Benevolent Sexism. a Correlations in this row are partial correlations controlling for Benevolent Sexism scores. b Correlations in this row are partial correlations controlling for Hostile Sexism scores. * p < .05; ** p < .01.

suggesting that, among female respondents, the overall ASI score is associated with ambivalence toward women. Although in Study 5 these correlations were significantly lower than those of male respondents (both zs > 2.08, p < .05), no significant sex differences emerged in the other two studies. For both nonstudent samples, the HS scale itself was related to ambivalence (see below for an explanation of this effect). Stereotypes about women We predicted that the HS scale would be related to viewing women in a negative light and that the BS scale would be related to viewing women more positively. Because nontraditional women are typically viewed as taking on masculine traits, we expected that the HS scale might predict the ascription of negative masculine, as well as negative feminine, traits to women. Men’s stereotypes about women. The relationships of the ASI scales to stereotypes about women were much as predicted for male respondents (see Table 6.12). The HS scale was significantly correlated with ascribing to women both negative feminine traits (all studies) and negative masculine traits (among the two nonstudent samples) but was not significantly related to the ascription of positive traits to women (except for a weak but significant relationship in Study 6). In contrast, the BS scale was significantly related to seeing women in a positive

144

Peter Glick and Susan T. Fiske

light; it showed significant correlations with positive traits in the two nonstudent samples (both masculine and feminine in Study 4; positive masculine traits only in Study 5). No significant correlations with BS emerged in Study 6 (student sample). Women’s stereotypes about women. Correlations between the ASI scales and trait ascriptions to women by female respondents are reported in Table 6.13. For the two nonstudent samples, HS was consistently related to ascribing both negative feminine and negative masculine traits to women, although no relationships involving HS emerged in the undergraduate sample. There were only two significant relationships that involved BS, both of which involved negative views of women: a negative relationship between BS and ascribing positive feminine traits (Study 4) and a positive relationship between BS and ascribing negative feminine traits (Study 6) to women. The former correlation significantly differed from that of male respondents (in Study 4) for whom BS was related to being more, rather than less, likely to ascribe positive feminine traits to women (z = 2.86, p < .01). Explication of the relationship of HS to stereotypic ambivalence. The consistent tendency of HS (once BS was partialed out) to correlate positively with ambivalence was unexpected. However, this relationship may have occurred because women are perceived highly favorably overall (Eagly & Mladinic, 1993). If people generally ascribe positive traits to women, but those high in HS also

Table 6.13 The Ambivalent Sexism Inventory (ASI) as a predictor of women’s stereotypes about women Feminine traits ASI scale

Positive

Masculine traits Negative

Positive

Negative

.03 .08 −.10

.26 .39* −.05

Nonstudent men, Study 4 (N = 22) ASI HSa BSb

.09 .34* −.36*

.41* .60** .21

Nonstudent men, Study 5 (N = 52) ASI HSa BSb

.14 −.10 .13

.46** .34** −.01

.08 .00 .07

.30* .35** −.15

Undergraduate men, Study 6 (N = 35) ASI HSa BSb

.06 −.10 .15

.53** .09 .41**

−.07 −.04 .04

.28 .16 .12

Note: HS = Hostile Sexism; BS = Benevolent Sexism. a Correlations in this row are partial correlations controlling for Benevolent Sexism scores. b Correlations in this row are partial correlations controlling for Hostile Sexism scores. * p < .05; ** p < .01.

The Ambivalent Sexism Inventory 145 perceive women as likely to possess negative traits, ambivalence scores would tend to be high for high HS scorers. For each study, we conducted a 2 (sex of respondent) × 2 (valence of traits) × 2 (gender type of traits) ANOVA of the trait ratings, which revealed that both men and women tended to ascribe positive feminine traits to women at a high rate in all three studies (means ranged from 65% to 76%), followed by masculine–positive traits (means ranged from 48% to 56%), feminine–negative traits (means ranged from 38% to 50%), and masculine–negative traits (means ranged from 32% to 41%). All three ANOVAs yielded significant valence of traits × gender type of traits interactions (all Fs > 15.93, p < .01). Because, especially for male respondents, high HS scorers tended to ascribe negative traits to women at higher rates but did not typically assign positive traits to them at a different rate (as indicated by the generally nonsignificant correlations between HS and positive traits), these high HS scorers would be expected to have high ambivalence scores. The high rate of ascribing women positive feminine traits, which restricted the range of this variable, may also have attenuated correlations between BS and the ascription of positive feminine traits to women.

Discussion The six investigations reported here provide strong support for our theory of ambivalent sexism and for the convergent, discriminant, and predictive validity of the ASI. Across five studies (involving men and women, undergraduates and two nonstudent samples), factor analysis repeatedly confirmed the existence of BS and HS, both of which are reliably measured by the two ASI subscales. In all of the factor analyses, a full model (HS and BS, with three BS subfactors) significantly outperformed a one-factor (sexism) and a simple two-factor (no BS subfactors) model. The positive correlation repeatedly found between the HS and BS scales (with the important exception of men from the nonstudent samples) supports the claim that these two forms of sexism tend to be related aspects of sexist ideology (see also Nadler & Morrow, 1959). Whereas the HS scale demonstrated convergent validity with other measures of sexism (and racism), the BS scale measured an aspect of sexism many other researchers apparently have missed or have only indirectly tapped. Finally, three predictive validity studies showed that, for both men and women, total ASI scores are related to ambivalence toward women, and HS predicts negative attitudes toward and stereotypes about women. That BS represents a subjectively positive orientation toward women was indicated by the findings that, for nonstudent men, BS scores predicted positive overall attitudes toward and positive images of women. These latter findings, however, did not occur among male undergraduate and female respondents (in both student and nonstudent samples). It is worth noting that although the ASI was initially developed with student samples, it showed its strongest predictive validity among men in the two nonstudent samples.

146

Peter Glick and Susan T. Fiske

Sources of sexism Although the tripartite structure of the BS scale confirmed our ideas about the three sources of ambivalence toward women, the HS scale turned out to be strongly unidimensional. We believe that this is not because the sources of hostile sexism differ from the sources of benevolent sexism but because the three sources of hostility are, in a psychological as well as empirical sense, more inextricably tied together. Dominative paternalism and competitive gender differentiation both result in the same impulse: a desire to dominate women. Furthermore, this impulse is related to sexual hostility as well. Recent research by Bargh and Raymond (1995) and by Pryor et al. (1995) strongly supports the notion that sexual hostility toward women is based on a tight link between desires for power and sex, a link that becomes automatic in men who are likely sexually to harass women. Despite the unidimensional nature of the HS scale, it represents a wide range of themes concerning hostility toward women: women exaggerate the existence of sexism, male–female relationships are characterized by a power struggle, women take advantage of men, and women use sexual relationships to manipulate and control men. These themes fit well with modern versions of dominative paternalism and competitive gender differentiation that do not include overt statements of female inferiority but are couched more in terms of a backlash against feminism and a concern with power relations between men and women (see Faludi, 1992). Heterosexual hostility is evident in the view that women are sexual teases and that they seek to gain power over men in intimate relationships. The diversity of themes represented in the HS and BS scales are part of their strength; they tap a wide range of issues involved in sexist beliefs. Consistencies and differences between male and female respondents With the exception of the magnitude of the relationship between HS and BS among nonstudent respondents, the factor structure of the ASI was similar for men and women. Two formal tests (in the two studies with sufficient sample sizes) confirmed this similarity. This finding was a welcome, though not wholly expected, result. The reader will note that the theory of ambivalent sexism is predicated on men’s ambivalence toward women. Although women also can be sexist, before beginning the present series of studies it was not a foregone conclusion that the structure of their beliefs in this area would match that of men’s. Indeed, the initial factor analytic studies of the AWS (Spence & Helmreich, 1972) revealed significant differences between the factors obtained for men and women. Pragmatically, the similarity in ASI factor structure across the sexes is highly convenient, indicating that similar constructs are being measured for both groups. Conceptually, this similarity argues for the importance of social learning when it comes to sexist beliefs. If sexist attitudes were solely the result of specifically male drives directed at women (e.g., to dominate, to have sex), the factor structure for women would not have been

The Ambivalent Sexism Inventory 147 similar. Although some aspects of sexist beliefs may originate in underlying male drives, these beliefs become culturally transmitted and can also be adopted by women. The cultural transmission of sexism to women, as opposed to the motivational origins we have argued for sexism in men, may account for the few exceptions to the similarity in the correlational structure of men’s and women’s responses. Although HS and BS were generally strongly and positively related, there were two exceptions to this: the correlations between the (raw score) HS and BS scales were nonsignificant (and, in both cases, negative in direction) for men in the two nonstudent samples; also, for one of the student samples, although the correlation was positive and significant for men it was significantly lower than the correlation between HS and BS for women. Why might these sex differences exist? Although explanations at this point can be only speculative, the differences make sense if men’s sexism reflects their motivational orientations toward women, whereas women’s adoption of sexist beliefs reflects their tendency to embrace or reject prevailing cultural norms. Hostile and benevolent motivations need not be tightly conjoined for individual men. For instance, sexual motivations may elicit benevolent sexism in many men (the desire to protect, idealize, and achieve intimacy with women) without engaging the explicitly hostile side of sexism. Although we have argued that high scores on the BS scale do reflect a kind of domination of women, the dominance is veiled, and individual men may be unaware of this component of their attraction toward women (see Bargh & Raymond, 1995; Pryor et al., 1995). In contrast, women may tend either to adopt the prevailing cultural beliefs about women or, if they question these attitudes, to confront simultaneously both hostile and benevolent sexism. In other words, women who reject traditional sexist roles and responsibilities are likely not only to consider the more obvious, hostile aspects of sexism but also to develop a sensitivity to the assumptions implicit in benevolent sexism (e.g., that the “protection” of women involves an assumption of lesser competence). The pattern of sex differences in the relationships between the ASI scales and other measures of prejudice seems generally consistent with this view. When HS was controlled, BS was associated with other measures of sexism (see note 9) and with modern racism among women but not among men (indicating that, for women, BS reflects the tendency to adopt other forms of prejudice prevalent in the culture). Differences between undergraduate men and nonstudent men The logic that applies to sex differences in the results also offers an explanation for the differences between men in the student samples, for whom HS and BS were always significantly and positively correlated, and men in the nonstudent samples, for whom HS and BS were consistently independent. The lack of correlation for nonstudent men reinforces the notion that HS and BS truly are separate components of sexism and offers opportunities for studies

148

Peter Glick and Susan T. Fiske

on a population of special interest (i.e., men outside of the standard undergraduate participant pool) in which the two factors are naturally unconfounded. Why did student and nonstudent men differ in this manner? Younger men, like women, may tend to adopt wholesale either generally sexist (both hostile and benevolent) or egalitarian beliefs, depending on how they are socialized. As they get more experience in relationships with women, however, men’s beliefs may become more experientially based and hence more differentiated (resulting in an absence of correlation between HS and BS). Indeed, the motivational underpinnings of BS (heterosexual intimacy needs that foster men’s dyadic dependency on women) are unlikely to develop before adolescence. Prior to adolescence, boys and girls tend to segregate themselves into separate groups whose relationships are predominantly characterized by competitive gender differentiation (Maccoby, 1988). With the onset of puberty and its attendant sexual urges, benevolent sexist motivations suddenly gain relevance. As adolescent males adjust to the more complex relationships they begin to have with women, their attitudes may undergo a transition from simple structures (e.g., an endorsement of both HS and BS based solely on unreflective adoption of cultural attitudes) to more complex, and perhaps more truly ambivalent, attitude structures. The older, nonstudent men in our samples undoubtedly had many more experiences with women that could account for more differentiated attitudes, with some experiences (e.g., divorce, losing a promotion to a woman) likely to generate greater hostility and others (e.g., raising a daughter) likely to foster more benevolent sexist attitudes. If our reasoning is correct, the college years are likely to be a time of transition for men, from simpler to much more complex attitudes toward women as they gain increasing experience in dealing with motives aroused by dyadic dependency on women, both hostile (e.g., resentment of women’s dyadic power over them) and benevolent (e.g., desire for heterosexual intimacy). The enterprising researcher could hardly construct a better laboratory for examining this transition than a setting in which male and female adolescents are allowed to interact freely in social, sexual, and competitive task-oriented situations. For whom is sexism ambivalent? Across two studies with nonstudent men, not only was the total ASI score correlated with indices of ambivalence toward women, but HS predicted negative attitudes toward and stereotypes about women, and BS predicted positive attitudes toward and stereotypes about women, providing highly consistent and unambiguous evidence of sexist ambivalence. Additionally, the absence of a correlation between HS and BS makes the structure of these men’s attitudes more consistent with others’ notions of ambivalence (e.g., Thompson et al., 1995), which predict negative correlations or independence between ambivalent attitudes. Undergraduate men, however, did not show unambiguous evidence of ambivalence, whereas they did show a consistent correlation between HS and

The Ambivalent Sexism Inventory 149 BS (which other theorists may take as evidence against the notion that they are ambivalent). More recent data, however, suggest that the methods used in the current study simply were not sensitive enough to demonstrate younger men’s ambivalence. Glick, Diebold, Bailey-Werner, and Zhu (1997) correlated undergraduate men’s ASI scores with ratings these men made of their own spontaneously generated subtypes of women. Total ASI scores were significantly correlated with the variance (i.e., polarization) of overall evaluations, extremity of separate positive and negative affect ratings, and extremity of separate positive and negative trait ratings across the eight subtypes generated by each participant. That is, in comparison with low scorers, men with high ASI scores spontaneously generated more extreme sets of female subtypes that included types of women they love and types they hate. Although the sexist undergraduate men showed more polarized ratings across the eight subtypes, the ASI was uncorrelated with overall evaluations averaged across their eight subtypes (i.e., they did not, on average, rate all women generally lower in terms of overall evaluation or ascription of negative traits and feelings). In contrast, HS and BS (with each scale partialed out of the other) did correlate significantly with ratings averaged across each man’s eight subtypes. HS correlated positively with ratings of negative feelings toward and ascription of negative traits to the subtypes but correlated negatively with ratings of positive feelings toward, ascription of positive traits to, and overall evaluations of the subtypes. BS showed precisely the opposite pattern of correlations. Consistent with the three validity studies presented here, Glick et al. (1995) found little evidence of ambivalence for high-ASI (in comparison to low-ASI) female participants. (The relationship between women’s total ASI scores and one measure of ambivalent attitudes in the current studies was probably due to the general tendency all participants had to rate women very favorably coupled with a greater hostility on the part of high scorers.) BS tends (among women) to be weakly, and HS more strongly, associated with negative attitudes toward women as a group. Sexist women seem to have picked up the overall sexist devaluation of women evident in HS, but their endorsement of BS was generally devoid of subjectively positive feelings toward other women. Such results make sense given our contention that, for women, the adoption of these beliefs is due more to socialization than to underlying ambivalent motives.13 Sexist beliefs, when endorsed by women, may serve mainly as justification for male structural control of society (hence their predominantly hostile tone). The nature of sexist ambivalence That a recent study (Glick et al., 1995) has obtained clear evidence of ambivalence toward women in a college male sample for which HS and BS were correlated (at .52) demonstrates that one form of ambivalence toward women does seem to be unconflicted; a positive correlation between HS and BS (which suggests that these are psychologically consistent belief systems) can coexist with ambivalence toward different female subtypes. These results

150

Peter Glick and Susan T. Fiske

convincingly demonstrate that younger men who score high on the ASI do experience at least one form of sexist ambivalence: the tendency to spontaneously (and, we therefore assume, habitually) divide women into in-groups and out-groups that receive polarized reactions from these men. Our failure to find evidence of this among undergraduate men in Study 6 may have been due to the well-known reluctance (in recent times) of college students to appear as though they are stereotyping an entire social group. Impression-management concerns of this type (which may crop up when undergraduates are asked to rate “women in general”) were not evident when undergraduate men were asked to generate and rate subtypes of women; derogatory epithets for subtypes of women (e.g., “slut,” “bitch”) were commonly used, as were extremely negative ratings of these types. This is consistent with our speculation that BS may serve to justify HS; because sexist men love some types of women they may feel less compunction about hating other types. Thus, we advise other researchers working with undergraduate populations that, although ambivalence toward women may be readily observed among college men, researchers must orient their measures toward various subtypes of women, not simply “women in general” if they are to find it. Was this unconflicted ambivalence the same sort of ambivalence evident among adult men in the predictive validity studies presented here? One argument in favor of such a conclusion is that HS was correlated with ascribing both negative feminine and negative masculine traits to women in general (and BS was correlated with ascribing both positive feminine and positive masculine traits to women), even though these traits are not typically seen as “going together” in the same person. One explanation of why, for example, HS was related to ascribing both traits such as “whiny” and “arrogant” to women is that high HS scoring men had in mind different subtypes of women to whom they ascribe these traits. Do sexist men also experience a more conflicted type of ambivalence toward particular female targets? In a second study that Glick et al. (1995) performed, there is evidence that although sexist men evaluate the subtype “sexy women” highly favorably overall, they are more likely than nonsexist men to ascribe negative traits to this subtype. We have suggested that this subtype may indeed be a prime candidate for the experience of conflicted ambivalence among sexist men whose sexual motives may be at odds with their fears that such women typically use their sexual allure to manipulate men. The ASI and impression management Although five studies (Studies 2–6) provided ample evidence for the convergent and discriminant validity of the ASI, one concern is that both subscales of the ASI were weakly but significantly related to the IM component of Paulhus’s (1988) BIDR. We do not see this as a devastating problem, for two reasons: (a) the college student sample for the study that included the BIDR represents a group likely to be particularly sensitive to issues of sexual politics; we suspect,

The Ambivalent Sexism Inventory 151 therefore, that the correlations found in this sample represent a ceiling unlikely to be exceeded; and (b) controlling for IM had absolutely no effect on the relationship between the ASI scales and the other measures included in the same study. Researchers who are not sanguine about this problem are urged to administer the IM scale of the BIDR along with the ASI so that the former may be partialed out of the results. The ASI and other measures of sexism The AWS has long been the standard measure of sexism among researchers. Although both the ASI and AWS measure aspects of sexism, it should be noted that (in spite of the strong correlation between the ASI and AWS) the two scales measure different constructs. The ASI not only includes a previously neglected aspect of sexism (BS), but it also measures attitudes toward women rather than attitudes toward equal rights for women. The AWS has often been misused as a measure of the former, even though its originators explicitly intended it as a measure of the latter (Eagly & Mladinic, 1989). Whereas Eagly and Mladinic found the AWS to be uncorrelated with general attitudes toward women (as measured by the same semantic differential scales used in Studies 4–6), we have demonstrated that the ASI subscales are related to both attitudes toward and stereotypes about women. Swim et al.’s (1995) Modern Sexism and Tougas et al.’s (1995) Neo-Sexism scales measure issues similar to those of the AWS, but in a more subtle manner that reflects the greater egalitarianism toward women that has come about in recent decades. Our finding that HS correlates strongly with the Modern Sexism scale, whereas BS does not, supports the notion that the beliefs the Modern Sexism scale taps do reflect hostility toward women. Despite the strong correlation between HS and the Modern Sexism scale (which presumably would also occur with Neo-Sexism), we believe that the HS scale is complementary to, rather than redundant with, these other scales. Recall that recognition of discrimination (items that tapped the same issues that are at the core of the Modern Sexism and Neo-Sexism scales) emerged as a separate factor in our exploratory analyses. Although HS does explore related issues (e.g., antifeminism), both HS and BS focus more on interpersonal relationships between men and women than on general political stances. Thus, the Modern Sexism and Neo-Sexism scales may have greater predictive utility for exploring genderrelated political attitudes, whereas the ASI may be of particular interest in the interpersonal relationships area (e.g., heterosexual romantic relationships, oneon-one stereotyping, sexual harassment). The other difference between the ASI and the Modern Sexism and Neo-Sexism scales is the ability to measure benevolent sexism, which offers opportunities to explore the subjectively positive aspects of sexism. Because of its relationship to the protector and provider aspects of traditional male gender identity, BS should be of special interest to those who study masculine identity (and its relationship to the treatment of women).

152

Peter Glick and Susan T. Fiske

The ASI as a research tool Researchers who wish to incorporate the ASI into experimental designs might consider two forms of participant selection: (a) selecting those who score highly on both HS and BS (ambivalent sexists) and those who score low on both scales (nonsexists) or, (b) if a sufficiently large sample can be examined, using all combinations formed by a 2 (HS: high and low) × 2 (BS: high and low) classification scheme. The latter would yield four categories: (a) ambivalent sexists (high HS, high BS), (b) hostile sexists (high HS, low BS), (c) benevolent sexists (low HS, high BS), and (d) nonsexists (low HS, low BS). In either case, the obvious next step in the use of the ASI is to demonstrate that ambivalent sexism (the high HS, high BS category) predicts polarized responses to women, particularly from men. Compared to men who are nonsexists, ambivalent sexist men are those who are likely to be patronizingly sweet or viciously hostile toward any particular woman at any given time; these men may exhibit volatile swings between these two poles, depending on the characteristics of the situation or of the women with whom they are interacting (for a discussion of situational and target traits that might elicit HS and BS in work settings, see Fiske & Glick, 1995). In contrast, the purely hostile-sexist subgroup (high HS, low BS) can be expected to be more consistently hostile, and men who are benevolent sexists may actually treat women more favorably than men on many counts (although many women may view this “favorable” treatment as patronizing). To complement the ASI as a research tool, we are currently developing a second instrument, the Ambivalence Toward Men Inventory (AMI). Using an approach parallel to the one used here (and with the same starting assumption that relationships between the sexes are, to a significant extent, conditioned by men’s structural power and women’s dyadic power) we are designing the AMI to tap women’s ambivalent feelings toward men. Concluding remarks The orientations toward women that we have labeled hostile sexism and benevolent sexism have ancient origins. Both sets of attitudes are clearly evident in Homer’s epic poem The Odyssey, composed almost three millennia ago. The poem chronicles Odysseus’s 10-year quest to reunite with his ever-constant wife, Penelope, who is presented as the Greek ideal of womanhood: beautiful, intelligent, and accomplished; the pillar of the home, yet circumspect, loyal, and subordinate to her husband. Until he can reunite with her, Odysseus, “sacker of cities,” is incomplete. Penelope, in turn, requires his protection from the unwanted suitors who besiege her during his long absence. In Odysseus’s relationship with Penelope, the components of benevolent sexism (and women’s dyadic power) are all in evidence. The obstacles that long delay their reunion come in an astonishing array of female forms, from the Sirens who would lure Odysseus to his doom on the rocks, to Circe, an enchantress

The Ambivalent Sexism Inventory 153 who uses her beauty to entice Odysseus’s crew to her home where, appropriately enough, she transforms the men literally into swine. Several such “dread goddesses with lovely hair” threaten to “unman” Odysseus, using their sexual allure in attempts to detain, dominate, or destroy him—fears that echo our hostile sexism items. Contemporary equivalents of these ancient images of the faithful wife versus the twisted, domineering seductress pervade our own popular images of women (see Faludi, 1992, for a discussion). Despite all that has changed since the Homeric age, the ancient roots of hostile and benevolent sexism entangle relationships between men and women to this day.

Notes 1

2

3

4

We anticipate that some readers might view the neologism benevolent sexism as oxymoronic. Although benevolent does not capture the underlying dominance inherent in this form of sexism, we were unable to discover a word that successfully combines connotations of dominance and the subjectively positive origins of this form of sexism (the term paternalism does so, but as the reader will see, we view paternalistic feelings as but one component of benevolent sexism). We hope that benevolent sexism, like the term benevolent dictator, successfully conveys the combination we intend. These distance, touching, and self-disclosure measures may well be confounded with issues of status and domination. This mixture of a desire for intimacy as well as an assumed dominance captures well both the “benevolent” and “sexist” aspects of benevolent sexism. It is certainly possible that hostile and benevolent sexism may be uncorrelated. It is conceivable that individual men could subscribe to one but not the other (e.g., one might believe that women are all “ladies” to be treated with chivalry, never hostility, by men). However, given the assumptions of male dominance common to hostile and benevolent sexist beliefs and earlier findings of Nadler and Morrow (1959), who showed that hostile beliefs toward women were positively correlated with men’s scores on a chivalry scale, our expectation was for a mildly positive relationship. Nadler and Morrow’s (1959) work, of which we belatedly became aware after conceptualizing our own theory and conducting the research reported here, attempted a similar endeavor to our own—measuring the negative and positive aspects of traditional attitudes toward women. They constructed two scales: (a) Open Subordination of Women (support for traditionally restrictive policies and negative stereotyping of women) and (b) Chivalry (protectiveness toward women, formalized rituals in interactions with women, idealization of women as pure). The content of their scales reflects the constructs we label paternalism and gender differentiation (indeed, in this respect their scales could be considered “old-fashioned” versions of our own), though they ignored heterosexuality. Nadler and Morrow did not present a theoretical analysis of the motivational sources of these attitudes, and they did not explore the factor structure of their scales. They did, however, view chivalry as supporting, rather than conflicting with, the subordination of women. Consistent with this view, which is remarkably like our own, they found (among their male participants) a positive correlation between their scales (r = .35). We did not exclude any participants on the basis of the validity items, because a number of participants seemed to have interpreted these items in unintended ways. For instance, several individuals wrote marginal comments on such items as “Men cannot bear children” that explained their disagreement with this statement.

154 5

6

7 8

9

Peter Glick and Susan T. Fiske

Similar factors emerged in the different college samples. In all cases, the first factor— a general HS factor—was by far the strongest (eigenvalues ranging from 23.6 to 29.2; percentage of variance accounted for ranging from 21% to 26%). Of the top 10 items to load on this factor within each sample, all loaded .48 or better on Factor 1 in all samples (with one exception the loadings were all .55 or greater). The next three factors in the Amherst College and Lawrence University samples and the next two factors in the University of Massachusetts sample corresponded to different aspects of benevolent sexism (the items that were spread across three factors for the former two samples were assimilated by only two factors in the latter sample). Eigenvalues for the BS factors ranged from 2.9 to 6.4; percentage of variance accounted for ranged from 3% to 6% (in each case, the BS factors together accounted for about 10% of the variance). Of the top four items within each sample to load on the BS factor, 83% were among the best loading items for at least two of the samples (close to 80% of these shared items were shared by all three samples). Given the relatively small size of each sample, we deemed the overlap impressive enough to warrant pooling the samples. After pooling the samples from different colleges, we separately analysed men’s and women’s responses. The overlap was again impressive, replicating a strong HS factor that accounted for 23% of the variance among male respondents (eigenvalue = 26.0) and 21% among female respondents (eigenvalue = 22.99), and the benevolent sexism items coalesced into a single BS factor, accounting for 5% of the variance among male respondents (eigenvalue = 5.74) and 6% among female respondents (eigenvalue = 6.68). Of the top 10 items to load on Factor 1 within each sex, all loaded .52 or higher on Factor 1 for the other sex. Of the top 10 items to load on Factor 2 within each sex, all loaded .44 or better on Factor 2 for the other sex. GFI and AGFI are monotonically related to chi-square but are standardized so that, unlike chi-square, their values are not directly determined by sample size. Chisquare is almost certain to be statistically significant when large samples are used (even when the model fits well, the small deviations between the observed and actual covariation matrices that invariably occur will be statistically significant with a large N). GFI and AGFI are therefore more useful for making comparisons across the five samples (which have varying Ns) as well as for judging the absolute fit of a model. In all five studies, each BS subfactor was more highly correlated with the other BS subfactors (rs ranged from .54 to .86) and with the superordinate BS factor than each was with HS (rs ranged from .27 to .73). When the ASI scales were averaged across male and female respondents, the mean rate of endorsement of ASI items generally ranged between 2.0 and 2.5 (the midpoint of the scale). The distributions of ASI total, HS, and BS scores are roughly bell shaped but deviate from the normal distribution in that they tend to be platykurtic (i.e., more scores toward the tails, perhaps indicating some tendency for there to be polarized responses on these issues) and, among the college student samples, exhibit a slight skew (with extreme low scores being more common than extreme high scores). These tendencies were not particularly pronounced and should present no problems for the use of the scales. Indeed, the ASI scales exhibit very little skew in comparison with the currently most popular measure of sexism, the AWS (Spence & Helmreich, 1972), which exhibited a J-curve distribution among college students in our Study 3 sample. We note, however, that despite its highly skewed distribution, the AWS still showed convergent validity with other measures of hostile sexism (including our HS scale). Separate analyses for male and female respondents generally yielded similar results, with the exception of the relationship of BS (once HS was partialed out) to the

The Ambivalent Sexism Inventory 155

10

11

12

13

AWS and Old-Fashioned Sexism (OFS) scales. BS was weakly negatively related to the AWS and OFS scales for men (rs = −.10 and −.20, respectively, p < .05 for the latter), whereas for women the relationships were weakly but significantly positive (r = .19 with AWS and .18 with OFS, ps < .05). Direct comparisons of the correlations revealed that the correlations for men and women were marginally significantly different for the AWS (z = 1.82, p = .07) and significantly different for the OFS scale (z = 2.39, p < .05). This pattern did not hold true for the Modern Sexism or Rape Myth Acceptance scales (both zs < 1.12, ns). This suggests that for women, but not for men, BS is related to the acceptance of traditional female roles and traditional, overt sexist beliefs, but not to rape myths or the more subtle form of sexism tapped by the Modern Sexism scale. Although the ASI was designed to measure attitudes toward women, it is clear that different stances toward women almost inevitably involve differing views of men. We performed a correlational analysis on attitudes toward and stereotypes about men in Study 4 (for which respondents rated men as well as women). The only significant correlation between the ASI and attitudes toward men occurred for women: BS was related to negative attitudes toward men (r = −.52). Analysis of the trait ratings reinforced this finding: For women, BS scores were related to seeing men as not possessing positive feminine traits (r = −.30) and as possessing negative masculine traits (r = .48). We note for readers who are concerned about the imbalance between the large size of the initial undergraduate samples on which the ASI was developed and the relatively small sample sizes for the predictive validity studies with adult community members that: (a) these studies have different purposes and, although the initial factor analyses on the larger item pool statistically required a large sample, subsequent factor analyses and correlational studies with the short version of the ASI did not statistically require larger samples; (b) the consistency and statistical significance of the correlational results across two quite different samples of adult men are particularly impressive precisely because of the small sample sizes; (c) the rate of endorsement of ASI items (although it varied by sex) is similar among undergraduates and adults in the community; (d) the factor analytic results with the community samples are highly consistent with those of the student samples: with the sole exception of the correlation between HS and BS, the same complex twofactor model of sexism (with three BS subfactors) provides the best fit to the data; and (e) the best evidence for the ASI’s predictive validity occurs with adult men from the community. Partial correlations were used once again to get “pure” measures of HS and BS, but it should be noted that because HS and BS were uncorrelated for men in the two nonstudent samples, the partial correlations of HS and of BS to the criterion variables are virtually identical to the first-order correlations of HS and BS to these variables for men in Studies 4 and 5. Interestingly, for undergraduate women it was BS, not HS, that was related to ambivalence toward women. Perhaps this reflects ambivalence about their own roles. Presumably, most college women feel that they are expected to pursue careers, yet those who endorse BS may desire more traditional roles in which they are “taken care of” by men. As a result, conflicting goals among benevolent sexist college women may create ambivalence about women’s roles and women’s traits.

References Adorno, T., Frenkel-Brunswik, E., Levinson, D., & Sanford, R. N. (1950). The authoritarian personality. New York: Harper.

156

Peter Glick and Susan T. Fiske

Allport, G. W. (1954). The nature of prejudice. Reading, MA: Addison-Wesley. Altemeyer, B. (1988). Enemies of freedom: Understanding right-wing authoritarianism. San Francisco, CA: Jossey-Bass. Bargh, J. A., & Raymond, P. (1995). The naive misuses of power: Non-conscious sources of sexual harassment. Journal of Social Issues, 51, 85–96. Beere, C. A., King, D. W., Beere, D. B., & King, L.A. (1984). The Sex-Role Egalitarianism Scale: A measure of attitudes toward equity between the sexes. Sex Roles, 10, 563–76. Berscheid, E., & Peplau, L. A. (1983). The emerging science of relationships. In H. H. Kelley, E. Berscheid, A. Christensen, J. H. Harvey, T. L. Huston, G. Levinger, E. McClintock, L. A. Peplau, & D. R. Peterson (Eds), Close relationships (pp. 1–19). New York: W. H. Freeman. Berscheid, E., Snyder, M., & Omoto, A. M. (1989). The Relationship Closeness Inventory: Assessing the closeness of interpersonal relationships. Journal of Personality and Social Psychology, 57, 792–807. Bollen, K. A. (1989). Structural equations with latent variables. New York: Wiley. Brehm, S. S. (1992). Intimate relationships. New York: McGraw-Hill. Burt, M. R. (1980). Cultural myths and support for rape. Journal of Personality and Social Psychology, 38, 217–30. Cacioppo, J. T., & Bernston, G. G. (1994). Relationship between attitudes and evaluative space: A critical review, with emphasis on the separability of positive and negative substrates. Psychological Bulletin, 115, 401–23. Check, J. V. P., Malamuth, N. M., Elias, B., & Barton, S. A. (1985, April). On hostile ground. Psychology Today, 56–61. Cozby, P. C. (1973). Self-disclosure: A literature review. Psychological Bulletin, 79, 73–91. Crosby, F. J., Bromley, F., & Saxe, L. (1980). Recent unobtrusive studies of black and white discrimination and prejudice: A literature review. Psychological Bulletin, 87, 546–63. Deaux, K., Winton, W., Crowley, M., & Lewis, L. L. (1985). Level of categorization and content of gender stereotypes. Social Cognition, 3(2), 145–67. Derlega, V. A., Winstead, B. A., Wong, P.T. P., & Hunter, S. (1985). Gender effects in an initial encounter: A case where men exceed women in disclosure. Journal of Social and Personal Relations, 2, 25–44. Eagly, A. H. (1987). Sex differences in social behavior: A social role interpretation. Hillsdale, NJ: Erlbaum. Eagly, A. H., & Chaiken, S. (1993). The psychology of attitudes. Fort Worth, TX: Harcout Brace Jovanovich. Eagly, A. H., & Crowley, M. (1986). Gender and helping behavior: A meta-analytic review of the social psychological literature. Psychological Bulletin, 100, 283–308. Eagly, A. H., Makhijani, M. G., & Klonsky, B. G. (1992). Gender and evaluation of leaders: A meta-analysis. Psychological Bulletin, 111, 3–22. Eagly, A. H., & Mladinic, A. (1989). Gender stereotypes and attitudes toward women and men. Personality and Social Psychology Bulletin, 15, 543–58. Eagly, A. H., & Mladinic, A. (1993). Are people prejudiced against women? Some answers from research on attitudes, gender stereotypes, and judgments of competence. In W. Stroebe & M. Hewstone (Eds), European review of social psychology (Vol. 5, pp. 1–35). New York: Wiley. Eagly, A. H., Mladinic, A., & Otto, S. (1991). Are women evaluated more favorably than men? Psychology of Women Quarterly, 15, 203–16.

The Ambivalent Sexism Inventory 157 Esses, V. M., Haddock, G., & Zanna, M. P. (1993). Values, stereotypes, and emotions as determinants of intergroup attitudes. In D. M. Mackie & D. L. Hamilton (Eds), Affect, cognition and stereotyping (pp. 137–66). San Diego, CA: Academic Press. Faludi, S. (1992). Backlash: The undeclared war against American women. New York: Doubleday. Fiske, A. P., Haslam, N., & Fiske, S. T. (1991). Confusing one person with another: What errors reveal about the elementary forms of social relations. Journal of Personality and Social Psychology, 60, 656–74. Fiske, S. T., & Glick, P. (1995). Ambivalence and stereotypes cause sexual harassment: A theory with implications for organizational change. Journal of Social Issues, 51, 97–115. Fiske, S. T., & Stevens, L. E. (1993). What’s so special about sex? Gender stereotyping and discrimination. In S. Oskamp & M. Costanzo (Eds), Gender issues in contemporary society: Applied social psychology annual (pp. 173–96). Newbury Park, CA: Sage. Fitzgerald, L. F., & Betz, N. E. (1983). Issues in the vocational psychology of women. In W. B. Walsh & S. H. Osipow (Eds), Handbook of vocational psychology (Vol. 1, pp. 83–159). Hillsdale, NJ: Erlbaum. Gaertner, S. L., & Dovidio, J. F. (1986). The aversive form of racism. In J. F. Dovidio & S. L. Gaertner (Eds), Prejudice, discrimination, and racism (pp. 61–89). San Diego, CA: Academic Press. Glick, P. (1991). Trait-based and sex-based discrimination in occupational prestige, occupational salary, and hiring. Sex Roles, 25, 351–78. Glick, P., Diebold, J., Bailey-Werner, B., & Zhu, L. (1997). The two faces of Adam: Ambivalent sexism and polarized attitudes toward women. Personality and Social Psychology Bulletin, 23 (12), 1323–1334. Green, D. P., Goldman, S. L., & Salovey, P. (1993). Measurement error masks bipolarity in affect ratings. Journal of Personality and Social Psychology, 64, 1029–41. Gutek, B. A. (1985). Sex and the workplace. San Francisco, CA: Jossey-Bass. Guttentag, M., & Secord, P. (1983). Too many women? Beverly Hills, CA: Sage. Harris, M. (1991). Cultural anthropology (3rd edn). New York: HarperCollins. Hatfield, E. (1988). Passionate and companionate love. In R. J. Sternberg & M. L. Barnes (Eds), The psychology of love (pp. 191–217). New Haven, CT: Yale University Press. Jöreskog, K., & Sörbom, D. (1993). LISREL 8: Structural equation modeling with the SIMPLIS command language. Hillsdale, NJ: Erlbaum. Katz, I., & Hass, R. G. (1988). Racial ambivalence and value conflict: Correlational and priming studies of dual cognitive structures. Journal of Personality and Social Psychology, 55, 893–905. Katz, I., Wackenhut, J., & Hass, R. G. (1986). Racial ambivalence, value duality, and behavior. In J. F. Dovidio & S. L. Gaertner (Eds), Prejudice, discrimination, and racism (pp. 35–59). San Diego, CA: Academic Press. Kinder, D. R., & Sears, D. O. (1981). Symbolic racism versus racial threats to “the good life.” Journal of Personality and Social Psychology, 40, 414–31. Maccoby, E. E. (1988). Gender as a social category. Developmental Psychology, 24, 755–65. Major, B., Schmidlin, A. M., & Williams, L. (1990). Gender patterns in social touch: The impact of setting and age. Journal of Personality and Social Psychology, 58, 634–43. McConahay, J. B. (1986). Modern racism, ambivalence, and the Modern Racism scale. In J. F.Dovidio & S. L. Gaertner (Eds), Prejudice, discrimination, and racism (pp. 91–125). San Diego, CA: Academic Press.

158

Peter Glick and Susan T. Fiske

Morton, T. U. (1978). Intimacy and reciprocity of exchange: A comparison of spouses and strangers. Journal of Personality and Social Psychology, 36, 72–81. Nadler, E. B., & Morrow, W. R. (1959). Authoritarian attitudes toward women and their correlates. Journal of Social Psychology, 49, 113–23. Paulhus, D. L. (1988). Assessing self-deception and impression management in self-reports: The Balanced Inventory of Desirable Responding. University of British Columbia, Vancouver, BC, Canada. Peplau, L. A. (1983). Roles and gender. In H. H. Kelley, E. Berscheid, A. Christensen, J. H. Harvey, T. L. Huston, G. Levinger, E. McClintock, L.A. Peplau, & D.R. Peterson (Eds), Close relationships (pp. 220–64). New York: W. H. Freeman. Pratto, F., Sidanius, J., & Stallworth, L. M. (1993). Sexual selection and the sexual and ethnic basis of social hierarchy. In L. Ellis (Ed.), Social stratification and socioeconomic inequality (Vol. 1, pp. 111–3 7). Westport, CT: Praeger. Pryor, J. B., Giedd, J. L., & Williams, K. B. (1995). A social psychological model for predicting sexual harassment. Journal of Social Issues, 51, 69–84. Random House College Dictionary. (1973). New York: Random House. Riess, M., & Salzer, S. (1981, August). Individuals avoid invading the space of males but not females. Paper presented at the 89th Annual Convention of the American Psychological Association, Los Angeles, CA. Rogers, R. W., & Prentice-Dunn, S. (1981). Deindividuation and anger-mediated interracial aggression: Unmasking regressive racism. Journal of Personality and Social Psychology, 41, 63–73. Sigall, H., & Page, R. (1971). Current stereotypes: A little fading, a little faking. Journal of Personality and Social Psychology, 18, 247–55. Snyder, M., & Ickes, W. (1985). Personality and social behavior. In G. Lindzey & E. Aronson (Eds), Handbook of social psychology (3rd edn, Vol. 2, pp. 883–948). Reading, MA: Addison-Wesley. Spence, J. T., & Helmreich, R. (1972). The Attitudes Toward Women Scale. JSAS Catalog of Selected Documents in Psychology, 2, MS 153. Spence, J. T., Helmreich, R., & Holahan, C. K. (1979). Negative and positive components of psychological masculinity and femininity and their relationships to selfreports of neurotic and acting out behaviors. Journal of Personality and Social Psychology, 37, 1673–82. Spence, J. T., Helmreich, R., & Stapp, J. (1973). A short version of the Attitudes Toward Women Scale (AWS). Bulletin of the Psychonomic Society, 2, 219–20. SPSSX [Computer software, Version 4.0]. (1993). Chicago, IL: SPSS. 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, 207–18. Stockard, J., & Johnson, M. M. (1992). Sex and gender in society (2nd edn). Englewood Cliffs, NJ: Prentice-Hall. Swim, J. K., Aikin, K. J., Hall, W. S., & Hunter, B. A. (1995). Sexism and racism: Old-fashioned and modern prejudices. Journal of Personality and Social Psychology, 68, 199–214. Tajfel, H. (1969). Cognitive aspects of prejudice. Journal of Social Issues, 25, 79–97. Tajfel, H. (Ed.) (1981). Social identity and intergroup relations. London: Cambridge University Press. Tavris, C., & Wade, C. (1984). The longest war (2nd edn). San Diego, CA: Harcourt Brace Jovanovich.

The Ambivalent Sexism Inventory 159 Taylor, S. E. (1981). A categorization approach to stereotyping. In D. Hamilton (Ed.), Cognitive processes in stereotyping and intergroup behavior (pp. 88–114). Hillsdale, NJ: Erlbaum. Thompson, M. M., Zanna, M. P., & Griffin, D. W. (1995). Let’s not be indifferent about (attitudinal) ambivalence. In R. E. Petty & J. A. Krosnick (Eds), Attitude strength: Antecedents and consequences (pp. 361–86). Hillsdale, NJ: Erlbaum. Tougas, F., Brown, R., Beaton, A. M., & Joly, S. (1995). Neo-sexism: Plus ça change, plus c’est pareil. Personality and Social Psychology Bulletin, 21, 842–9. Unger, R., & Crawford, M. (1992). Women and gender: A feminist psychology. New York: McGraw-Hill. Woolf, V. (1981). A room of one’s own. New York: Harcourt Brace Jovanovich. (Original work published 1929) Zillmann, D., & Weaver, J. B. (1989). Pornography and men’s sexual callousness toward women. In D. Zillmann & J. Bryant (Eds), Pornography: Research advances and policy considerations (pp. 95–125). Hillsdale, NJ: Erlbaum.

Appendix The Ambivalent Sexism Inventory (ASI) Relationships between men and women Below is a series of statements concerning men and women and their relationships in contemporary society. Please indicate the degree to which you agree or disagree with each statement using the following scale: 0 = disagree strongly; 1 = disagree somewhat; 2 = disagree slightly; 3 = agree slightly; 4 = agree somewhat; 5 = agree strongly. 1 No matter how accomplished he is, a man is not truly complete as a person unless he has the love of a woman. 2 Many women are actually seeking special favors, such as hiring policies H that favor them over men, under the guise of asking for “equality”. B(P)* 3 In a disaster, women ought not necessarily to be rescued before men. H 4 Most women interpret innocent remarks or acts as being sexist. 5 Women are too easily offended. H 6 People are often truly happy in life without being romantically involved B(I)* with a member of the other sex. 7 Feminists are not seeking for women to have more power than men. H* B(G) 8 Many women have a quality of purity that few men possess. 9 Women should be cherished and protected by men. B(P) H 10 Most women fail to appreciate fully all that men do for them. 11 Women seek to gain power by getting control over men. H 12 Every man ought to have a woman whom he adores. B(I) B(I)* 13 Men are complete without women. 14 Women exaggerate problems they have at work. H H 15 Once a woman gets a man to commit to her, she usually tries to put him on a tight leash. B(I)

160 H B(P) H* B(G) B(P) H* B(G)

Peter Glick and Susan T. Fiske 16 When women lose to men in a fair competition, they typically complain about being discriminated against. 17 A good woman should be set on a pedestal by her man. 18 There are actually very few women who get a kick out of teasing men by seeming sexually available and then refusing male advances. 19 Women, compared with men, tend to have a superior moral sensibility. 20 Men should be willing to sacrifice their own well being in order to provide financially for the women in their lives. 21 Feminists are making entirely reasonable demands of men. 22 Women, as compared with men, tend to have a more refined sense of culture and good taste.

Note: Copyright 1995 by Peter Glick and Susan T. Fiske. Use of this scale requires permission of one of the authors. A Spanish-language version of the ASI is available from the authors. H = hostile sexism, B = benevolent sexism, (P) = protective paternalism, (G) = complementary gender differentiation, (I) = heterosexual intimacy, * = reverse-scored item.

Scoring instructions The ASI may be used as an overall measure of sexism, with hostile and benevolent components equally weighted, by simply averaging the score for all items after reversing the items listed below. The two ASI subscales (hostile sexism and benevolent sexism) may also be calculated separately. For correlational research, purer measures of HS and BS can be obtained by using partial correlations (so that the effects of the correlation between the scales are removed). Reverse the following items (0 = 5, 1 = 4, 2 = 3, 3 = 2, 4 = 1, 5 = 0): 3, 6, 7, 13, 18, 21. Hostile sexism score = average of the following items: 2, 4, 5, 7, 10, 11, 14, 15, 16, 18, 21. Benevolent sexism score = average of the following items: 1, 3, 6, 8, 9, 12, 13, 17, 19, 20, 22.

Part III

Twenty-first-century activated actors Social brain and social mind

7

A model of (often mixed) stereotype content Competence and warmth respectively follow from perceived status and competition Susan T. Fiske, Amy J. C. Cuddy, Peter Glick, and Jun Xu

Not all stereotypes are alike. Some stereotyped groups are disrespected as incapable and useless (e.g., elderly people), whereas others are respected for excessive, threatening competence (e.g., Asians). Some stereotyped groups are liked as sweet and harmless (e.g., housewives), whereas others are disliked as cold and inhuman (e.g., rich people). Surely, such differences matter. However, social psychology of late has eschewed the study of stereotype content, focusing instead on stereotyping processes (for reviews, see Brown, 1995; Fiske, 1998; Leyens, Yzerbyt, & Schadron, 1994; Macrae & Bodenhausen, 2000). And for good reason. Stereotyping processes respond to systematic principles that generalize across different specific instances of stereotypes, so the processes invite social–psychological investigation, because they are presumably stable over time, place, and out-group. If the contents of stereotypes come and go with the winds of social pressures, then no single stereotype remains stable and predictable from theoretical principles. Alternatively, if stereotypes do come and go with the winds of social pressures, maybe we can understand those wind patterns and, thus, some origins of stereotype content. In short, perhaps we need a model that predicts the intergroup weather: Stereotype content may respond to systematic principles, just as stereotyping processes do. If stereotype content responds to principles, then the first principle must identify common dimensions of content. Following Allport (1954), social psychologists have typically viewed only unflattering stereotypes as indicating prejudice, where prejudice is a uniform antipathy or contempt toward an outgroup across a variety of dimensions. Flattering stereotypes have presumably targeted in-groups or, when they target out-groups, have presumably indicated compunction stemming from modern egalitarian ideals. We argue instead that stereotypes are captured by two dimensions (warmth and competence) and that subjectively positive stereotypes on one dimension do not contradict prejudice but often are functionally consistent with

164

Susan T. Fiske et al.

unflattering stereotypes on the other dimension. Moreover, we argue that two variables long identified as important in intergroup relations—status and competition—predict dimensions of stereotypes. We suggest that for subordinate, noncompetitive groups (e.g., elderly people), the positive stereotype of warmth acts jointly with the negative stereotype of low competence to maintain the advantage of more privileged groups. For high-status, competitive out-groups (e.g., Asians), the positive stereotype of their competence justifies the overall system but acts jointly with the negative stereotype of low warmth to justify the in-group’s resentment of them. Finally, we argue that different combinations of stereotypic warmth and competence result in unique intergroup emotions—prejudices—directed toward various kinds of groups in society. Pity targets the warm but not competent subordinates; envy targets the competent but not warm competitors; contempt is reserved for out-groups deemed neither warm nor competent. Each of these issues—focus on dimensions of content, mixed (but functionally consistent) content, predictions of that content, and ensuing types of prejudice —follows precedents set by previous literature. Our innovation is to synthe-size these insights into a model of stereotype content that cuts across out-groups.

Focus on content: Competence and warmth Unencumbered by theory, the classic study of stereotype contents (Katz & Braly, 1933) was replicated at Princeton over about 20-year intervals (Bergsieker, Leslie, Constantine, & Fiske, 2012; G. M. Gilbert, 1951; Karlins, Coffman, & Walters, 1969). These studies document changes in the favorability (mostly improving) and uniformity (decreasing) of stereotypes over time but do not uncover dimensions or principles therein. Although the Katz–Braly checklist method has limitations (Devine & Elliot, 1995; Madon et al., 2001), it does provide one of the few consistently documented measures of stereotypes across groups.1 However, the Katz–Braly lineage does not claim theoretical roots. From a functional, pragmatic perspective (Fiske, 1992, 1993b), we suggest that dimensions of stereotypes result from interpersonal and intergroup interactions. When people meet others as individuals or group members, they want to know what the other’s goals will be vis-à-vis the self or in-group and how effectively the other will pursue those goals. That is, perceivers want to know the other’s intent (positive or negative) and capability; these characteristics correspond to perceptions of warmth and competence, respectively. A variety of work on intergroup and interpersonal perception suggests the relevance of these two dimensions in social perception. In the intergroup domain, early on, one ethnic out-group (i.e., Jews) was viewed as competent but not warm, and another (i.e., “Negroes”) was viewed as warm but not competent (Allport, 1954; Bettelheim & Janowitz, 1950). Curiously, this older ethnic-group distinction echoes modern-day views about perceived subgroups of women (Deaux, Winton, Crowley, & Lewis, 1985; Eckes, 1994; Noseworthy & Lott, 1984; Six & Eckes, 1991): disliked, dominant, competent,

A model of (often mixed) stereotype content

165

nontraditional women (e.g., career women, feminists, lesbians, athletes) versus likable, dependent, incompetent, traditional women (e.g., housewives, sometimes “chicks”). Overall, the ethnic and gender distinctions both fit our hypothesized dimensions of competence and warmth. From various out-group stereotypes, Fiske and Glick (Fiske, 1998, p. 380; Fiske, Xu, Cuddy, & Glick, 1999; Glick & Fiske, 1999, 2001b) constructed a preliminary model of stereotype content: Stereotype content may not reflect simple evaluative antipathy but instead may reflect separate dimensions of (dis)like and (dis)respect. Some out-group stereotypes (e.g., housewives, disabled people, elderly people) elicit disrespect for perceived lack of competence; other out-group stereotypes elicit dislike for perceived lack of warmth (e.g., Asians, Jews, career women). Although some groups may elicit both dislike and disrespect (e.g., welfare recipients), qualitative differences among stereotypes are captured by the crucial dimensions of competence and warmth. The plausibility of competence and warmth as core dimensions also springs from person perception research: Asch’s (1946) warm–cold versus competence-related adjectives (Hamilton & Fallot, 1974; Zanna & Hamilton, 1977) and multidimensional scaling of trait descriptions (Rosenberg, Nelson, & Vivekanathan, 1968; see also Jamieson, Lydon, & Zanna, 1987; Lydon, Jamieson, & Zanna, 1988). Perceptions of individuals in groups also vary along a task dimension and a social dimension (Bales, 1970). Relatedly, Peeters (1983, 1992, 1995) has argued for the dimensions of self-profitability (e.g., confident, ambitious, practical, intelligent)—akin to competence—and other-profitability (e.g., conciliatory, tolerant, trustworthy)—akin to warmth. The Peeters distinction has been applied to national stereotypes (Peeters, 1993; Phalet & Poppe, 1997; Poppe & Linssen, 1999),2 values (Wojciszke, 1997), and evaluations of social behavior (Vonk, 1999). Across racial prejudice, gender subgroups, national stereotypes, and person perception, thus, come two dimensions. They fit the functional idea that people want to know others’ intent (i.e., warmth) and capability to pursue it (i.e., competence). Groups (like individuals) are distinguished according to their potential impact on the in-group (or the self). Our stereotype content model’s first hypothesis hence holds that perceived competence and warmth differentiate out-group stereotypes.

Mixed stereotype content Across out-groups, stereotypes often include a mix of more and less socially desirable traits, not just the uniform antipathy so often assumed about stereotypes. Specifically, we suggest that mixed stereotypes for some out-groups include low perceived competence but high perceived warmth. These paternalistic stereotypes portray out-groups that are neither inclined nor capable to harm members of the in-group. Another equally important mixture depicts out-groups that are seen as competent but not warm, resulting in envious stereotypes. These groups are acknowledged to be doing well (for themselves), but their intentions

166

Susan T. Fiske et al.

toward the in-group are presumed not to be positive. Consistent with this idea, Phalet and Poppe’s (1997) multidimensional scaling of Central and Eastern European stereotypes revealed the majority (37 out of 58) in two quadrants: incompetent but moral/social (e.g., Byelorussians, Bulgarians, Czechs) and competent but immoral/unsocial (e.g., Germans, Jews). Paternalistic and envious stereotypes result from the combination of two separate dimensions, which also allows for the more traditional kinds of prejudice, uniform derogation of a disliked and disrespected out-group and pure in-group favoritism toward the competent and warm in-group. But our model emphasizes the mixed combinations, the off-diagonal cells of a theoretical competence × warmth matrix. We argue that these mixed combinations are frequent because they are functional. Our second hypothesis holds that many stereotypes are mixed on competence and warmth, as defined by low ratings on one dimension coupled with high ratings on the other. Paternalistic stereotypes Paternalistic mixed stereotypes show up in race, age, dialect, and gender prejudice. Ambivalent racism (I. Katz & Hass, 1986) depicts a mix of anti-Black attitudes (e.g., perceived incompetence and laziness, violating the work ethic) and paternalistic pro-Black attitudes (e.g., perceived pitiful disadvantage, deserving help). Overall, paternalistic mixed stereotypes portray a group disrespected but pitied, which carries overtones of compassion, sympathy, and even tenderness, under the right conditions.3 In ageism, dominant views of older people as not competent but kind suggest a similarly ambivalent dynamic (Cuddy & Fiske, 2002). Linguistic out-groups provide another example: Speakers of nonstandard dialects (e.g., Scottish accents in Great Britain, Chicano accents in the United States) are perceived as less competent but simultaneously friendly (Bradac, 1990; Ruscher, 2001). Paternalism appears prominently in gender stereotypes. The Ambivalent Sexism Inventory (ASI; Glick & Fiske, 1996, 2001a, 2001b) measures, in part, subjectively benevolent sexism (BS), which includes paternalistic power relations; BS is directed toward traditional women (homemakers), who are viewed as warm but not competent outside the home. When people rate women in general, traditional homemakers serve as the paternalistic default (Haddock & Zanna, 1994); this generates the “women are wonderful” effect: positive ratings of generic women (Eagly & Mladinic, 1989), but primarily on communal (i.e., warm), not agentic (i.e., competent), qualities. All four paternalistic stereotypes (regarding disadvantaged Blacks, elderly people, nonstandard speakers, and traditional women) describe out-groups perceived as low on competence but high on warmth.4 Envious stereotypes In contrast stands a different set of out-groups stereotyped as highly competent but not warm (Glick & Fiske, 2001a, 2001b): nontraditional women,

A model of (often mixed) stereotype content

167

Jews, and Asians. The ASI in part measures hostile sexism (HS), which includes competitive gender roles; HS is directed toward nontraditional women (e.g., career women, feminists, lesbians, athletes), who are viewed as task competent but not warm (see also Eagly, 1987; Glick, Diebold, Bailey-Werner, & Zhu, 1997; MacDonald & Zanna, 1998). Anti-Semitic notions of a Jewish economic conspiracy exaggerate Jews’ stereotypically feared competence, whereas views of them as self-serving portray them as not warm (Glick, 2002). The modern American equivalent, Asians—who are viewed as the model minority—are seen as too competent, too ambitious, too hardworking, and, simultaneously, not sociable (Hurh & Kim, 1989; Kitano & Sue, 1973; Sue & Kitano, 1973; Sue, Sue, & Sue, 1975). The Anti-Asian-American Prejudice scale measures dislike for this perceived lack of sociability along with envious respect of perceived competence (Lin, Kwan, Cheung, & Fiske, 2005). Thus, nontraditional women, Jews, and Asians elicit a shared stereotype as being too competent and not at all nice. Why mixed stereotypes occur Although isolated analyses of specific out-groups suggest mixed competence –warmth ascriptions, the present research aims to examine whether these mixed stereotypes are sustained across a wider variety of out-groups, all compared at once. Our approach emphasizes a 2 × 2 (warmth × competence) interaction (see Table 7.1). The mixed stereotypes hypothesis predicts that many out-group stereotypes fall into two cells: high warmth but low competence for compliant subordinates, and low warmth but high competence for successful competitors. For paternalized out-groups, the mixed stereotype justifies their Table 7.1 Four types of out-group, combinations of status and competition, and corresponding forms of prejudice as a function of perceived warmth and competence Competence Warmth

Low

High

High

Paternalistic prejudice Low status, not competitive Pity, sympathy (e.g., elderly people, disabled people,housewives)

Admiration High status, not competitive Pride, admiration (e.g., in-group,close allies)

Low

Contemptuous prejudice Low status, competitive Contempt, disgust, anger, resentment (e.g. welfare recipients, poor people)

Envious prejudice High status, competitive Envy, jealousy (e.g. Asians, Jews, rich people, feminists)

168

Susan T. Fiske et al.

subordination (i.e., low competence) and encourages their compliance (i.e., high warmth). They are seen as having no intent to harm societal reference groups and no ability to do so, in any case. The mixed stereotype functions to promote existing systems of privilege and to placate the nonthreatening but disadvantaged outgroups by assigning them socially desirable, though subordinating, traits (Ridgeway, 2001). Socioeconomically successful outgroups, however, pose a competitive threat, and their success elicits envy. For envied out-groups, the mixed stereotype explains their apparent success, thereby justifying the system of meritocracy that benefits societal reference groups and dominant ingroups. Stereotypes of low warmth justify taking action against envied groups by casting the groups as being concerned only with furthering their own goals. Thus, envied groups may be appropriately resented and socially excluded. Because these mixed stereotypes involve two separate dimensions, they are not psychologically inconsistent—one may view a group as warm but not competent (e.g., the elderly as nice but dotty) or as competent but not warm (e.g., Asians as cold but efficient) without experiencing discomfort. Furthermore, the functional perspective suggests that both envious and paternalistic stereotypes maintain the status quo and defend the position of societal reference groups. We hypothesize that many out-groups are stereotyped as high on either competence or warmth but low on the other, precisely because these combinations are functionally consistent for perceivers. These mixed combinations have been neglected by prior treatments that focus on uniformly negative stereotypes (see Glick and Fiske, 2001b). Of course, out-groups do not fall into only these two mixed cells. Lowstatus groups viewed as openly parasitic (i.e., opportunistic, freeloading, exploitative) underlings are banished to the not warm, not competent cell. These groups are rejected for their apparent negative intent toward the rest of society (i.e., not warm) and for their apparent inability to succeed on their own (i.e., not competent). At the opposite extreme, who is favored as both warm and competent? We suggest three possible inhabitants of this cell: Through in-group favoritism, the in-group may be rated both warm and competent. Close allies in a hostile world might also be allowed a purely positive stereotype. And the cultural default (e.g., middle class) may be viewed in an unmixed, positive way. We refer to both in-groups and societal reference groups because in the United States, at least, many groups view themselves as part of the societal ideal; for instance, most Americans identify themselves as middle class (even if qualified by lower or upper). Similarly, Whites and Christians, even where they are not a local majority, may be viewed as culturally dominant, societywide reference groups. Even groups who acknowledge their own exclusion from the cultural ideal may still identify with aspects of the societal reference group. Hence, people’s understanding of culturally shared stereotypes takes the perspective of society’s dominant reference groups.

A model of (often mixed) stereotype content

169

Predicting stereotype content If stereotype contents systematically vary along competence and warmth, with many stereotypes falling in the mixed combinations, the question follows, what predicts where groups fall on these mixed dimensions? In their 1933 study, D. Katz and Braly noted that the degree of agreement among students in assigning characteristics . . . seems too great to be the sole result of the students’ contacts with members of those races. . . . Prejudice of this kind seems largely a matter of public attitude toward a race name or symbol. (pp. 288, 290) Stereotype content may result from shared public views of groups. Hence, we focus on perceived cultural—that is, shared—stereotypes. Why the consensus on groups’ warmth and competence? We suggest that cultural stereotypes result from the social structural relations between groups in two primary ways. Specifically, the social structural hypothesis proposes, first, that outgroups are perceived as more competent to the extent that they are perceived as powerful and high status or as less competent to the extent that they are perceived as powerless and low status. The perceived link between a group’s societal outcomes and its perceived competence serves several functions. This link may represent a form of correspondence bias, namely, that people’s behavior (in this case, their position) reflects their traits (D. T. Gilbert & Malone, 1995). Or it might reflect just-world thinking, namely, that people get what they deserve (Lerner & Miller, 1978). At the level of groups, it justifies the system (Jost & Banaji, 1994) and legitimates power–prestige rankings (Berger, Rosenholtz, & Zelditch, 1980; Ridgeway & Berger, 1986). The opposite viewpoint is conceivable: Cultural stereotypes could instead reflect group-level sour grapes (with a bigot reasoning that the out-group may have high status, but they inherited it, lucked out, or cheated, so they do not deserve it, and they actually are stupid). However, we suggest that intergroup stereotypes turn in part on consciousness of power relations; stereotypes function to justify the status quo (Berger et al., 1980; Fiske, 1993a; Glick & Fiske, 2001b; Jost & Banaji, 1994; Jost, Burgess, & Mosso, 2001; Ridgeway & Berger, 1986). Envious stereotypes devolve on that high competence but low warmth lot who seem to be doing better than others. This prediction receives support from findings that perceived power strongly predicted perceived competence in Central and Eastern European stereotypes (Phalet & Poppe, 1997; Poppe & Linssen, 1999). The second part of the social structure hypothesis holds that out-groups are seen as relatively warm and nice to the extent that they do not compete with others. Compliant subordinate groups fulfill a convenient role, so they receive paternalistic prejudice, which disrespects their competence but simultaneously likes the qualities that keep them subordinated as long as they do not pose a threat. Warmth-related identities placate subordinates by assigning them socially

170

Susan T. Fiske et al.

desirable traits that conveniently also imply deference to others (Glick & Fiske, 2001b; Ridgeway, 2001). Negative intentions are not attributed to noncompetitive outgroups, and attributions of warmth help to maintain the status quo with a minimum of conflict (Jackman, 1994). In contrast, competitive out-groups frustrate, tantalize, and annoy, so they are viewed as having negative intent. Out-group goals presumably interfere with in-group goals, so they are not warm. A primary source of negative affect toward out-groups results from perceived incompatibility of their goals with in-group goals (Fiske & Ruscher, 1993). If out-groups are successful, they receive grudging respect for their envied control over resources but never are liked as warm. Low–low groups (e.g., welfare recipients), viewed as parasites in the system, also compete with other groups, not for status but for resources nonetheless. In allegedly draining economic and political capital from society, they supposedly compete in a zero-sum allocation of resources. Their goals are incompatible with others (and in that sense are competitive), so they are not warm. Finally, of course, the in-group, its allies, and reference groups do not compete with themselves, so they are acknowledged as warm. The cultural default groups (middle class, Christian, heterosexual) may not be viewed as competitive, precisely because they possess cultural hegemony. Support for the competition → warmth prediction also comes from the Phalet and Poppe (1997) and Poppe and Linssen (1999) studies, in which perceived inter-nation conflict negatively predicted socially desirable traits (i.e., morality or warmth). Generally parallel efforts to predict intergroup images from structural relations show up in previous work: for example, enemy images in political psychology (Alexander, Brewer, & Herrman, 1999),5 the social role theory of gender stereotypes (Eagly, 1987; Eagly, Wood, & Diekman, 2000),6 and analyses of citydweller and rural-dweller stereotypes (Campbell, 1967; LeVine & Campbell, 1972).7 Both the Eagly (1987) and the Campbell (1967) role analyses focus on characterizing behaviors that result from roles, hence their social utility. Nevertheless, our view is more general, at once applying to many more social groups and going beyond analyses of specific roles. We also emphasize the functional compatibility of combinations that mix perceived competence and warmth, whereby the high–low combination justifies resentment, the low–high combination justifies subordination, and both maintain the status quo.

Review of hypotheses The goals of this research are to investigate our proposals regarding stereotype content: 1. Perceived competence and warmth differentiate out-group stereotypes. 2. Many stereotypes include mixed ascriptions of competence and warmth, as defined by low ratings on one dimension coupled with high ratings on the other.

A model of (often mixed) stereotype content

171

3. Stereotypes depict out-groups as competent to the extent that they are perceived as powerful and high status; stereotypes depict out-groups as relatively warm and nice to the extent that they do not compete with others.

Research strategy A preliminary study and three of the current studies (on eight samples) address these hypotheses. Each study uses a sample of 6–25 out-groups, which come primarily from judges’ nominations of out-groups that are important in the current U.S. scene. Participants rated cultural stereotypes of the out-groups on a series of trait adjectives derived from previous work. We then separately factor analyzed each group’s trait ratings and isolated those that loaded distinctly on competence and warmth dimensions. Traits that loaded consistently across groups constituted two common dimensions, which provides an initial evaluation of the hypothesis that competence and warmth differentiate out-groups. Each group, with its score on the common competence and warmth dimensions, became a unit in cluster analyses. Reasonable cluster solutions derive from standard decision rules. We compared clusters for distributions of groups across the entire space to examine further the dimensional hypothesis. For the mixed stereotypes hypothesis, we examined (a) proportions falling into the mixed, off-diagonal combinations, (b) between-clusters group differences on competence and on warmth, (c) within-cluster group differences between competence and warmth, and (d) individual within-group competence and warmth differences. Participants also rated each group on items assessing perceived status and competition, with specific items again derived from their reliability across a new set of factor analyses within each rated individual group. Correlations of status and competition scales with competence and warmth scales assess the third, social structural hypothesis. A fourth study, on a ninth sample, examines unique affective responses for each of the four competence–warmth combinations. Elaboration of that hypothesis appears later.

Preliminary evidence Previous studies have lacked theory, cross-groups comparison, or generalizable samples. To examine the mixed content of stereotypes, as predicted by social structural variables of status and competition, we undertook some preliminary studies (Fiske et al., 1999). Forty-two undergraduates rated consensual stereotypes of 17 groups on competence and warmth traits.8 A first study indicated that many groups fell along the diagonal from being relatively high on competence but low on warmth to being relatively low on competence but high on warmth, forming two predominantly mixed clusters. A second study examined social structure correlates of stereotypic competence and warmth, with the same groups rated on the single traits of competence and likability (for warmth) along with the hypothesized social structural correlates,

172

Susan T. Fiske et al.

status and competition.9 Perceived status did predict perceived competence, and perceived competition predicted perceived (lack of) warmth. Although they are generally supportive of our framework, these preliminary studies were theoretically undeveloped (i.e., did not include the functional analysis developed here), focused on a broad-brush description that has proven insufficiently sensitive (i.e., only two clusters), and did not include emotional reactions (i.e., prejudices). Moreover, the preliminary studies have several methodological shortcomings. First, they used groups that are certainly current on the U.S. scene but that were selected by our own judgment. Thus, a critic could argue that the results fit the hypotheses because the groups were selected to fit the model. Second, the entire trait scale appeared in the first study only, so the second study’s social structural correlates tested only one trait for each dimension, which is hardly ideal but was necessary to prevent participant fatigue. A critic could argue that this creates a weak test of the hypotheses, generalizing inappropriately from one study to another without completely overlapping scales. Third, the respondents were University of Massachusetts undergraduates, so if they accorded some positive attributes to any given outgroup (i.e., not rating any minorities as completely without positive attributes), perhaps this derived from their liberal political orientation, northeast subculture, or college egalitarianism. Fourth, a salient American out-group, Blacks, fell unaccountably in the middle on warmth and competence.

Current studies The current full-scale studies, long surveys on four samples and short surveys on five samples, formally test our hypotheses. To avoid potential bias in sampling out-groups, in our pilot studies we checked the selection of groups to be included in the surveys. To avoid separating the trait and social structure scales, we included both scales on each questionnaire. To include varied samples, we ensured that five out of nine samples comprised adult respondents, whereas four samples went outside Massachusetts to diverse locations across the United States. To address the puzzlingly nondescript stereotypes of Blacks, we better specified that out-group in terms of commonly used subgroups. This research fills a gap in studies of stereotype content by simultaneously examining groups that cut across gender, age, race, ethnicity, nationality, social class, and disability. It investigates stereotypes that do not neatly fit into the antipathy model of prejudice. It also examines prejudices that correspond to different types of out-groups. Moreover, it offers theoretically guided social structure correlates as predictors of stereotype content. In addition, it taps a wide variety of respondents in the United States.

Pilot Study: Selecting representative and relevant groups for Study 1 The pilot study sought a more representative array than the groups in our initial studies.

A model of (often mixed) stereotype content

173

Method Participants University of Massachusetts undergraduates (24) and nonstudent Amherst, Massachusetts, residents (7) volunteered to complete the questionnaire (15 women, 12 men, 4 unknown; mean age = 21.5 years). They were completely unaware of our hypotheses and unacquainted with stereotyping research. Questionnaire and procedure Participants completed a self-administered, open-ended questionnaire at home, reading the following: Off the top of your head, what various types of people do you think today’s society categorizes into groups (i.e., based on ethnicity, race, gender, occupation, ability, etc.)? In the space below, please list between eight and sixteen such groups. Most participants finished the questionnaire in less than 10 min. Results and discussion The most frequently listed groups were Blacks (74%), Hispanics (45%), rich people (45%), poor people (42%), gay men (39%), Asians (32%), elderly people (29%), blue-collar workers (23%), Jews (23%), disabled people (19%), retarded people (16%), poor Whites (13%), physically attractive people (13%), professionals (13%), southerners (10%), welfare recipients (10%), business or professional women (10%), and housewives (3%). Of the 17 groups used in the preliminary studies, 12 were listed by at least 1 person in our new sample, which suggests that the preliminary list was not too biased by our hypotheses. Nevertheless, these responses—as well as the prior results—changed some of the groups considered. The new set included 23 groups, 12 of which appeared in both our preliminary studies and the pilot sample: rich people, gay men, Asians, elderly people, Jews, disabled people, retarded people, southerners, welfare recipients, businesswomen, housewives, and Latinos (which we changed to Hispanic to reflect respondents’ own terms). The pilot study added blue-collar workers and poor Whites, which makes 14 groups that directly fit the pilot study. Five groups were included for purely theoretical reasons. Because of the gender subgrouping literature, which indicates four consistently replicated subtypes (i.e., housewives, career women, feminists, and sex objects), feminists were retained, although they were not mentioned in the pilot, and sexy women were added. Because of our interest in locating Blacks more precisely, we tried separating Black subgroups by social class on the basis of our pilot sampling listing poor

174

Susan T. Fiske et al.

Blacks among poor people, our own judgment, and prior studies (Bayton, McAlister, & Hamer, 1956; Smedley & Bayton, 1978): We chose Black professionals and poor Blacks. If respondents had been combining these two groups previously, the averaged response might land generic Blacks in the middle. If we were wrong to divide them, professional and poor Blacks should end up in the same middle location as before. We added poor Whites to examine race– class stereotypes suggested by this division of Blacks and also to fit the pilot study item poor people. Finally, four groups resulted from psychometric concerns. Because of our interest in retaining groups that might be significant in the United States outside the northeast, we kept migrant workers and house cleaners and added Arabs. For continuity, we also retained blind people. Thus, the new set of groups, although it was not entirely determined by our pilot sample’s response, included the major groups mentioned by them as well as some other theoretically and politically interesting ones. In any event, the essential sample was not determined a priori by our specific hypotheses.

Study 1, long survey: Competence, warmth, mixed stereotypes, and their predictors Students and nonstudents were surveyed about society’s perceptions of social groups’ traits and the structural relationships of status and competition. An adult and a student sample, both from Massachusetts, completed a questionnaire on which they rated 23 groups on warmth and competence traits and on social structure variables representing status and competition. Method Participants Students. University of Massachusetts undergraduates, recruited from various psychology courses, completed the questionnaire for course credit (50 women, 23 men, 1 who did not indicate gender; mean age = 19.4). Of the 74 participants, 58 (78%) identified themselves as White or Caucasian, 6 (8%) as Black or African American, 4 (5%) as Asian, 3 (4%) as multiethnic, and 2 (3%) as European, leaving 1 (1%) unknown. Participants completed the questionnaires in groups of 10–20, using an empty classroom and taking less than half an hour. One questionnaire was eliminated because it had a completion rate of less than one fifth, which left us with n = 73. Nonstudents. Fifty nonstudents (25 women, 13 men, and 12 who did not indicate gender; mean age = 35.2), recruited by undergraduate psychology students, completed the questionnaires in their own home on a volunteer basis. Most of the adults were friends or family of University of Massachusetts students. Two thirds of the participants identified themselves as White. The students who recruited participants received extra course credit for their involvement.

A model of (often mixed) stereotype content

175

Because of the unmonitored conditions under which the questionnaires were completed and some of the sample’s apparent inexperience with questionnaires, 12 questionnaires were omitted because respondents failed to follow the instructions, which left us with n = 38. Questionnaire and procedure The questionnaire named the same 23 groups listed on the second pilot questionnaire. Participants rated these groups on scales reflecting warmth, competence, perceived status, and perceived competition (see Table 7.2); items were scrambled. Participants were instructed to make the ratings, using 5-point scales (1 = not at all to 5 = extremely), on the basis of how the groups are viewed by American society. They read, “We are not interested in your personal beliefs, but in how you think they are viewed by others.” As in all our studies, this instruction was intended to reduce social desirability concerns and to tap perceived cultural stereotypes. Students received written feedback, and nonstudents received oral feedback. Results This study tests the introduction’s three hypotheses. To test the utility of warmth and competence in describing out-groups, we examined their two-dimensional array in cluster analyses. To test the frequency of mixed combinations, we examined the distribution of groups into various clusters and assessed differences Table 7.2 Scales, Study 1 Construct

Items

Competence

As viewed by society, how . . . are members of this group? [competent, confident, independent, competitive, intelligent]

Warmth

As viewed by society, how . . . are members of this group? [tolerant, warm, good-natured, sincere]

Status

How prestigious are the jobs typically achieved by members of this group? How economically successful have members of this group been? How well-educated are members of this group?

Competition

If members of this group get special breaks (such as preference in hiring decisions), this is likely to make things more difficult for people like me. The more power members of this group have, the less power people like me are likely to have. Resources that go to members of this group are likely to take away from the resources of people like me.

Note: For the Competence and Warmth Scales, the points of ellipsis were replaced by the words in brackets for each question.

176

Susan T. Fiske et al.

in warmth and competence ratings for each group. To test the structural hypotheses, we examined correlations of status with competence and competition with (lack of) warmth. Perceived competence and warmth differentiate among out-group stereotypes To construct trait and predictor scales, we needed ones that worked for each group separately but that overlapped across groups. We calculated 23 factor analyses (one per group) examining all 26 response items; these typically yielded five–eight factors with eigenvalues greater than 1.0. Across groups, five similar factors emerged consistently, and these formed the scales of competence, warmth, status, and competition (as noted in Footnote 2, we omitted cooperation). Each participant rated the 23 groups according to the competence scale (competent, confident, independent, competitive, intelligent; student α = .90, nonstudent α = .85) and warmth scale (tolerant, warm, good-natured, sincere; student α = .82, nonstudent α = .82). For each of the 23 groups, the competence and warmth ratings each were averaged across participants, so the means supplied competence and warmth scores for each group. According to these means, the 23 groups arrayed on a two-dimensional Competence × Warmth space (see Figures 7.1 and 7.2). As predicted, the two dimensions differentiated the groups. To examine the structure of this two-dimensional space, we conducted two types of cluster analyses of the 23 groups. Following Hair, Anderson, Tatham, and Black (1995), we first conducted hierarchical cluster analyses (Ward’s, 1963, method, which minimizes within-cluster variance) to determine the best fitting number of clusters. We then conducted k-means cluster analyses (with the parallel threshold method) to determine which groups fell into which clusters. The distinction between the two analyses roughly parallels stepwise and simultaneous multiple regression. To decide the number of clusters that best reflect the data, we examined agglomeration statistics from the hierarchical analysis. Using Blashfield and Aldenderfer’s (1988) guidance, we interpreted the hierarchical cluster analyses with a twofold approach. First, we identified a plausible number of clusters using typical decision rules, and, second, we validated that solution several ways.10 For both student and nonstudent samples in this study, the last large change came in the break between three and four clusters, so we adopted a four-cluster solution. As in note 10, this decision rule resembles the scree test in factor analysis, whereby researchers have typically cut the number of factors at the bend in the eigenvalues, below which lies statistical rubble. Next, we turned to the k-means cluster analysis to examine which groups fit into which cluster. For both the student sample (see Figure 7.1) and the nonstudent sample (see Figure 7.2), one cluster comprised seven groups: Asians, Black professionals, businesswomen, feminists, Jews, northerners, and

A model of (often mixed) stereotype content

177

High Housewives

Elderly Blind

Warmth

Retarded Disabled House cleaners

Hispanics Poor whites

Gay men Blue-collar Southerners Migrant workers Sexy Arabs women

Black professionals Northerners Business women Asians Jews Feminists Rich

Poor blacks Welfare recipients

Low Low

Competence

High

Figure 7.1. Four-cluster solution, Study 1, long survey, student sample

High

Housewives Elderly Retarded Blind

Gay men

Warmth

Southerners Disabled Migrant workers House cleaners Poor whites Welfare recipients

Poor blacks

Blue-collar Hispanics Sexy women Arabs

Black professionals Northerners Asians Business Feminists women Jews Rich

Low Low

Competence

High

Figure 7.2. Four-cluster solution, Study 1, long survey, nonstudent sample

rich people. These groups also clustered together in the less useful three- and two-cluster solutions in both samples (Table 7.3), so this cluster was stable across samples and across solutions. Another cluster comprised three groups: blind people, elderly people, and housewives; for both samples, these were groups that clustered together also

178

Susan T. Fiske et al.

Table 7.3 Group cluster assignments in two-, three-, and four-cluster solutions, students and nonstudents, Study 1 Students

Nonstudents

Group

4a

3

2

4

3

2

Asians Black professionals Businesswomen Feminists Jews Northerners Rich people

2 2 2 2 2 2 2

2 2 2 2 2 2 2

2 2 2 2 2 2 2

2 2 2 2 2 2 2

2 2 2 2 2 2 2

2 2 2 2 2 2 2

Blind people Elderly people Housewives Retarded people Disabled people House cleaners

4 4 4 4 4 4

3 3 3 3 3 3

1 1 1 1 1 1

3 3 3 3 4 4

3 3 3 1 1 1

1 1 1 1 1 1

Poor Blacks Poor Whites Welfare recipients Hispanics

1 1 1 1

1 1 1 1

1 1 1 1

4 4 4 1

1 1 1 1

1 1 1 1

Migrant workers Blue-collar workers Southerners Gay men Arabs Sexy women

3 3 3 3 3 3

3 2 2 3 1 3

1 2 2 2 1 1

4 1 1 1 1 1

1 3 3 3 2 2

1 2 2 2 2 2

Note: Groups indicated in boldface showed the most stable respective clusters, across solutions and across samples. Breaks between clusters indicate student solutions; nonstudent solutions differed only slightly, as indicated in the right three columns. a Indicates the number of clusters in the solution.

in the four-, three-, and two-cluster solutions, making these stable solutions. The student sample added to this cluster disabled people, house cleaners, and retarded people, who appeared with the others in all three student cluster solutions, making this addition a stable result for the student but not the nonstudent sample. Another cluster also included, for both students and nonstudents, three groups: poor Blacks, poor Whites, and welfare recipients, groups that appeared together in all three cluster solutions for each sample, making this a stable result. Students consistently added Hispanics to this trio in all cluster solutions, making this group stable in this cluster for students. Nonstudents included house cleaners and disabled people here rather than in the previous cluster, in which the students had placed them; nonstudents also added migrant workers here; the last three groups remained in all nonstudent solutions.

A model of (often mixed) stereotype content

179

The final cluster included only two groups that consistently appeared together across solutions and across samples: blue-collar workers and southerners. Across samples and across solutions, the remaining groups (Arabs, gay men, sexy women, and, for students, migrant workers) did not reliably cluster with these two or with each other. In short, competence and warmth dimensions differentiated among four stable clusters that meaningfully and reliably accounted for 16 of the 23 groups (70%) across solutions and samples. Many stereotypes include mixed competence and warmth We defined mixed stereotypes as low ratings on one dimension coupled with high ratings on the other; our hypothesis holds that a substantial number of out-group stereotypes will prove high on either competence or warmth but low on the other. Three analyses address this hypothesis. First, compare the means for the four cluster centers (Table 7.4). In both samples, the cluster with the highest competence ratings (student M = 4.04,

Table 7.4 Competence and warmth means for each cluster, Study 1 Students (n = 73)

Nonstudents (n = 38)

Cluster

Competence

Warmth

Competence

Warmth

Asians, Black professionals, businesswomen, feminists, Jews, northerners, rich people

4.04a

>

3.12b

3.78b

>

2.94b

Housewives, elderly people, blind people, retarded people (student sample adds house cleaners, disabled people)

2.49c




2.48 2.75 3.37 2.83b

1.86 0.94 0.46 1.10

Elderly peoplea Housewives Cluster

2.74 3.06 2.94b

< <
.88. None of the other dimensions showed any significant differences between any of the SCM quadrants. The stimuli in Study 2 were eight images of objects. They were selected using the same pretesting procedures used for the stimuli in Study 1. Scanning parameters All fMRI scanning was conducted at Princeton’s Center for the Study of Brain, Mind, and Behavior, which uses a 3.0-T Siemens Allegra head-dedicated MR scanner. The stimuli were presented by a Dell computer projecting to a screen mounted at the rear of the scanner bore. While supine, participants viewed the screen through a series of mirrors. Responses were recorded using bimanual fiberoptic response pads (Current Designs, Inc., http:// www.curdes.com/response). Prior to functional echoplanar imaging (EPI), subjects received a short series of structural MRI scans to allow for subsequent functional localization. These scans took approximately 12 min and included (a) a brief scout for landmarking and (b) a high-resolution whole-brain magnetization-prepared rapid gradient-echo (MPRAGE) sequence for later localization and intersubject registration. Functional imaging then proceeded using an EPI sequence that allowed for whole-brain coverage in a relatively short period of time (thirty-two 3-mm axial slices, 1-mm gap, repetition time = 2 s, echo time = 30 ms). In-plane resolutions were 3 mm × 3 mm (196-mm field of view, 64 × 64 matrix). Procedure The two studies followed similar procedures; differences are noted. Before entering the scanner, each participant practiced the task on a computer by rating a number of neutral pictures (landscapes) on each of the four emotions (pride, envy, pity, disgust). This provided the opportunity for participants to familiarize themselves with the task. Inside the scanner, participants saw the photographs in a series of six runs of 10 photographs each. In Study 1, they saw each photo only once. In Study 2, the selected experimental pictures were presented three times each, and each run contained 6 filler pictures of neutral stimuli randomly selected from the pretest pool. Data from the filler images are not relevant here and are not presented. All pictures were randomly sequenced for each run, and run order was randomized for each participant. Each picture appeared for 6 s in Study 1 and 4 s in Study 2, followed by a response screen asking participants to indicate which of the four emotions they most felt toward the picture just displayed. This response screen was presented for 2 s in Study 1 and 4 s in Study 2. The interstimulus interval was 12 s after response. During this time, a black screen displayed a green fixation cross. Then the green cross turned red for 1 s, signaling that the next picture was about to appear.

220

Lasana T. Harris and Susan T. Fiske

After the scanning session, outside the scanner, participants again saw the stimuli, but in gray scale on paper. Their task replicated the pretest: rate each of the images on each of the four emotions, along a 5-point scale. Also, the participants indicated whether they felt any additional emotion beyond the four options provided. Finally, they were probed for suspicion; none were suspicious. The participants were then thoroughly debriefed, given credit, and thanked. Preprocessing Both image preprocessing and statistical analysis used Brain Voyager 2000, version 4.8. Before statistical analysis, image preprocessing consisted of (a) correction of slice acquisition order, (b) 3-D rigid-body motion correction, (c) voxel-wise linear detrending across time, and (d) temporal band-pass filtering to remove low- and high-frequency (scanner- and physiology-related) noise. Distortions of EPI images were corrected with a simple affine transformation. Functional images were registered to the structural images and interpolated to cubic voxels. After co-registering participants’ structural images to a standard image using a 12-parameter spatial transformation, we transformed functional data similarly, with a standard moderate degree of spatial smoothing (Gaussian 8-mm full-width/half-maximum filter). Data analysis Data were analyzed using the general linear model available in the Brain Voyager software package. A series of regressions examined BOLD brain activity in response to each of the four kinds of stimuli (i.e., those hypothesized to elicit pride, envy, pity, and disgust). Contrast maps were then created for each participant by simply subtracting the activation during exposure to each picture from the activation during the fixation-cross display. The resulting maps were then averaged across participants and registered to a standard Talairach brain, with coordinates based on this standard system (Talairach & Tournoux, 1988). Additionally, each cell was compared with the other three cells in a contrast analysis (+3 −1 −1 −1). These results are discussed where appropriate. Random-effects analyses were performed on all imaging data.

Results Judgments made by Study 1 participants while in the scanner revealed that they identified the predicted emotions for the pictures of the social groups at rates well above chance (i.e., .25; see Table 8.2), which indicated that they agreed with the pretest data for the stimuli. Further, a point-biserial correlation revealed that while outside the scanner, participants rated a photo in the lowwarmth/low-competence cell higher on disgust, using a 5-point Likert scale, if they had, as predicted, categorized the picture as disgusting while they were inside the scanner (i.e., scored a “hit”) than if they had identified the picture as eliciting another emotion, r(120) = .65, p < .05, prep > .88.

Dehumanizing the lowest of the low 221 Table 8.2 Proportion of the initial ratings indicating that the pictures in each quadrant elicited the predicted emotion Quadrant Pride

Envy

Pity

Disgust

.70 (.05)

.52 (.10)

.83 (.05)

.64 (.06)

Note: The pictures in each quadrant were rated as eliciting the predicted emotion at a rate well above chance (.25). Standard errors are in parentheses.

All reported imaging results are significant at α = .01 (prep > .95) unless otherwise noted. The Study 1 subtractive analysis against fixation completely supported the dehumanization hypothesis. Significant mPFC activity (see Fig. 8.1) was revealed for pride, t(9) = 3.48 at x = −2, y = 48, z = −7 (29 voxels); envy, t(9) = 4.89 at x = −7, y = 51, z = −1 (427 voxels); and pity, t(9) = 4.40 at x = −5, y = 53, z = −1 (31 voxels); but no significant mPFC activity beyond a significant threshold emerged for the disgust cell (see Fig. 8.2). The disgust cell instead was associated with activations in left insula, t(9) = 6.69 at x = −41, y = 13, z = 0 (318 voxels), and right amygdala, t(9) = 3.72 at x = 22, y = −2, z = 17 (22 voxels); these activations are consistent with disgust responses to objects (Schafer, Schienle, & Vaitl, 2005).5 In addition, the activated mPFC anatomical locations did overlap. Fourteen voxels were common for the three areas of mPFC activation, and an examination of the conjunction of these three sets of activation revealed an area of 432 voxels (see Table 8.3 for a region-of-interest, ROI, analysis of these areas). Finally, a 3-versus-1 contrast analysis revealed no mPFC activity beyond a significant threshold for the pictures of low-warmth/low-competence (disgust) social groups, compared with the average activation to pictures of social groups in the other three SCM cells.6 In Study 2, analysis comparing activation in response to pictures of objects relative to fixation also revealed no mPFC activity above baseline for objects pretested as eliciting disgust (see Fig. 8.2). Furthermore, a 3-versus-1 contrast analysis revealed no differential mPFC activation for these disgusting objects, compared with objects in the other three cells. Therefore, we could infer that the emotion of disgust itself was not sufficient to generate mPFC activity. Two of the remaining three cells also did not activate mPFC in either the subtractive analysis or the 3-versus-1 contrast. In fact, only a subtractive analysis of the imaging data in Study 2 revealed small yet significant mPFC activation for the objects pretested as eliciting envy, t(11) = 3.13 at x = −9, y = 49, z = 3 (five voxels). Pride, envy, and pity are social emotions felt only via the presence, implied or actual, of another person (Fiske, Cuddy, & Glick, 2002). Careful debriefing of participants after both the pretest and the imaging session made it apparent that they would report feeling envy in particular toward objects only if the presence of a person was implied. For instance, a stack of money (one of the stimuli representing this quadrant) implied wealth, and

222

Lasana T. Harris and Susan T. Fiske

Figure 8.1. Results of a subtractive analysis of blood-oxygen-level-dependent (BOLD) activations in Study 1. Activation during fixation was subtracted from activation while participants viewed pictures of social groups representing the four quadrants of the stereotype content model (SCM). Results for the three SCM cells (pride, envy, and pity) showing significant activation in medial prefrontal cortex (mPFC) are shown here. The coordinates are from Talairach and Tournoux (1988). The circled areas indicate mPFC activation. Positive t values indicate greater activation to the out-group pictures in the indicated quadrant, and negative t values indicate greater activation to the fixation cross. A = anterior; R = right.

Figure 8.2 Blood-oxygen-level-dependent (BOLD) activations when participants viewed (a) pictures of low-competence/low-warmth groups in Study 1 and (b) pictures of disgusting objects in Study 2. The Y coordinates are from Talairach and Tournoux (1988). Positive t values indicate greater activation to the out-group pictures in the indicated quadrant, and negative t values indicate greater activation to the fixation cross. R = right.

Dehumanizing the lowest of the low 223 Table 8.3 Estimated effect sizes and t values for clusters of overlapping voxels in medial prefrontal cortex (mPFC) in Study 1 Voxels activated by all three emotions activating mPFC (common overlap)

Voxels activated by any of the three emotions activating mPFC (additive overlap)

Emotion

Maximal t value

ηp2

Maximal t value

ηp2

Pride Envy Pity Disgust

3.96 4.27 3.55 2.79

.47 .50 .41 .30

4.03 4.89 4.39 3.06

.47 .57 .52 .34

participants reported feeling envious of the wealthy person, not the object. The subtractive analysis showed no significant mPFC activation to the objects in the pride and pity cells.

General discussion As hypothesized, members of some social groups seem to be dehumanized, at least as indicated by the absence of the typical neural signature for social cognition, as well as the exaggerated amygdala and insula reactions (consistent with disgust) and the disgust ratings they elicit. This conclusion is supported by the relative lack of mPFC activation when participants viewed pictures of low–low social groups. It is also supported by the finding of mPFC activation when these same participants viewed pictures of social groups that did not fall into this cell of the SCM. Additional research also supports the idea that not just any out-group is dehumanized: Significant mPFC activity emerged when White participants looked at Black faces in an age-categorization task (Harris & Fiske, 2003; Wheeler & Fiske, 2005).7 In addition, the low–low quadrant differentially elicits neural patterns consistent with disgust (insula) and fear (amygdala), according to meta-analyses (Murphy, Nimmo-Smith, & Lawrence, 2003; Phan, Wager, Taylor, & Liberzon, 2002). The current results empirically support the idea of dehumanization and are consistent with verbal reports. By providing neural evidence of the phenomenon, these data go beyond verbal reports, which may be subject to selfpresentational concerns. Furthermore, if replicated and extended, this kind of evidence could begin to help explain the all-too-human ability to commit atrocities such as hate crimes, prisoner abuse, and genocide against people who are dehumanized (Allport, 1954; Fiske, Harris, & Cuddy, 2004).

Notes 1

The U.S. Constitution, Article 1, Section 2, considered an African American slave to be three fifths of a person, implying that these people were perceived as less than human at the time the Constitution was written.

224 2

3 4 5

6

7

Lasana T. Harris and Susan T. Fiske

We are not implying that the function of mPFC is solely social cognition. The evidence as to its exact functions is still being gathered. However, the literature indicates that mPFC activation reliably covaries with social cognition, that is, thinking about people, compared with thinking about objects. Space precludes our listing another 10 to 15 studies; those references can be obtained from us upon request. Note that the entirely repeated measures design characteristic of social neuroscience compensates for the fact that the study involved fewer participants than in most social psychological research. The other cells activated the same insula and amygdala areas, relative to fixation, but a 3-versus-1 contrast indicated that only the hypothesized disgust cell activated the amygdala more than the remaining three cells, and that the hypothesized disgust and envy cells activated the insula in this contrast whereas the remaining two cells did not. Although there was not a complete double dissociation, the disgust cell showed more arousal or disgust generally than the others, while failing to activate the mPFC as much as the others. The high-warmth/low-competence (pity) social groups also did not show more mPFC activity than the other social groups in the 3-versus-1 contrast. However, note that mPFC activation in the pity cell (unlike the disgust cell) was significant, relative to fixation, in the main analysis. The 3-versus-1 contrast analysis compares activation for the focal cell with activation across the average of the remaining three cells. In our case, if the mean BOLD signal change of the pride and envy cells was significantly bigger than that of the disgust cell, and the pity cell showed only a moderate effect, then the resulting t test for the pity cell would reveal a significant difference. Thus, although the 3-versus-1 contrasts were significant for both the disgust cell and the pity cell, only in the case of the disgust cell does this result converge with the results of the initial analysis against fixation in showing insufficient activation above baseline. Analyses of the effect size for the overlapping voxels in mPFC in the fixation contrast analysis showed that the disgust cell had the smallest effect, for either common areas or additive areas (see Table 8.3). Finally, this finding for the pity cell fits with the continuum model of impression formation (Fiske, Lin, & Neuberg, 1999), within which the SCM is couched. The continuum model allows for varying degrees of categorical and individuated perception. In this paradigm, White participants were primed to individuate or categorize Black faces, in separate scanning runs. The results replicated the published effects of activation in amygdala and insula (vigilance-related and arousal-related areas) when the faces were treated categorically, but those effects were eliminated when the faces were individuated.

References Allport, G. W. (1954). The nature of prejudice. Reading, MA: AddisonWesley. Amodio, D.M., & Frith, C.D. (2006). Meeting of minds: The medial frontal cortex and social cognition. Nature Reviews Neuroscience, 7, 268–277. Bar-Tal, D. (1989). Delegitimization: The extreme case of stereotyping and prejudice. In D. Bar-Tal, C. Graumann, A. Kruglanski, & W. Stroebe (Eds.), Stereotyping and prejudice: Changing conceptions (pp. 169–182). New York: Springer-Verlag. Castelli, F., Happe, F., Frith, U., & Frith, C. (2000). Movement and mind: A functional imaging study of perception and interpretation of complex intentional movement patterns. NeuroImage, 12, 314–325.

Dehumanizing the lowest of the low 225 Cuddy, A.J., Fiske, S.T., & Glick, P. (2007). The BIAS map: Behaviors from intergroup affects and stereotypes. Journal of Personality and Social Psychology, 92, 631–648. Fiske, S.T., Cuddy, A.J., & Glick, P. (2002). Emotions up and down: Intergroup emotions result from perceived status and competition. In D.M. Mackie & E.R. Smith (Eds.), From prejudice to intergroup emotions: Differentiated reactions to social groups (pp. 247–264). Philadelphia: Psychology Press. Fiske, S.T., Cuddy, A.J., Glick, P., & Xu, J. (2002). A model of (often mixed) stereotype content: Competence and warmth respectively follow from perceived status and competition. Journal of Personality and Social Psychology, 82, 878–902. Fiske, S.T., Harris, L.T., & Cuddy, A.J. (2004). Why ordinary people torture enemy prisoners. Science, 306, 1482–1483. Fiske, S.T., Lin, M., & Neuberg, S. (1999). The continuum model: Ten years later. In Y. Trope & S. Chaiken (Eds.), Dual-process theories in social psychology (pp. 231–254). New York: Guilford Press. Frith, U., & Frith, C. (2001). The biological basis of social interaction. Current Directions in Psychological Science, 10, 151–155. Gallagher, H.L., & Frith, C.D. (2002). Functional imaging of ‘theory of mind.’ Trends in Cognitive Sciences, 7, 77–83. Greene, J.D., Sommerville, R.B., Nystrom, L.E., Darley, J.M., & Cohen, J.D. (2001). An fMRI investigation of emotional engagement in moral judgment. Science, 293, 2105–2108. Harris, L.T., & Fiske, S.T. (2003). [BOLD activations to Black and White faces under different social goal conditions]. Unpublished raw data. Harris, L.T., Todorov, A., & Fiske, S.T. (2005). Attributions on the brain: Neuro-imaging dispositional inferences, beyond theory of mind. NeuroImage, 28, 763–769. Haslam, N. (2005). Dehumanization: An integrative review. Manuscript submitted for publication. Haxby, J.V., Gobbini, M.I., & Montgomery, K. (2004). Spatial and temporal distribution of face and object representations in the human brain. In M. Gazzaniga (Ed.), The cognitive neurosciences (pp. 889–904). Cambridge, MA: MIT Press. Kumaran, D., & Maguire, E.A. (2005). The human hippocampus: Cognitive maps or relational memory? Journal of Neuroscience, 25, 7254–7259. Leibenluft, E., Gobbini, M.I., Harrison, T., & Haxby, J.V. (2004). Mothers’ neural activation in response to pictures of their children and other children. Biological Psychiatry, 56, 225–232. Leyens, J.-P., Cortes, B.P., Demoulin, S., Dovidio, J., Fiske, S.T., Gaunt, R., Paladino, M.-P., Rodriguez-Perez, A., Rodriguez-Torres, R., & Vaes, V. (2003). Emotional prejudice, essentialism, and nationalism. European Journal of Social Psychology, 33, 703–718. Leyens, J.-P., Rodriguez-Perez, A., Rodriguez-Torres, R., Gaunt, R., Paladino, M.P., Vaes, J., & Demoulin, S. (2001). Psychological essentialism and the differential attribution of uniquely human emotions to ingroups and outgroups. European Journal of Social Psychology, 31, 395–411. Mackie, D.M., & Smith, E.R. (2002). From prejudice to intergroup emotions: Differentiated reactions to social groups. New York: Psychological Press. Macrae, C.N., Heatherton, T.F., & Kelley, W.M. (2004). A self less than ordinary: The medial prefrontal cortex and you. In M. Gazzaniga (Ed.), The cognitive neurosciences (pp. 1067–1075). Cambridge, MA: MIT Press.

226

Lasana T. Harris and Susan T. Fiske

Mason, M.F., & Macrae, C.N. (2004). Categorizing and individuating others: The neural substrates of person perception. Journal of Cognitive Neuroscience, 16, 1785–1795. McCabe, K., Houser, D., Ryan, L., Smith, V., & Trouard, T. (2001). A functional imaging study of cooperation in two-person reciprocal exchange. Proceedings of the National Academy of Sciences, USA, 98, 11832–11835. Mitchell, J.P., Banaji, M.R., & Macrae, C.N. (2005). The link between social cognition and self-referential thought in the medial prefrontal cortex. Journal of Cognitive Neuroscience, 17, 1306–1315. Murphy, F.C., Nimmo-Smith, I., & Lawrence, A.D. (2003). Functional neuroanatomy of emotions: A meta-analysis. Cognitive, Affective, and Behavioral Neuroscience, 3, 207–233. Ochsner, K.N. (2005, May). The ME in MEdial prefrontal cortex. Paper presented at the Social-Cognitive-Affective Neuroscience Conference, Princeton, NJ. Ochsner, K.N., Knierim, K., Ludlow, D.H., Hanelin, J., Ramachandran, T., Glover, G., & Mackey, S.C. (2004). Reflecting upon feelings: An fMRI study of neural systems supporting the attribution of emotion to self and other. Journal of Cognitive Neuroscience, 16, 1–27. Opotow, S. (1990). Moral exclusion and injustice: An introduction. Journal of Social Issues, 46, 1–20. Ostrom, T.M., Bond, C.F., Krosnick, J.A., & Sedikides, C. (1994). Attitude scales: How we measure the unmeasurable. In S. Shavitt & T. Brock (Eds.), Persuasion: Psychological insights and perspectives (pp. 15–42). Needham Heights, MA: Allyn & Bacon. Phan, K.L., Wager, T., Taylor, S.F., & Liberzon, I. (2002). Functional neuroanatomy of emotion: A meta-analysis of emotion activation studies in PET and fMRI. NeuroImage, 16, 331–348. Rosenberg, S., Nelson, C., & Vivekananthan, P.S. (1968). A multidimensional approach to the structure of personality impressions. Journal of Personality and Social Psychology, 9, 283–294. Sanfey, A.G., Rilling, J.K., Aronson, J.A., Nystrom, L.E., & Cohen, J.D. (2003). The neural basis of economic decision-making in the Ultimatum Game. Science, 300, 1755–1758. Saxe, R., Carey, S., & Kanwisher, N. (2004). Understanding other minds: Linking developmental psychology and functional neuroimaging. Annual Review of Psychology, 55, 87–124. Saxe, R., & Wexler, A. (2005). Making sense of another mind: The role of the right temporo-parietal junction. Neuropsychologia, 43, 1391–1399. Schafer, A., Schienle, A., & Vaitl, D. (2005). Stimulus type and design influence hemodynamic responses towards visual disgust and fear elicitors. International Journal of Psychophysiology, 57, 53–59. Staub, E. (1989). The roots of evil: The origins of genocide and other group violence. New York: Cambridge University Press. Struch, N., & Schwartz, S.H. (1989). Intergroup aggression: Its predictors and distinctness from in-group bias. Journal of Personality and Social Psychology, 56, 364–373. Talairach, J., & Tournoux, P. (1988). Co-planar stereotaxic atlas of the human brain. New York: Thieme. Wheeler, M.E., & Fiske, S.T. (2005). Controlling racial prejudice: Social-cognitive goals affect amygdala and stereotype activation. Psychological Science, 16, 56–63.

Part IV

Inequality enablers Social cognition and social relevance

9

A prescriptive intergenerational-tension ageism scale Succession, identity, and consumption (SIC) Michael S. North and Susan T. Fiske

Ageism is a peculiar prejudice. Despite the reality that every living person potentially joins every age group, ageism remains relatively underresearched (comparatively rare in prejudice literature), underappreciated (overlooked as a prejudice), and under the radar (subtle in nature; North & Fiske, 2012). Nevertheless, a rapidly growing older population necessitates increased understanding of ageism—for both social psychology and society at large. In the current article, we analyze potential intergenerational tensions over practical and symbolic resources and introduce a measure of ageism with contemporary relevance. Although prior scales focus mainly on what older people allegedly “are” (descriptive stereotypes), the current analysis centers on the role of more controlling, “should”-based, prescriptive beliefs. This approach proposes three prescriptive dimensions that younger generations are particularly likely to endorse: (a) active succession of enviable positions and influence, (b) ageappropriate, symbolic identity maintenance, and (c) minimizing passive sharedresource consumption (SIC). We argue that a rapidly growing older population —intensifying potential intergenerational tensions—necessitates the new, prescriptive ageism scale presented here.

The potential rise of prescriptive (hostile) ageism Demographic shifts render ageism a particularly ripe research topic. Already the largest proportion in history—currently 13% of the U.S. populace—the older population is expected to compose almost 20% by 2030 (U.S. Census Bureau, 2012). Though prevailing stereotypes place elders outside mainstream consciousness—spurring negative (or at best, mixed) descriptive elder stereotypes of ineptness, illness, and irrelevance—an era of more conspicuous older age is forthcoming.1 How increased visibility will change elder images is an empirical question (North & Fiske, 2012). An optimistic standpoint posits a more visible older age debunking negative elder stereotypes. The pessimistic counterpoint cites

230

Michael S. North and Susan T. Fiske

the potential for hostile ageism to brew among younger generations if elders do not step aside and cede resources in the traditional manner (e.g., if they postpone retirement or reap disproportionate government benefits). Theoretically, backlash for overstepping societal boundaries is particularly likely when group outcomes are interdependent (as younger and older age groups are). For instance, because the genders are intimately interconnected in everyday outcomes, women face negative repercussions for violating expectations (e.g., by being too agentic; Glick & Fiske, 1996). Such controlling, prescriptive (“should”-based) stereotypes aim to dictate other groups’ behavior so as to benefit ingroup outcomes (Prentice & Carranza, 2002; Rudman & Glick, 2001). Thus, prescriptive expectations yield far greater between-group differences in endorsement than do descriptive stereotypes. But despite the reality that age groups coexist within society, and the fact that everyone (with luck) eventually joins each age group, different ages’ inherent interdependence has not been considered as an integral factor in driving ageism. Therefore, even though age groups largely agree about descriptive elderstereotype content (Greenberg, Schimel, & Mertens, 2004; Nosek, Banaji, & Greenwald, 2002), because of resource interdependence, younger people theoretically should endorse prescriptive stereotypes more than older people do. Ageism measures have largely overlooked these prescription-based possibilities, as we discuss next.

Extant (descriptive) ageism measures Although useful, extant prejudice measures focus primarily on content (“are”)based notions about older people. Various scales gauge ageist sentiment indirectly, focusing on descriptive aging-process knowledge. A prominent example is the Facts on Aging Quiz (Palmore, 1998). Other instruments focus more directly on prejudicial attitudes. Two early scales—Tuckman and Lorge’s (1953) Attitudes Toward Old People measure and the Negative Attitudes Toward Old People Scale (KOPS; Kogan, 1961)— both gauge agreement with descriptive elder statements. The Aging Semantic Differential (Rosencranz & McNevin, 1969) organizes descriptive statements into three overarching constructs. The three-factor Fraboni Scale of Ageism (FSA; Fraboni, Saltstone, & Hughes, 1990) aims to “measure the affective component of [ageist] attitude, to supplement the cognitive aspect measured by other instruments” (p. 56). It includes both attitudinal beliefs—via descriptive, antilocution items (e.g., “Many old people just live in the past”)—and discriminatory behavior (avoidance; e.g., “I sometimes avoid eye contact with old people when I see them”). Though the word “should” does appear in a third subscale, discrimination (e.g., “Old people should be encouraged to speak out politically”), these prescriptions tend to focus more on what society as a whole should do, rather than expectations targeting elders themselves. Moreover, prescription is not a conceptual focus of the largely descriptive measure. Nevertheless, to demonstrate the current (SIC) measure’s divergence from the FSA, we conducted a study with circumstances

Succession, identity, and consumption in ageism

231

in which we expected the SIC measure to have greater predictive power than the FSA.

Prescriptive domains: Succession, identity, and consumption Age differs from any other social category in its permeability: provided they live long enough, all people eventually join each group. Another way to think of this is that age groups take turns along a hypothetical age queue—with younger people entering, middle-agers enjoying, and elders exiting (e.g., retiring). Although societal allocation of practical and figurative resources tends to favor the middle-aged (North & Fiske, 2012), as long as the line keeps moving, everyone generally gets his or her privileged turn. However, those at the back of the line are dependent on those at the front transitioning away in order to keep the line moving. Thus, we posit three key ways that older people particularly are expected to relinquish resources (North & Fiske, 2013), each pertinent to blocking a different aspect of the theoretical queue. Although not the only possible prescriptive dimensions, we focus on these three as central ones. Succession-based prescriptions derive from expectations surrounding enviable resources and societal positions. Although middle-agers predominately hold the greatest societal influence, younger people’s opportunities more realistically depend on the older people’s stepping aside—primarily in employment (where retirement opens up jobs for the young) and political influence (where older voters form a powerful bloc, while minors face age restrictions). In other words, acceding to succession means allowing those waiting to move predictably toward their turn at the front. Consumption-based prescriptive stereotypes center on passive depletion of currently shared resources. Elder violations derive from apparent exploitativeness, reaping more than a fair share of allotted societal resources—characterized by dilemmas involving government money (e.g., health care) and shared public space (e.g., the highway). Put another way, sharing consumption of societal resources means not using up everything before others get there. Identity involves resources more symbolic than succession or consumption, limiting elder participation in activities usually reserved for younger people. Although in this case those who are considered “old” often depends on context, elders are particularly barred from youth culture (Greenberg et al., 2004). Unwanted intrusions into young ingroup territory can be both direct (e.g., frequenting youth-centered hangouts) and indirect (e.g., attempting to act “cool”). Thus, avoiding identity invasion means not trying to go back through the line again by adopting youth’s territory.

Research overview and hypotheses A priori, we aimed to test empirically a three-factor model of prescriptive ageism. The methodological foundation for the proposed scale utilized four samples, totaling 2,010 participants.

232

Michael S. North and Susan T. Fiske

In Study 1 (scale development), participants rated their agreement with pilotgenerated SIC-based statements; exploratory factor analysis (EFA) ascertained the number of latent factors underlying these items. In Study 2 (initial scale validation), we tested the convergent and divergent validity of the scale, comparing it with other measures of prejudice and potentially related factors. Study 3 tested elder-focused SIC’s divergent validity from antiyoung ageism. Finally, Study 4 tested the scale’s predictive validity, exploring whether high scorers— controlling for another ageism measure (the FSA)—would exhibit the most bias toward older prescription violators. This also served as a test of divergent validity from the descriptive-focused FSA, exploring which scale best predicts reactions to older people who do and do not “know their place” (i.e., adhere to versus violate expectations). Using structural equation modeling (SEM), we conducted in each study a confirmatory factor analysis (CFA), testing the proposed model’s fit to the data. We hypothesized a three-factor solution to be the best fit via both EFA and CFA. We also expected younger people to score the highest on the scale, given SIC’s emphasis on prescriptive tensions arising from generational interdependence.

Study 1 Method Item generation. Forty-one potential scale items derived from lab and participant samples’ open-ended reports. Responses generally answered the question, “What are things older people should or shouldn’t do?” Participants. Participants (N = 427; 264 women; mean age = 32.9 years, median = 32, range 16–81) included 397 online participants from Amazon Mechanical Turk (mTurk) and 30 undergraduates. Participants were primarily White (74.5%); 6.3% or less were East Asian, African American, Latino, South Asian, and “other”/mixed. Procedure. As part of a “social statements survey,” participants rated the SICbased items from 1 (strongly disagree) to 6 (strongly agree).2 Online participants received a nominal payment; undergraduate lab participants received appropriate course credit. Results EFA was used to examine the intercorrelation pattern among the 41 preliminary items, utilizing principal components extraction and a varimax rotation.3 All items with loadings below .40 on their respective factors were discarded, as were strongly double-loading items. Based on the scree plot of variance explained, three overall factors were specified for subsequent extraction (explaining 46.51% of the variance). Factor 1 apparently represented Consumption, Factor 2 reflected Succession, and Factor 3 comprised Identity items.

Succession, identity, and consumption in ageism

233

The final total number of items across these three factors was 20 (Table 9.1 shows item factor loadings across studies), with an overall alpha reliability of .90 and substantial subscale reliabilities (Table 9.2). These three subscales correlated moderately with each other, with Pearson rs ranging from .46 to .61. Descriptive statistics for this and all subsequent studies appear in Table 9.3. An initial CFA was used to examine whether the proposed model (Figure 9.1)4 fit well with the current data set. We used an SEM technique with AMOS Version 7.0 (Arbuckle, 2006). Given that the model was created from this data set, this was not an independent test of fit (this analysis was conducted in Studies 2–4). Nevertheless, across various standard indices, initial evidence emerged for good three-factor model fit (Table 9.4).5 Moreover, in line with comparative practices utilized in other SEM scale-development studies (e.g., Glick & Fiske, 1996; Luhtanen & Crocker, 1992), these same fit indices were considerably worse for a comparative one-factor model (Table 9.4).

Study 2 Measuring potential convergence and divergence, participants in Study 2 completed the SIC scale, in addition to measures of prejudice, social control orientation, and political ideology. We expected the SIC dimensions to correlate with each other (as in Study 1). We also hypothesized SIC to correlate relatively highly with another ageism measure (due to measuring the same general construct), moderately with other types of prejudice (because biases tend to correlate), and slightly with general social control measures (given SIC’s emphasis on control-oriented stereotypes). Finally, we expected political ideology to be uncorrelated with SIC-based bias because ageism has not provoked partisan debate, not yet having featured a salient civil rights movement; thus (unlike sexism and racism) strong political correctness norms have not developed regarding ageism. Method Participants. Participants (N = 93; 69 women; M age = 25.11 years, median = 21, range 18–60) were online paid university participants in a “social statements survey.” Participants were 60.9% White, and 12.0% or less were of the previously noted minority or mixed groups. Procedure. Participants completed the SIC measure using the same 6-point scale as Study 1. Additional prejudice measures included the 29-item FSA (Fraboni et al., 1990), the 22-item Ambivalent Sexism Inventory (ASI; Glick & Fiske, 1996), and an eight-item version of the Symbolic Racism Scale (Henry & Sears, 2002). Intergroup threat/control scales included the 16-item Social Dominance Orientation scale (SDO; Pratto, Sidanius, Stallworth, & Malle, 1994) and the 20-item Right-Wing Authoritarianism scale (RWA; Altemeyer, 1998). Finally, participants reported their political orientation (1 = definitely

Table 9.1 Item factor loadings across studies for the succession, identity, and consumption subscales Item Factor 1: Consumption Doctors spend too much time treating sickly older people.* Older people are too big a burden on the healthcare system.* Older people are often too much of a burden on families. At a certain point, older people’s maximum benefit to society passing along their resources. Older people shouldn’t be so miserly with their money; younger relatives need it. Older people don’t really need to get the best seats on if buses and trains. AARP (American Association of Retired Persons) wastes charity money. Factor 2: Succession If it weren’t for older people opposed to changing the way things are, we could probably progress much more rapidly as a society. The older generation has an unfair amount of political power compared with younger people. Most older people don’t know when to make way for younger people.+ Most older workers don’t know when it’s time to make way for the younger generation.+ Older people are often too stubborn to realize they don’t function like they used to. Younger people are usually more productive than older people at their jobs. Job promotions shouldn’t be based on older workers’ experience rather than their productivity. It is unfair that older people get to vote on issues that will impact younger people much more. Factor 3: Identity Older people typically shouldn’t go to places where younger people hang out.x In general, older people shouldn’t hang out at places for younger people.x Generally older people shouldn’t go clubbing. Older people probably shouldn’t use Facebook. Older people shouldn’t even try to act cool.

S1

S2

S3

S4

.69

.74

.71

.76

.64

.73

.67

.72

.62

.67

.56

.72

.60

.59

.51

.65

.57

.43

.42

.55

.53

.45

.68

.63

.48

.30

.38

.38

.66

.65

.69

.77

.65

.80

.65

.74

.58

.60

.75

.65

.55

.53

.74

.67

.53

.33

.62

.45

.50

.44

.59

.54

.50

.28

.47

.54

.50

.58

.65

.71

.81

.80

.81

.83

.78 .70 .66 .62

.80 .74 .74 .82

.84 .72 .47 .63

.79 .82 .57 .60

Note: S = Study. * + x indicate similar items denoted as covarying in the structural equation model.

Succession, identity, and consumption in ageism

235

Table 9.2 Alpha reliabilities across studies for the succession, identity, and consumption subscales and total scale Scale

S1 (N = 427)

S2 (N = 92)

S3 (N = 1,283) S4 (N = 207)

Succession Identity Consumption Total scale

.84 .85 .83 .90

.84 .87 .75 .90

.85 .83 .83 .91

.85 .84 .86 .91

Note: S = Study.

Table 9.3 Means, standard deviations, and range across studies for the succession, identity, and consumption subscales and total scale Subscale/descriptor S1

S2

S3

S4

Succession Mean SD Minimum Maximum

2.97 0.92 1.00 5.88

3.01 0.82 1.00 5.75

2.96 0.96 1.00 6.00

3.24 0.93 1.38 6.00

Identity Mean SD Minimum Maximum

2.62 1.09 1.00 6.00

2.84 1.20 1.00 6.00

2.65 1.05 1.00 6.00

2.63 1.02 1.00 5.80

Consumption Mean SD Minimum Maximum

2.30 0.88 1.00 5.86

2.31 0.72 1.00 5.00

2.34 0.91 1.00 6.00

2.44 0.90 1.00 5.71

Total scale Mean SD Minimum Maximum

2.65 0.79 1.00 5.65

2.72 0.74 1.00 5.25

2.66 0.83 1.00 5.50

2.81 0.79 1.25 5.85

Note: Each scale ranged from 1 (strongly disagree) to 6 (strongly agree). S = Study.

liberal, 7 = definitely conservative) and party affiliation (1 = strongly Democrat, 7 = strongly Republican). After the SIC scale, the sequence of the other scales was counterbalanced to avert potential order effects. Participants were entered into a monetary-prize lottery. Results Interitem reliability was once again high for the 20-item SIC scale (.90) and each subscale (Table 9.2).

236

Michael S. North and Susan T. Fiske S1

e1

S2

e2

S3

e3

S4

e4

S5

e5

S6

e6

S7

e7

Succession

Identity

C1

e8

I2

e9

I3

e10

I4

e11

I5

e12

e13

C2

e14

C3

e15

C4

e16

C5

e17

C6

e18

C7

e19

C8

e20

Consumption

I1

Figure 9.1 Three-factor structural equation model

Model fit. Like Study 1, CFA found the three-factor model to provide good structural fit for the new data set, and a better fit across all indices than a onefactor model comprising all 20 items (see Table 9.4). Divergent and convergent validity. Alpha reliabilities were high for all comparative scales: FSA (.86), ASI–Hostile (.93), ASI–Benevolent (.88), SDO (.93), RWA (.94), and Symbolic Racism (.81). As expected, the SIC subscales correlated with each other moderately to strongly (rs = .48–.65). Also anticipated, SIC correlated most strongly with

Succession, identity, and consumption in ageism

237

Table 9.4 Full (three-factor) and restricted (one-factor) model fit indices across studies for the succession, identity, and consumption scales Fit index

S1

S2

S3

S4

Three-factor model Chi-square-to-degrees-of-freedom ratio Incremental fit index Comparative fit index Tucker–Lewis index coefficient Root-mean-square error of approximationa

1.94 .96 .96 .95 .047

1.38 .92 .92 .91 .065

4.36 .95 .95 .94 .05

1.87 .93 .93 .92 .065

One-factor model Chi-square-to-degrees-of-freedom ratio Incremental fit index Comparative fit index Tucker–Lewis index coefficient Root-mean-square error of approximationa

4.19 .85 .85 .85 .087

1.90 .81 .81 .78 .10

9.14 .87 .87 .85 .08

3.05 .83 .82 .80 .10

Note: S = Study. a See Kenny (2011) for explanations.

the FSA (r = .70, p < .001) and then more moderately with Symbolic Racism, ASI–Hostile, and ASI–Benevolent (rs = .32–40, ps < .003). Though total SIC correlated significantly with intergroup-hierarchy-focused SDO (r = .31, p = .003), it did not do so with intergroup-value-conflict-based RWA (r = .15, p = .14). As predicted, correlations with political and party affiliations were nonsignificant (both rs < .02). Notably, with the exception of the FSA (β = −.24, p = .02) and its avoidance subscale (β = −.27, p = .009), rater age did not predict scores on any comparative scale (all ps > .05). This contrasts with consistent age trends for the SIC scale (see Demographic Analyses section).

Study 3 Method Participants. Participants (N = 1,283; 808 women) were recruited from Princeton University’s paid participant pool (N = 97) and mTurk (N = 1,186; M age across samples = 33.23 years, median = 30, range 18–81). Participants were 75.4% White, and 6.0% or less were of the noted minority groups. Measures and procedure. Participants completed the 20 validated SIC items using the same 6-point scale; in addition, a subset of the sample completed 21 items reflecting prescriptive stereotypes of younger people taken from a separate project by the current authors (e.g., “Young people shouldn’t use so much foul language”; “Today’s youth are too idealistic”).

238

Michael S. North and Susan T. Fiske

Results Substantial reliability again emerged among all 20 SIC scale items (α = .91), as well as each subscale (Table 9.2), and the 21 antiyoung items (α = .90). As with the prior two studies, the Succession, Identity, and Consumption subscales correlated with one another (rs ranging from .53 to .66, all ps < .001). Model fit. The three-factor model again provided good structural fit for this data set, despite a large chi-square/degrees of freedom ratio (as noted by Kenny, 2011, and others, this is a poor test of fit for data sets with more than 400 cases); see Table 9.4. A three-factor model again bested a one-factor version on all noted indices. Divergence from antiyoung ageism. Though the total SIC scale did correlate significantly with the antiyoung items, the effect size was small (r = .14, p = .01). Moreover, only the Identity subscale significantly correlated with the antiyoung scale (r = .19, p = .001); Consumption was a marginal correlate (r =.10, p =.08), and Succession was nonsignificant (r = .07, p = .22). Demographic analysis also underscored divergence between SIC and antiyoung items. Participant age was not a significant predictor of antiyoung ageism (β = .02, p = .68). Moreover, the two genders did not differ in their endorsement of the antiyoung items, t(332) < 1, nor did Whites and non-Whites differ, t(332) = 1.17, p = .24. These null results contrast with predicted, consistent, significant trends for the SIC scale (see Demographic Analyses section).

Study 4 Method Participants. These mTurk participants (N = 207; 88 women, mean age = 26.87 years, median age = 25, age range 18 –59) were 70.0% White; 9.2% or less were of the noted minority groups. Procedure. Participants completed a study ostensibly connecting mTurk participants with professionals from an “online career network.” This involved reading a randomly assigned profile of another person’s brief self-description. Participants read about a 74-year-old with the name “Max [last name withheld]” who either violated or adhered to SIC prescriptions. Six between-subjects versions were possible: Three different targets violated, respectively, Succession (refusing to retire despite blocking younger potential hires), Identity (displaying affinity for the latest pop music), and Consumption (undergoing a resourceintensive medical procedure). Three adhering counterparts, respectively, stated that “It’s probably time to step aside,” enjoyed oldies music, and decided against the burdensome treatment. After viewing one of the six profiles, participants provided two types of target ratings: Three separate items composed an overall measure of perceived warmth

Succession, identity, and consumption in ageism

239

(α = .88); three other items formed a competence measure (α = .69). These two dimensions, considered by many to be fundamental dimensions of person perception, serve as useful valence snapshots of target judgments, at both the individual and group level (Fiske, Cuddy, & Glick, 2007). The dimensions also hold particular relevance in perceptions of and selection bias toward older workers (Krings, Sczesny, & Kluge, 2011). Participants also responded to six behavioroid items (α = .80) indicating their desire to interact with Max (Table 9.5). Afterward, participants completed both the SIC scale and the FSA. Participants were thanked, debriefed, and compensated. Results The total SIC scale exhibited strong reliability (α = .91), as did each subscale (Table 9.2). The FSA also had high reliability (α = .91). Model fit. The three-factor model again fit the data, and outperformed a onefactor counterpart per all noted indices (Table 9.4). Table 9.5 Study 4 elder target ratings as a function of succession, identity, and consumption scale and Fraboni scale of ageism Standardized β Succession, identity, and consumption scalea

Fraboni scale of ageismb

Warmthc – Violators – Adherers

−.32† −.03

.17 −.27*

Competenced – Violators – Adherers

−.33† .12

.07 −.45***

Interaction desiree – Violators – Adherers

−.46** −.15

.03 −.21†

Participant age

−.22*

−.05

Rated qualities

Note: Fraboni Scale of Ageism from Fraboni, Saltstone, & Hughes (1990). a Controlling for Fraboni Scale of Ageism. b Controlling for the Succession, Identity, and Consumption Scale. c “Warmth” comprises three items: “Overall, how warm might people think Max is?”; “Overall, how kind might people think Max is?”; and “Overall, how good-natured might people think Max is?” d “Competence” comprises three items: “Overall, how competent might people think Max is?”; “Overall, how capable might people think Max is?”; and “Overall, how confident might people think Max is?” e “Interaction desire” represents six items: “Would you be willing to interact further with Max after the study is over?”; “Would you recommend other participants in this survey to interact with Max?”; and “Would you be willing to write and send Max a supportive message?”; reverse-scored: “Would you prefer to ignore Max altogether?”; “If you were to interact further, how likely would you be to say mean things to Max?”; and “Would you suggest to other participants in this survey that they ignore Max?” † p < .10. * p < .05. ** p < .01. *** p < .001.

240

Michael S. North and Susan T. Fiske

Main effects. As the prescriptive framework predicts, SIC-violating targets were perceived as less warm (M = 3.16, SD = 0.93) than adherers (M = 3.96, SD = 0.70) and less worthy of interaction (violator M = 3.23, SD = 0.82; adherer M = 3.58, SD = 0.73), both ts > 3.20, both ps < .003. No differences in competence emerged (violator M = 3.54, SD = 0.72; adherer M = 3.57, SD = 0.80), t < 1. Predictive ability of SIC versus FSA. The framework also predicts that SIC scores will predict reactions to SIC-violating older people; thus, multiple regressions tested the ability of each scale to predict prescriptive reactions toward targets. Indeed, SIC score (controlling for FSA) marginally predicted violating targets’ perceived warmth (β = −.32, p = .08) and competence (β = −.33, p = .06) and significantly predicted interaction desire (β = −.46, p = .007). However, FSA (controlling for SIC) did not significantly predict any of these dependent variables (all ps> .34; Table 9.5). As expected, for adhering targets, SIC score (controlling for FSA) did not significantly predict target ratings (Table 9.5). Controlling for SIC, FSA did significantly predict perceived warmth and competence and marginally predicted interaction desire, but in the opposite direction: The more ageist participants were, the more negatively they viewed the older, adhering targets.6 Rater age significantly predicted FSA score (β = −.21, p = .002). However, controlling for FSA, rater age still significantly predicted SIC (β = −.22, p = .04), whereas the reverse was nonsignificant (FSA controlling for SIC, β = −.05, p = .67). A cross-study demographic analysis on SIC, including rater age, appears next.

Demographic analyses (all studies) Pooling all 2,010 responses, consistent and predicted demographic trends emerge for age, gender, and race/ethnicity. Younger people and men should show more prescriptive ageism; the ethnicity results are unexpected. Age As expected, participant age was a significant inverse predictor of total SIC score (β = −.31, p < .001) and each subscale: Succession (β = −.37, p < .001), Identity (β = −.22, p < .001), and Consumption (β = −.17, p < .001). Gender Men (M = 2.94, SD = 0.82) scored significantly higher overall than women (M = 2.51, SD = 0.76), t(1558.99) = 11.84, p < .001. This held for subscores of Succession (male M = 3.24, SD = 0.99; female M = 2.83, SD = 0.89), t(1530.72) = 9.41, p < .001; Identity (male M = 2.89, SD = 1.09; female M = 2.50, SD = 1.02), t(1587.79) = 8.03, p < .001; and Consumption (male

Succession, identity, and consumption in ageism

241

M = 2.64, SD = 0.92; female M = 2.15, SD = 0.82), t(1519.42) = 12.20, p < .001. Race/Ethnicity A one-way analysis of variance (ANOVA) found significant ethnic differences on the total SIC scale, F(4, 1883) = 13.84, p < .001, partial η2 = .03, and each subscale, all Fs > 6.90, all ps < .001. On the total scale, East Asians (M = 3.07, SD = 0.71) and South Asians (M = 3.02, SD = 0.84) scored the highest. Post hoc Tukey comparisons found both groups’ scores to be significantly higher than Whites’ (M = 2.63, SD = 0.80) and Blacks’ (M = 2.56, SD = 0.86), all ps < .001. Latinos (M = 2.87, SD = 0.93) scored marginally higher than both Whites and Blacks (both ps = .06), but not significantly different from East or South Asians (ps > .38). Subscales showed similar trends, with East Asians scoring significantly higher than Whites and Blacks on all three subscales, South Asians outpacing Whites and Blacks on the Succession and Consumption subscales, and Latinos similarly scoring higher than Blacks and Whites on the Succession subscale and lower than East Asians on the Consumption subscale (all ps < .05). All other comparisons were nonsignificant.

General discussion Four studies created a three-factor (SIC) scale of prescriptive ageism, affirming its convergent validity (with other ageism and prejudice measures), divergent validity (from other types of prejudice, group hierarchy endorsement, political orientation, and antiyoung ageism), and predictive validity (in a SIC-based experiment, doubling as evidence for divergence from FSA). Though overall scale means emerged toward the lower portion of the scale, demographic differences contribute to the larger body of ageism research, helping answer unresolved questions. The younger participants were, the more they endorsed prescriptive, ageist stereotypes across all three SIC domains. Similar young-as-ageist results have emerged with the FSA (Rupp, Vodanovich, & Credé, 2005); however, the opposite trend appears on other descriptive measures (e.g., the KOPS; Hellbusch, Corbin, Thorson, & Stacy, 1995). Such conflicting findings reflect the ageism literature as a whole, which has left largely unresolved the question of whether the young are the most ageist (North & Fiske, 2012). The current prescription-based findings implicate the young as focal proponents of this explicitly controlling prejudice. Moreover, men scored consistently higher than women. Men do tend to exhibit higher levels of prejudice than women in various other domains (e.g., racism; Ekehammar & Sidanius, 1982; Nosek et al., 2002), and the FSA yields the same pattern (Fraboni et al., 1990).

242

Michael S. North and Susan T. Fiske

That Asians scored consistently higher than other groups on the SIC scale contrasts with lay beliefs that traditions of Eastern filial piety reduce ageism (Ng, 1998). Admittedly, our United-States-only samples do not necessarily speak directly to this issue; to clarify this picture, future studies might employ the SIC scale using non-American Asian samples. Meanwhile, Whites and Blacks tended to score at or among the lowest (with Latinos falling somewhere in between). This latter result echoes prior findings of young African Americans’ respect for their elders (Fiske, Bergsieker, Russell, & Williams, 2009). One potential limitation is that the current measure’s development relied heavily on mTurk samples. Though initial testimony touted the website as an inexpensive, reliable source of data (Buhrmester, Kwang, & Gosling, 2011; Paolacci, Chandler, & Ipeirotis, 2010), others have expressed concerns about reduced quality (Reips, Buffardi, & Kuhlmann, 2012). Less debatable is mTurk’s wider age range of participants than undergraduate samples offer—particularly relevant for the current research’s focus on age-based bias and the moderating impact of perceiver age on such perceptions. Another potential limitation is the current research’s focus on a Western population. This is an intentional starting point; many of the issues concerning intergenerational resource distribution seem particularly salient in industrialized societies, where modernization perhaps has made elders’ utility less obvious than in the past (Nelson, 2005). Nevertheless, it remains to be seen whether prescriptive ageism exists at equally high levels in other cultures that are at least believed to be more reverent of their elders. Despite limitations and unanswered questions, the utility of the current ageism scale lies in its departure from existing scales—which tend to focus on infirm irrelevant “elderly.” Given an increasingly visible and relevant older population, the measure’s divergence presents greater contemporary relevance for research in a number of key areas. In gauging attitudes toward Succession of envied resources, the measure can be used to investigate shifting workplace age dynamics and attitudes toward mandatory retirement. The scale’s shared resource-Consumption subscale factors in philosophical and psychological work (e.g., “trolley problem” studies) on people’s valuing of the lives of older adults—particularly important in current hot-button debates over Social Security and health care. The symbolic Identity subscale can be included to buttress both psychological theory on ageism (e.g., social identity theory perspectives, which characterize ageism as deriving from young ingroup identity motivation) and work in more applied domains (e.g., the need for firms marketing products to the fast-growing older demographic to bear in mind limitations for the amount of “cool” older people are allotted). Taken together, the current measure’s focus on expectations concerning proper roles, behaviors, and resources seems vital, as the population ages and notions of what signifies “old” changes with it (North & Fiske, 2012). It is possible that such prescriptions will shift over time, and the scale can be used to document that change.

Succession, identity, and consumption in ageism

243

Conclusion Departing from extant ageism scales, which focus primarily on descriptive elder stereotypes, the SIC scale introduces a novel, prescriptive ageism measure. In incorporating intergenerational tensions over practical and symbolic resources— and finding the young to most endorse ageist prescriptions—the SIC measure diverges from prior measures in both focus and findings. Given its relevance to current real-world resource concerns (such as health care and employment), the scale is a promising tool for cutting-edge ageism research, as the population grays and generational equity concerns grow more salient.

Notes 1

2 3

4

5 6

Underlying the numerous issues arising from a more visible older age is the question of, “How old is old?” On one hand, social policies still conceptualize 65 as senior; on the other hand, an ever-growing, healthier older population might be antiquating this idea. Admittedly, we do not speak to this question in this article (for one helpful discussion, see Dychtwald, 1999, Chap. 4), but it’s an important consideration for researchers, psychologists, and policymakers. We used a 6-point scale so as to force participants to take a stand one way or the other. Varimax rotation presents well-documented strengths, such as simplifying factor structure and aiding interpretability (Abdi, 2003). It is also the method typically used if the proposed factors are expected to represent generally separate constructs (Rattray & Jones, 2007), as the current (SIC) framework does. In addition to the first-order model illustrated here, we also evaluated the fit of a second-order model (one overarching factor with three underlying subfactors). However, fit statistics did not change meaningfully to include the model in this article, so for simplicity’s sake we report only the first-order model. Three pairs of highly similar items—marked in Table 9.1 and delineated in Figure 9.1—were considered intercorrelated to assist with overall model fit. We interpret this unexpected finding as adhering elder targets most resembling the default. Because (like other ageism measures) FSA captures general resentment toward elders, predicting hostility toward default elder targets is not shocking.

References Abdi, H. (2003). Factor rotations. In M. Lewis-Beck, A. Bryman, & T. Futing (Eds.), Encyclopedia for research methods for the social sciences (pp. 978–982). Thousand Oaks, CA: Sage. Altemeyer, B. (1998). The other “authoritarian personality.” In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 30, pp. 47–92). San Diego, CA: Academic Press. Arbuckle, J. L. (2006). Amos (Version 7.0) [Computer software]. Chicago, IL: SPSS. Buhrmester, M., Kwang, T., & Gosling, S. D. (2011). Amazon’s Mechanical Turk: A new source of inexpensive, yet high-quality, data? Perspectives on Psychological Science, 6, 3–5. doi:10.1177/1745691610393980 Dychtwald, K. (1999). Age power: How the 21st century will be ruled by the new old. New York, NY: Tarcher/Putnam.

244

Michael S. North and Susan T. Fiske

Ekehammar, B., & Sidanius, J. (1982). Sex differences in sociopolitical ideology: A replication and extension. British Journal of Social Psychology, 21, 249–257. doi:10.1111/j.2044–8309.1982.tb00546.x Fiske, S. T., Bergsieker, H., Russell, A. M., & Williams, L. (2009). Images of Black Americans: Then, “them” and now, “Obama!” DuBois Review: Social Science Research on Race, 6, 83–101. doi:10.1017/ S1742058X0909002X Fiske, S. T., Cuddy, A. J. C., & Glick, P. (2007). Universal dimensions of social perception: Warmth and competence. Trends in Cognitive Sciences, 11, 77–83. doi: 10.1016/j.tics.2006.11.005 Fraboni, M., Saltstone, R., & Hughes, S. (1990). The Fraboni Scale of Ageism (FSA): An attempt at a more precise measure of ageism. Canadian Journal on Aging, 9, 56–66. doi:10.1017/S0714980800016093 Glick, P., & Fiske, S. T. (1996). The Ambivalent Sexism Inventory: Differentiating hostile and benevolent sexism. Journal of Personality and Social Psychology, 70, 491–512. doi:10.1037/0022–3514.70.3.491 Greenberg, J., Schimel, J., & Mertens, A. (2004). Ageism: Denying the face of the future. In T. D. Nelson (Ed.), Ageism: Stereotyping and prejudice against older persons (pp. 27–48). Cambridge, MA: MIT Press. Hellbusch, J. S., Corbin, D. E., Thorson, J. A., & Stacy, R. D. (1995). Physicians’ attitudes towards aging. Gerontology and Geriatrics Education, 15, 55–66. doi:10.1300/ J021v15n02_06 Henry, P. J., & Sears, D. O. (2002). The Symbolic Racism 2000 Scale. Political Psychology, 23, 253–283. doi:10.1111/0162–895X.00281 Kenny, D. A. (2011). Measuring model fit. Retrieved September 4, 2011, from http:// davidakenny.net/cm/fit.htm Kogan, N. (1961). Attitudes toward old people: The development of a scale and an examination of correlates. Journal of Abnormal and Social Psychology, 62, 44–54. doi:10.1037/h0048053 Krings, F., Sczesny, S., & Kluge, A. (2011). Stereotypical inferences as mediators of age discrimination: The role of competence and warmth. British Journal of Management, 22, 187–201. doi:10.1111/j.1467–8551.2010.00721.x Luhtanen, R., & Crocker, J. (1992). A collective self-esteem scale: Self-evaluation of one’s social identity. Personality and Social Psychology Bulletin, 18, 302–318. doi: 10.1177/0146167292183006 Nelson, T. D.(2005). Ageism: Prejudice against our feared future self. Journal of Social Issues, 61, 207–221. doi:10.1111/j.1540–4560.2005.00402.x Ng, S. H. (1998). Social psychology in an ageing world: Ageism and intergenerational relations. Asian Journal of Social Psychology, 1, 99–116. doi:10.1111/1467–839X. 00007 North, M. S., & Fiske, S. T. (2012). An inconvenienced youth? Ageism and its potential intergenerational roots. Psychological Bulletin, 138, 982–997. doi:10.1037/a0027843 North, M. S., & Fiske, S. T. (2013). Act your (old) age: Prescriptive, ageist biases over succession, consumption, and identity. Personality and Social Psychology Bulletin, 39, (6), 720–734. Nosek, B. A., Banaji, M. R., &, Greenwald, A. G. (2002). Harvesting implicit group attitudes and beliefs from a demonstration web site. Group Dynamics: Theory, Research, and Practice, 6, 101–115. doi:10.1037/1089–2699.6.1.101 Palmore, E. (1998). The Facts on Aging Quiz (2nd ed.). New York, NY: Springer.

Succession, identity, and consumption in ageism

245

Paolacci, G., Chandler, J., & Ipeirotis, P. G. (2010). Running experiments on Amazon Mechanical Turk. Judgment and Decision Making, 5, 411–419. 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 and Social Psychology, 67, 741–763. doi:10.1037/0022–3514.67.4.741 Prentice, D. A., & Carranza, E. (2002). What women and men should be, shouldn’t be, are allowed to be, and don’t have to be: The contents of prescriptive gender stereotypes. Psychology of Women Quarterly, 26, 269–281. doi:10.1111/1471–6402. t01–1–00066 Rattray, J., & Jones, M. C. (2007). Essential elements of questionnaire design and development. Journal of Clinical Nursing, 16, 234–243. doi:10.1111/j.1365–2702. 2006.01573.x Reips, U., Buffardi, L., & Kuhlmann, T. (2012, November). Why NOT to use Amazon Mechanical Turk for the recruitment of participants. Poster presented at the Society for Judgment and Decision Making 33rd annual conference, Minneapolis, MN. Rosencranz, H. A., & McNevin, T. E. (1969). A factor analysis of attitudes toward the aged. The Gerontologist, 9, 55–59. doi:10.1093/geront/9.1.55 Rudman, L. A., & Glick, P. (2001). Prescriptive gender stereotypes and backlash toward agentic women. Journal of Social Issues, 57, 743–762. doi:10.1111/0022–4537.00239 Rupp, D. E., Vodanovich, S. J., & Credé, M. (2005). The multidimensional nature of ageism: Construct validity and group differences. The Journal of Social Psychology, 145, 335–362. doi:10.3200/SOCP.145.3.335–362 Tuckman, J., & Lorge, I. (1953). Attitudes toward old people. The Journal of Social Psychology, 37, 249–260. U.S. Census Bureau. (2012). State and county quickfacts. Retrieved from http:// quickfacts.census.gov/qfd/states/00000.html

10 Nations’ income inequality predicts ambivalence in stereotype content How societies mind the gap Federica Durante, Susan T. Fiske, Nicolas Kervyn, Amy J. C. Cuddy, Adebowale (Debo) Akande, Bolanle E. Adetoun, Modupe F. Adewuyi, Magdeline M. Tserere, Ananthi Al Ramiah, Khairul Anwar Mastor, Fiona Kate Barlow, Gregory Bonn, Romin W. Tafarodi, Janine Bosak, Ed Cairns, Claire Doherty, Dora Capozza, Anjana Chandran, Xenia Chryssochoou, Tilemachos Iatridis, Juan Manuel Contreras, Rui Costa-Lopes, Roberto González, Janet I. Lewis, Gerald Tushabe, JacquesPhilippe Leyens, Renée Mayorga, Nadim N. Rouhana, Vanessa Smith Castro, Rolando Perez, Rosa Rodríguez-Bailón, Miguel Moya, Elena Morales Marente, Marisol Palacios Gálvez, Chris G. Sibley, Frank Asbrock, and Chiara C. Storari

Inequality corrodes human relations. As Alexis de Tocqueville (1835/2003) noted, material differences divide people socially and obstruct empathy, favouring exploitation and slavery. Coming from aristocratic France, in 1831, de Tocqueville travelled the United States, impressed by the ‘equality of conditions’ (p. 11), which, in his opinion, helped Americans to trust each other. Indeed, for thousands of years the quality of human life has improved by raising material living standards, but nowadays for rich countries to get richer adds nothing to quality of life (Wilkinson & Pickett, 2010). What instead seems to matter the most in developed nations is the level of inequality in society, namely, the size of income disparities. Many problems plague more unequal societies: The more inequality, the more health problems, social tensions, and the lower life expectancy, social mobility,

Nations’ income inequality and ambivalence 247 education, trust, happiness, and well-being (see Wilkinson & Pickett, 2009, for a review). If on the one hand, both history and recent events (e.g., the Arab spring, the Occupy movement) argue in favour of people’s need for justice and therefore for fighting against inequality, on the other hand, both history and recent events (e.g., the economic crisis) show the existence of a perplexing degree of acquiescence that contributes to the maintenance of unequal systems. Certainly, collective actions have played a critical role in reducing inequality. However, when considering the level of persistent disparity within and between societies, the relative lack of collective actions may seem surprising. The American historian Howard Zinn (1968) claimed that ‘society’s tendency is to maintain what has been’ and ‘rebellion is only an occasional reaction’ (p. 16; cited in Jost, Banaji, & Nosek, 2004). Social psychologists Jost and Banaji (1994) offer an explanation for such a tendency by arguing that individuals are inclined to rationalize the status quo, thus perceiving the existing social arrangements that affect them as fair, legitimate, and justified. The stereotype content model (SCM; Fiske, Cuddy, Glick, & Xu, 2002) suggests that depicting societal groups in ambivalent ways—such as fortunate in one sphere while unfortunate in another—may mask socio-economic disparities, facilitating, as a consequence, the rationalization and maintenance of the status quo. Using the SCM, the present work begins to investigate the relationship between ambivalent societal stereotypes and income inequality across nations, hypothesizing that the more unequal a society is, the more ambivalence appears as a rational buffer that helps to conceal inequality and maintain the system.

Ambivalence and the SCM Among the functions of stereotypes, Tajfel (1981) argued that stereotypes contribute to the maintenance of the system ‘explaining or justifying a variety of social actions’ (p. 146). More recently, ambivalent stereotypes especially appear to serve this function because they paint both advantaged and disadvantaged groups as possessing distinctive but counterbalanced strengths and weaknesses, as if every ‘class gets its share’ (Lane, 1959, p. 39), leading people to perceive society as fair (Glick & Fiske, 2001; Kay & Jost, 2003). Underlying ambivalent stereotypes, favourable and unfavourable biases co-exist, beyond outgroup antipathy (e.g., Eagly & Kite, 1987; Eagly & Mladinic, 1989; Glick & Fiske, 1996, 2001, 2011; Katz & Hass, 1988). In this context, the SCM (Fiske et al., 2002) innovates by looking at various stereotypes simultaneously and from society’s perspective, as shared, cultural, public images. Not only are many societal stereotypes ambivalent, combining both hostile and favourable beliefs about a group, but also warmth and competence (W–C) are the two basic dimensions capturing cultural contents. Although labels differ (socially vs. intellectually good–bad, Rosenberg, Nelson, & Vivekananthan, 1968; communion vs. agency, Bakan, 1966; see also Abele & Wojciszke, 2007), W–C repeatedly appear as basic dimensions of social judgment (see Cuddy, Fiske, & Glick, 2008; Fiske, Cuddy, & Glick, 2007) and intergroup behaviour (Cuddy, Fiske, & Glick,

248

Federica Durante et al.

2007) because they help individuals’ social interactions, indicating how helpful or harmful a target may be (Fiske et al., 2002). Ambivalent combinations of competence and warmth also emerge in compensation-hypothesis studies (Judd, James-Hawkins, Yzerbyt, & Kashima, 2005), revealing trade-offs of competence and warmth when people rate individuals or groups in a comparative context. A compensatory process occurs when a positive perception on one dimension is offset by a negative perception on the other: Participants tried to ‘rectify this disparity [on one dimension] by asserting that the situation must be reversed on the other dimension of social judgment’ (Judd et al., 2005, p. 910), but only on competence and warmth (see Kervyn, Yzerbyt, & Judd, 2010, for a review). According to the SCM, competence and warmth judgments, respectively, stem from perceived socio-economic status (high–low) and perceived interdependence (cooperative–competitive). These socio-structural factors predict groups’ location on the competence–warmth stereotype map (Fiske, 2012; Fiske et al., 2002, 2007). Crossing status and interdependence, four kinds of stereotype content emerge: High-status, cooperative groups seem both competent and warm, a univalent and positive stereotype; low-status, competitive groups receive univalent but negative stereotypes, seeming both cold and incompetent. Highstatus, competitive groups are ambivalently judged as competent, but not warm; whereas low-status, cooperative groups are ambivalently considered as warm, but incompetent. Stereotype content model hypotheses—perceived competence and warmth differentiate groups’ stereotypes, many ambivalent (or mixed), with status predicting competence, and competition predicting (low) warmth—have extensive support, using a wide range of target groups: Occupations, nationalities, ethnicities, socio-economic groups, religions, and gender subtypes (see Cuddy et al., 2008; for a review). Most relevant here, Cuddy et al. (2009) tested SCM hypotheses using eight European (mostly individualistic) and three East Asian (collectivistic) samples, finding cross-cultural similarities on the main SCM hypotheses, with cross-cultural differences (more modest, collectivistic cultures do not locate reference groups in the high–high cell). SCM provides a pancultural tool for predicting group stereotypes from structural relations with other groups in society, and for comparing across societies. As noted, individuals are inclined to maintain the status quo rather than to subvert it and this might be particularly true for high-status people (e.g., Schmitt, Branscombe, & Kappen, 2003). However, favourable attitudes towards the preservation of the social order, albeit unjust, are also shared, at least under certain conditions, by disadvantaged groups (Jost et al., 2004; Kay et al., 2009; Lane, 1959; Stott & Drury, 2004; Tajfel, 1981). Holding ambivalent beliefs about social categories may help people (especially the more deprived) to tolerate their situation (e.g., Jackman, 1994) because when one’s group is low on one dimension, it is rewarded on the other. For this reason, ambivalent stereotypes can legitimate the status quo in ways that purely hostile stereotypes cannot (Fiske et al., 2002). Because people are more likely to endorse collective actions

Nations’ income inequality and ambivalence 249 only when injustice is relatively clear (see Ellemers & Barreto, 2009), these subtle, ambivalent forms of prejudice may discourage people from challenging unequal systems. Becker and Wright (2011) have indeed recently shown that exposure to benevolent sexism (Glick & Fiske, 1996, 2011), a paternalistic belief that portrays women as ‘wonderful but incompetent’, hence best suited for low-status positions, decreases women’s engagement in collective action, while exposure to hostile sexism increases it (both effects were mediated by system justification motives). Ambivalent gender stereotypes are in fact more prevalent in countries with higher gender inequality at a societal level (Brandt, 2011; Glick et al., 2000, 2004). Furthermore, social problems more reliably associate with income distribution ‘when income differences are measured across nation-states and other large geo-political units’ (Wilkinson & Pickett, 2007, p. 1966). Hence, to explore the ambivalence–inequality association, our cross-national study used SCM theory and method to investigate the ambivalent warmth–competence relationship and its relationship with an income-inequality measure, namely, the Gini index (Brandolini & Smeeding, 2007).

Hypotheses To establish comparability with earlier efforts, we first tested the four SCM hypotheses, namely, how social groups were rated in warmth, competence, status, and competition, expecting that: Societal groups would array on perceived W–C (Hypothesis 1); many groups would appear either more competent or more warm (but not necessarily both or neither; Hypothesis 2); perceived status would positively correlate with competence (Hypothesis 3), and competition negatively with warmth (Hypothesis 4). The inequality hypotheses investigated first overall correlations between W–C, and whether these co-varied with Gini inequality coefficients. The Gini index measures the degree of inequality in the distribution of income within a society. As calculated by the American Central Intelligence Agency, the cumulative family income is plotted against the number of families arranged from the poorest to the richest. Low Gini coefficients indicate a more equal distribution, with 0 corresponding to complete equality, and 100 corresponding to complete inequality.1 We considered a society as more ambivalent when the overall W–C correlation, calculated across societal groups within each sample, was around zero: The less correlated the W–C dimensions, the more the society’s groups appear as a cloud of points, rather than a vector, as they would under a high warmth–competence correlation. The more circular cloud reflects the distribution of many groups into the ambivalent quadrants of the space; the vector shape would show most groups being univalent, low–low or high–high. Thus, we expected higher inequality (Gini) to be associated with lower W–C correlation coefficients (Hypothesis 5). This would fit greater inequality requiring more compensation.

250

Federica Durante et al.

The inequality focus also compared how stereotypic traits and their respective socio-structural predictors (status–competence, competition–warmth) vary across societies. Perhaps viewing some groups ambivalently helps mask the status–competence dimension as the main source of social difference, emphasizing the competition–(low)warmth dimension as an additional source of social difference. A group might appear low status and low competence, but compensated by stereotypically appearing cooperative and friendly. This mechanism, that inequality could be masked by emphasizing another dimension besides status–competence, suggests the hypothesis that correlations between inequality (Gini) and ambivalence (the warmth–competence correlation) will themselves be linked to variations in the competition–warmth correlation: In other words, the inequality–ambivalence association will depend on the way in which competitive groups are perceived in terms of warmth (Hypothesis 6). Alternatively, correlations between inequality and ambivalence could result from compressing or exaggerating perceived status differences (vs. actual differences measured by the Gini). If societies tend to conceal inequality, then the competence attributed to high-status groups could be either minimized or alternatively inflated. Therefore, the inequality–ambivalence correlation would itself correlate with the status–competence correlation, showing either compression, as when people endorse egalitarian principles (negative correlation), or exaggeration, as when people endorse meritocracy (positive correlation; Hypothesis 7). Method Data were collected in: Australia, Belgium, Bolivia, Canada, Chile, Costa Rica, England, Greece, India, Israel, Italy, Malaysia, Mexico, Northern Ireland, New Zealand, Peru, Portugal, South Africa, Spain, Switzerland, and Uganda. Data from South Korea, Japan, Hong Kong, and United States (US) were retrieved from previous studies (Cuddy et al., 2009; Fiske et al., 2002, Study 1) and reanalysed here. Preliminary groups-listing study Following Fiske et al. (2002) and Cuddy et al. (2009), in each country2 a preliminary study identified societal groups considered most salient. In their respective native languages, approximately 1,379 participants, mostly students, 55.14% female, sample sizes between 28 and 100, averaging 25.07 years, voluntarily completed a self-administered, open-ended questionnaire listing: What various types of people their society categorizes into groups; which groups were considered to be of very low status; and of which groups they consider themselves to be member. These questions aimed to identify relevant social groups in the least constrained way, ensuring that all types of groups would be mentioned. Groups listed by at least 15% of participants then appeared in that country’s main survey questionnaire. Across samples, the number of distinct groups ranged between

Nations’ income inequality and ambivalence 251 14 (Chile) and 33 (Bolivia, UPB-CB; Table SI.1, posted online under Supporting Information, doi: 10.1111/bjso.12005, presents demographic information for each sample). Overall, 235 different societal groups were listed, many of which were context specific (i.e., 140 social categories were mentioned only in one preliminary study). Given present purposes, we checked for the possibility that specific target groups would be chosen only in more equal versus unequal countries. We considered the societal groups listed in at least 10 of 292 preliminary studies (i.e., 19 societal groups: Blacks, Catholic people, children, Christians, disabled people, gays, immigrants, Jews, men, middle-class, Muslims, old people, poor people, rich people, students, unemployed people, women, working class, young people). For each target group, we took into consideration the Gini coefficients of the countries in which it was listed, and then calculated the median, thus obtaining 19 Gini medians. They ranged from 33.70 to 39.20, just on either side of the overall median Gini (37.60) of the countries in which the 29 preliminary studies were carried out. Results showed that all these groups were mentioned about equally in both high versus low equal countries. Main survey SAMPLES AND PARTICIPANTS

Thirty-seven samples were recruited, one from each of the countries mentioned, with the exceptions of: Australia (two samples, Asian and EuropeanAustralians), Bolivia (four different Bolivian universities’ campuses2), Israel (two samples, Israeli-Jews and -Arabs), Italy (two samples, students and non-students), New Zealand (two samples, European and Maori-New Zealanders), Northern Ireland (two samples, Catholic and Protestant Irish), Switzerland (four samples, Swiss-German students, Swiss-Italian students, Swiss-French students, and Swiss-French non-students), United States (two samples, students and nonstudents). Respondents (N = 3,229) voluntarily participated in the main survey. Sample sizes varied (n = 30–272), mostly students, 61.04% female, mean age 23.7 years (Table SI.2, posted online under Supporting Information, doi: 10.1111/ bjso.12005, presents demographic information for each sample). QUESTIONNAIRE AND PROCEDURE

In their native languages, participants rated the groups from their countries’ respective preliminary studies on items reflecting warmth, competence, status, and competition.3 Two items measured each construct (Appendix SI.A, posted online under Supporting Information, doi: 10.1111/bjso.12005, presents all items) on 5-point scales (1 = not at all to 5 = extremely). As in previous SCM studies, participants rated how the groups are viewed at a cultural level: ‘We intend to investigate the way societal groups are viewed by the [. . .] society. Thus, we are not interested in your personal beliefs, but in how you think

252

Federica Durante et al.

they are viewed by others.’ This instruction aimed to reduce social desirability concerns and to detect socially shared group stereotypes, in effect collective lay theories about how groups interrelate. Results Reliabilities, for each construct, within each sample, were generally sufficient: Competence α = .64–.93 (median = 0.85); warmth α = .54–.93 (median = 0.75); status α =.60–.95 (median = 0.86); and competition α =.56–.95 (median = 0.71).4 SCM hypotheses As in previous SCM studies, to test whether groups’ stereotypes fell along the two primary dimensions, competence and warmth scores for each societal group were averaged across participants within each sample. These means indeed arrayed groups in a two-dimensional Competence × Warmth space (Hypothesis 1). Two types of cluster analysis examined its structure: Hierarchical cluster analysis (Ward’s, 1963, method, which minimizes within-cluster variance and maximizes between-cluster variance) helped determine the best-fitting number of clusters; then, k-means cluster analyses examined which groups fit into which cluster. For 25 of 37 samples (67%), agglomeration statistics from the hierarchical cluster analysis pointed to a four-cluster solution. Five clusters were, instead, the best fit for the 12 remaining samples—Asian and WhiteAustralian, Canadian, English, Greek, Indian, European and Maori-New Zealand, Malaysian, Japanese, Spanish, and Ugandan samples. To test for ambivalent stereotypes (Hypothesis 2), in each sample, competence and warmth means were compared within (paired t-test) and between (independent t-test) clusters. To be identified as ambivalent (either highcompetence/low-warmth [HC–LW] or low-competence/high-warmth [LC–HW]), a cluster had to meet two conditions established previously (Cuddy et al., 2009; Fiske et al., 2002): W–C means differed significantly; a cluster’s mean for the high dimension was higher than a cluster low on that dimension, and its mean for the low dimension was lower than a cluster high on that dimension. As expected, the majority of groups ended up in ambivalent clusters in 20 of 37 samples. Three exceptions (Portugal and Northern Ireland— Catholic sample) were all univalent or all ambivalent (Japan). Two samples showed an almost equivalent number of groups contained in univalent versus ambivalent clusters (Indian, 17 vs. 16 groups, and Swiss German, 13 vs. 14 groups, respectively), whereas the remaining 12 samples showed more groups gathered in univalent than ambivalent clusters (Appendix SI.B, posted online under Supporting Information, doi: 10.1111/bjso.12005, presents cluster analysis results for each sample). Although in each country participants evaluated their own societal groups, some groups present in most societies, such as immigrants (labelled also as illegal immigrants or migrant workers), and unemployed people consistently ended

Nations’ income inequality and ambivalence 253 up in the low-competence/low-warmth (LC–LW) cluster across cultures. We could furthermore notice some regional idiosyncrasies: For instance, in the European samples, the group Gypsies is included in the LC–LW cluster, whereas in the South American samples we found the group illiterates in that quadrant. The high-competence/high-warmth cluster, instead, comprised the ingroups and the reference groups, which vary from society to society. The LC–HW cluster consistently included old people, children, and disabled people, as the HC–LW consistently included rich people. To test SCM structure–trait predictions (Hypotheses 3 and 4), status and competition scores for each societal group were averaged across participants within each sample, and their means correlated with competence and warmth means, respectively. As expected, perceived status positively correlated with competence (rs = .74–.99, all ps < .001; average r = .90). Perceived competition– warmth correlations averaged r = .32: Ranging from r = .42 to .92, ps < .05, in 15 of 36 samples3; three samples were marginal, ps < .07: r = .37, .35, and .36. Unexpectedly, perceived competition correlated positively with warmth in the Israeli-Arab sample (r = .45, p < .05). The remaining competition– warmth correlations were nonsignificant, but 11 were in the predicted negative direction whereas 6 were not (see Table 10.1). As in previous data sets using these methods, the average status–competence correlation showed twice the effect size of the average (negative) competition–warmth correlation. Overall, Hypotheses 1–4 were supported in all our samples. Because each society rated its own social categories, these results suggest that evaluating different target groups does not affect the SCM basic tenets. Ambivalence and inequality hypotheses The distribution of target groups, within each sample, in the Competence × Warmth space suggested degrees of ambivalence: Higher degrees in some samples (a circular cloud of points, showing a zero warmth–competence correlation; e.g., Figure 10.1), lower degrees in others (a vector, from bottom left to top right, showing a positive warmth–competence correlation; e.g., Figure 10.2). Thus, W–C relate to each other in different ways, in different societies. As said, to measure different patterns across nations, we considered the overall W–C correlation within each sample as an index of ambivalence: The lower the W–C correlation, the greater the ambivalence; the higher a positive W–C correlation, the lower the ambivalence. Warmth and competence correlations were calculated at the target-group level within each sample. As Table 10.1 illustrates, correlations ranged between −.19 (ns) and .91(p < .001), average r = .40. Of 37 samples, 16 showed a small, non-significant W–C correlation, that is, a higher degree of ambivalence according to our definition. Because the number of groups evaluated by participants in each sample varied (i.e., from 14 to 33), we also considered R2 as an estimate of W–C correlations’ effect size. The R2 of the aforementioned 16 coefficients were also small, ranging from .0002 to .1283. The remaining 21 W–C correlation coefficients were positive and significant, ranging from

254

Federica Durante et al.

High

Women Elderly HC-HW Northerners

Disabled people Poor people

Warmth

Hippies

Catholics

Southerners

Christians

HW-LC Natives

Blacks Gays

Intellectuals

LC-LW

Jews

HC-LW

Nerds

Rich people

Politicians Fresas

Low Low

Competence

High

Figure 10.1 Mexican sample, four-cluster solution. Stars indicate cluster centres. H and L, respectively, indicate high and low; W, warmth; C, competence. The term ‘fresas’ refers to rich, spoiled kids.

Women

hi

LC-M/HW

Middle class

W

HC-HW

te

Au

st

ra

Australians

Elderly

White

Buddhists

Children

Warmth

lia

ns

High

Christians Gay

Disabled people

Blue collar

Catholics Students Tradies

Men White collar

Fat people Poor

Asian Australians

Indians

Immigrants

Asian

Teenagers

Black Muslims

HC-MW Rich

LC-M/LW

Refugees LC-LW Unemployed

Aboriginal Australians

Low Low

Competence

High

Figure 10.2 European Australian sample, five-cluster solution. Stars indicate cluster centres. H, M, and L, respectively, indicate high, medium, and low; W, warmth; C, competence.

Nations’ income inequality and ambivalence 255 Table 10.1 Stereotypic traits correlations, social-structure stereotype correlations, Gini coefficients, all samples Sample

Warmth– Status– Competition– Competition– Status– competence r competence r warmth r competence r warmth r

Gini

Mexico US (non-student) South Korea Canada US (student) Peru Japan Israel (Jews) South Africa Hong Kong Italy (student) Chile Bolivia (UMSA) Switzerland (German) Italy (non-student) Costa Rica Bolivia (UCB) Bolivia (UMSA) England Belgium New Zealand (Maori) New Zealand (European) Bolivia (UPB-CB) Australia (European) Uganda Spain Greece Portugal Australia (Asian) Switzerland (French student) India Malaysia Switzerland (Italian) Northern Ireland (Catholic) Switzerland (French non-student) Israel (Arabs) Northern Ireland (Protestant)

−.19 −.09 −.07 −.03 −.03 −.01 .03 .08 .11 .12 .23 .24 .26 .28

UNA .97*** .91*** .91*** .98*** .97*** .88*** .99*** .78*** .99*** .79*** .97*** .90*** .93***

UNA −.53** −.42* −.56** −.67*** −.10 −.39 −.59** −.18 −.37† −.70*** −.14 .03 −.67***

UNA .55** .83*** .19 .33 .97*** .06 .48* .19 .26 .15 .75** .76*** −.10

UNA −.10 −.30 −.18 .01 .10 .03 .07 −.25 .12 −.14 .11 .02 .03

48.2 45.0 31.4 32.1 45.0 49.6 37.6 39.2 65.0 53.3 32.0 52.4 58.2 33.7

.31 .36 .39* .41* .46* .48* .48**

.75*** .89*** .94*** .94*** .93*** .95*** .96***

−.68*** −.36†† −.17 −.11 −.53** −.92*** −.02

.14 .66*** .48** .49** −.35† −.45† .40*

−.10 .05 .25 .36 .48** .65** .42*

32.0 48.0 58.2 58.2 34.0 28.0 36.2

.51**

.91***

−.30

.34

36.2

.54** .56** .57** .60** .61** .61** .65*** .65**

.91*** .94*** .92*** .92*** .81*** .74*** .95*** .87***

.25 −.73*** .41 −.35†† −.14 −.67** .01 −.80***

.64*** −.34 .89*** .09 .51** −.21 .45* −.48*

.42* .45* .40 .53** .37 .31 .63*** .34

58.2 30.5 45.7 32.0 33.0 38.5 30.5 33.7

.72*** .74*** .76*** .77***

.87*** .93*** .76*** .95***

.01 .04 −.18 −.69***

.51** .49** .00 −.33

.53*** .69*** .39* .70***

36.8 44.1 33.7 34.0

.80***

.84***

−.66***

−.40†

.46*

33.7

.81*** .91***

.80*** .92***

.45* −.25

.66*** −.07

.60** 39.2 .78*** 34.0

.27

Note: Data are reported according to the warmth and competence correlations: Ascending order. UNA = Unavailable; *p < .05; **p < .01;***p < .001; †p between .052 and .058; ††p = .07.

256

Federica Durante et al.

r = .39 to .91 (all ps < .05; R2 from .15 to .83), that is, a lower degree of ambivalence. As expected, the percentage of target groups gathered in ambivalent clusters was significantly higher in samples with small, non-significant W–C correlations (on average, 60.82%) than in samples with positive, significant ones (on average, 44.72%), t(35) = 2.03, p = .05. Further correlations compared the percentages of HC–LW and LC–HW groups in each sample with the W–C correlations. The W–C coefficients significantly correlated with percentages of LC–HW groups, r = .48, p < .01, but did not with percentages of HC–LW groups, r = .09, p = .58, suggesting that our ambivalence index was mostly driven by the LC–HW combinations. Next, Fisher’s z-transformation normalized the distribution of W–C coefficients, allowing us to correlate the standardized coefficients with the Gini ones. A significant correlation, r = .34, p < .05, indicated that more ambivalent societies are also generally more unequal, supporting Hypothesis 5 (see Figure 10.3). Inequality emphasizes more than one dimension of intergroup perception (not just unequal status–competence but also differentiated competition– warmth) to compensate.5 The robustness of the inequality–ambivalence association was tested in several ways. First, in a regression analysis, Gini coefficients (centred around the mean), the number of groups rated by each sample (centred around the mean), and their interaction were regressed onto W–C correlations (Fisher standardized) to rule out the possibility that the number of target groups evaluated in each sample had an impact on the W–C correlation’s size and, therefore, on the inequality–ambivalence association. The model explained 15.2% of variance. Gini was the best and only predictor (β = .33, p = .05); neither the number of groups (β = .09, p = .60) nor the interaction term (β = .16, p = .36) was significant. Second, we further tested Hypothesis 5 by controlling for other potential related factors. Wilkinson and Pickett (2009) suggest that when inequality is measured across whole societies versus small areas, its association with social problems is ‘stronger with inequality than with average income, and, in most cases, controlling for average income strengthens the associations with inequality’ (p. 498). Therefore, in a regression analysis, both our measure of income inequality (Gini) and a measure of average income (GDP per capita 2009)6 were regressed onto the Fisher standardized W–C correlations. Results showed that Gini predicted W–C correlations (β =.39, p = .08) whereas GDP did not (β = .09, p = .69; R2 = 11.6%). Furthermore, arguably, the ideology concerning power and inequality in a given society (and not income inequality per se) might be responsible for the prevalence of ambivalent stereotypes. To control for the role of ideology, the Distance Power Index (PDI; Hofstede, 1980)7 was considered. PDI indexes the extent to which the less powerful members of a society accept and expect power to be distributed unequally, in other words, indexes the extent to which society’s inequality is endorsed by those at the bottom of the social ladder as much as by those at the top. ‘The fundamental issue here is how a society handles

Nations’ income inequality and ambivalence 257

Northern Ireland (Protestant)

Switzerland (French non-student)

Israel (Arabs)

W-C correlation

Northern Ireland (Catholic) Switzerland (Italian) Australia (Asians)

Malaysia

India

Switzerland (French student)

Spain Greece Portugal Australia (Europeans) New Zealand (European) Belgium New Zealand (Maori) England Italy (non-student)

Uganda

Bolivia (UPB-BC) Bolivia (UPB) Bolivia (UCB)

Costa Rica

Switzerland (German) Italy (student)

Chile

Israel (Jews)

South Africa

Japan Canada South Korea

Bolivia (UMSA)

Hong Kong

US (student)

Peru

US (non-student) Mexico

GINI

Figure 10.3 Gini coefficients and warmth–competence (W–C) Fisher standardized correlations, all samples.

inequalities among people’ (http://geert-hofstede.com/dimensions.html). Gini and PDI coefficients were regressed onto the Fisher standardized W–C correlations, and results showed that Gini predicted W–C correlations (β = .39, p < .05) whereas PDI did not (β =.06, p = .74; R2 = 16.9%).8 Status–competence and competition–warmth correlations were also Fisher standardized and correlated with Gini coefficients. Testing Hypothesis 6, competition–warmth and Gini coefficients significantly correlated, r = .48, p < .01, indicating that more equal societies show stronger negative competition– warmth associations; in other words, more equality, more dislike for competitive groups. In unequal societies, competition is more acceptable. Testing Hypothesis 7, no significant pattern was found for the Gini and the status– competence correlations (r = .21, p =.23). Finally, we checked for the unpredicted structure–trait combinations: Status–warmth and competition–competence correlations. As previously, they were calculated at the level of target groups within each sample; both unpredicted patterns were inconclusive: For competition–competence, average r = .26, 16 positive correlations (range .40–.97, all ps < .05), four negative correlations (r =.48, p < .05); three marginally significant, r = −.45, −.40, −.35, all ps < .06), the remaining 16 were non-significant. The status–warmth average r = .27, 14 positive correlations (range .39–.78, all ps < .05), the remaining non-significant (see Table 10.1). Fisher standardized correlations with Gini coefficients showed that Gini coefficients significantly correlated with the competition–competence correlations (r = .49, p < .01), indicating that in more

258

Federica Durante et al.

Table 10.2 Across samples correlations among stereotype content model (SCM) indices and Gini coefficients SCM indicies

1

2

1. Warmth–competence 2. Status–competence 3. Competition–warmth 4. Competition–competence 5. Status–warmth 6. Gini

– −.38* – −.19 −.07 −.34* .22 .87*** −.01 −.34* .21

3

4

5

– .68*** .15 .48***

– – −.29† .49*** −.29†

6

–

Note: The SCM correlations used to compute this table were Fisher standardized. *p ⱕ .05; **p ⱕ .01; ***p ⱕ .001; †p = .09.

equal societies being competitive does not necessarily imply being competent. Gini and status–warmth coefficients were only marginally associated (r = .29, p = .09), suggesting a tendency, in more egalitarian settings, to perceive highstatus groups as warm. All the SCM correlations illustrated above and the Gini coefficients were finally correlated across samples. As Table 10.2 shows, when ambivalence is higher (i.e., low W–C correlations), both structural factors (i.e., status and competition) are more strongly related to competence. For lower degrees of ambivalence instead (i.e., high W–C correlation), status is associated with warmth. This latter result is not surprising given the very high status–competence correlation that we consistently find, which implies that any dimension correlating with competence (warmth in this case) will also correlate with status. Considering the Gini index, the cross-sample correlations therefore suggest that less egalitarian societies show more ambivalent stereotypes, and both high status and competition lead to perceiving groups as competent. More egalitarian societies, instead, have fewer ambivalent stereotypes, status and competition do not necessarily imply competence, and high-status groups tend to be perceived as warm. Finally, given the cross-cultural nature of our data and the fact that scales’ reliabilities and sample sizes varied noticeably across countries, the SCM correlations and the moderating role played by income inequality were checked using meta-analytic techniques, which allow weighting correlations to minimize the variance between samples, and to correct for the unreliability of measures. Hedges and colleagues’ method (Hedges & Olkin, 1985; Hedges & Vevea, 1998) was applied, and random-effects models were performed.9 On each SCM index, two meta-analyses were carried out: First, as suggested by Hedges and Vevea (1998), on Fisher standardized correlations; second, as recommended by Hunter and Schmidt (2004), on correlations corrected for unreliability (see Lipsey & Wilson, 2001 for the formulas used). Inversevariance-weighted mean effect sizes concerning the SCM correlations are summarized in Table 10.3. The significance of the average effect sizes can be inferred from the boundaries of the 95% confidence intervals constructed around

Nations’ income inequality and ambivalence 259 Table 10.3 Meta-analysis results. Random-effects models I2 (r1) (%)

95% CI (r2)

I2(r2) (%)

.51

.38, .65

88.6

93.2

1.09

1.04, .1.13

0.6

−.49,−.25

93.6

−.44

−.58, .30

87.3

.29

.17, .40

91.4

.34

.20, .48

88.9

.34

.18, .48

95.5

.38

.21, .55

92.3

Stereotype content model correlations

N

K

r1

95% CI (r1)

Warmth– competence Status– competence Competition– warmth Status– warmth Competition– competence

3,229

37

.44

.33, .55

93

3,139

36

.92

.90, .94

3,139

36 −.38

3,139

36

3,139

36

r2

Note: N = total sample size for the given meta-analysis; K = number of samples included in the meta-analysis; r1 = inverse-variance-weighted mean effect size calculated on Fisher standardized correlations; 95% CI = 95% confidence interval for the inverse-variance-weighted mean effect size (r1 and related 95% CI reported in the table were converted back to r); r2 = inverse-varianceweighted mean effect size on correlations corrected for unreliability; I2 = index of heterogeneity (Higgins & Thompson, 2002): It is based on the Q homogeneity statistic (goodness of fit), I2 = (Q (K − 1))/Q, multiplied by 100 to express it as percentage. Larger values of I2, more heterogeneity.

the mean effect size, which in all cases presented in Table 10.3 do not contain zero. Next, inverse-variance-weighted regressions (random-effects model) assessed the relationship between SCM correlations and income inequality. Weighted regressions were performed on both Fisher standardized and corrected for unreliability correlations. Results, summarized in Table 10.4, corroborated our findings and provided some support for our Hypothesis 7. In fact, when the Table 10.4 Inverse-variance-weighted regression results (random-effects model). Moderating role of Gini coefficients on stereotype content model (SCM) correlations Fisher standardized r SCM correlations

b (95% CI)

Warmth– competence Status– competence Competition– warmth Status– warmth Competition– competence

−.01 (−0.03, −0.001) .009 (−0.005, 0.022) .02 (0.009, 0.03) .01 (−0.02, 0.001) .03 (0.01, 0.04)

r Corrected for unreliability β

p

−.33