Volume 91 Number 5 Published monthly by the American Psychological Association
November 2006
ISSN 0022-3514
Journal of
Personality and Social Psychology ATTITUDES AND SOCIAL COGNITION
Charles M. Judd, Editor Dacher Keltner, Associate Editor Anne Maass, Associate Editor Bernd Wittenbrink, Associate Editor Vincent Yzerbyt, Associate Editor INTERPERSONAL RELATIONS AND GROUP PROCESSES
John F. Dovidio, Editor Daphne Blunt Bugental, Associate Editor Jacques-Philippe Leyens, Associate Editor Antony Manstead, Associate Editor Cynthia L. Pickett, Associate Editor Jeffry A. Simpson, Associate Editor Scott Tindale, Associate Editor Jacquie D. Vorauer, Associate Editor PERSONALITY PROCESSES AND INDIVIDUAL DIFFERENCES
www.apa.org/journals/psp
Charles S. Carver, Editor Tim Kasser, Associate Editor Mario Mikulincer, Associate Editor Eva M. Pomerantz, Associate Editor Richard W. Robins, Associate Editor Gerard Saucier, Associate Editor Thomas A. Widiger, Associate Editor
The Journal of Personality and Social Psychology publishes original papers in all areas of personality and social psychology. It emphasizes empirical reports but may include specialized theoretical, methodological, and review papers. The journal is divided into three independently edited sections: f ATTITUDES AND SOCIAL COGNITION addresses those domains of social behavior in which cognition plays a major role, including the interface of cognition with overt behavior, affect, and motivation. Among topics covered are the formation, change, and utilization of attitudes, attributions, and stereotypes, person memory, self-regulation, and the origins and consequences of moods and emotions insofar as these interact with cognition. Of interest also is the influence of cognition and its various interfaces on significant social phenomena such as persuasion, communication, prejudice, social development, and cultural trends. f INTERPERSONAL RELATIONS AND GROUP PROCESSES focuses on psychological and structural features of interaction in dyads and groups. Appropriate to this section are papers on the nature and dynamics of interactions and social relationships, including interpersonal attraction, communication, emotion, and relationship development, and on group and organizational processes such as social influence, group decision making and task performance, intergroup relations, and aggression, prosocial behavior and other types of social behavior. f PERSONALITY PROCESSES AND INDIVIDUAL DIFFERENCES publishes research on all aspects of personality psychology. It includes studies of individual differences and basic processes in behavior, emotions, coping, health, motivation, and other phenomena that reflect personality. Articles in areas such as personality structure, personality development, and personality assessment are also appropriate to this section of the journal, as are studies of the interplay of culture and personality and manifestations of personality in everyday behavior. Manuscripts: Submit manuscripts to the appropriate section editor according to the above definitions and according to the Instructions to Authors. Section editors reserve the right to redirect papers among themselves as appropriate unless an author specifically requests otherwise. Rejection by one section editor is considered rejection by all; therefore a manuscript rejected by one section editor should not be submitted to another. The opinions and statements published are the responsibility of the authors, and such opinions and statements do not necessarily represent the policies of APA or the views of the editors. Section editors’ addresses appear below:
ATTITUDES AND SOCIAL COGNITION Charles M. Judd, Editor c/o Laurie Hawkins Department of Psychology University of Colorado UCB 345 Boulder, CO 80309
INTERPERSONAL RELATIONS AND GROUP PROCESSES John F. Dovidio, Editor Department of Psychology University of Connecticut 406 Babbidge Road Storrs, CT 06269-1020
PERSONALITY PROCESSES AND INDIVIDUAL DIFFERENCES Charles S. Carver, Editor ATTN: JPSP: PPID Department of Psychology University of Miami P.O. Box 248185 Coral Gables, FL 33124-0751 Change of Address: Send change of address notice and a recent mailing label to the attention of the Subscriptions Department, American Psychological Association, 30 days prior to the actual change of address. APA will not replace undelivered copies resulting from address changes;
journals will be forwarded only if subscribers notify the local post office in writing that they will guarantee periodicals forwarding postage. Electronic access: APA members who subscribe to this journal have automatic access to a 3-year file of the journal in the PsycARTICLES姞 full-text database. See http://members.apa.org/access. Reprints: Authors may order reprints of their articles from the printer when they receive proofs. Single Issues, Back Issues, and Back Volumes: For information regarding back issues or back volumes write to Order Department, American Psychological Association, 750 First Street, NE, Washington, DC 20002-4242. Microform Editions: For information regarding microform editions, write to University Microfilms, Ann Arbor, MI 48106. Copyright and Permission: Those who wish to reuse APAcopyrighted material in a non-APA publication must secure from APA and the author of reproduced material written permission to reproduce a journal article in full or journal text of more than 500 words. APA normally grants permission contingent upon like permission of the author, inclusion of the APA copyright notice on the first page of reproduced material, and payment of a fee of $20 per page. Permission from APA and fees are waived for those who wish to reproduce a single table or figure from a journal for use in a print product, provided the author’s permission is obtained and full credit is given to APA as copyright holder and to the author through a complete citation. (Requesters requiring written permission for commercial use of a single table or figure will be assessed a $25 service fee.) Permission and fees are waived for authors who wish to reproduce their own material for personal use; fees only are waived for authors who wish to use more than a single table or figure of their own material commercially (but for use in edited books, fees are waived for the author only if serving as the book editor). Permission and fees are waived for the photocopying of isolated journal articles for nonprofit classroom or library reserve use by instructors and educational institutions. A permission fee may be charged to the requester if students are charged for the material, multiple articles are copied, or large-scale copying is involved (e.g., for course packs). Access services may use unedited abstracts without the permission of APA or the author. Libraries are permitted to photocopy beyond the limits of U.S. copyright law: (1) post-1977 articles, provided the per-copy fee in the code for this journal (0022-3514/06/ $12.00) is paid through the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923; (2) pre-1978 articles, provided that the per-copy fee stated in the Publishers’ Fee List is paid through the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923. Address requests for reprint permission to the Permissions Office, American Psychological Association, 750 First Street, NE, Washington, DC 20002-4242. APA Journal Staff: Susan J. A. Harris, Senior Director, Journals Program; Skip Maier, Director, Journal Services; Paige W. Jackson, Director, Editorial Services; Becky Shaw, Account Manager; Jodi Ashcraft, Advertising Sales Manager.
Journal of Personality and Social Psychology (ISSN 0022-3514) is published monthly in two volumes per year by the American Psychological Association, 750 First Street, NE, Washington, DC 20002-4242. Subscriptions are available on a calendar year basis only (January through December). The 2007 rates follow: Nonmember Individual: $442 Domestic, $488 Foreign, $515 Air Mail. Institutional: $1,349 Domestic, $1,443 Foreign, $1,470 Air Mail. APA Member: $208. Write to Subscriptions Department, American Psychological Association, 750 First Street, NE, Washington, DC 20002-4242. Printed in the U.S.A. Periodicals postage paid at Washington, DC, and at additional mailing offices. POSTMASTER: Send address changes to Journal of Personality and Social Psychology, 750 First Street, NE, Washington, DC 20002-4242.
The paper in this journal meets or exceeds EPA guidelines for recycled paper. Since 1986, this journal has been printed on acid-free paper.
NEW RELEASES from the American Psychological Association
Psychological Treatment of Obsessive-Compulsive Disorder Fundamentals and Beyond
Edited by Martin M. Antony, Christine Purdon, and Laura J. Summerfeldt 2007. 328 pages. Hardcover. List: $59.95 APA Member/Affiliate: $49.95 ISBN 1-59147-484-1 ISBN-13: 978-1-59147-484-5 Item # 4317123
Termination in Psychotherapy A Psychodynamic Model of Processes and Outcomes
Anthony S. Joyce, William E. Piper, John S. Ogrodniczuk, and Robert H. Klein 2007. 224 pages. Hardcover. List: $59.95 APA Member/Affiliate: $49.95 ISBN 1-59147-730-1 ISBN-13: 978-1-59147-730-3 Item # 4316086
Why Aren’t More Women in Science? Top Gender Researchers Debate the Evidence
Edited by Stephen J. Ceci and Wendy M. Williams 2007. 248 pages. Hardcover. List: $59.95 APA Member/Affiliate: $49.95 ISBN 1-59147-485-X ISBN-13: 978-1-59147-485-2 Item # 4316085
Psychology and the Department of Veterans Affairs A Historical Analysis of Training, Research, Practice, and Advocacy
Rodney Baker and Wade Pickren 2007. 200 pages. Hardcover. List: $59.95 APA Member/Affiliate: $49.95 ISBN 1-59147-453-1 ISBN-13: 978-1-59147-453-1 Item # 4316084
Sensation Seeking and Risky Behavior Marvin Zuckerman 2007. 326 pages. Hardcover. List: $59.95 APA Member/Affiliate: $49.95 ISBN 1-59147-738-7 ISBN-13: 978-1-59147-738-9 Item # 4317124
Career Paths in Psychology Where Your Degree Can Take You Second Edition
Edited by Robert J. Sternberg 2007. 443 pages. Paperback. List: $29.95 APA Member/Affiliate: $24.95 ISBN 1-59147-732-8 ISBN-13: 978-1-59147-732-7 Item # 4313008
AD0485
To O r d e r : 8 0 0 - 374 -2721 • w w w. a p a .or g / b o o ks
apa books
Available July 2006
A Landmark Reference That Defines the Lexicon of Psychology
APA Dictionary of Psychology Editor-in-Chief: Gary R. VandenBos, PhD The American Psychological Association is proud to announce the publication of an invaluable addition to your reference shelf, one that represents a major scholarly and editorial undertaking. With over 25,000 terms and definitions, the APA Dictionary of Psychology encompasses all areas of research and application, and includes coverage of concepts, processes, and therapies across all the major subdisciplines of psychology. Ten years in the making and edited by a distinguished editorial board of nearly 100 psychological scholars, researchers and practitioners, the Dictionary is destined to become the most authoritative reference of its kind. Academicians, researchers, clinicians, undergraduates and graduate students, and professionals in allied mental health, education, medicine, and law, as well as academic and public libraries, will find the Dictionary essential. 2006. Hardcover. 1,008 pages. List: $49.95 | APA Member/Affiliate: $39.95 ISBN 1-59147-380-2 | Item # 4311007
The APA Dictionary of Psychology includes •
25,000 entries offering clear and authoritative definitions
•
Thousands of incisive cross-references directing the user to synonyms and antonyms, acronyms and abbreviations, and related terms and concepts that deepen the user’s understanding of related topics
•
Balanced coverage of over 100 subject areas across the field of psychology including clinical, experimental, neuropsychology, cognitive, personality and social, developmental, health, psychopharmacology, methodology and statistics, and many others
• 800-374-2721 www.apa.org/books
Entries include nearly 8,000 terms from the APA’s Thesaurus of Psychological Index Terms® which helps students and researchers refine their APA database searches (such as the flagship PsycINFO® bibliographic database’s 2+ million records)
• •
“A Guide to Use” and “Quick Guide to Format” that together explain important stylistic and format features to help readers most effectively use the Dictionary Each of four appendices gathers terms into a thematic summary listing, covering (1) biographies; (2) institutions, associations and organizations; (3) psychological therapies and interventions; and (4) psychological tests and assessment instruments
AD0470
Journal of
Personality Social Psychology and
www.apa.org/journals/psp November 2006 VOLUME 91 NUMBER 5
Copyright © 2006 by the American Psychological Association
Attitudes and Social Cognition 797
Automatic and Controlled Components of Judgment and Decision Making Mario B. Ferreira, Leonel Garcia-Marques, Steven J. Sherman, and Jeffrey W. Sherman
814
Stereotypes: Static Abstractions or Dynamic Knowledge Structures? Leonel Garcia-Marques, A. Sofia C. Santos, and Diane M. Mackie
832
Terror Management and Attributions of Blame to Innocent Victims: Reconciling Compassionate and Defensive Responses Gilad Hirschberger
845
Transcending the “Here”: The Effect of Spatial Distance on Social Judgment Marlone D. Henderson, Kentaro Fujita, Yaacov Trope, and Nira Liberman
Interpersonal Relations and Group Processes 857
The Costs and Benefits of Undoing Egocentric Responsibility Assessments in Groups Eugene M. Caruso, Nicholas Epley, and Max H. Bazerman
872
When Perspective Taking Increases Taking: Reactive Egoism in Social Interaction Nicholas Epley, Eugene M. Caruso, and Max H. Bazerman
890
Cultural Affordances and Emotional Experience: Socially Engaging and Disengaging Emotions in Japan and the United States Shinobu Kitayama, Batja Mesquita, and Mayumi Karasawa
904
Will You Be There for Me When Things Go Right? Supportive Responses to Positive Event Disclosures Shelly L. Gable, Gian C. Gonzaga, and Amy Strachman
918
When Inclusion Costs and Ostracism Pays, Ostracism Still Hurts Ilja van Beest and Kipling D. Williams
929
When Sex Is More Than Just Sex: Attachment Orientations, Sexual Experience, and Relationship Quality Gurit E. Birnbaum, Harry T. Reis, Mario Mikulincer, Omri Gillath, and Ayala Orpaz
Personality Processes and Individual Differences 944
Social Role and Birth Cohort Influences on Gender-Linked Personality Traits in Women: A 20-Year Longitudinal Analysis Stephanie Kasen, Henian Chen, Joel Sneed, Thomas Crawford, and Patricia Cohen
(contents continue)
959
Personality Development in Emerging Adulthood: Integrating Evidence From Self-Ratings and Spouse Ratings David Watson and John Humrichouse
975
Nostalgia: Content, Triggers, Functions Tim Wildschut, Constantine Sedikides, Jamie Arndt, and Clay Routledge
Other 993 871 958
Call for Nominations New Editors Appointed, 2008–2013 Subscription Order Form
ii
ATTITUDES AND SOCIAL COGNITION CHARLES M. JUDD, Editor University of Colorado at Boulder ASSOCIATE EDITORS DACHER KELTNER University of California, Berkeley ANNE MAASS Universita` di Padova, Padova, Italy BERND WITTENBRINK University of Chicago VINCENT YZERBYT Catholic University of Louvain, Louvain-la-Neuve, Belgium CONSULTING EDITORS ICEK AJZEN University of Massachusetts
ALICE H. EAGLY Northwestern University
NIRA LIBERMAN Tel Aviv University, Tel Aviv, Israel
LINDA SKITKA University of Illinois at Chicago
NICHOLAS EPLEY University of Chicago
DIANE M. MACKIE University of California, Santa Barbara
JOHN SKOWRONSKI Northern Illinois University
RUSSELL H. FAZIO Ohio State University
NEIL MACRAE Dartmouth College
ELIOT R. SMITH Indiana University Bloomington
LISA FELDMAN BARRETT Boston College
TONY MANSTEAD Cardiff University, Cardiff, Wales
SUSAN T. FISKE Princeton University
THOMAS MUSSWEILER Universita¨t Ko¨ln, Cologne, Germany
DIEDERIK STAPEL University of Groningen, Groningen, the Netherlands
BARBARA L. FREDRICKSON University of Michigan
JAMES M. OLSON University of Western Ontario, London, Ontario, Canada
WENDI GARDNER Northwestern University
MAHZARIN BANAJI Harvard University
BERNADETTE M. PARK University of Colorado at Boulder
DANIEL GILBERT Harvard University
MONICA BIERNAT University of Kansas
RICHARD E. PETTY Ohio State University
THOMAS GILOVICH Cornell University
IRENE V. BLAIR University of Colorado at Boulder
NEAL J. ROESE University of Illinois at Urbana– Champaign
ANTHONY G. GREENWALD University of Washington
GALEN V. BODENHAUSEN Northwestern University
DAVID L. HAMILTON University of California, Santa Barbara
MARKUS BRAUER LAPSCO, Universite´ Blaise Pascal Clermont-Ferrand, France
EDWARD R. HIRT Indiana University Bloomington
MARILYNN B. BREWER Ohio State University
TIFFANY ITO University of Colorado at Boulder
JOHN T. CACIOPPO University of Chicago
YOSHIHISA KASHIMA University of Melbourne, Victoria, Australia
OLIVIER CORNEILLE Catholic University of Louvain, Louvain-la-Neuve, Belgium
KARLE CHRISTOPHE KLAUER Albrecht-Ludwigs-Universita¨t Freiburg, Freiburg, Germany
PATRICIA DEVINE University of Wisconsin—Madison AP DIJKSTERHUIS University of Amsterdam, Amsterdam, the Netherlands DAVID DUNNING Cornell University
MYRON ROTHBART University of Oregon LAURIE RUDMAN Rutgers, The State University of New Jersey MARK SCHALLER University of British Columbia, Vancouver, British Columbia, Canada TONI SCHMADER University of Arizona NORBERT SCHWARZ University of Michigan
ARIE W. KRUGLANSKI University of Maryland
GU¨N R. SEMIN Free University, Amsterdam, the Netherlands
ALAN LAMBERT Washington University in St. Louis
JEFFREY W. SHERMAN University of California, Davis
JENNIFER LERNER Carnegie Mellon University
STEVEN J. SHERMAN Indiana University Bloomington
FRITZ STRACK Universita¨t Wu¨rzburg, Wu¨rzburg, Germany ABRAHAM TESSER University of Georgia YAACOV TROPE New York University THERESA K. VESCIO Pennsylvania State University WILLIAM VON HIPPEL University of New South Wales, Sydney, Australia DUANE T. WEGENER Purdue University DANIEL M. WEGNER Harvard University DIRK WENTURA Saarland University, Saarbru¨cken, Germany DANIEL WIGBOLDUS Radboud University Nijmegen, Nijmegen, the Netherlands TIMOTHY D. WILSON University of Virginia PIOTR WINKIELMEN University of California, San Diego MARK P. ZANNA University of Waterloo, Waterloo, Ontario, Canada
ASSISTANT TO THE EDITOR—LAURIE HAWKINS
INTERPERSONAL RELATIONS AND GROUP PROCESSES JOHN F. DOVIDIO, Editor University of Connecticut ASSOCIATE EDITORS DAPHNE BLUNT BUGENTAL University of California, Santa Barbara BEVERLEY FEHR University of Winnipeg, Winnipeg, Manitoba, Canada JACQUES-PHILIPPE LEYENS Catholic University of Louvain, Louvain-la-Neuve, Belgium ANTONY MANSTEAD Cardiff University, Cardiff, United Kingdom JEFFRY A. SIMPSON University of Minnesota, Twin Cities Campus
ARTHUR ARON State University of New York at Stony Brook
RUPERT BROWN The University of Kent at Canterbury, Canterbury, England
XIMENA ARRIAGA Purdue University
LORNE CAMPBELL University of Western Ontario, London, Ontario, Canada
WINTON W. T. AU The Chinese University of Hong Kong, Shatin, Hong Kong MARK BALDWIN McGill University, Montreal, Quebec, Canada KIM BARTHOLOMEW Simon Fraser University, Burnaby, British Columbia, Canada C. DANIEL BATSON University of Kansas
SCOTT TINDALE Loyola University Chicago
B. ANNE BETTENCOURT University of Missouri—Columbia
JACQUIE D. VORAUER University of Manitoba, Winnipeg, Manitoba, Canada
GERD BOHNER Universita¨t Bielefeld, Bielefeld, Germany
CONSULTING EDITORS DOMINIC ABRAMS University of Kent at Canterbury, Canterbury, England
NIALL BOLGER Columbia University
CHRIS AGNEW Purdue University
JONATHON D. BROWN University of Washington
NYLA R. BRANSCOMBE University of Kansas
SERENA CHEN University of California, Berkeley MARGARET CLARK Yale University CARSTEN DE DREU University of Amsterdam, Amsterdam, the Netherlands STE´PHANIE DEMOULIN Catholic University of Louvain Louvain-la-Neuve, Belgium, and Belgan National Fund for Scientific Research, Brussels, Belgium
KLAUS FIEDLER University of Heidelberg, Heidelberg, Germany GARTH FLETCHER University of Canterbury, Christchurch, New Zealand SHELLY GABLE University of California, Los Angeles LOWELL GAERTNER University of Tennessee, Knoxville SAMUEL L. GAERTNER University of Delaware ADAM GALINSKY Northwestern University PETER GLICK Lawrence University STEPHANIE A. GOODWIN Purdue University
DAVID DESTENO Northeastern University
MARTIE G. HASSELTON University of California, Los Angeles
STEVE DRIGOTAS Johns Hopkins University
S. ALEXANDER HASLAM University of Exeter, Exeter, United Kingdom
ELISSA S. EPEL University of California, San Francisco VICTORIA ESSES University of Western Ontario, London, Ontario, Canada
(editors continue)
VERLIN HINSZ North Dakota State University GORDON HODSON Brock University, St. Catherine’s, Ontario, Canada
MICHAEL A. HOGG University of Queensland, Brisbane, Australia
LAURA J. KRAY University of California, Berkeley
ANDREA B. HOLLINGSHEAD University of Southern California JOHN G. HOLMES University of Waterloo, Waterloo, Ontario, Canada RICK H. HOYLE University of Kentucky
JAMES R. LARSON JR. University of Illinois at Chicago COLIN WAYNE LEACH University of Sussex, Sussex, United Kingdom JOHN LEVINE University of Pittsburgh JOHN E. LYDON McGill University, Montreal, Quebec, Canada
JOLANDA JETTEN University of Exeter, Exeter, United Kingdom
JON K. MANER Florida State University
JAMES D. JOHNSON University of North Carolina at Wilmington TATSUYA KAMEDA Hokkaido University, Sapporo, Japan BENJAMIN R. KARNEY RAND Corporation, Santa Monica, California YOSHI KASHIMA University of Melbourne, Victoria, Australia
BRENDA MAJOR University of California, Santa Barbara CRAIG MCGARTY Australian National University, Canberra, Australia WENDY BERRY MENDES Harvard University RICHARD MORELAND University of Pittsburgh
DEBORAH A. KASHY Michigan State University
SABINE OTTEN University of Gro¨ningen, Gro¨ningen, the Netherlands CRAIG D. PARKS Washington State University LOUIS A. PENNER Wayne State University PAULA PIETROMONACO University of Massachusetts at Amherst
CHRISTINE SMITH Grand Valley State University HEATHER J. SMITH Sonoma State University RUSSELL SPEARS Cardiff University, Cardiff, Wales CHARLES STANGOR University of Maryland GARY L. STASSER Miami University—Ohio
TOM POSTMES University of Exeter, Exeter, United Kingdom
WALTER STEPHAN New Mexico State University
FELICIA PRATTO University of Connecticut
WILLIAM B. SWANN JR. University of Texas at Austin
HARRY T. REIS University of Rochester
JANET SWIM Pennsylvania State University
W. STEVEN RHOLES Texas A&M University
LEIGH L. THOMPSON Northwestern University
JENNIFER A. RICHESON Northwestern University
TOM TYLER New York University
MARK SCHALLER University of British Columbia, Vancouver, British Columbia, Canada
JEROEN VAES University of Padova, Padova, Italy
BRIAN MULLEN KERRY KAWAKAMI University of Kent at Canterbury, York University, Toronto, Ontario, Canada Canterbury, England JANICE R. KELLY AME´LIE MUMMENDEY Purdue University Friedrich-Schiller-Universita¨t, Jena, DACHER KELTNER Jena, Germany University of California, Berkeley MARK MURAVEN DAVID A. KENNY University at Albany, State University University of Connecticut of New York
DAVID A. SCHROEDER University of Arkansas
KEES VAN DEN BOS University of Utrecht, Utrecht, the Netherlands
CONSTANTINE SEDIKIDES University of Southampton, Southampton, England
PAUL A. M. VAN LANGE Free University, Amsterdam, Amsterdam, the Netherlands
PHILLIP R. SHAVER University of California, Davis
LAURIE R. WEINGART Carnegie Mellon University
J. NICOLE SHELTON Princeton University
GWEN M. WITTENBAUM Michigan State University
DOUGLAS T. KENRICK Arizona State University
SANDRA L. MURRAY State University of New York at Buffalo
MARGARET SHIH University of Michigan
NORBERT L. KERR Michigan State University
STACEY SINCLAIR LISA A. NEFF University of Virginia University of Toledo ASSISTANT TO THE EDITOR—CHRISTINE KELLY
WENDY L. WOOD Texas A&M University MICHAEL ZA´RATE University of Texas at El Paso
PERSONALITY PROCESSES AND INDIVIDUAL DIFFERENCES CHARLES S. CARVER, Editor University of Miami ASSOCIATE EDITORS TIM KASSER Knox College
GEORGE A. BONANNO Teachers College, Columbia University
AVSHALOM CASPI MARIO MIKULINCER Bar-Ilan University, Ramat-Gan, Israel King’s College, London EDWARD C. CHANG EVA M. POMERANTZ University of Michigan University of Illinois at Urbana– Champaign RICHARD W. ROBINS University of California, Davis GERARD SAUCIER University of Oregon THOMAS A. WIDIGER University of Kentucky
SERENA CHEN University of California, Berkeley A. TIMOTHY CHURCH Washington State University JAMES COAN University of Wisconsin—Madison M. LYNNE COOPER University of Missouri—Columbia
EDDIE HARMON-JONES Texas A&M University
DANIEL W. RUSSELL Iowa State University
TODD HEATHERTON Dartmouth College
OLIVER C. SCHULTHEISS University of Michigan
JUTTA HECKHAUSEN University of California, Irvine
SUZANNE C. SEGERSTROM University of Kentucky
STEVEN J. HEINE University of British Columbia, Vancouver, British Columbia, Canada
KENNON M. SHELDON University of Missouri—Columbia
RICHARD KOESTNER McGill University Montreal, Quebec, Canada
C. R. SNYDER University of Kansas SANJAY SRIVASTAVA University of Oregon
DAVID LUBINSKI Vanderbilt University
TIMOTHY STRAUMAN Duke University
MICHAEL EID University of Geneva, Geneva, Switzerland
RICHARD E. LUCAS Michigan State University
MICHAEL J. STRUBE Washington University
ROBERT R. MCCRAE National Institute on Aging, Baltimore
JERRY SULS University of Iowa
ANDREW J. ELLIOT University of Rochester
WENDY BERRY MENDES Harvard University
WILLIAM B. SWANN JR. University of Texas at Austin
LISA FELDMAN BARRETT Boston College
RODOLFO MENDOZA-DENTON University of California, Berkeley
HOWARD TENNEN University of Connecticut Health Center
WILLIAM FLEESON Wake Forest University
DANIEL K. MROCZEK Fordham University
MICHAEL C. ASHTON Brock University, St. Catherines, Ontario, Canada
SUZANNE THOMPSON Pomona College
R. CHRIS FRALEY University of Illinois at Chicago
STEPHEN A. PETRILL Pennsylvania State University
OZLEM AYDUK University of California, Berkeley
ANTONIO L. FREITAS State University of New York at Stony Brook
RALPH L. PIEDMONT Loyola College in Maryland
ROBERT J. VALLERAND Universite´ du Que´bec a` Montre´al Montreal, Quebec, Canada
CONSULTING EDITORS STEPHAN A. AHADI American Institutes for Research, Washington, DC JAMIE ARNDT University of Missouri—Columbia JENS B. ASENDORPF Humboldt-Universita¨t Berlin Berlin, Germany
E. ASHBY PLANT Florida State University
ROY F. BAUMEISTER Florida State University VERO´NICA BENET-MARTI´NEZ University of California, Riverside
DAVID C. FUNDER University of California, Riverside STEVEN W. GANGESTAD University of New Mexico
BRENT ROBERTS University of Illinois at Urbana–Champaign
APRIL L. BLESKE-RECHEK University of Wisconsin—Eau Claire
CAROL L. GOHM University of Mississippi
MICHAEL D. ROBINSON North Dakota State University
ASSISTANT TO THE EDITOR—BARBARA ADEWUSI
KATHLEEN D. VOHS University of Minnesota DAVID WATSON University of Iowa BARBARA WOIKE Columbia University REX A. WRIGHT University of Alabama at Birmingham
ATTITUDES AND SOCIAL COGNITION
Automatic and Controlled Components of Judgment and Decision Making Mario B. Ferreira and Leonel Garcia-Marques
Steven J. Sherman
University of Lisbon
Indiana University
Jeffrey W. Sherman University of California, Davis The categorization of inductive reasoning into largely automatic processes (heuristic reasoning) and controlled analytical processes (rule-based reasoning) put forward by dual-process approaches of judgment under uncertainty (e.g., K. E. Stanovich & R. F. West, 2000) has been primarily a matter of assumption with a scarcity of direct empirical findings supporting it. The present authors use the process dissociation procedure (L. L. Jacoby, 1991) to provide convergent evidence validating a dual-process perspective to judgment under uncertainty based on the independent contributions of heuristic and rule-based reasoning. Process dissociations based on experimental manipulation of variables were derived from the most relevant theoretical properties typically used to contrast the two forms of reasoning. These include processing goals (Experiment 1), cognitive resources (Experiment 2), priming (Experiment 3), and formal training (Experiment 4); the results consistently support the author’s perspective. They conclude that judgment under uncertainty is neither an automatic nor a controlled process but that it reflects both processes, with each making independent contributions. Keywords: dual-process approach, heuristic reasoning, rule-based reasoning, process dissociation
For example, it is quite trivial to calculate that the likelihood of picking the one red ball in an urn out of 10 balls is 10% and that the likelihood of picking a red ball from another urn when there are 8 red balls out of 100 is only 8%. Yet, even knowing this, when we are asked from which urn we would prefer to sample and try to get a red ball and win $100, many of us have a compelling desire to choose the urn with 100 balls (and in fact, do make such a choice if asked to use their gut feelings), despite the fact that we “know” this is an irrational choice (Denes-Raj & Epstein, 1994). Similarly, even though we know rationally that the two lines in the Mu¨llerLyer illusion are the same length, we cannot escape the feeling, and the perception, that they are different. Some judgments seem to come to us (and stay with us) independently of any logical considerations. From our perspective, the greatest contribution of more than 30 years of research concerning the use of heuristics and biases is not so much the realization that intuitive judgments are often governed by heuristics that do not follow probability rules but it is the revelation of a gap, within our own heads, between “natural assessments” such as availability or representativeness and the deliberate application of a justifiable set of inductive rules. In recent years, dual-process approaches of judgment under uncertainty (e.g., Chaiken & Trope, 1999; Kahneman & Frederick, 2002; Kirkpatrick & Epstein, 1992; Sloman, 1996; Sloman & Rips, 1998; Stanovich & West, 2000) have categorized the cognitive processes underlying inductive reasoning into two basic forms of reasoning: largely automatic associative processes (here referred to as heuristic reasoning [H]) and controlled analytical processes
Think for a moment about all of the relevant factors involved in daily judgments such as the likelihood of a current relationship leading to marriage or a sports team winning a game. This mental experience is usually enough to make us aware of the simple fact that the world is too complex to predict accurately. Perhaps we can rely on resource-consuming decision rules based on formal theories of probability, but even those may be unsatisfactory and are not always consensual. Of course, they do often work and lead to accurate judgments. Alternatively, instead of the deliberate use of algorithms, sometimes a judgment or prediction seems to come to us, rather spontaneously and quickly, and a feeling of relative certainty (or uncertainty) will “pop-out.” Even when basing judgments on such a simple process, we sometimes make probability judgments that are relatively well calibrated. In addition, the two kinds of judgment processes often occur together. When they suggest the same answer, there is no problem or conflict. However, a good deal of tension and anxiety may come about when deliberate rule-based reasoning and intuitive heuristics produce contradictory outputs within our own heads.
Mario B. Ferreira and Leonel Garcia-Marques, Department of Psychology and Education, University of Lisbon; Steven J. Sherman, Department of Psychology, Indiana University; Jeffrey W. Sherman, Department of Psychology, University of California, Davis. Correspondence concerning this article should be addressed to Mario B. Ferreira, University of Lisbon, Department of Psychology and Education, Alameda de Universidade, 1649-013 Lisbon, Portugal. E-mail:
[email protected]
Journal of Personality and Social Psychology, 2006, Vol. 91, No. 5, 797– 813 Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.5.797
797
798
FERREIRA, GARCIA-MARQUES, SHERMAN, AND SHERMAN
(rule-based reasoning [RB]).1 Although this characterization has led to several property lists contrasting the two reasoning modes (Sloman, 1996; Smith & DeCoster, 1999; Stanovich & West, 2000), such theoretical descriptions have been primarily a matter of assumption with a scarcity of direct empirical findings supporting it. In this article, we report theoretically based process dissociations between the two forms of reasoning, obtained by experimentally manipulating variables derived from the most relevant theoretical properties typically used to contrast the two forms of reasoning. As Kahneman (2003) has noted: “There is considerable agreement on the characteristics that distinguish the two types of cognitive processes” (p. 698). Answering the what, how, and when of H and RB seems a sensible starting point to describe the general view that has motivated most dual-model research in reasoning and decision making. In other words, what are these two forms of reasoning? How do they work? When do they become active? The what: H refers to inferences based on simplifying principles such as similarity and contiguity, whereas RB refers to symbolically represented inferential rules structured by logic. The how: H operates intuitively in the sense that once triggered it gives rise to an autonomous process without further control until an end response pops out into consciousness. RB’s operation involves the deliberate application of rules that are put to work strategically according to the person’s goals. The when: H’s activation depends only on appropriate triggering cues (e.g., similarity matching involved in the representativeness heuristic), whereas RB’s activation depends on recognizing the applicability of an abstract rule (based on the verification of formal conditions) as well as on the availability of cognitive resources and motivation.
The Dual Nature of Judgment Under Uncertainty A successful account of judgment under uncertainty must be capable of retaining the explanatory power of the past research on simplified heuristics and biases (for reviews see Kahneman, Slovic, & Tversky, 1982; Sherman & Corty, 1984), but it must also be able to delineate the conditions underlying inductive judgment based on deliberate RB (e.g., Fong, Krantz, & Nisbett, 1986). Thus, we argue that human inductive reasoning has a dual nature: one aspect operates by heuristic principles such as similarity and contiguity, and the other operates by the use of deliberate analytic rules (Sloman, 1996; Smith & DeCoster, 1999). Such an approach describes several existing dual-process models of judgment under uncertainty (Epstein, 1994; Griffin, Gonzalez, & Varey, 2001; Kahneman & Frederick, 2002; Stanovich & West, 1999, 2000). However, none of these models has attempted to derive, in the area of reasoning under uncertainty, independent estimates of these two processes and to observe independent effects of manipulated variables on the two processes. Such evidence would clearly demonstrate the dual-process nature of judgments under uncertainty. The successful modeling of dual-process approaches typically involves two steps. First, one must establish a one-to-one relation between processing modes and participants’ responses to inferential tasks. That is, the adoption of the H process must be associated with a particular response, and the adoption of the RB process must be associated with a particular response. Second, one must demonstrate and understand how empirical variables selectively affect the two processes. Research on judgment under uncertainty has traditionally used errors and biases in answers to inferential problems to characterize the
underlying heuristic principles and their consequences (Kahneman & Tversky, 1972, 1973; Tversky & Kahneman, 1971, 1973, 1974). However, researchers readily note that, although heuristics play a major role in judgment, reasoning based on the purposeful application of some statistical concepts is also a part of people’s judgmental repertoire (e.g., Ginossar & Trope, 1987; Jepson, Krantz, & Nisbett, 1983; Kruglanski, Friedland, & Farkash, 1984; Nisbett, Krantz, Jepson, & Kunda, 1983). In such research, RB is typically gauged in terms of correct responses (defined by applicable probability or statistical rules) or calibrated responses (defined by ecological considerations or objective criteria) to inferential problems, whereas associative inferential processes (H) are usually estimated by incorrect or badly calibrated responses to the same kinds of inferential tasks. This approach contrasts with our own both conceptually and methodologically. At the conceptual level, the above approach implies a zero-sum or hydraulic relation between the RB and the H processes. As correct responses increase, incorrect responses necessarily decrease. Our dual-process approach conceives of the two processing modes as contributing independently to the judgment. At the methodological level, the above approach assumes that inferential problems or tasks are pure measures of underlying processes (rule-based and associative processes, respectively). However, such a processpure assumption may be troublesome to maintain because tasks differ in a number of ways beyond the extent to which they tap automatic (heuristic) versus controlled (rule-based) processes. In the same vein, different levels of a manipulated variable may differ in ways other than simply mapping onto RB and H. Several lines of research have led to the conclusion that there is often not a sharp dissociation between analytic and heuristic reasoning (Ajzen, 1977; Bar-Hillel, 1979, 1980; Tversky & Kahneman, 1974, 1982). The more general point is that no task is “process pure.” An inferential task that depends entirely on heuristic processes and not at all on rule-based processes is technically unattainable. An inferential task that depends entirely on rule-based processes and not at all on heuristic processes is highly unlikely. Rather, most, if not all, judgments under uncertainty are influenced by simultaneously occurring heuristic and rule-based processes. The process-pure problem is present, to a smaller or greater extent, in research involving other variables that are known to affect respondents’ performance on inferential tasks such as time available for deliberation (Finucane, Alhakami, Slovic, & Johnson, 2000), intelligence (Stanovich & West, 1998, 2002), mood (Bless & Schwarz, 1999), opportunities to apply intuitive representations of statistical rules (Ginossar & Trope, 1987), presentation format (Gigerenzer, 1991; Tversky & Kahneman, 1983), and perceptual salience of randomness (Ferreira & Garcia-Marques, 2003). The process-pure problem is not specific to the study of inferential processes, but it emerges whenever processes are to be measured in terms of particular experimental tasks (Hollender, 1986; Jacoby, 1991). As a consequence, selective influences of 1
We consider heuristic reasoning to be based on natural associative assessments such as similarity matching (representativeness) and memory fluency (availability). We also recognize that some heuristics involve meta-cognitive activity (e.g., the ease-of-retrieval heuristic) that reflect judgments about the validity of activated associations rather than associative processes per se. Both involve automatic rather than reasoned, analytic processes. In our studies, the heuristics do involve associative processes (see our subsequent definitions of H and RB), and we thus use the term associative in describing heuristic reasoning.
AUTOMATIC AND CONTROLLED COMPONENTS OF JUDGMENT
empirical variables cannot be measured directly. Therefore, it is important to use uncontaminated measures of processes through procedures that do not require or assume a one-to-one relation between tasks and processes. We use one such solution by applying the process dissociation framework to judgments under uncertainty.
Process Dissociation Procedure (PDP) and Judgments Under Uncertainty The PDP was originally designed to separate automatic and conscious contributions to memory task performance (Jacoby, 1991). However, its logic may be applied to different experimental contexts as a general methodological tool for separating contributions of automatic and controlled processes. The procedure makes use of a facilitation paradigm or inclusion condition in which automatic and controlled processes act in concert and an interference paradigm or exclusion condition in which the two processes act in opposition. Assuming that both processes contribute to performance and operate independently, estimates of each can be obtained by comparing performance across the two conditions. Consider a fame judgment task in which you are asked to study a list of names (Jacoby, Kelley, Brown, & Jasechko, 1989). At test you are presented with another list of names, half of which have been included in the study list (old), and the other half of which are new. In the inclusion condition you are told to decide whether each name is of a famous or a nonfamous person. Furthermore, you are informed that all the names included in the study list were of famous persons. Thus, if you happen to remember that a name was included in the study list, you should judge it to be of a famous person. Some names may also simply “sound” familiar even if you don’t remember whether they were included in the study list or not. In such cases, old names judged as famous may have been consciously recollected (C) or they may have come to mind automatically (A). The probability of judging an old name as famous is given by: C ⫹ A (1 ⫺ C). In this case, the use of either process would lead to the same result. In the exclusion condition, you are told to decide whether each name is of a famous or a nonfamous person, and you are further informed that all the names included in the study list were of nonfamous persons. Thus, if you happen to remember that a name was included in the list, you should judge it to be of a nonfamous person. In such a case, judging an old name as famous would happen only if conscious recollection failed and as a result of automatic influences of memory: A(1 – C). In this case, the two processes work in opposition to each other. Given these two equations, one can derive estimates of automatic and controlled processes. The difference between performance in the inclusion and exclusion conditions provides an estimate of C (C ⫽ Inclusion–Exclusion); similarly the estimate of the automatic component can be obtained in the following way: A ⫽ Exclusion/(1 – C). Now, suppose you are asked to respond to the following version of the lawyer– engineer problem (Kahneman & Tverky, 1972): Several psychologists interviewed a group of people. The group included 30 engineers and 70 lawyers, The psychologists prepared a brief summary of their impression of each interviewee. The following description was drawn randomly from the set of descriptions: Dan is 45. He is conservative, careful, and ambitious. He shows no interest in political issues and spends most of his free time on his many hobbies, which include carpentry, sailing, and mathematical puzzles. Which of the following is more likely? (a) Dan is an engineer (b) Dan is a lawyer
799
In this problem, Dan’s description is closer to that of an engineer but not highly diagnostically so. Thus, a judgment by representativeness (Kahneman & Tverky, 1972; Tversky & Kahneman, 1971), based on the similarity between the description and the prototypes of engineer and lawyer, is in opposition to a response based on the application of a sampling rule (taking into consideration the prior probabilities of being an engineer or a lawyer). As such, choosing the response option Dan is an engineer is assumed to happen only if conscious application of a relevant inferential rule (C) fails and as a result of the automatic influences of heuristic processing: A(1 – C). The lawyer– engineer problem, as well as other inferential problems possessing the same basic structure (opposing heuristic and rule-based judgments), may be considered good instantiations of an interference paradigm or exclusion condition. However, one can also develop an inclusion condition for the same problem. An easy way to obtain such an inclusion version of the above lawyer– engineer problem is to simply invert the base rates. That is, you now consider a group of interviewees composed of 70 engineers and 30 lawyers. Therefore, the response option Dan is an engineer may be chosen as a consequence of applying a sampling rule by using base rates or simply because it was automatically computed as more similar to Dan’s description. The proportion of responses Dan is an engineer is given by C ⫹ A(1 ⫺ C). In most judgment research, the problems that are used are almost always exclusion problems. That is, the rule-based and the automatic processes lead to different responses. However, as in the lawyer– engineer problem above, it is possible to develop inclusion versions very similar to the original versions (exclusion versions) except that rule-based and heuristic judgments coincide in leading to the same response output. Once one develops comparable inclusion and exclusion conditions, it is possible to apply the logic of the PDP to obtain separate estimates of RB and H. As indicated, we begin with a dual-process approach to judgment under uncertainty that postulates the existence of two different processing modes, RB (involving explicit and controlled rule application) and H (based on automatic processing). We assume that RB and H processes operate in parallel and that they contribute to judgment independently of each other.
The Present Experiments The PDP provides a way to investigate theory-driven process dissociations underlying the current approach to reasoning under uncertainty. We report four experiments exploring how different independent variables influence the estimates of RB and H. Each manipulation is historically relevant to the distinction between automatic and controlled processes. Our main goal is to determine whether derived estimates of RB and H will show expected trends based on our assumptions. Literature involving judgments under uncertainty has traditionally assumed that performance based on H (but not based on RB) is unaffected by participants’ intentions or goals (Kahneman & Frederick, 2002; Sherman & Corty, 1984). Although some research has suggested that goals such as incentives to be accurate do not reduce heuristically driven biases (Camerer, 1987; Tversky & Kahneman, 1974), there is no direct evidence supporting this notion. Experiment 1 sought such evidence by manipulating participants’ goals through instructions to answer the inferential problems in an intuitive or in a rational way. RB is believed to be under
800
FERREIRA, GARCIA-MARQUES, SHERMAN, AND SHERMAN
participants’ control, whereas H is assumed to be largely automatic. Accordingly, varying participants’ goals should affect RB but leave H unchanged. One of the main reasons usually put forward for the ubiquitous use of H in everyday inductive judgments is its efficiency. In contexts of high cognitive load, the use of heuristics produces fast and effortless responses that often conform to the outcome of deliberate and effortful statistical judgments (RB). If the characterization of H and RB processes in terms of their demand of cognitive resources is accurate, then manipulations of cognitive load to induce simple or resource-demanding processing (e.g., Chen & Chaiken, 1999) should dissociate the two processing modes. Experiment 2 places some participants under a cognitive load, and this depletion of attentional resources is expected to decrease RB but to leave H estimates largely invariant. Processing a particular stimulus in a particular way facilitates the subsequent repetition of the same processing with new stimuli (Smith, 1994). This facilitation is generally independent of any explicit memory of the previously presented stimuli. Accordingly, priming the use of heuristics is expected to dissociate the two reasoning modes by increasing H but leaving RB invariant. Experiment 3 primed participants with inferential problems designed to facilitate H. In addition, the priming problems in Experiment 3 and the subsequent target problems shared the same superficial structure and were presented sequentially. Because H reasoning is often based on associative assessments of similarity, the use of highly similar problems for priming and target stimuli should also facilitate H. On the other hand, RB processes were expected to be invariant because they correspond to a reasoning mode governed by explicit application of rules, largely insensitive to the automatic processing principles underlying H. In contrast, priming of rule-based processing should affect the subsequent use of RB but leave H unaffected. In fact, the mere induction of formal thought has been shown to generally improve subsequent measures of abstract reasoning (e.g., LaRue & Olejnik, 1980). As a case of abstract reasoning, RB should be responsive to formal training. In contrast, given its heuristic nature and the considerable degree of independence between the two processing modes, H is expected to be insensitive to any rule-governed formal activity prior to an inferential task. Experiment 4 primed participants with formal rule-based problems before they responded to subsequent target problems. The induction of formal thought was expected to increase RB processes but have no effect on H. Experiments 1, 2, 3, and 4 were thus designed to validate the PDP as a suitable method to study heuristic and rule-based processes underlying judgments under uncertainty and to provide evidence for the simultaneous operation of both RB and H. More important, they directly test a number of predictions logically derived from the literature on judgments under uncertainty (Epstein, 1994; Kahneman & Frederick, 2002; Sloman, 1996; Stanovich & West, 1998, 2002) that, to our knowledge, have received no direct empirical support to date.
Experiment 1 In Experiment 1, we obtained estimates of RB and H for participants’ performance under two different processing goals. Specifically, goals were manipulated through instructions to answer a set of inclusion and exclusion problems in an intuitive way or to answer it in a more reflective way. Given RB’s controlled and intentional nature and H’s autonomous and more spontaneous nature, RB is expected to be greater when participants are
given reflective instructions compared with when they are given intuitive instructions. Because H processes are automatic and unaffected by goals, H is expected to be largely unchanged across the instructions sets.
Method Participants. The participants were 40 students (29 women and 11 men) at the University of Lisbon who participated to fulfill a requirement of their introductory educational psychology course. Procedure and material. For Experiment 1, and for the rest of the experiments here reported, participants were given a brief oral introduction to the experiment on arrival at the laboratory, and they were then escorted to a room equipped with PC-compatible computers. Experimental sessions comprised between 1 and 6 participants. Written instructions followed by a list of problems were presented, and responses were collected on the computers. Each problem was followed by two response options. Participants had to choose one option before they could go on to the following problem. They responded to the problems at their own pace and waited in their places until everyone in their session had finished. In all four experiments, two lists of problems (List 1 and List 2) were created such that inclusion problems in List 1 became exclusion problems in List 2 and vice versa. In order to guarantee that participants never saw the inclusion and exclusion version of the same problem, List 1 and List 2 were manipulated between participants. In each of these lists, problems were sorted differently to control for order effects, leading to Lists 1A and 1B and Lists 2A and 2B, respectively. Order of presentation of the problems was random with the restriction that not more than two problems of the same version type or involving the same statistical principle could be presented in a row. Two experimental conditions, corresponding to two instruction sets, were used in Experiment 1. In one condition, referred to as the intuitive condition, the experiment was introduced as a study of human intuition. The study’s goal was to evaluate personal intuition and sensibility when one has to make choices on the basis of incomplete information. Participants were encouraged to base their answers to the problems on their intuition and personal sensitivity. In the other condition, referred to as the rational condition, the experiment was introduced as a study on human rationality. The study’s goal was to evaluate scientific reasoning ability when one has to make choices on the basis of incomplete information. Participants were encouraged to behave like scientists and to base their answers on rational and reflective thinking. Half of the participants were randomly assigned to the intuitive condition and the other half to the rational condition. Problems used in Experiment 1 include base-rate problems, conjunction problems, and ratio-bias effect problems. Base-rate problems are equivalent to the classical lawyer– engineer problem (Kahneman & Tversky, 1972). Specifically, participants have to choose between two opposing response options, one that is favored by the base rates (reflecting rule-based processing) and the other that is favored by the description of the target (reflecting heuristic processing). The base rates used were more extreme than in the original Kahneman and Tversky (1972) problems and were expressed in absolute numbers instead of percentages (e.g., 85 lawyers and 15 engineers out of 100 persons). Furthermore, individuating information was less diagnostic of a given category (e.g., engineer) than in the original problems, and the stories always made explicit reference to some kind of random process by which the target individual (the specific person described in the problem) was chosen.2 These changes gave rise to equivalent but “easier” base-rate problems, allowing for a larger proportion of statistical answers when compared with the original problems. Problems involving the conjunction rule appeared in a format not used in previous research. Participants were presented with two alternative solutions. The single-case solution was associated with a certain probabil-
2 By mistake, the reference to the random process was omitted in one of the base-rate problems.
AUTOMATIC AND CONTROLLED COMPONENTS OF JUDGMENT ity of success, whereas the compound-case solution involved two different stages with independent probabilities of success. Each one of these independent probabilities was higher than the probability of the single solution but the conjunction of the two was lower. For instance, one single agent can accomplish a certain activity within a specified time period with a probability of 60% (single case). Alternatively, two independent agents can divide that activity in two parts and finish them within a specified time period with probabilities of 70% and 80%, respectively (compound case). Note that the mean probability of success of the two agents is 75%, but the probability of both agents finishing their parts in time is only 56% (lower than the 60% probability of success of the single agent). If our participants consider only how large each independent probability is and neglect the consequences of set intersection (conjunction) for the compound case, this leads to a statistically incorrect answer. On the contrary, the consideration of the relative magnitude of the intersection between the two sets considered in the compound case leads to the statistically correct answer.3 The ratio-bias effect refers to the preference for equally small or even smaller probabilities for success when they are based on a larger sample size (Kirkpatrick & Epstein, 1992; Miller, Turnbull, & McFarland, 1989). For instance, there is a tendency to intuitively prefer a probability of success of 10 out of 100 when compared with 1 out of 10. The higher absolute number of favorable cases in the first ratio renders it more attractive. Kirkpatrick and Epstein (1992) reported that 9 out of 100 is frequently preferred to 1 out of 10 probability of success, showing that this bias even extends to cases in which the ratio of the larger sample actually represents a lower probability of success than the ratio of the smaller sample. In the ratio-bias effect problems used here, participants had to choose between two probabilities of success presented in the form of large and small samples. For the large samples, the absolute number of favorable cases is obviously larger than in the smaller samples. However, in the exclusion cases the smaller samples correspond to a higher probability of success. In past work applying the PDP, the response based on RB in the exclusion version is the correct response. The response based on H in the exclusion version is the incorrect response. The designation of correct and incorrect for RB and H responses makes sense for previous studies that investigated memory (Jacoby, 1991), fame judgments (Jacoby et al., 1989), Stroop task responses (Lindsay & Jacoby, 1994), or stereotypes (Payne, 2001). In all of these cases, responses based on RB are in fact the correct ones. In the present studies, the designation of correct and incorrect for the exclusion problems is not so clear. For some of the problems (the ratio-bias effect problems), the RB response is the correct response. Two of 10 envelopes (the response based on the RB process) gives one a better chance of winning than 19 of 100 (the response based on the H process). For other problems, the correct response is indeterminate. For example, in the baserate problems, whether the base-rate derived response (the RB response) is correct depends on the real or perceived diagnosticity of the information that describes the social target. If that information is highly diagnostic, the H response may in fact be more likely to be correct. However, in the decision-making arena, there is a clear difference between statistically based (RB) problem solving and judgmental heuristics (H). To be precise, statistical response alternatives (to exclusion problems) reflect extensional reasoning and nonstatistical response alternatives (to exclusion problems) reflect nonextensional reasoning. Extensional reasoning involves taking into consideration set inclusion and/or intersection (e.g., the consideration of base rates, proportionality, conjunction). Nonextensional reasoning corresponds to the neglect of these problem features (cf. Tversky & Kahneman, 1983). Note that taking extensional features into account does not necessarily guarantee a normatively suitable answer. For ease of presentation (and in the absence of more clear nomenclature), we refer to responses reflecting RB and H in the exclusion problems as statistical and nonstatistical responses, respectively. The PDP is applicable whether or not there is a clear correct response, so long as RB and H lead to the same response in the inclusion version and to different re-
801
sponses in the exclusion version. We return to this issue in the presentation of Experiment 3. All problems in Experiment 1 (and in the other experiments) had an inclusion and an exclusion version. The exclusion versions or exclusion problems (described in the preceding paragraph) correspond to the format traditionally used in research in judgments under uncertainty. It is customary in these problems that the statistical and nonstatistical answers correspond to different responses or alternative response options. The inclusion versions (or inclusion problems) were the equivalent of exclusion versions except that the statistical information was inverted so that both RB and H produced the same response option, the dominant response. In base-rate problems, base rates and individuating information point to the same answer. In conjunction problems, the response option based on the conjunction of two items is not only less probable but also less representative than the single-response option. In the ratio-bias effect problems, the larger sample is also a higher probability than the smaller one.4 Nondominant responses to inclusion problems do not correspond to either RB or H reasoning. Instead, they are likely to be based on some other kind of idiosyncratic associations (e.g., I have a cousin named Dan, and he is a lawyer). The dominant answers in the inclusion version of a given problem correspond to the added contribution of H and RB processes, whereas the statistical answers to the exclusion version of the same problem reflect only the RB contribution. Therefore, the proportion of dominant answers in the inclusion version should be greater than the proportion of statistical answers in the exclusion version. Problems were pretested and selected to meet this criterion. Data analysis of Experiment 1 considered participants’ responses to 10 problems (5 base-rates problems, 2 conjunction problems, and 3 ratio-bias effect problems). Dependent measures. To arrive at the H and RB estimates used as dependent measures, the proportions of nonstatistical answers to exclusion problems and dominant answers to inclusion problems were obtained for each participant across problems and were then used to compute individual RB and H estimates from PDP Equations 1 and 2 (Jacoby, 1991): RB ⫽ P共dominant answers inclusion problems 兲 ⫺ P共nonstatistical answers exclusion problems 兲, H ⫽ P共nonstatistical answers exclusion problems 兲/共1 ⫺ RB兲.
(1) (2)
3 We also used traditional conjunction problems similar to the “Linda problem” (Tversky & Kahneman, 1983). However, for these problems we did not accurately follow the logic of opposition underlying the PDP. An important assumption of the PDP approach is that levels of controlled and automatic processes do not change across inclusion and exclusion conditions (Jacoby, 1991; Jacoby, Toth, & Yonelinas, 1993). The logic is that the H process always works in the same direction with the same strength (although leading to correct answers on inclusion trials and incorrect answers on exclusion trials). The RB process should work in opposite directions for the inclusion and exclusion cases but should have equal strength in the two cases. All our problems except the Linda-type problems follow this logic. For those problems, it was RB that worked in the same direction for the inclusion and exclusion cases, and H that led to different answers in the two cases. In these cases, we cannot guarantee that H operates with the same strength across inclusion and exclusion problems only in different directions. Thus, the PDP assumption that H (and RB) contributions are equally strong in inclusion and exclusion trials is likely to be violated. In light of this, we discarded responses to conjunction problems of this type in the data analyses. Such conjunction problems were eliminated from consideration in all four experiments. We thank Larry Jacoby for calling our attention to this point. 4 See Appendix A for examples of target problems used in Experiments 1, 2, and 4.
FERREIRA, GARCIA-MARQUES, SHERMAN, AND SHERMAN
802
Estimation of the experimental parameters H and RB is dependent on a minimum level of errors in exclusion tasks. Perfectly statistical performance (i.e., no nonstatistical answers to exclusion problems) mathematically constrains individual estimates of H to be zero (H ⫽ 0/[1 – RB] ⫽ 0). As a precaution, participants with zero nonstatistical answers to exclusion problems were discarded for purposes of analyses (see Jacoby, Toth, & Yonelinas, 1993). Dependent measures for Experiments 2, 3, and 4 were obtained in the same manner. This issue will be addressed in the General Discussion. Design. The design is a 2 ⫻ 2 ⫻ 2 ⫻ 2 factorial with instruction type (intuitive and rational conditions), problem version (List 1 and List 2), and problem order (List A and List B) as the between-subjects variables and type of problem (inclusion and exclusion problems) as the within-subjects variable.
Results Several separate one-way ANOVAs showed neither version effects nor order effects on the RB and H estimates. Therefore these results are not discussed further. H and RB estimates across intuitive and rational conditions are shown in Table 1. The increase in the proportion of dominant answers (inclusion problems) and the decrease in nonstatistical answers (exclusion problems) from the intuitive to the rational condition indicate that the instructions to consider the problems as a scientist have enhanced participants’ performance. An ANOVA was performed with instruction type as a between-subjects variable and the RB and H estimates as repeated measures. The analysis revealed a reasoning mode main effect, indicating that H is greater than RB, F(1, 36) ⫽ 127,89, MSE ⫽ 0.04, p ⫽ .00, and an instruction Type ⫻ Reasoning mode interaction, F(1, 36) ⫽ 3.75, MSE ⫽ 0.04, p ⫽ .06, reflecting the differential impact of instruction type on H and RB. Changing from “rational” instructions to “intuitive” instructions produced a strong reduction of RB, t(36) ⫽ 2.02, SD ⫽ 0.12, p ⫽ .02 (one-tailed planned comparisons), while leaving H constant, t(37) ⬍ 1, SD ⫽ 0.05 (two-tailed planned comparisons).5
Discussion Experiment 1 examined the impact of processing goals on the contributions of RB and H to inductive judgment tasks. Participants were instructed to answer a set of inclusion and exclusion problems in an intuitive way or to answer them in a more reflective way. Given RB’s controlled and intentional nature and H’s spontaneous nature, we predicted that RB would be greater for rational instructions when compared with intuitive instructions, whereas because H is an automatic process unaffected by goals, we preTable 1 Mean Proportion of Dominant Answers for Inclusion Problems, Nonstatistical Answers for Exclusion Problems, and Estimates of H and RB Across Instruction Conditions Problem version
Estimate
Condition
Inclusion
Exclusion
H
RB
Intuitive Rational
.69 .80
.59 .47
.70 .76
.10 .33
Note. For both conditions, n ⫽ 19. H ⫽ heuristic reasoning; RB ⫽ rule-based reasoning.
dicted that it would be largely unchanged across instructions sets. Results corroborated our hypotheses. The invariance of H across instructions is in line with previous research on heuristics as natural assessments, showing heuristic-based reasoning to be insensitive to incentives to respond more thoroughly such as the use of pay-off matrices (e.g., Tversky & Kahneman, 1974).
Experiment 2 In Experiment 1, we explored the impact of intentional goals on RB and H by obtaining process-dissociation estimates of the two reasoning modes. In Experiment 2, we examined the effect of cognitive load on both processes as another way to dissociate H from RB. Automatic processes such as H should be unaffected by the depletion of attentional resources under cognitive load. In contrast, given its resource demanding nature, RB should compete for cognitive resources with a digit rehearsal task, resulting in a decrease of RB estimates.
Method The inductive judgment task of Experiment 2 was memory-based. That is, the question for each problem did not directly follow the problem’s text. Instead it was delayed in time and appeared later when the problem’s text was no longer available. Thus, participants initially read a problem, but no question was asked. Only afterwards, when the problem’s text was no longer available, did participants respond to the question. Therefore, participants’ responses were based on their memory of the problems. A second memory task competed with the judgment task for cognitive resources. Participants were asked to memorize a number and keep rehearsing it while choosing one of the two response options. The cognitive load conditions differed in the length of the number to be rehearsed (one-digit or sevendigit numbers). Participants. The participants were 112 students at Indiana University who participated to partially fulfill a requirement of their introductory psychology course. They were randomly assigned to one of the experimental conditions. Procedure and material. After the initial instructions, participants began by reading a problem displayed on the computer screen. When finished, they pressed a “continue” key and proceeded to a second screen, which presented a number (consisting of either one or seven digits) with instructions to memorize it. Numbers were displayed for a 5-s period. The following screen displayed the problem’s corresponding question with two alternative response options. Participants were asked to choose one of the alternative response options while still rehearsing the number. A final screen prompted participants to enter the memorized number. This sequence was repeated for each problem. Recollection errors were used as a manipulation check. Participants were also presented with 10 filler problems along with the experimental problems. These fillers were used in order to guarantee that participants could not anticipate the exact question corresponding to one of the types of experimental problems.
5
In the experiments here reported, it is hypothesized that the manipulations affect one of the reasoning modes in a given direction, leaving the other invariant. To test for these hypotheses, we used planned comparisons that are one-tailed tests for the changes of the reasoning mode estimates in the predicted direction and two-tailed tests for the invariance of the other reasoning mode. In other words, the hypotheses receive empirical support if Ho is rejected in the first case and if Ho is accepted in the second case. To decrease the probability of committing a Type II error when accepting Ho, the value of ␣ (probability of making a Type I error) is set to .10. Thus, when predicting change (one-tailed tests), Ho will be rejected for ␣ ⬍ . 05; when predicting invariance (two-tailed tests), Ho will be rejected for ␣ ⬍ .10.
AUTOMATIC AND CONTROLLED COMPONENTS OF JUDGMENT Besides base-rate problems, conjunction problems, and ratio-bias effect problems, Experiment 2 included a new type of problem based on the law of large numbers (LLN), inspired by problems proposed by Nisbett et al. (1983; Fong et al., 1986). In the exclusion version of LLN problems, participants were asked to choose between two alternative response options, one of which was favored on the basis of a large sample (indicating statistical reasoning), and the other of which was favored by evidence from a much smaller sample (the choice of which would indicate nonstatistical processing based on representativeness). In the inclusion versions of these problems, both H and RB processes favored the same option. Filler problems were identical to experimental problems except for the question and alternative response options. For instance, following a ratiobias effect problem in which Mark has to draw a ticket from one of two bowls containing 1 out of 9 and 9 out of 91 “winner” tickets, participants were asked: How many “winner” tickets are contained in the first bowl? (a) 1 ticket (b) 5 tickets Data analysis of Experiment 2 considered participants’ responses to eight problems (four base-rates problems, one conjunction problem, two law of large numbers problems, and one ratio-bias effect problem). Design. The design is a 2 ⫻ 2 ⫻ 2 ⫻ 2 factorial with cognitive load (high and low), problem versions (List 1 and List 2), and problem order (List A and List B) as the between-subjects variables and type of problem (inclusion and exclusion problems) as the within-subjects variable.
Results Several separate one-way ANOVAs showed no order effects on the RB and the H estimates. Therefore this factor is not discussed further. However, significant version effects were obtained for both RB estimates, F(1, 102) ⫽ 21.77, MSE ⫽ 0.08, p ⫽ .00, and H estimates, F(1, 102) ⫽ 4.24, MSE ⫽ 0.05, p ⫽ .05. Nevertheless, the following analysis reports results collapsed across this variable because data analysis performed by version showed the same pattern of results and a difference in degree only. Inclusion and exclusion results as well as H and RB estimates are presented in Table 2. The proportion of dominant answers (inclusion problems) was quite stable across cognitive load conditions. The decrease of nonstatistical answers from the high-load to the low-load condition for exclusion problems indicates that the secondary memory task interfered with participants’ performance. An ANOVA, with the two cognitive load conditions as a between-subjects variable, was performed on the RB and H estimates as repeated measures. The analysis revealed a reasoning mode main effect, indicating that H is greater than RB, F(1, Table 2 Observed Mean Proportions of Dominant Answers for Inclusion Problems, Nonstatistical Answers for Exclusion Problems, and Estimates of H and RB Across Cognitive Load Problem version
Estimate
Inclusion
Exclusion
H
RB
High loada Low loadb
.79 .80
.57 .46
.78 .77
.23 .35
Note. H ⫽ heuristic reasoning; RB ⫽ rule-based reasoning. n ⫽ 51. b n ⫽ 52.
101) ⫽ 368,81, MSE ⫽ 0.03, p ⫽ .00, and an interaction between cognitive load conditions and reasoning modes, F(1, 101) ⫽ 6.31, MSE ⫽ 0.03, p ⫽ .01. This interaction reflects the differential impact of cognitive load on H and RB estimates. Increasing the difficulty of the competing memory task (changing from one- to seven-digit numbers) produced a reduction in RB, t(101) ⫽ 2.06, SD ⫽ 0.30, p ⫽ .02 (one-tailed planned comparisons), whereas it left H largely unchanged, t(101) ⬍ 1, SD ⫽ 0.23 (two-tailed planned comparisons). Participants who made many errors in the memory task for seven-digit numbers may not have been engaged in the cognitive load memory task. Hence, the statistical analysis was redone, eliminating participants who made more than three errors on the memory task. The pattern of results was basically unchanged, showing once more an interaction between cognitive load conditions and reasoning modes, F(1, 77) ⫽ 4.74, MSE ⫽ 0.03, p ⫽ .04. Planned comparisons also revealed a significant reduction of RB, t(77) ⫽ 1.99, SD ⫽ 0.28, p ⫽ .02 (one-tailed), and no significant change for H, t(77) ⬍ 1, SD ⫽ 0.25 (two-tailed).
Discussion Cognitive load was manipulated through the introduction of a secondary memory task, which was meant to interfere with the primary judgment task. Cognitive load, as expected, had a differential effect on RB and on H by affecting the former but not the latter. The dissociation between H and RB again supports the operation of both processes for these judgments and supports the assumption that H is an efficient and effortlessly activated process, whereas RB is a controlled and resource consuming cognitive activity.
Experiment 3 The process dissociations reported so far resulted from the use of variables that are traditionally considered to affect controlled processes (intentional and resource consuming) such as RB and to have little impact on automatic processes (autonomous and efficient) such as H. Experiment 3 involves a manipulation known to affect H. Dual-process models of judgment under uncertainty assume that H is often based on associative principles of similarity and temporal structure, whereas RB involves the cognitive manipulation of symbolic rules and is not expected to be affected by manipulations involving the priming of associative principles (Epstein, 1994; Epstein, Donovan, & Denes-Raj, 1999; Sloman, 1996). On the basis of this assumption, Experiment 3 used heuristic priming problems that, besides sharing the same statistical principle as the target problem, were very similar to target problems in terms of their superficial structure (subject matter and story outline) within each problem’s type. The combination of the priming of heuristic use plus the similarity in superficial structure of the priming and target problems should increase the activation and the use of the H process. Consequently, estimates of H should increase in the priming condition. Furthermore, given that RB is expected to be largely insensitive to the problems’ superficial structure, it should be unaffected by heuristic priming.
Method
Condition
a
803
Participants. The participants were 95 students (26 men and 69 women) at Indiana University who participated to partially fulfill a requirement of their introductory psychology course. Procedure and material. The material included target problems, heuristic priming problems (all with an exclusion problem format), and neutral problems. Because H is expected to be induced for later problems that have
804
FERREIRA, GARCIA-MARQUES, SHERMAN, AND SHERMAN
similar superficial structures to the initial heuristic priming problems, new target problems with similar superficial structure to the initial heuristic priming problems were developed. There are two main differences between priming problems and target problems. One is that priming problems do not have inclusion versions; they are all exclusion problems. The other is that the target description information of priming problems is so diagnostic that, even in the face of opposing statistical information, the nonstatistical response option is more appropriate than the statistical response option. In this case, the nonstatistical response is actually the more correct response. In order to prime heuristic use, we needed to make the H-based information so strong that it would dominate the judgment. As an example, consider a population that consists of 80 men and 20 women (high base rate of men). One person is randomly chosen. This person likes modern art, is fashion aware, breast-fed the children, and a DNA test shows the presence of XX chromosomes. Is the person a woman or a man? Despite the high base rate of men, the information is even more diagnostic, and H-based judgments yield the better answer. Our target problems, however, maintain the same form as in the other experiments. Thus, we continue to use the terms statistical and nonstatistical for RB-based versus H-based answers, respectively, for target exclusion problems. Neutral problems neither involve inductive reasoning nor share similar superficial structures with priming and target problems. They are small texts followed by a question about mundane aspects of life. For instance, one neutral problem tells the story of Chad, who went to New York, loved it, but realized he would not like to live in such a big city. The subsequent question was as follows: Where would you prefer to live? (a) In a big city like New York. (b) In a small city like Bloomington.6 Participants were randomly assigned to a priming condition or a neutral problem control condition. Problems were organized in four blocks, one for each type of target problem (base-rate problems, conjunction problems, ratio-bias effect problems, and problems based on the law of large numbers). In the priming condition, each block was composed of six priming problems followed by two target problems (one exclusion problem and one inclusion problem) that were very similar to the priming problems in terms of their superficial features. The control condition was equivalent to the priming condition, except that priming problems were replaced by neutral problems. Participants responded to a set of 8 target problems following 24 priming problems (or 24 neutral problems). Data analysis considered participants’ responses to 6 target problems (2 base-rate problems, 2 law of large numbers problems, and 2 ratio-bias effect problems). Design. The design is a 2 ⫻ 2 ⫻ 2 ⫻ 2 factorial with priming manipulation (heuristic priming and control condition), problem versions (List 1 and List 2), and problem order (List A and List B) as the betweensubjects variables and type of problem (inclusion and exclusion problems) as the within-subjects variable.
Results Several separate one-way ANOVAs showed neither version nor order effects on the RB and H estimates. Therefore these factors are not discussed further. Inclusion and exclusion results as well as H and RB estimates are presented in Table 3. In the heuristic priming condition, the proportion of both dominant answers for inclusion problems and nonstatistical answers for exclusion problems increased. An ANOVA was performed, with heuristic priming as a between-subjects variable and RB and H estimates as repeated measures. The analysis revealed a reasoning mode main effect, indicating that H is greater than RB, F(1, 75) ⫽ 163,69, MSE ⫽ 0.05, p ⫽ .00, and a Heuristic Priming ⫻ Reasoning Mode interaction, F(1, 75) ⫽ 3.87, MSE ⫽ 0.05, p ⫽ .05. As predicted, planned comparisons indicated that priming H pro-
Table 3 Observed Mean Proportions of Dominant Answers for Inclusion Problems, Nonstatistical Answers for Exclusion Problems, and Estimates of H and RB Across Conditions Problem version Condition
Estimate
Inclusion
Exclusion
H
RB
.74 .83
.42 .53
.70 .83
.32 .30
a
Control Primingb
Note. H ⫽ heuristic reasoning; RB ⫽ rule-based reasoning. n ⫽ 37. b n ⫽ 40.
a
duced an increase in H, t(75) ⫽ 2.278, SD ⫽ 0.24, p ⫽ .01, (one-tailed), whereas it left RB largely unchanged, t(75) ⬍ 1, SD ⫽ 0.18 (two-tailed). For participants who made wrong answers to priming problems, heuristic processing was likely not primed. To check for this possibility, we redid the above analyses including only participants with three or fewer errors to priming problems. The resulting Heuristic Priming ⫻ Reasoning Mode interaction, F(1, 62) ⫽ 4.12, MSE ⫽ 0.05, p ⫽ .05, reflects the same pattern of results as the analysis conducted with all participants. As predicted, planned comparisons revealed an increase in H from the control to the priming condition, t(62) ⫽ 1.90, SD ⫽ 0.25, p ⫽ .03 (one-tailed), as well as an invariance of RB estimates across control and priming conditions, t(62) ⬍ 1, SD ⫽ 0.38 (two-tailed).
Discussion Experiment 3 was designed to test the impact of heuristic priming on RB and H processes. As predicted, heuristic priming problems with highly similar superficial structures to the target problems facilitated subsequent H processes without affecting RB. Heuristic priming seems to be an effective way to increase H. The individuating information of the target problems used in the present experiments was less diagnostic than in the original problems used by others (e.g., Tversky & Kahneman, 1974). It is likely that, at least for some participants, this individuating information was not diagnostic enough to trigger the automatic associative process that characterizes H. Thus, the priming manipulation of Experiment 3 increased H’s activation level enough so as to augment heuristic-based responses to subsequent target problems that had weak individuating information. Similarly, priming problems may have also promoted output convergence to the expected heuristics-based response option (i.e., nonstatistical responses to exclusion problems and dominant responses to inclusion problems) by reducing the frequency of idiosyncratic answers. On the other hand, the same priming manipulation did not affect RB because this reasoning mode is a deliberate activity governed by cognitive representations of inductive rules and is not based on the automatic processing principles underlying H. These results provide supporting evidence for dual-process accounts for the judgments involved in the problems used here that postulate the existence of two qualitatively different modes of reasoning, one based on the symbolic operation of localized representations of inductive 6
See Appendix B.
AUTOMATIC AND CONTROLLED COMPONENTS OF JUDGMENT
rules (RB), and the other based on the parallel operation of associative principles of similarity and temporal structure (H; e.g., Sloman, 1996).
Experiment 4 Research on formal operations has shown that the mere induction of formal thought (as opposed to concrete thought) generally increases the subsequent use of abstract reasoning (LaRue & Olejnik, 1980). Experiment 4 primed participants with formal problems (derived from Graduate Record Exam [GRE] problems) before they responded to the same kind of target problems used in Experiments 1, 2, and 3. Because RB is usually described as abstract inductive reasoning, it should be responsive to formal pretraining. However, the formal problems were completely different from the target problems in terms of their superficial structure (i.e., the subject matter and the story outline) as well as in terms of their deep structure (i.e., the reasoning rules involved). Hence, a reasoning mode based on principles of similarity and temporal structure, such as H involves, cannot be affected by such pretraining. Thus, it is predicted that formal training will promote subsequent RB, whereas it will leave H largely unchanged.
Method Participants. The participants were 107 students (52 men and 55 women) at Indiana University who participated to partially fulfill a requirement of their introductory psychology course. Procedure and material. The material included target problems, formal problems, and neutral problems. The formal problems were based on actual GRE problems. Finding the correct answer to these problems implies different sorts of formal reasoning. Specifically, the problems involve conditional reasoning (e.g., to find out which of two response options is in agreement with a set of conditional rules), abstract reasoning (e.g., to choose between two interpretations of proverbs, an abstract and a concrete one), and semantic reasoning (e.g., to choose between pairs of words that best express a relationship that is a better analogy to a previously presented pair of words). Neutral problems were equivalent to formal problems except that the reasoning questions were replaced by trivial questions concerning participants’ interests, opinions, or preferences related to mundane aspects of their lives.7 Target problems were equivalent to the experimental problems of Experiments 1 and 2 and thus did not share a similar superficial structure with the formal problems. Participants were randomly assigned to the formal training condition or to the neutral problem control condition. Problems were organized in four blocks, one for each type of target problem [base-rate problems, conjunction problems, problems based on the law of large numbers and ratio-bias effect problems). Within each block, participants in the formal condition responded to 8 formal problems followed by two target problems (one inclusion problem and one exclusion problem). Participants in the neutral condition responded to 8 neutral problems followed by the same two target problems. Thus, the control condition was the same as the training condition, except that priming problems were replaced by neutral problems. Thus, in total, participants responded to a set of 8 target problems plus either 32 initial priming problems or 32 initial neutral problems. Data analysis of Experiment 4 considered participants’ responses to 6 target problems (2 base-rate problems, 2 LLN problems, and 2 ratio-bias effect problems). Design. The design is a 2 ⫻ 2 ⫻ 2 ⫻ 2 factorial with priming manipulation (formal reasoning and control conditions), problem version (List 1 and List 2), and problem order (List A and List B) as the betweensubjects variables and type of problem (inclusion and exclusion problems) as the within-subjects variable.
Results Several separate one-way ANOVAs revealed no version or order effects on the RB and H estimates, except for a significant
805
version effect for the H estimate, F(1, 68) ⫽ 4.29, MSE ⫽ 0.06, p ⫽ .04. The statistical analysis that follows reports results collapsed across this variable because data analysis comparing Version 1 with Version 2 showed the same pattern of results and a difference in degree only. Inclusion and exclusion results as well as H and RB estimates are presented in Table 4. Formal training affected responses to exclusion problems in the expected direction, leading to an increase in statistical answers. However, this training had virtually no effect on responses to inclusion problems. An ANOVA was performed with formal training (vs. control) as a between-subjects variable and the RB and H estimates as repeated measures. The analysis revealed a reasoning mode main effect, indicating that H is greater than RB, F(1, 68) ⫽ 283,48, MSE ⫽ 0.02, p ⫽ .00, and a formal Training ⫻ Reasoning Mode interaction, F(1, 68) ⫽ 4.07, MSE ⫽ 0.02, p ⫽ .05. As predicted, planned comparisons showed that formal training promoted RB, although this promotion did not reach conventional levels of statistical significance, t(68) ⫽ 1.43, SD ⫽ 0.30, p ⫽ .08 (one-tailed), and it left H unaltered, t(68) ⬍ 1, SD ⫽ 0.25 (two-tailed). The above analysis was redone to include participants with four or fewer errors to the formal problems (this corresponds to an average of fewer than one error for each block of formal problems).8 The pattern of results was the same as the analysis conducted with all participants, with a significant Formal Training ⫻ Reasoning Mode interaction, F(1, 58) ⫽ 5.94, MSE ⫽ 0.02, p ⫽ .02. Planned comparisons revealed an increase in RB from the control to the training condition (although it did not quite reach conventional levels of statistical significance), t(58) ⫽ 1.49, SD ⫽ 0.30, p ⫽ .07 (one-tailed), as well as an invariance of H estimates across control and priming conditions, t(58) ⬍ 1, SD ⫽ 0.25 (two-tailed).
Discussion Experiment 4 directly tested a prediction stemming from related literature showing that priming formal thought is sufficient to induce subsequent abstract reasoning (LaRue & Olejnik, 1980). Given that RB is a form of rule-based symbolic reasoning, as proposed by different dual-process models of judgment under uncertainty (e.g., Stanovich & West, 2002), it was predicted and was found to be responsive to formal training that involves abstract reasoning. H, however, does not involve rule application or a complete formal description of concepts. Instead, it is based on automatic responses involving simplifying principles such as similarity or availability. Accordingly, it was insensitive to the priming effect of formal problems. Thus, results support the prediction that the priming of formal thought induces general, abstract rule application regardless of any specific operations attributable to the features of problems.
General Discussion Judgment under uncertainty has recently been approached from the perspective of dual-process models (Griffin, Gonzalez, & 7
See Appendix C. Note that four or fewer errors in Experiment 4 and three or fewer errors in Experiment 3 correspond to the same criterion of 87.5% or more of correct answers for both cases. 8
FERREIRA, GARCIA-MARQUES, SHERMAN, AND SHERMAN
806
Table 4 Mean Proportion of Dominant Answers for Inclusion Problems, Nonstatistical Answers for Exclusion Problems, and Estimates of H and RB Across Formal Training and Control Conditions Problem version Condition a
Control Trainingb
Estimate
Inclusion
Exclusion
H
RB
.83 .84
.50 .40
.81 .81
.33 .44
Note. H ⫽ heuristic reasoning; RB ⫽ rule-based reasoning. n ⫽ 31. b n ⫽ 39.
a
Varey, 2001; Kahneman & Frederick, 2002; Stanovich & West, 2000, 2002). These models converge in postulating that inductive judgment may be based on heuristic (H) and/or on analytical (RB) processing modes. According to these models, H, as a largely automatic, fast, and effortless process, consists of the spontaneous activation of simplifying principles such as similarity and temporal structure (e.g., the representativeness heuristic), induced by situational as well as internal factors. In contrast, RB is a controlled process involving the intentional and effortful activation of a sequence of symbolically represented information (inductive rules). The above characterization of H and RB processing modes as they apply specifically to judgments under uncertainty has been mostly a matter of assumption, with surprisingly little direct empirical support. The intent of the work reported here is to change this state of affairs by providing direct evidence of the dualprocess nature of judgment under uncertainty on the basis of the involvement of both H and RB as independent processes. Specifically, we used the PDP to assess both H and RB and to demonstrate theoretically derived process dissociations. These experiments convincingly show that variables traditionally associated with controlled processes affected RB but not H processes. Conversely, a variable already known to affect automatic processes affected H but left RB unchanged. Results across the four experiments strongly support the proposal that automatic versus controlled processes in judgments are not an either/or proposition but rather that both operate in an independent and parallel way such that an increase in one type of process does not indicate a decrease in the other type. In addition, the results demonstrate that simply assessing statistical or nonstatistical responses cannot reveal the level of rational or heuristic processing. Experiment 1 manipulated participants’ goals through instructions to answer a set of inferential problems in an intuitive way, or to answer these problems in a more reflective way. Instructions affected RB in the predicted direction, whereas they left H unchanged. The results of Experiment 1 thus indicate that RB is sensitive to goals to respond accurately, whereas H is insensitive to such goals or to incentives to respond accurately (e.g., Tversky & Kahneman, 1974). Past research is consistent with our findings. Zukier and Pepitone (1984) found greater attention to base-rates when participants were instructed to think like scientists (as opposed to thinking like clinical psychologists). Ginossar and Trope (1987) found greater use of the sampling rule (in base-rate problems) to the extent that it was instrumental to reach previously defined goals. The present results suggest that these effects (i.e., decreasing judgmental errors [or biases] by role-playing, increasing the instrumental value of
goals or financial incentives) are independent of H and are exclusively due to an increase in RB. In Experiment 2, we obtained H and RB estimates under low versus high cognitive load, testing the effect of a resourcedemanding task on the two processing modes. As expected, an increase in cognitive load reduced RB processes but left H processes invariant, confirming the efficient operation of heuristicbased judgments. Experiment 3 primed participants with inferential problems designed to facilitate H that shared the same superficial structure as the target problems and were presented sequentially prior to the target problems. Such heuristic priming significantly increased H processes, but it did not affect RB because this reasoning mode is a deliberate activity governed by cognitive representations of inductive rules and is not based on the more automatic processing principles underlying H. As a type of abstract reasoning, RB should be responsive to formal training (e.g., LaRue & Olejnik, 1980). However, H is expected to be insensitive to formal training given its automatic nature. Experiment 4 used formal problems as primes for subsequent target problems. These priming problems were completely different from the target problems in terms of their superficial structure as well as their deep structure (i.e., no statistical rules were involved). As expected, H was not affected by these primes. However, rule-governed thinking involved in solving the priming problems led to an increase of RB. Because different types of inferential problems were used (involving base-rates, the conjunction rule, the ratio-bias effect, and the LLN), it is important to provide some indication that the results were not due to responses to a specific type of problem in particular but were the outcome of all types of problems. Unfortunately, for most cases there were not enough problems of each type to compute reliable process estimates. Instead, data from each experiment were reanalyzed by problem (aggregating across participants). H and RB estimates were then computed for all combinations of two types of problems at a time. Mean results for H and RB obtained in this manner show exactly the same predicted result pattern for all problem types within each of the four experiments.
Findings of Invariance and the Assumption of Independence As already noted, demonstrating situations in which H and RB contribute independently to judgments under uncertainty is a prerequisite for avoiding a process-pure assumption and does not reflect any theoretical claim about the possible modal interaction between these two processes. The independent dual-process model assumed by the PDP appears to be justified within the present paradigm given that the correlation between H and RB estimates across all studies was near zero (r ⫽ ⫺.08, ns),9 which strongly suggests functional independence.
Automatic and Controlled Influences in Social Judgments Much social cognition research has been concerned with the interaction between cognitive control and automatic bias in social 9
Before computing this mean, correlations of RB and H were first computed for each experiment and transformed into Fisher’s significance test scores.
AUTOMATIC AND CONTROLLED COMPONENTS OF JUDGMENT
judgments. The bias component of social judgment has gained importance over the years. At first, it was viewed as response error that should be corrected in order to better estimate “true” social judgments (e.g., Cohen, 1981; Hartwick, 1979); next, it came to be recognized as an important side-effect of the normal use of cognitive schemas (e.g., W. Brewer & Nakamura, 1984; Rumelhart, 1984); and most recently, the bias component is considered the implicit process constituent of dual-process approaches (Chaiken & Trope, 1999). It is important to note that although work in several areas of social cognition assumes the operation of both heuristic and systematic processes, until recently there was little in the way of independent assessment of these two processes as aspects of social cognition. Thus, regardless of the way in which automatic and controlled influences on judgment are interpreted, recent methodological tools that allow for the measurement of both influences within the same task have contributed greatly to our understanding of the two influences. Among these are signal detection theory (Correl, Park, Judd, & Wittenbrink, 2002; Green & Swets, 1967), the PDP, and other polynomial models (see Batchelder & Riefer, 1999) such as the guessing-corrected, multinomial process-dissociation analysis developed by Buchner and Wippich (1996); Klauer and Wegener’s (1998) model of multiple memory discriminability and bias parameters; and the recently proposed Quad model (Conrey, Sherman, Gawronski, Hugenberg, & Groom, 2005). The PDP in particular has been applied in different domains with implications for several areas of social cognition such as perception (Debner, & Jacoby, 1994), habit and recollection (Hay & Jacoby, 1996), proactive interference (Jacoby, Debner, & Hay, 2001), explicit versus implicit memory in judgments of fame (Jacoby et al. 1989; Jennings & Jacoby, 1993), and most recently the controlled and automatic influences of prejudice and social stereotypes (Lambert et al., 2003; Payne, 2001; Sherman, Groom, Ehrenberg, & Klauer, 2003).
Automaticity and Heuristic Reasoning H was regarded as an instance of Sloman’s (1996) associative system. Furthermore, this reasoning mode was considered to share at least some of the key aspects of automatic processes (Bargh, 1994). In fact, H appears to involve few cognitive resources and to operate with little awareness or process control. H also seems to lack some of the defining features of control, namely the ability to monitor information processing so as to flexibly vary it in response to feedback (Wegner & Bargh, 1998). In the case of Jacoby’s (1991) process dissociation procedure, automatic processes are typically viewed as a bias in facilitating certain responses that becomes apparent when controlled processes fail. The exact nature of this bias varies. In Jacoby’s model, the bias is familiarity bias or processing fluency bias (Jacoby, Toth, & Yonelinas, 1993), whereas in our case it is heuristic processing bias. Moreover, the use of the PDP experimentally constrains the automatic nature of H, defining it by the relation between performance in inclusion problems and that in exclusion problems. As a consequence, to be automatic, H must have an obligatory nature in that it remains the same regardless of whether its influence facilitates or hampers performance. Other uses of the terms heuristics or heuristic reasoning that do not accommodate this conception of automaticity refer to reasoning forms that could not be separated from (controlled) RB by using the PDP and as such are beyond the
807
scope of the present definition of H.10 Other dual-process approaches to reasoning adopt a conception of automaticity that is similar to our own (Evans & Over, 1996; Kahneman & Frederick, 2002; Stanovich & West, 2000).
The Perfect Performance Problem The PDP assumes independence between automatic (H) and controlled (RB) processes and uses the proportion of nonstatistical answers in the exclusion test to estimate the strength of H. Thus, if no nonstatistical answers are given in the exclusion test (exclusion ⫽ 0), the heuristic component cannot be estimated. The relative number of cases discarded due to exclusion ⫽ 0 was very low for Experiment 1 (2.5%) and Experiment 2 (8.0%), moderate for Experiment 3 (18.9%), and relatively high for Experiment 4 (34.6%). Exclusion of these data has no effect on the process dissociations here reported (based on the PDP estimates) as long as the H and RB processes are independent. However, if the two processes are positively correlated, H will be underestimated by the PDP, and this underestimation will be higher for higher levels of RB. Why? Because, if a positive correlation between H and RB exists, then H processes will no longer be equivalent under higher and lower levels of RB. In fact, when a positive correlation exists, H will be higher when RB is higher and lower when RB is lower, and as we are estimating H from the proportion of nonstatistical answers to exclusion problems (i.e., when RB is lower), we will necessarily underestimate H. Under these circumstances, eliminating participants with perfect performances would facilitate the emergence of artifactual dissociations because we are eliminating the participants for whom RB is highest. In any case, H and RB estimates calculated from data collapsed by participants and by problems (the grand means) are unaffected by exclusion ⫽ 0 and provide a way to test for the above possibility (Curran & Hintzman, 1995; Jacoby et al., 1993; Toth, Reingold, & Jacoby, 1994). Thus, we recomputed H and RB in this way and used the standard error of the proportion to derive 95% confidence intervals around the grand means. The pattern of results coincides with that of the original data analyses (although collapsing across participants and problems precludes the computation of statistical interactions). Multinomial modeling analyses that included participants with perfect performance (exclusion ⫽ 0) was also used and the fit of the model estimates was good, reproducing therefore the pattern of results we obtained in our original analyses. Regardless, exclusion performance ⫽ 0 should be avoided in future studies that test new hypotheses involving the interaction of the two processing modes, perhaps by increasing the number of target problems. 10
Because RB does not capture all forms of rule-governed cognitive activity but only the deliberate use of certain statistical principles, other controlled processes not anticipated by us may have also contributed to the dominant answers to inclusion problems and nonstatistical answers to exclusion problems. Nevertheless, a nonrandom distribution of such types of bias would certainly affect the PDP estimates, rendering findings of invariance highly unlikely. For instance, choosing the nonstatistical option based on deliberate and controlled reasoning would imply that fewer resources would produce fewer errors to exclusion problems, canceling the predicted decrease of RB as a function of cognitive load or even inverting this tendency.
808
FERREIRA, GARCIA-MARQUES, SHERMAN, AND SHERMAN
Relationship to Other Dual-Process Approaches The PDP approach differs in two significant ways from other dual-process approaches to understanding judgment and decision making. First, the PDP conceptualization of RB and H processes as two independent and parallel reasoning modes differs from models that conceive of them as mutually exclusive. For example, a number of dual-process models have argued that heuristic and rule-based processes represent distinct alternatives and that the processes do not co-occur (e.g., M. B. Brewer, 1988; Fazio, 1990; Kahneman & Tversky, 1972, 1973; Petty & Cacioppo, 1981). Other models (e.g., Fiske, Lin, & Neuberg, 1999; Fiske & Neuberg, 1990) have argued that RB and H represent two ends of a continuum and that movement toward one end of the continuum (e.g., H or RB) necessarily coincides with diminished activity on the other end (e.g., RB or H). In contrast, the PDP approach assumes that all judgments reflect the joint and independent contributions of RB and H. Increases in one process do not imply decreases in the other. Other dual-process models do emphasize the simultaneous influences of heuristic and systematic processes (Epstein, 1991; Kahneman, 2003; Sloman, 1996; Smith & DeCoster, 2000; Strack & Deutsch, 2005). However, only the PDP approach also offers a means for independently assessing the joint contributions of these processes to performance on a single task (see the following paragraph). The second significant difference between the PDP approach and other dual-process approaches concerns the means of estimating RB and H. Many dual-process approaches have relied on content dissociations to infer the extent to which a judgment reflects relatively heuristic or systematic processing. For example, the classical approach to understanding judgment under uncertainty is to associate the influence of one kind of information (e.g., base-rate information) with systematic processing (RB) and the influence of another kind of information (e.g., stereotypic target descriptions) with the use of heuristics (e.g., Kahneman & Tversky, 1972, 1973). Other dual-process approaches have attempted to estimate RB and H by administering two separate measures, one aimed at tapping an automatic process (reflecting H) and one aimed at tapping a controlled process (reflecting RB). A significant drawback to content dissociations is that they incorporate a confound between content and process. Though it may be the case that some kinds of information (e.g., heuristic cues) are typically applied with more ease and with less intent than other kinds of information (e.g., base rates), it is not difficult to find or induce exceptions to this state of affairs (e.g., Krull & Dill, 1996; Kunda & Thagard, 1996; Trope & Alfieri, 1997). Administering separate measures confounds the processing style (H vs. RB) with the particular measurement task (e.g., IAT vs. questionnaire). This is a problem because the chosen tasks may differ in a number of ways beyond the extent to which they tap automatic versus controlled processes (e.g., Jacoby, 1991; Roediger, 1990; Sherman, 2006). From the PDP perspective, the application of any type of information always reflects a combination of heuristic and systematic processes, and these processes can be estimated simultaneously from a single response, independently of particular content (for a review, see Sherman, 2006). Indeed, the PDP was developed specifically to overcome task confounds in the implicit– explicit memory literature (e.g., Jacoby, 1991).
Thus, the PDP approach adopted for understanding judgment under uncertainty offers important theoretical and methodological advantages over other dual-process approaches. Theoretically, it avoids fragile assumptions of process purity and process exclusivity, embracing the view that all judgments recruit parallel and independent heuristic and systematic processes that interact to produce output. Methodologically, it offers a means for measuring the independent contributions of the two processes within a single task, thereby avoiding content and task confounds that threaten the validity of dual-process conclusions.
On the Relationship Between H and RB Processes The C-first model. In the PDP model that we applied in our data analyses (Jacoby, 1991), the RB process constrains the influence of the H process. That is, the equations are such that the influence of H is observed only in cases in which RB does not provide a response. Note that designation of C-first is not meant to reflect temporal sequence. As Payne, Jacoby, and Lambert (2005) explained [T]he order depicted in this model refers to logical priority, not temporal ordering. The model is not sequential. We assume both processes begin at the same time and proceed simultaneously and independently. The priority of one process over the other means that the second process can drive the behavior only in the absence of the first. If both processes occur, then the first one dominates and determines the response. (p. 412)
Rather, this describes the mathematical relationship between the processes in producing judgments. Thus, the RB and H processes are thought to occur simultaneously. However, in determining a response on a trial of a given task, the influence of H is seen only in cases in which RB fails to provide a response. In this way, the RB process dominates or constrains the H process in this model. As such, this model may be called the C-first (control) model, and it has been applied to separate the automatic and controlled components of behavior in many areas of research, including perception (Debner & Jacoby, 1994), habit and recollection (Hay & Jacoby, 1996), proactive interference (Jacoby et al., 2001), judgments of fame (Jacoby et al., 1989; Jennings & Jacoby, 1993), and stereotyping (Lambert et al., 2003; Payne, 2001; Sherman et al., 2003). The A-first model. However, it is clear that automatic and controlled processes do not always interact in this C-first fashion. Instead, in some cases, it is the automatic process that dominates and constrains the application of control. For example, on incompatible trials in the Stroop task (i.e., the word blue written in red ink; Stroop, 1935), the automatic habit to read the word captures attention and interferes with the more controlled process of naming the color of the ink. Lindsay and Jacoby (1994) proposed a variation on the original C-first PDP model to account for such situations. This may be referred to as the A-first model. The logic of the A-first model is identical to that of the C-first model (Jacoby, 1991), except that the roles of A and C have been reversed, resulting in slightly different equations in solving for A and C. In the A-first model, A and C are solved by comparing correct responses on inclusion (the word blue in blue ink) trials to correct responses on exclusion (the word blue in red ink) trials. The probability of a correct response on inclusion trials is A ⫹ C (1 ⫺ A; the same as for the C-first model). The probability of a correct response on exclusion trials is C (1 ⫺ A)—when control drives the response in the absence of the auto-
AUTOMATIC AND CONTROLLED COMPONENTS OF JUDGMENT
matic process. A is solved by subtracting the probability of correct responses on exclusion trials from the probability of correct responses on inclusion trials. C then equals the probability of correct responses on exclusion trials divided by (1 ⫺ A). Choosing a model. In applying the PDP approach, it is not always perfectly clear a priori whether the C-first or the A-first model is more appropriate. As a consequence, the question of which model to apply is often treated as an empirical question (e.g., Batchelder & Riefer, 1999). For example, in Payne’s (2001) research on automatic and controlled components of stereotyping, participants are required to identify objects as either guns or tools in the presence of Black and White faces. Conceivably, one could view this as a case in which the automatic influence of stereotypes activated by the faces would always influence responses, unless the stereotype either was not activated or it was somehow overcome. However, because the C-first model has consistently provided a better account of results with this task than the A-first model (e.g., Payne et al., 2005), it is the C-first analyses that are reported in this research. It is important to note that, although the choice of which model to apply may be an empirical one, that choice does constrain subsequent interpretation of the data. For example, Payne and his colleagues were careful to interpret their A parameter as a stereotypic bias that influences judgments only when controlled efforts to identify the object fail. Similarly, the controlled process is identified as a process that attempts to correctly identify the object (gun or tool) rather than a process that attempts to correct for the influence of automatic stereotypes after they have already been activated (that would be according to Payne et al., 2005, an A-first depiction of control). One way to empirically distinguish between the models is to run separate analyses on the A and C estimates derived from the two models and to observe whether one model provides a more theoretically parsimonious accounting of the data than does the other model. In addition to the C-first model analyses reported in the Results sections, we also conducted analyses for each study on the basis of the A-first equations. In Experiment 1, although the results showed the same basic pattern as the C-first analysis, the A-first model did not demonstrate clear support for the theoretically derived hypotheses. The key interaction between instruction (rational vs. intuitive) and processing mode (H vs. RB) was not reliable (F ⬍ 1). Planned comparisons demonstrated marginally greater RB in the rational than in the intuitive condition, t(35) ⫽ 1.44, p ⬍ .09, and no differences in H between the two conditions (t ⬍ 1). Contrary to our hypothesis, and contrary to the C-first analysis, the A-first analysis of Experiment 2 showed no effect of the load manipulation on RB. There was no significant interaction between processing load (high vs. low) and processing mode (H vs. RB; F ⬍ 1). Neither H nor RB was affected by the load manipulation (both ts ⬍ 1, planned comparisons). The A-first analysis of Experiment 3 demonstrated no significant interaction between priming (heuristic priming vs. control) and processing mode (H vs. RB), F(1, 83) ⫽ 1.35, p ⬍ .25. Instead both estimates increased as a function of the priming. The increase in H, t(83) ⫽ 2.93, p ⬍ .05, is expected and consistent with the C-first model, but the increase in RB, t(83) ⫽ 2.49, p ⬍ .05, is not. Finally, the A-first analysis of Experiment 4 demonstrated a significant interaction between training (formal training vs. control) and processing mode (H vs. RB), F(1, 103) ⫽ 5.10, p ⬍ .05. However, planned comparisons showed that formal training marginally reduced H, t(103) ⫽ 1.67, p ⬍ .10, but had no effect on RB
809
(t ⬍ 1). Neither finding was predicted or consistent with results from the C-first model. In summary, the C-first model provides a more theoretically parsimonious account of the data across the four experiments than does the A-first model. In the C-first analyses, the manipulations influenced the appropriate processing modes in theoretically consistent ways. In contrast, in the A-first analyses, the manipulations did not influence the processing modes in a predictable fashion. Given the wealth of past process dissociation research validating the automatic and controlled nature of H and RB (A and C), this suggests that the C-first model provides a better account of the current data than does the A-first model. This does not necessarily imply that all judgment and decision making proceeds in a C-first rather than A-first fashion. Further research will be needed to delineate the particular types of judgments and contexts associated with the two different relationships between automatic and controlled processing. Another way to empirically distinguish between the models is to compare their ability to account for the data through the use of maximum likelihood statistics and multinomial modeling. We applied such techniques to both C-first and A-first models of each experiment. Results showed that both models provided adequate fits of the data for Experiments 1, 2 and 4, but only the C-first model provided an adequate fit of data for Experiment 3. Thus, although this modeling does not clearly distinguish whether the C-first or A-first model is most appropriate for most of the current judgment tasks, the C-first model is the only one that accounts for the entire result patterns. Further information on the modeling is available from the authors.
Conclusion One of the most important contributions of the heuristics and biases research programs was to bring the study of judgments under uncertainty into the realm of cognitive psychology and social cognition. However, apart from a few exceptions (see Weber, Goldstein, & Barlas, 1995; Weber, Goldstein, & Busemeyer, 1991), the promised integration has not lived up to expectations. In fact, most models of inductive judgment do little to explicitly combine considerations of memory processes, representation of information, categorization, and so forth, with other stages of the decision process. Now, as before, there is a need to articulate research in inductive judgment with the general conceptual framework of social cognitive research. In applying the PDP to inductive judgment, the present work aims to contribute a clearer definition of the automatic and intentional processes involved in inductive judgment. The resulting dual-process approach makes use of several psychological distinctions such as symbolic, localized representations usually coupled with rule-governed processing versus distributed representations of information generally attached to associative processing (Smith & DeCoster, 1999). In essence, we aimed to explore the operating principles and representational nature of human inferences in light of advances in the social cognitive literature toward a better and more articulated comprehension of judgments under uncertainty. This work is far from being completed.
References Ajzen, I. (1977). Intuitive theories of events and the effects of base-rate information on predictions. Journal of Personality and Social Psychology, 35, 303–314.
810
FERREIRA, GARCIA-MARQUES, SHERMAN, AND SHERMAN
Bargh, J. A. (1994). The four horsemen of automaticity: Awareness, intention, efficiency and control in social cognition. In R. S. Wyer, Jr., & T. K. Srull (Eds.), Handbook of social cognition: Basic processes, Vol. 1 (2nd ed., pp. 1– 40). Mahwah, NJ: Erlbaum. Bar-Hillel, M. (1979). The role of sample size in sample evaluation. Organizational Behavior and Human Decision Processes, 24, 243–257. Bar-Hillel, M. (1980). The base-rate fallacy in probability judgments. Acta Psychologica, 44, 211–233. Batchelder, W. H., & Riefer, D. M. (1999). Theoretical and empirical review of multinomial process tree modeling. Psychonomic Bulletin & Review, 6, 57– 86. Bless, H., & Shwarz, 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 Press. 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). Mahwah, NJ: Erlbaum. Brewer, W., & Nakamura, G. V. (1984). The nature and functions of schemas. In R. S. Wyer & T. K. Srull (Eds.), Handbook of social cognition (Vol. 1, pp. 119 –160). Mahwah, NJ: Erlbaum. Buchner, A., & Wippich, W. (1996). Unconscious gender bias in fame judgments? Consciousness and Cognition, 5, 197–220. Camerer, C. F. (1987). Do bias in probability judgments matter in markets? Experimental evidence. American Economic Review, 77, 981–997. Chaiken, S., & Trope, Y. (1999). Dual-process theories in social psychology. New York: Guilford Press. Chen, S., & Chaiken, S. (1999). The heuristic-systematic model in its broader context. In S. Chaiken & Y. Trope (Eds.), Dual-process theories in social psychology (pp. 73–96). New York: Guilford Press. Cohen, C. E. (1981). Person categories and social perception: Testing some boundaries of the processing effects of prior knowledge. Journal of Experimental Social Psychology, 40, 441– 452. Conrey, F. R., Sherman, J. W., Gawronski, B., Hugenberg, K., & Groom, C. (2005). Separating multiple processes in implicit social cognition: The quad-model of implicit task performance. Journal of Personality and Social Psychology, 89, 469 – 487. Correl, J., Park, B., Judd, C. M., & Wittenbrink, B. (2002). The police officer’s dilemma: Using ethnicity to disambiguate potentially threatening individuals. Journal of Personality and Social Psychology, 83, 1314 –1329. Curran, T., & Hintzman, D. L. (1995). Violations of the independence assumption in process dissociation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 531–547. Debner, J. A., & Jacoby, L. L. (1994). Unconscious perception: Attention, awareness, and control. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20, 304 –317. Denes-Raj, V., & Epstein, S. (1994). Conflict between intuitive and rational processing: When people behave against their better judgment. Journal of Personality and Social Psychology, 66, 819 – 829. Epstein, S. (1991). Cognitive-experiential self theory: An integrative theory of personality. In R. Curtis (Ed.), The self with others: Convergences in psychoanalytical, social, and personality psychology (pp. 111–137). New York: Guilford Press. Epstein, S. (1994). An integration of the cognitive and psychodynamic unconscious. American Psychologist, 49, 709 –724. Epstein, S., Donovan, S., & Denes-Raj, V. (1999). The missing link in the paradox of the Linda conjunction problem: Beyond knowing and thinking of the conjunction rule, the intrinsic appeal of heuristic processing. Personality and Social Psychology Bulletin, 25, 204 –214. Evans, J. St. B. T., & Over, D. E. (1996). Rationality and reasoning. Hove, England: Psychology Press. Fazio, R. H. (1990). A practical guide to the use of response latency in social psychological research. In C. Hendrick & M. S. Clark (Eds.), Review of personality and social psychology (Vol. 11, pp. 74 –97). Newbury Park, CA: Sage.
Ferreira, M. B., & Garcia-Marques, L. (2003). O papel do reconhecimento do acaso no raciocı´nio indutivo. The role of perception of randomness on inductive reasoning. Ana´lise Psicolo´gica, 3, 353–374. Finucane, M. L., Alhakami, A., Slovic, P., & Johnson, S. M. (2000). The affect heuristic in judgments of risks and benefits. Journal of Behavioral Decision Making, 13, 1–17. Fiske, S. T., Lin, M., & Neuberg, S. L. (1999). The continuum model: Ten years later. In S. Chaiken & Y. Trope (Eds.), Dual-process theories in social psychology (pp. 231–254). New York: Guilford Press. Fiske, S. T., & Neuberg, S. E. (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. Fong, G. T., Krantz, D. H., & Nisbett, R. E. (1986). The effects of statistical training on thinking about everyday problems. Cognitive Psychology, 18, 293–328. Gigerenzer, G. (1991). How to make cognitive illusions disappear: Beyond “heuristics and biases”. In W. Stroebe & M. Hewstone (Eds.), European review of social psychology (Vol. 2, pp. 83–115). Chichester, England: Wiley. Ginossar, Z., & Trope, Y. (1987). Problem solving in judgment under uncertainty. Journal of Personality and Social Psychology, 52, 464 – 474. Green, D. M., & Swets, J. A. (1967). Signal detection theory and psychophysics. New York: Wiley. Griffin, D., Gonzalez, R., & Varey, C. (2001). The heuristics and biases approach to judgment under uncertainty. In A. Tesser & N. Schwarz (Eds.), Blackwell handbook of social psychology: Intraindividual processes (Vol. 1, pp. 207–235). London: Blackwell. Hartwick, J. (1979). Memory for trait information: A signal detection analysis. Journal of Experimental Social Psychology, 15, 533–552. Hay, J. F., & Jacoby, L. L. (1996). Separating habit and recollection: Memory slips, process dissociations, and probability matching. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 1323–1335. Hollender, D. (1986). Semantic activation without conscious identification in dichotic listening, parafoveal vision, and visual masking: A survey and appraisal. Behavioral and Brain Sciences, 9, 1– 66. Jacoby, L. L. (1991). A process dissociation framework: Separating automatic from intentional uses of memory. Memory and Language, 30, 513–541. Jacoby, L. L., Debner, J. A., & Hay, J. F. (2001). Proactive interference, accessibility, and process dissociations: Valid subjective reports of memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27, 686 –700. Jacoby, L. L., Kelley, C., Brown, J., & Jasechko, J. (1989). Becoming famous overnight: Limits on the ability to avoid unconscious influences of the past. Journal of Personality and Social Psychology, 46, 326 –338. Jacoby, L. L., Toth, J. P., & Yonelinas, A. P. (1993). Separating conscious and unconscious influences of memory: Measuring recollection. Journal of Experimental Psychology: General, 122, 139 –154. Jennings, J. M., & Jacoby, L. L. (1993). Automatic versus intentional uses of memory: Aging, attention, and control. Psychology and Aging, 8, 283–293. Jepson, D., Krantz, D. H., & Nisbett, R. E. (1983). Inductive reasoning: Competence or skill? Behavioral and Brain Sciences, 6, 494 –501. Kahneman, D. (2003). A perspective on judgment and choice. American Psychologist, 58, 697–720. Kahneman, D., & Frederick, S. (2002). Representativeness revisited: Attribute substitution in intuitive judgment. In T. Gilovich, D. Griffin, & D. Kahneman (Eds.), Heuristics and biases: The psychology of intuitive judgment (pp. 49 – 81). New York: Cambridge University Press. Kahneman, D., Slovic, P., & Tversky, A. (1982). Judgment under uncertainty: Heuristics and biases. New York: Cambridge University Press. Kahneman, D., & Tversky, A. (1972). Subjective probability: A judgment of representativeness. Cognitive Psychology, 3, 430 – 454.
AUTOMATIC AND CONTROLLED COMPONENTS OF JUDGMENT Kahneman, D., & Tversky, A. (1973). On the psychology of prediction. Psychological Review, 80, 237–251. Kirkpatrick, L. A., & Epstein, S. (1992). Cognitive– experiential selftheory and subjective probability: Further evidence for two conceptual systems. Journal of Personality and Social Psychology, 63, 534 –544. Klauer, K. C., & Wegener, I. (1998). Unraveling social categorization in the “who said what?” paradigm. Journal of Experimental Social Psychology, 75, 1155–1178. Kruglanski, A. W., Friedland, N., & Farkash, E. (1984). Lay persons’ sensitivity to statistical information: The case of high perceived applicability. Journal of Personality and Social Psychology, 46, 503–518. Krull, D. S., & Dill, J. C. (1996). On thinking first and responding fast: Flexibility in social inference processes. Personality and Social Psychology Bulletin, 22, 949 –959. 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., Payne, B. K., Jacoby, L. L., Shaffer, L. M., Chasteen, A. L., & Khan, S. R. (2003). Stereotypes as dominant responses: On the “social facilitation” of prejudice in anticipated public contexts. Journal of Personality and Social Psychology, 84, 277–295. LaRue, A., & Olejnik, A. B. (1980). Cognitive “priming” of principles of moral thought. Personality and Social Psychology Bulletin, 6, 413– 416. Lindsay, D. S., & Jacoby, L. L. (1994). Stroop process dissociations: The relationship between facilitation and interference. Journal of Experimental Psychology: Human Perception and Performance, 20, 219 –234. Miller, D. T., Turnbull, W., & McFarland, C. (1989). When a coincidence is suspicious: The role of mental simulation. Journal of Personality and Social Psychology, 57, 581–589. Nisbett, R. E., Krantz, D. H., Jepson, C., & Kunda, Z. (1983). The use of statistical heuristics in everyday inductive reasoning. Psychological Review, 90, 339 –363. Payne, B. K. (2001). Prejudice and perception: The role of automatic and controlled processes in misperceiving a weapon. Journal of Personality and Social Psychology, 81, 181–192. Payne, B. K., Jacoby, L. L., & Lambert, A. J. (2005). Attitudes and accessibility bias: Dissociating automatic and controlled processes. In R. R. Hassin, J. S. Uleman, & J. A. Bargh (Eds.), The new unconscious (pp. 393– 420). New York: Oxford University Press. Petty, R. E., & Cacioppo, J. T. (1981). Attitudes and persuasion: Classic and contemporary approaches. Dubuque, IA: Brown. Roediger, H. L. (1990). Implicit memory. American Psychologist, 45, 1043–1056. Rumelhart, D. E. (1984). Schemata and the cognitive system. In R. S. Wyer & T. K. Srull (Eds.), Handbook of social cognition (Vol. 1, pp. 161– 188). Mahwah, NJ: Erlbaum. Sherman, J. W. (2006). On building a better process model: It’s not only how many, but which ones and by which means? Psychological Inquiry, 17, 173–184. Sherman, J. W., Groom, C. J., Ehrenberg, K., & Klauer, K. C. (2003). Bearing false witness under pressure: Implicit and explicit components of stereotype-driven memory distortions. Social Cognition, 21, 213–246. Sherman, S. J., & Corty, E. (1984). Cognitive heuristics. In R. S. Wyer & T. K. Srull (Eds.), Handbook of social cognition (Vol. 1, pp. 189 –286). Mahwah, NJ: Erlbaum. Sloman, S. A. (1996). The empirical case for two systems of reasoning. Psychological Bulletin, 199, 3–22. Sloman, S. A., & Rips, L. J. (1998). Similarity as an explanatory construct. Cognition, 65, 2–3, 87–101. Smith, E. R. (1994). Procedural knowledge and processing strategies in
811
social cognition. In R. S. Wyer & T. K. Srull (Eds.), Handbook of social cognition (Vol. 1, pp. 99 –151). Mahwah, NJ: Erlbaum. Smith, E. R., & DeCoster, J. (1999). A connectionist interpretation of dual-process models. In S. Chaiken & Y. Trope (Eds.), Dual-process theories in social psychology (pp. 323–336). New York: Guilford Press. Smith, E. R., & DeCoster, J. (2000). Dual-process models in social and cognitive psychology: Conceptual integration and links to underlying memory systems. Personality and Social Psychology Review, 4, 108 –131. Stanovich, K. E., & West, R. F. (1998). Individual differences in rational thought. Journal of Experimental Psychology: General, 127, 161–188. Stanovich, K. E., & West, R. F. (1999). Discrepancies between normative and descriptive models of decision making and the understanding/ acceptance principle. Cognitive Psychology, 38, 349 –385. Stanovich, K. E., & West, R. F. (2000). Individual differences in reasoning: Implications for the rationality debate. Behavioral and Brain Sciences, 23, 645– 665. Stanovich, K. E., & West, R. F. (2002). Individual differences in reasoning: Implications for the rationality debate. In T. Gilovich, D. Griffin, & D. Kahneman (Eds.), Heuristics and biases (pp. 421– 440). New York: Cambridge University Press. Strack, F., & Deutsch, R. (2005). Reflection and impulse as determinants of conscious and unconscious motivation. In J. Forgas, K. Williams, & S. Laham (Eds.), Social motivation: Conscious and unconscious processes (pp. 91–112). Cambridge, England: Cambridge University Press. Stroop, J. R. (1935). Studies on the interference in serial verbal reactions. Journal of Experimental Psychology, 59, 239 –245. Toth, J. P., Reingold, E. M., & Jacoby, L. L. (1994). Toward a redefinition of implicit memory: Process dissociations following elaborative processes and self-generation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20, 290 –303. Trope, Y., & Alfieri, T. (1997). Effortfulness and flexibility of dispositional inference processes. Journal of Personality and Social Psychology, 73, 662– 675. Tversky, A., & Kahneman, D. (1971). Belief in the law of small numbers. Psychological Review, 2, 105–110. Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5, 207–232. Tversky, A., & Kahneman, D. (1974, September). Judgment under uncertainty: Heuristics and biases. Science, 185, 1124 –1131. Tversky, A., & Kahneman, D. (1982). Evidential impact of base-rates. In D. Kahneman, P. Slovic, & A. Tversky (Eds.), Judgment under uncertainty: Heuristics and biases (pp. 153–160). Cambridge, England: Cambridge University Press. Tversky, A., & Kahneman, D. (1983). Extensional versus intuitive reasoning: The conjunction fallacy in probability judgment. Psychological Review, 90, 293–315. Weber, E. U., Goldstein, W. M., & Barlas, S. (1995). Let us not forget about memory: The role of memory processes and techniques in the study of judgment and choice. The Psychology of Learning and Motivation, 32, 33– 81. Weber, E. U., Goldstein, W. M., & Busemeyer, J. R. (1991). Beyond strategies: Implications of memory representation and memory processes for models of judgment and decision making. In W. E. Hockley & S. Lewandowsky (Eds.), Relating theory and data: Essays on human memory in honor of Bennet B. Murdock (pp. 75–100). Mahwah, NJ: Erlbaum. Wegner, D. M., & Bargh, J. A. (1998). Control and automaticity in social life. In D. T. Gilbert, S. T. Fiske, & G. Lindzey (Eds.), Handbook of social psychology (4th ed., pp. 446 – 496). New York: McGraw-Hill. Zukier, H., & Pepitone, A. (1984). Social roles and strategies in prediction: Some determinants of the use of base-rate information. Journal of Personality and Social Psychology, 47, 349 –360.
(Appendixes follow)
FERREIRA, GARCIA-MARQUES, SHERMAN, AND SHERMAN
812
Appendix A Examples of the Problems UsedA1 Base-Rates Problem One hundred undergraduate students applied for a part-time job. Of the applicants, 15 (85) were humanities students, and 85 (15) were science students. Mike was one of the 100 students who applied for the job. His name was randomly chosen by computer to participate in the first day of interviews. Mike is 23 years old, he likes to travel, and he was quite a good student in high school. His preferred subjects were English poetry, modern art, and sports. Which of the following is more likely? (a) Mike is one of the humanities students. (b) Mike is one of the science students.
Conjunction Problem A company that makes beauty products is about to launch a new product line. The marketing department wanted to begin with the promotion of this new line as quickly as possible. To do this they can either deliver all the promotional work to a large publicity agency, or they can divide the promotional work between two smaller publicity agencies. The large agency has a record of meeting deadlines of 60% (90%). One of the smaller agencies has a record of meeting deadlines of 80%, and the other has a record of meeting deadlines of 70%. The marketing department can begin the promotion only when all the promotional work is ready to be used. Which of the following is more likely? (a) The best possibility of starting the promotion sooner would be to deliver all the promotional work to the larger publicity agency. (b) The best possibility of starting the promotion sooner would be to divide the promotional work between the two smaller agencies.
Ratio-Bias Effect Problem Elaine was on a TV show where she had to choose an envelope from one of two sets of envelopes. In the first set, there were 100
envelopes, 21 (19) of which contained a prize ticket of $5,000. In the second set, there were 10 envelopes, 2 of which contained a prize ticket of the $5,000. If you were Elaine, what would you do? (a) I would choose an envelope from the first set of envelopes. (b) I would choose an envelope from the second set of envelopes.
Law of Large Numbers Problem In their graduation year, students in theatre are chosen to act in a play to be presented at the end of the year. For this year’s play, the professors have to decide between two students (Suzanne and Amy) for the main role in the play. Suzanne played brilliantly in several main roles during the 3 years of the theatre course, but (and) her audition for the present main role was mediocre (also brilliant). Amy’s performance in several main roles during the course was mediocre but (and) her audition to the present main role was brilliant (also mediocre). What do you think is more likely? (a) Amy has a better possibility of being selected. (b) Suzanne has a better possibility of being selected.
Filler Problem Sylvia is a 35-year-old woman. She is intelligent, pretty, an excellent debater, and she is very captivating. Since she was a girl, Sylvia was interested in journalism. After finishing high school, Sylvia majored in Social Communication and Journalism with excellent grades. Given that she was one of the best students in her major, she rapidly initiated a very successful professional career. Since she was a girl, Sylvia was interested in what? (a) science fiction (b) journalism
Appendix B Example of a Heuristic Priming Problem, a Neutral Problem, and a Target Problem Used in Experiment 3 (for Base-Rates Problems)B1 Priming Problem One hundred men from the U.S. Army Special Forces were selected for a dangerous secret mission in South America. Ten of these men are officers, and 90 are privates. Bob is a veteran from the Vietnam War. He is often called for special missions, and he is used to commanding men under extremely difficult situations. Last year he was promoted and was decorated by the U.S. president for his accomplishments in the army and for his exceptional qualities of leadership. Which of the following is more likely?
(a) Bob is one of the 10 officers in special forces selected for the mission. (b) Bob is one of the 90 privates in the special forces selected for the mission.
A1 Numbers in parentheses were used in the inclusion versions of the problems. B1 Numbers in parentheses were used in the inclusion versions of the problems.
AUTOMATIC AND CONTROLLED COMPONENTS OF JUDGMENT
Neutral Problem The first time Chad went to New York he was very impressed with the city: the huge buildings, the nonstop activity, everybody always rushing, et cetera. Chad spent 2 weeks there and he loved it. However he also realized that he would not like to live in such a big city. There was too much confusion for him. Where would you prefer to live? (a) In a big city like New York. (b) In a smaller city like Bloomington.
Target Problem In the year 2467, after 50 years of war against the Cyclons (an alien species), the human race is about to be defeated. One hundred
813
men from the U. N. Army Special Forces were selected for a dangerous secret mission in a last war effort. Twenty (80) of these men are majors, and 80 (20) are sergeants. Amos is a veteran from several battles against the Cyclons. He is often called for special missions, and he is used to commanding men under extremely difficult situations. Last year he was promoted and was decorated by the U. N. president for his accomplishments in the army and for his exceptional qualities of leadership. Which of the following is more likely? (a) Amos is one of the 20 (80) majors in special forces selected for the mission. (b) Amos is one of the 80 (20) sergeants in the special forces selected for the mission
Appendix C Examples of Graduate Record Exam Problems Used in Experiment 4 Three-person work crews are to be chosen from among two groups totaling seven people. Group I consists of A, B, C, and D. Group II consists of E, F, and G. Each group must have at least one representative in any possible work crew. C refuses to work unless E works. G will not work if A works. Which of the following crew may not be assembled? (a) A, C, E (b) A, F, G Building B is taller than Building C; Building A is taller than Building B. Which is the shorter Building? (a) Building B (b) Building A Which of the following pairs of words expresses a relationship that is a best analogy for the pair RACE–FATIGUE? (a) TRACK–ATHLETE (b) FAST–HUNGER What is the best interpretation of the following proverb, “out of the pot, into the fire”? (a) Sometimes people escape from a bad situation just to get into a worse one.
(b) If you put too much food in a cooking pot, food will spill out of the pot and fall right into the fire. Which of the following pairs of words expresses a relationship that is the best analogy for the pair BIRTHMARK–CONGENITAL? (a) BEAUTY SPOT–FACIAL (b) BALDNESS–HEREDITARY
Example of a Neutral Problem Three-person work crews are to be chosen from among two groups totaling seven people. Group I consists of A, B, C, and D. Group II consists of E, F, and G. With which of the following do you agree more? (a) Each group must have at least one representative in any possible work crew. (b) It is better to form work crews with workers coming from the same group. Received May 1, 2005 Revision received January 20, 2006 Accepted February 3, 2006 䡲
Journal of Personality and Social Psychology 2006, Vol. 91, No. 5, 814 – 831
Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.5.814
Stereotypes: Static Abstractions or Dynamic Knowledge Structures? Leonel Garcia-Marques and A. Sofia C. Santos
Diane M. Mackie
Universidade de Lisboa
University of California, Santa Barbara
Stereotypes have been assumed to be long-lasting knowledge structures that persist even in the face of contrary evidence. However, there is almost no within-participant research relevant to this assumption. The authors describe 4 studies (N ⫽ 267), the first 3 of which assessed within-participant stereotype stability over a few weeks with measures of stereotypic trait verification, typicality ratings of exemplar sets, and exemplar retrieval. In the 4th study, the authors manipulated context stability. Overall, results showed only low-to-moderate stereotype stability. The stability obtained was a function of the perceived centrality of traits or exemplars and of context constancy. The authors discuss the implications of these results for abstractionist, exemplar, mixed, and connectionist models and identify possible mechanisms that underlie within-participant stereotype instability. Keywords: malleability of knowledge structures, context sensitivity, stereotypes
already raised the question of within-individual representational stability. In this article, we first briefly describe abstractionist, exemplar, mixed models, and connectionist perspectives on the question of within-individual stability of stereotypes. We then review relevant findings from studies of nonsocial common concepts and present four studies. The first three were designed to assess the stability of social category representations in three crucial domains: stereotypic property verification, graded structure, and exemplar retrieval. In the fourth study, we directly manipulated context stability and evaluated its effects on stereotype stability.
A stereotype is a fixed impression, which conforms very little to the fact that it pretends to represent, and results from our defining first and observing second. (Katz and Braly, 1935, p. 181, emphasis added) [S]tereotypes have been regarded as rigid because they are believed to be persistent over time. (Ashmore & Del Boca, 1981, p. 18)
Are stereotypes stable over time? For many, including the authors, the answer to this question seems, at first sight, quite obvious. Just think of the “Princeton trilogy” studies, for example (Gilbert, 1951; Karlins, Coffman, & Walters, 1969; Katz & Braly, 1933). Although the level of consensus decreased somewhat across studies, successive generations of Princeton University students conveyed only slightly more benevolent versions of basically the same stereotypes. But are stereotypes stable over time within the same individual? Again, our gut feeling, and we suppose we are not alone, suggests a positive answer. But, in fact, we simply lack the relevant empirical evidence (for one exception, see unpublished longitudinal study, described in Rothbart & John, 1993). At best, we might derive theoretical expectancies from what we currently know about knowledge structures, including stereotypes. In particular, we might look to studies of nonsocial categorization, which have
Classic Abstractionist Perspectives According to early abstractionist positions, stereotypes play an important role in achieving cognitive economy (e.g., Crocker, Fiske, & Taylor, 1984; Fiske, 1980; Fiske & Taylor, 1984; Taylor, 1981). Fulfilling such a role demands cognitive structures that are both constant and persistent. In fact, according to these views, the need for cognitive stability coupled with the scarcity of cognitive resources forces social information processors to neglect much of the detail about individual members of social groups or categories. Perceivers do best by judiciously ignoring the least relevant characteristics of individual targets and going beyond the information given: in short, by becoming chronic abstractionists. Such abstractionist tendencies should in turn make stereotypes selfperpetuating and highly resistant to change (e.g., Hamilton & Trolier, 1986; MacArthur, 1982; Snyder, 1981). In fact, the many ingenious ways stereotypes resist change are among our discipline’s most well-documented findings and popular class anecdotes. Early abstractionist views thus envisaged stereotypes, like mental representations of other objects, as enduring mental entities that exhibit an impressive degree of constancy in the face of environmental turmoil. Theoretically, then, stereotypes, at least under ideal measurement conditions, should exhibit high reliability across relatively extended periods of time within the same individual. Such views inspired decades of theory and research on stereotypes in the social cognition tradition (Abelson, 1994).
Leonel Garcia-Marques and A. Sofia C. Santos, Faculdade de Psicologia e de Cieˆncias da Educac¸a˜o, Universidade de Lisboa, Lisboa, Portugal, Diane M. Mackie, Department of Psychology, University of California, Santa Barbara. This research was supported in part by Grant POCTI/PSI/47252/2002 from the Sapiens Program from Faculdade de Cieˆncias e Tecnologia da Universidade Nova de Lisboa to Leonel Garcia-Marques and A. Sofia C. Santos and by Public Health Service Grant MH63762 to Diane M. Mackie. We thank Lawrence Barsalou, Markus Brauer, and David L. Hamilton, who provided valuable commentary and criticism regarding a previous version of the article. Correspondence concerning this article should be addressed to Leonel Garcia-Marques, Faculdade de Psicologia, e de Cieˆncias da Educac¸a˜o, Alameda da Universidade, 1649-013, Lisboa, Portugal. E-mail:
[email protected] 814
STEREOTYPES AS DYNAMIC KNOWLEDGE STRUCTURES
The Priority of the Specific: The Exemplar View What perceivers actually reported about social categories soon triggered challenges to the abstractionist stance (i.e., Kahneman & Miller, 1986; Linville, Fischer, & Salovey, 1989; E. R. Smith, 1988). People can retrieve much more information about specific social category exemplars—including the contingency, range, and variability of group members’ attributes (Linville & Fischer, 1993)—than should be included in any fully developed abstract stereotype. Thus it appears that the information gained about specific exemplars is never completely discarded but remains available to be sampled when required. According to the exemplar perspective, stereotypes, group judgments, and group generalizations are the result of an exhaustive parallel search of a retrieved subset of stored exemplar representations. How much withinindividual stability should we expect according to this view? Not much, because exemplar retrieval is at least partially guided by the context the judgment is made in (Garcia-Marques & Mackie, 1999; Kahneman & Miller, 1986; E. R. Smith, 1988). When the judgment context changes, so too does the relative impact of specific exemplars (Sia, Lord, Blessum, Thomas, & Lepper, 1999). Thus according to exemplar views, within-individual stability of stereotypes should be relatively low—a clear disavowal of the abstractionist position.
The Eclectic Edge: Mixed Models of Social Categorization Contemporary views have more often sought the conditions under which group judgments reflect either abstractions or exemplars, rather than pitting one perspective against the other in a quest for the ultimate truth (McGuire, 1983). Many current researchers adopted hybrid or mixed model views that suggest that group concepts or stereotypes include both abstractions and specific exemplars (Babey, Queller, & Klein, 1998; Brewer, 1988; Fiske & Neuberg, 1990; Hamilton & Sherman, 1994; Judd & Park, 1988; Sherman, 1996; Zarate & Smith, 1990). Actually, most mixed models confer primary status on abstractions and assign exemplars more of a last-resource role in judgments (e.g., Babey et al., 1998; Brewer, 1988; Fiske & Neuberg, 1990; Hastie & Park, 1986; for a similar argument, see E. R. Smith, 1998). Abstractions play the primary role because they allow cognitive economy, but abstractions are complemented with exemplar information either because abstractions take time to develop (e.g., Sherman, 1996) or because abstractions are sometimes too narrow to accommodate the idiosyncrasies of atypical group members (e.g., Babey et al., 1998). How much within-individual stability would mixed models lead us to expect? It would depend, of course, on the relative weight of abstraction and exemplar components. The abstraction component should be relatively stable over time (or it loses its cognitive efficiency edge), but if exemplar retrieval changes with the context, then only moderate stability might be expected.
Patterns That Connect: The Connectionist Perspective The connectionist perspective is a relatively new contender in the literature, but a number of connectionist models of stereotyping have already emerged (Kashima, Woolcock, & King, 1998; Queller & Smith, 2002; E. R. Smith & DeCoster, 1998; Van Rooy, Van
815
Overwalle, Vanhoomissen, Labiouse, & French, 2003). According to this perspective, stereotypes are represented by dynamic activation patterns that occur in networks formed by simple and undifferentiated nodes. Each node receives positive or negative activation from neighboring nodes according to their respective connection weights. These connection weights derive from the previous history of activation from neighboring nodes. As all information is represented in the same network by different activation patterns, connectionist representations are superimposed and/or distributed (but see Van Rooy et al., 2003). Note that the information learned by such a network is not stored and consequently cannot be retrieved. It must be reconstructed from connection weights in response to input activation cues. Although network learning is preserved in connection weights, this reconstructive process is strongly affected by the immediate context because both the immediate context and previous learning are being represented at the same time in the same network. It is obvious that the dynamic nature of connectionist representations is at odds with abstractionist predictions about within-individual stability of stereotypes. Of course, whereas activation patterns can change very rapidly from moment to moment, connection weights change only very slowly and keep the network from behaving erratically. Even so, from the connectionist perspective, stereotypes are certainly not rigid knowledge structures and withinindividual stability in stereotyping is expected to be only moderate at best. In sum, although classic abstractionist perspectives would predict a high degree of within-individual stability in social category representation and use, mixed model and connectionist approaches suggest greater variability, and exemplar models predict maximal context-driven variability.
The Empirical Case of Common Concepts Empirical evidence from studies assessing the stability of common concepts and categories also found much more instability than a classic abstractionist position would suggest. The same individual on two different occasions (24 hr or 2 months apart) exhibited only modest reliability in defining and characterizing common concepts (Barsalou, Spindler, Sewell, Ballato, & Gendel, 1987; Bellezza, 1984a, 1984b), retrieving exemplars from common categories (Bellezza, 1984c), classifying instances into categories (McCloskey & Glucksberg, 1978), and rating the typicality of instances relative to their parent categories (Barsalou, Sewell, & Ballato, 1986). Other research showed that common categories are largely context sensitive, in that the immediate linguistic context biases both how typical an instance is judged to be as well as how fast it can be accessed (Roth & Shoben, 1983).
Stability in Social Category Representation Thus the preponderance of theoretical expectation and relevant empirical evidence suggests a considerable degree of variability in social category representation, even within the same individual. Should we expect mental representations of social categories to show the same degree of fluidity as nonsocial categories? Stereotypes share crucial cognitive features with other mental representations and so might be expected to show the same degree of malleability and context sensitivity. At the same time, there are reasons to think that social stereotypes might be more abstract in
GARCIA-MARQUES, SANTOS, AND MACKIE
816
nature than common categories and, therefore, less fluid. After all, stereotype content is often developed from hearsay without direct intergroup contact (Linville & Fischer, 1993); such information comes in an abstract linguistically encoded form from the start. In addition, stereotypes bias information processing in a number of self-perpetuating ways, increasing stereotype stability (see Hamilton & Sherman, 1994; E. R. Smith & Mackie, 1995, for reviews). For these reasons, stereotype representations may well be more stable than those of common concepts. Thus, we assessed as one of the primary goals of the first three studies to be reported here whether stereotypes evidenced the same type of context sensitivity and malleability demonstrated for mental representations of nonsocial categories by using longitudinal methodology (for a review, see Barsalou, 1987, 1989; Barsalou & Medin, 1986). The studies reported here belong to the very few that directly examine the within-participant stability of social stereotypes (for a precedent, see Rothbart & John, 1993).
ates stereotype instability is a crucial one to examine in the case of social categories. Finally, we also addressed the question of perspective. Barsalou and his colleagues (1987; see also Barsalou & Sewell, 1984) asked participants to generate the most important attributes of a number of categories from different points of view (e.g., from the perspective of the self, a suburban housewife, a country redneck, etc.) and found similar degrees of within-participant instability. We chose to manipulate point of view in a way that closely corresponds to an important variable in the social stereotype literature—the difference between cultural and individual stereotypes. Devine and Elliot (1995) suggested that cultural stereotype content—what “people in general” think—seems to be more stable than individual or personal stereotypes. However, because their data were not longitudinal, it remains to be seen whether cultural stereotypes evidence greater stability than do personal stereotypes in a withinparticipant sense.
Method
Study 1 How likely is a given individual to select the same attributes to characterize a social category at two different points in time (the property verification task, Barsalou et al., 1987)? This seems a particularly appropriate way to assess stereotype stability because attribute or property selection was the first procedure used to study stereotypes empirically (i.e., the adjective checklist of Katz & Braly, 1933) and has remained quite popular (for a recent review see Dovidio, Brigham, Johnson, & Gaertner, 1996). The checklist was of course the procedure used in the series of studies that assessed stereotypes in different generations of the same student population to infer the temporal persistence of stereotypes (Gilbert, 1951; Karlins et al., 1969; Katz & Braly, 1933). Our primary goal in Study 1 was to assess stereotype stability as reflected in stability of selection of attributes as stereotypic within individuals over time. We also assessed whether all stereotype content showed equal levels of stability or instability across time. Most current theories of representation conceive of conceptual cores, comprising more central or important, and more stable, attributes or features. Some proposed that conceptual cores are definitional (Armstrong, Gleitman, & Gleitman, 1983; Osherson & Smith, 1981; E. E. Smith & Medin, 1981). According to core-plus-identification views, mental representations of categories contain definitional cores and identification procedures based on typicality. To preserve the existence of definitional cores, the core-plus-identification view allows for categorization instability in terms of identification procedures— such procedures focus on the attributes that proved helpful in previous categorization tasks and reflect idiosyncratic experience, resulting in less stability. The definitional cores, however, reflect natural and logical invariants and should be perfectly stable. Other authors proposed that conceptual cores contain intuitive theories (Murphy & Medin, 1985). On the other hand, Barsalou (1982, 1987, 1989) saw conceptual cores as based on categorization experience. Certain properties become core properties because they are processed in conjunction with a category on so many occasions that they become automatically associated with it. Regardless of the mechanism(s) that might produce conceptual cores, the question of whether attribute centrality or importance moder-
Participants and Design Participants were 46 University of Lisbon students (26 female and 20 male) who volunteered for the study at the request of the researcher. The design of the study was a 3 (social category: gypsy, gay, and African immigrant) ⫻ 2 (perspective: cultural and individual) ⫻ 2 (session: 1 and 2) within-participants factorial.
Pretesting the Adjective Checklist A different group of 31 psychology sophomores were asked to give descriptions of three social groups (gypsy, gay, and African immigrant).1 Participants were instructed to generate a list of attributes for each social category on the basis of cultural stereotypes. From these data and for each social group, the nine most frequently mentioned attributes were selected (excluding those overlapping in meaning). The nine traits more frequently identified are in bold in Appendixes A, B, and C for gypsy, gay, and African immigrant social categories, respectively. Whenever possible, attribute antonyms were added to the list. Because in this free-response task participants almost always generated personality traits, the final list included only this type of characteristic. This task produced a final list of 43 personality traits.
Procedure All participants were tested twice as a group, with the second session following the first session by 2 weeks. To identify a participant’s answers across sessions, we asked each participant to indicate his or her birthday date and that of his or her mother, assuring anonymity. Participants were given a booklet containing the instructions and experimental materials. The instructions for the individual stereotype trait selection task read as follows:
1
In previous research (Santos, 2001), we used occupational groups (medical doctors, computer programmers, and disco bouncers) as targets. The level of instability we obtained in stereotype assembling was considerable, but it could plausibly be argued that these stereotypes are weak or relatively benevolent. We chose the target social categories because they were consensually identified by pretest participants as strong, pervasive, and clear-cut stereotypes in contemporary Portuguese society and thus provided a conservative test of the fluidity hypothesis.
STEREOTYPES AS DYNAMIC KNOWLEDGE STRUCTURES Society is composed of many different groups about whom we usually have some knowledge. In fact, the ease with which we form relatively well-defined impressions about the individuals and social groups who surround us greatly simplifies our social life. These impressions about groups are often generic, and don’t apply to every member of the group but only to a percentage of them. For instance, when we say that computer programmers are intelligent we are not saying that every computer programmer is intelligent, but that a significant percentage of them are. These generic impressions are, obviously, simplifications. They are not judgments based in objective data. And even when we recognize differences between groups, that doesn’t mean that one group is better or worse than another group. In this study, you will be asked to give impressions about some social groups. Naturally, we don’t all have the same ideas about such groups. There are no right or wrong answers. We are interested in your personal impressions, your intuitions, and your gut reactions, and not so much in what you think it is proper to say. The instructions for the cultural stereotype version of the trait selection task read as follows: Society is composed of many different groups about whom people in general have some knowledge. In fact, the ease with which people form relatively well-defined impressions about the individuals and social groups that surround them greatly simplifies their social life. On many occasions, either through hearsay or direct contact, we find out something about the impressions that people in general have about social groups. In this study, you will be asked to give your opinion about what people in general think about some social groups. Naturally, the impressions that people in general have about social groups may or may not reflect your personal beliefs. So give your answer based on what you know to be the culturally shared beliefs people in general have about those social groups, whether or not you believe those ideas to be true. In both perspective conditions, participants then had to choose and write down from the full list of 43 traits the 5 that best described each of the target groups (following Katz & Braly, 1933). Participants then made ratings of trait centrality and importance of the traits they chose. Evaluations were made on two types of 9-point scales, one ranging from 1 (not at all central) to 9 (very central), and the other ranging from 1 (not at all important) to 9 (very important). Ratings of centrality and importance were highly correlated (correlations across group replications ranged from .62 to .77, p ⬍ .001, at least) and were averaged to form a single centrality index. At the second session approximately 2 weeks later, participants were again given the same perspective instruction they received in the first session and completed the checklist again (but they did not make new centrality ratings), and they were fully debriefed and thanked.
Results and Discussion Aggregate Sample (Within-Item) Stability When the checklist methodology is used, stereotype stability is typically assessed by the correspondence between the attributes chosen to describe the social category across different studies (e.g., Devine & Elliot, 1995). We followed this procedure to compare attributes chosen across the two sessions (see Appendixes A, B, and C). Across sessions, agreement was very high (the within-item correlations varied from .91 to .97, including both cultural and individual stereotypes). A similar degree of agreement was found by Rothbart and John (1993; average group stereotype across session 4 years apart, r ⫽ .95).2 Note that, if we used only this analysis to assess stereotype stability, as previous studies have
817
done, we would find results that vindicated the abstractionist position.
Within-Participants Stability To determine the degree of overlap in the attributes used to describe social groups by any one participant across sessions, we used a common-element correlation (Bellezza, 1984a, 1984b, 1984c). To compute this value, by participant, we divided the number of common attributes generated in both sessions by the square root of the product of the total number of attributes generated in first session and the total number of attributes generated in second session (the geometric mean). This measure of correlation represents the proportion of common to total items and varies between the values of 0 and 1. The overlap values indicated that there was only moderate correspondence between the category’s attributes selected in the two sessions. Mean overlap scores ranged in value from .48 to .60 (see Table 1), indicating that only approximately half a participant’s trait selections for a category in one session were also chosen in the second session. A similar result was obtained by Rothbart and John (1993; average personal stereotypes across sessions 4 years apart, r ⫽ .50). These levels of within-participant stability in the content of social categories were generally similar to those found with nonsocial categories, using a similar experimental paradigm. In fact, when we computed the 5% confidence intervals of these commonelements correlations (following Schmidt, 1996), they overlap those found in the common objects category domain (see Table 1). Thus, despite the factors noted earlier that might have conferred even greater stability on social stereotypes, social stereotypes were no more stable than common objects concepts. These results stand in stark contrast to abstractionist expectations and reverse the picture we obtained when stability was assessed within-item aggregating across participants. A 3 (social category) ⫻ 2 (perspective) analysis of variance (ANOVA) on the mean overlap scores revealed only a significant effect for social group, F(2, 88) ⫽ 4.06, p ⫽ .021, MSE ⫽ .046, with less overlap in across-session responses about the African immigrant group (M ⫽ .49) than about the gypsy group (M ⫽ .58), with responses about the gay group falling in between (M ⫽ .53). Note that although we anticipated higher levels of stability for cultural relative to individual stereotypes, we did not find that to be the case.
Impact of Centrality on Stability As a first assessment of whether centrality affected attribute stability, we compared the average centrality of traits common across sessions versus traits unique to sessions. The results from t tests for dependent samples are shown in Table 2. 2 Rothbart and John (1993) did not assess the stability across time in the choice of traits that best describe a social group but the stability across time of ratings of how characteristic of a social group a set of traits is. Rothbart and John’s (1993) methods thus approximate a mix of the different aspects of stereotypes we studied separately in Studies 1 and 2. Although we included the reference to Rothbart and John (1993) in Study 1, their results also compare and converge with the results of Study 2.
GARCIA-MARQUES, SANTOS, AND MACKIE
818
Table 1 Within-Participant Trait Overlap Across Sessions for Personal and Cultural Stereotypes by Social Category Personal stereotype
Cultural stereotype
Social category
M
5% CI
M
5% CI
Gypsy Gay African immigrant Common taxonomic categories (from Barsalou et al., 1987)
.56 .56 .50
.36–.76 .36–.76 .38–.72
.60 .49 .48
.41–.79 .37–.71 .26–.60
Note.
.43 .24–62
CI ⫽ confidence interval.
Mean centrality was generally higher for common than for unique traits although the magnitude of these differences was relatively small and the differences did not always reach conventional statistical significance. The traits that were stable across sessions were considered more central and important, providing some evidence for the core attributes idea. To explore the role of centrality further, we performed a median split on the centrality ratings of the traits chosen by each participant and computed separate common-element correlations. To distinguish between central and peripheral traits, we computed, for each participant, the median of the centrality ratings made in the first session. Traits with ratings below the median were considered peripheral traits, and traits with ratings above the median were considered central traits. Traits with evaluations equal to the median were excluded. The total number of attributes selected in the first session was used to compute the common-element correlations. Mean overlap scores by condition are shown in Table 3.
Impact of Centrality on Accessibility Traits high in centrality were more stable across sessions than traits low in centrality. These highly central traits appear to be responsible within individuals for the moderate level of stability found across sessions. Barsalou (1982, 1987, 1989) referred to such attributes as context independent and proposed that, rather Table 2 Differences in Mean Evaluations of Centrality and Importance (Averaged) for Common and Unique Traits by Social Category, Study 1 Variable Personal Common traits Unique traits t p Cultural Common traits Unique traits t p
Gypsy
Gay
African immigrant
5.92 5.47 1.63, df ⫽ 43 .110
6.03 5.29 2.08, df ⫽ 43 .042*
5.96 5.11 2.12, df ⫽ 43 .038*
6.24 5.44 2.02, df ⫽ 43 .049*
5.75 5.46 0.74, df ⫽ 42 .462
6.12 6.20 ⫺0.26, df ⫽ 43 .79
Note. Asterisks indicate differences that reach conventional statistical significance.
Table 3 Degree of Within-Participant Trait Overlap by Social Category, Type of Stereotype, and Trait Centrality Study 1 Stereotype and trait
Gypsy
Gay
African immigrant
.38 .64
.37 .59
.54 .69
.44 .78
.32 .51
.50 .49
Personal Low centrality High centrality Cultural Low centrality High centrality
Note. A 3 (social category) ⫻ 2 (perspective) ⫻ 2 (centrality/importance) analysis of variance was performed on the mean overlap scores. Only a significant effect of centrality emerged, F(1, 12) ⫽ 7.98, p ⫽ .015, MSE ⫽ .180, indicating that, as expected, traits high in centrality were significantly more stable (M ⫽ .62) than traits low in centrality (M ⫽ .43).
than being definitional, these are simply properties that have been processed frequently with the category. If that is the case, they should be highly accessible and therefore occur early in participants’ protocols. That is, a trait’s degree of centrality should negatively predict its ordinal output position. The output position of each trait in each participant’s protocol for each social category for each type of belief was correlated with the trait’s centrality, by condition. One Pearson correlation was computed for each participant’s protocol for each social group and for each type of belief. Each correlation was then transformed to Fisher Z scores and back to Pearson coefficients so that the average correlation could be calculated. Correlations in all conditions (see Table 4) indicated a tendency for highly central traits to be retrieved early, as expected.
Impact of Stability on Accessibility Are the traits that are more accessible in the first session more stable across sessions? If so, then traits with earlier output positions in the first session should be more likely to be also generated in the second session than traits with later output positions. Bellezza (1984a) found that exemplars retrieved earlier were more likely to be common to both sessions than exemplars retrieved later. Barsalou et al. (1987) also reported that common traits appeared earlier in participants’ protocols than did unique properties. To explore this idea, we performed a median split on the output position of the traits chosen by each participant and computed separate common-element correlations. To distinguish between traits with early and late output positions, we computed for each participant the median on the basis of the output positions of the traits chosen in the first session. Traits with output positions below the median were considered early traits, and traits that appeared Table 4 Average Correlation Between Trait Centrality and Trait Output Position by Social Category and Stereotype Stereotype
Gypsy
Gay
African immigrant
Personal Cultural
⫺.12 ⫺.25
⫺.18 ⫺.29
⫺.13 ⫺.19
STEREOTYPES AS DYNAMIC KNOWLEDGE STRUCTURES
after the median were considered late traits. Traits at the median position were excluded. The total number of attributes selected in the first session was used to compute the common-element correlations. Mean overlap scores by condition are shown in Table 5. A 3 (social category) ⫻ 2 (perspective) ⫻ 2 (output position) ANOVA on the mean overlap scores revealed a significant effect of output position, F(1, 43) ⫽ 17.30, p ⫽ .0001, MSE ⫽ .138. As expected, traits with early output positions were significantly more stable (M ⫽ .60) than traits with late output positions (M ⫽ .46). More accessible traits were more stable across sessions. Output position also interacted with perspective, F(1, 43) ⫽ 8.25, p⫽ .006, MSE ⫽ .083. Accessible traits were more stable (M ⫽ .63) and less accessible traits less stable (M ⫽ .42) for cultural stereotypes than for individual stereotypes (.57 and .50, respectively). The results of Study 1 revealed about the same moderate level of trait stability in social category representations as other researchers have found with nonsocial categories (for reviews, see Barsalou, 1987, 1989; Barsalou & Medin, 1986). When a participant was asked to choose five traits that best describe a group from a list of 43 traits, the chance that he or she would include one trait across time was approximately equal to the probability of including it at only one time. This result held for both personal and cultural stereotypes. These results are clearly at odds with early abstractionist assertions, which might predict perfect reliability. Moreover, had we looked only at across-participants within-item agreement, as previous research did (Devine & Elliot, 1995), our conclusions would be quite the opposite. Although these results clearly clash with abstractionist positions about stereotype stability, they fit quite well with exemplar, mixed model, and connectionist alternative views. Not all kinds of stereotype content were equally unstable, however. In fact, more central attributes were found to be significantly more stable than peripheral ones, though the stability of even highly central traits was still far from what a full-fledged abstractionist perspective would lead us to expect. Traits with early output positions were also more stable within participants than traits with late output positions, indicating that accessible traits within sessions were also accessible across sessions for a given individual. Most of the instability of stereotype content across sessions was due to attributes considered to be less central and/or important. Such traits appear to be relatively inaccessible and thus have late output positions. These results are consistent with theories that postulate the existence of a category conceptual core, whether this core is definitional in nature (Armstrong et al., 1983; Osherson & Smith, 1981) or more experience-based (Barsalou, 1982, 1987,
Table 5 Within-Participant Stability by Social Group, Type of Stereotype, and Trait Output Position Stereotype and output position Personal Early trait Late trait Cultural Early trait Late trait
Gypsy
Gay
African immigrant
.57 .54
.63 .47
.51 .51
.72 .43
.58 .43
.59 .40
819
1989). We defer further consideration of this issue to the General Discussion.
Study 2 Having demonstrated only moderate stability across time in stereotype content, we next assessed the degree of stability in the graded structure of social stereotypes. Graded structure (also known as typicality or exemplar goodness) reflects the extent to which different exemplars represent the category. Within a category, graded structure refers to the gradient of representativeness that exists across category members. Graded structure has been found in a wide variety of categories, and a natural category without such structure has yet to be identified. Common taxonomic categories (e.g., fruit, birds, furniture; Rosch and Mervis, 1975), formal categories (e.g., odd numbers, square root; Armstrong et al., 1983), and ad hoc categories and goal-derived categories (Barsalou, 1983, 1985) all show graded structure. E. E. Smith and Medin (1981) and Medin and Smith (1984) have shown that graded structure predicts performance in a number of fundamental categorization tasks. The graded structure of nonsocial categories has been shown to be dynamic. Depending on the population, individual, or context, the perceived typicality of category exemplars can vary widely (e.g., Barsalou, 1987). Barsalou and Medin (1986; see also Barsalou, 1985, 1987) argued that changes in the graded structure of a category provide further evidence for the fluid nature of category representation. Because an exemplar’s typicality rating reflects its similarity to the category representation, changes in category representation would cause instability in category– exemplar similarity and thus instability in typicality ratings and thus graded structure (Barsalou, 1989). Barsalou and Medin (1986) explored the extent to which any individual produces the same graded structure for a category across time but in the same context. To assess agreement, they correlated participants’ graded structure on 1 day with their graded structure in the same context a few weeks later. Participants’ exemplar typicality ratings in one session accounted for only about 64% of the variance in the later session. Because individuals showed much higher stability after only 1 hour’s delay, the instability at longer delays could not be attributed to measurement error or some other source of random variability. Instead, it appeared to reflect changes in the individual’s representation of the category between the sessions. Is graded structure important for stereotypes? The answer is clearly yes. Such a typicality gradient is a major component of a very influent exemplar theory of stereotype representation and change (Rothbart & John, 1985). According to this theory, differences in typicality help to perpetuate stereotypes because typical exemplars have much higher probability of being retrieved and used in categorization-based tasks (Rothbart & Lewis, 1988; Rothbart, Sriram, & Davis-Stitt, 1996). In addition, changes in graded structure (i.e., an increase in the perceived typicality of atypical exemplars) are considered the most reliable signs of stereotype change (Maurer, Park, & Rothbart, 1995; see also Garcia-Marques & Mackie, 1999). On this basis, we again expected that graded structure in stereotypes might exhibit higher within-participant stability than the graded structure of common objects concepts.
GARCIA-MARQUES, SANTOS, AND MACKIE
820
Our second study assessed the extent to which social categories show within-participant stability in graded structure across time. By including manipulations of exemplar typicality, we could also assess whether typicality moderated this stability, replicating the moderating effects of attribute centrality found in Study 1. We did not manipulate belief type. Given that in Study 1 we found few differences in personal and cultural stereotypes, we asked only about cultural stereotypes because many participants feel more comfortable reporting cultural than individual stereotypes (Devine & Elliot, 1995).
ings in Session 1 and the same typicality ratings made in the same context 4 weeks later. The 32 participant correlations, computed for each social group, were transformed to Fisher Z scores and back to Pearson coefficients so that average correlations could be calculated. Across the three social groups, we found that the average correlation was around .67. This moderate level of agreement across time was not higher than the level found with nonsocial categories (Barsalou et al., 1986), such that the respective 5% confidence intervals overlap. Table 6 shows the average correlation scores by social group.
Method Participants and Design Participants were 32 (23 female and 9 male) University of Lisbon students who volunteered for the study. The design was a 2 (social category: gypsy, gay, and African immigrant) ⫻ 2 (session: 1 and 2) within-participants factorial.
Pretest Sixty participants first generated short descriptions of the first five exemplars, either real or imagined, that came to mind for each social category. Forty other participants then rated the typicality of each exemplar description on a 9-point rating scale ranging from 1 (extremely uncharacteristic of the group) to 9 (extremely characteristic of the group).3 Participants were told to use the cultural stereotype—the views that people in general had about these groups—in making their ratings. On the basis of these ratings, we then selected the eight most consensually agreed atypical, eight moderately typical, and eight typical exemplars for each category for use in the study.
Procedure Participants received a booklet with each of the 24 exemplars descriptions typed on a single page, in one set order. Participants judged the typicality of each exemplar with the following instructions: Society is composed of many different groups about whom we usually have some knowledge. Most of us have a certain mental image about what is typical or characteristic of each group, and, generally, we are able to identify examples that are typical of the group. Some members can be more similar to what is typical of the group than others. In this study, some examples of a social group’s members will be presented and you will be asked to evaluate how good an example each exemplar is of his or her group. We are not so much interested in your personal opinions but in what are considered true by people in general. So it is absolutely essential that you assume the point of view of people in general while judging the typicality of exemplars. There are no right or wrong answers. We are interested in your first impressions, intuitions, and gut reactions, and not so much in what you think it is proper to say. Participants were given as much time as necessary to complete the task. Participants were asked to rate the typicality of each exemplar of each of the three target groups on the 9-point rating scale. Participants performed the same tasks again approximately 4 weeks later and were fully debriefed after the session.
Results and Discussion Within-Participants Correlation by Social Group To assess within-participant agreement, we calculated for each participant Pearson correlation coefficients between typicality rat-
Within-Participants Correlation by Exemplars Levels of Typicality To distinguish between atypical, moderately typical, and typical exemplars, we computed, for each exemplar, the 33rd and 66th percentile of the distribution of typicality judgments made by all participants in the first session. Exemplars with a typicality rating below the 33rd percentile were considered atypical, exemplars between 33rd and 66th percentiles (inclusively) were considered moderately typical, and exemplars with a typicality rating above the 66th percentile were considered typical. To assess the degree of correlation between the graded structures generated in the first and second sessions, we calculated within-participants Pearson correlation coefficients by degree of typicality. Average within-participant correlation scores increased from .54 for atypical exemplars, through .58 for moderately typical exemplars, to .62 for typical exemplars, indicating that withinparticipant stability of typicality ratings increased with exemplar typicality. In sum, participants showed only moderate reliability in the exemplar typicality ratings they generated 4 weeks apart. These results indicate levels of instability in the graded structure of social categories similar to those found with nonsocial categories. Stability increased linearly with exemplar typicality, a finding that diverged from Barsalou et al.’s (1986) finding that typicality instability peaked with moderately typical exemplars. This may suggest that slightly different mechanisms underlie perceived typicality of social and nonsocial exemplars. Alternatively, the slightly different methodology used in the two studies may be responsible for this difference. Regardless, it is clear that instability in graded structure does not come only from perceptions of atypical exemplars— considerable instability was found even in ratings of typical exemplars.
Study 3 As a third assessment of fluidity in social category representation, we examined the reliability of retrieval of social category exemplars. In the nonsocial domain, Bellezza (1984a) demon3 Instead of asking participants to judge how typical each exemplar is, some researchers (Barsalou, 1985) ask participants to rate how good an example a given examplar is of its category because typicality might bias participants toward using frequency of instantiation (people’s estimates of how often they have encountered an exemplar as a category member). However, asking how bad or good something is might also have evaluative connotations.
STEREOTYPES AS DYNAMIC KNOWLEDGE STRUCTURES
Table 6 Within-Participant Correlation Across Sessions by Social Category Within-participant correlation Social category
M
5% CI
Gypsy Gay African immigrant Common taxonomic categories (Barsalou et al., 1986)
.72 .58 .70
.55–.89 .35–.81 .52–.88
.70
.54–.86
821
Typicality judgments. After completing the 15 descriptions, participants evaluated each exemplar’s typicality on a 9-point rating scale ranging from 1 (extremely uncharacteristic of the group) to 9 (extremely characteristic of the group). Second exemplar retrieval task. Approximately 2 weeks later, participants returned and completed the exemplar retrieval task again. At the end of the second session, each participant was given the descriptions he or she had generated in the first session and asked to say which across-sessions descriptions depicted the same category member. Thus, we coded a second session description as repeated across sessions whenever the participant who generated it identified it that way. Note that this procedure set a conservative standard for instability, working against our hypothesis.
Results and Discussion
Note. CI ⫽ confidence interval.
Number of Exemplars Generated strated some representational fluidity when he asked participants to generate category instances in two recall tests 1 week apart. He found that only about 69% of the category instances generated in the first session were also generated in the second session. Does exemplar retrieval from stereotypical social category representations show a similar degree of instability? This question is critical for exemplar-based categorization models (e.g., E. R. Smith & Za´rate, 1992) because these models argue that judgments about a category’s most characteristic attributes are based on the attributes of exemplars activated at the time (i.e., Bodenhausen, Schwarz, Bless, & Wanke, 1995). Mixed model perspectives also of course understand instability in stereotypes as deriving from instability in exemplar retrieval rather than from changes in the abstracted components of the representation. Thus if exemplar retrieval within-participants exhibited a much higher degree of stability than the level already found for property verification (Katz & Braly’s (1933) task) or graded structure, exemplar and mixed model views would be hard pressed to account for such findings. Abstractionist views would of course have little to say about the reliability of exemplar retrieval. Thus, on the basis of the results of Studies 1 and 2, we expected a similar degree of instability in exemplar retrieval as we found in stereotype attribute assignment (Study 1) and graded structure (Study 2). We again assessed this index of representational stability at various levels of exemplar typicality.
Although participants were asked to generate about five exemplars, some generated more than this and some fewer. An ANOVA on the mean number of exemplars retrieved by session and category (see Table 7, Columns 1 and 2) revealed a significant main effect for social category, F(2, 128) ⫽ 13.47, p ⫽ .0001, MSE ⫽ .609. Participants produced more exemplars for the gypsy target group (M ⫽ 4.80) than for the other two social categories, F(1, 64) ⫽ 17.69, p ⫽ .0001, MSE ⫽ .925, which did not differ one from another (M ⫽ 4.38 vs. M ⫽ 4.36), F(1, 64) ⫽ .118, p ⫽ .731, MSE ⫽ .292.
Within-Participants Reliability To determine the degree of overlap in the exemplars retrieved by any one participant across the two sessions, we computed common-element correlations for each social group by participant (Bellezza, 1984a, 1984b, 1984c). The number of common exemplars (as identified by the participant) retrieved in both sessions was divided by the square root of the product of the total number of exemplars retrieved in the first session and the total number of exemplars retrieved in the second session. Mean overlap scores ranged from .55 to .62 (see Table 7, Column 3), indicating the modest reliability with which social category exemplars are retrieved from memory. Again, this level of within-participant instability was about the same as that found with nonsocial categories, using an equivalent experimental paradigm (Bellezza, 1984a),
Method Participants and Design Participants were 65 University of Lisbon students (40 women and 25 men) who volunteered for the study in return for partial course credit. The design was a 3 (gypsy, gay, and African immigrant) ⫻ 2 (Session 1 and Session 2) within-participants factorial.
Table 7 Mean Number of Exemplars Generated and Across Session Stability by Social Category Withinparticipant overlap
Procedure Participants were tested in small group sessions of up to 10 members. General instructions were the same as in Study 1. Participants were debriefed after Session 2. First exemplar retrieval task. Participants were asked to generate descriptions of five (more or less) different members of each of the three social categories, always in the same order: gypsy, gay, and African immigrant. Participants described each exemplar, which could be a real or imaginary person, in a few sentences.
Social category
Session 1
Session 2
M
5% CI
Gypsy Gay African immigrant Common taxonomic categories (Bellezza, 1984a)
4.83 4.37 4.34
4.77 4.38 4.37
.55 .62 .55
.38–.72 .47–.77 .38–.72
.69
.51–.87
Note.
CI ⫽ confidence interval.
GARCIA-MARQUES, SANTOS, AND MACKIE
822
such that their respective 5% confidence intervals overlap (see Table 7).
Typicality Judgments of Retrieved Exemplars Paralleling the procedures used in Study 1, we assessed whether the exemplars that were common versus unique across sessions differed in typicality. To do so, arithmetic typicality means were computed for the set of common exemplars and also for the set of unique exemplars. The results of t tests for dependent samples are shown in Table 8. Results supported the findings from Study 1: Exemplar typicality was a good predictor of stability in exemplar production, at least in two of the groups. For the gypsy and gay categories, common exemplars (generated in both sessions) were judged as significantly more typical than uniquely generated instances.
Within-Participants Reliability Concerning High and Low Typical Exemplars Each participant’s median typicality rating in the first session was computed. Exemplars with typicality ratings below the median were considered atypical exemplars and exemplars with typicality ratings above the median were considered typical exemplars. Exemplars with ratings equal to the median were excluded. Commonelement correlations were then computed as described earlier. A 3 (social category) ⫻ 2 (exemplar) ANOVA revealed a significant effect of typicality, F(1, 27) ⫽ 18.81, p ⫽ .0002, MSE ⫽ .158, showing that highly typical descriptions (.77) were more likely to be generated across sessions than less typical descriptions (.50; see Table 9).
Correlation of Degree of Typicality Attributed to Exemplars and Output Position Following the same reasoning applied to traits in Study 1, we expected highly typical exemplars to be more accessible and therefore to be generated earlier than less typical exemplars, causing exemplar typicality and exemplar output position to be negatively related. The output position of each exemplar generated by each participant in the first session was correlated with that exemplar’s typicality rating for the relevant social group. Each correlation computed was transformed to Fisher Z scores and back to Pearson coefficients so that average correlations could be calculated. Correlations in all conditions indicated that, as expected, typical exemplars tend to be retrieved early (r ⫽ ⫺.59 for gypsy exemplars, r ⫽ ⫺.62 for gay exemplars, and r ⫽ ⫺.36 for African immigrant exemplars). Table 8 Mean Typicality of Common and Unique Exemplar Descriptions by Social Category, Study 3 Exemplar descriptions
Gypsy
Gay
African immigrant
Common Unique t (difference) p
6.32 5.51 1.98, df ⫽ 64 .052
6.02 4.28 4.47, df ⫽ 64 .000
6.37 5.94 1.17, df ⫽ 64 .243
Table 9 Within-Participant Stability by Social Category and Exemplar Typicality Exemplar typicality
Gypsy
Gay
African immigrant
Low High
.52 .65
.46 .88
.53 .78
Within-Participant Reliability for Accessible and Less Accessible Exemplars Following Bellezza (1984a), we also tested the idea that accessible exemplars (those output early) in the first session were more likely to be generated in both sessions. To distinguish exemplars with early and late output positions, we computed, for each participant, the median output positions of the exemplars generated in the first session. Exemplars with output positions below the median were considered accessible, and exemplars with output positions above the median were considered less accessible. Common exemplars with evaluations equal to the median were excluded. The total number of exemplars generated in the first session was used to compute the common-element correlations. Mean stability scores by condition are shown in Table 10. A 3 (social category) ⫻ 2 (exemplar) ANOVA indicated that, as expected, exemplars with early output positions had a significantly higher stability (.63) than exemplars with late output positions (.51), F(1, 64) ⫽ 11.33, p ⫽ .001, MSE ⫽ .126. In sum, the results of the exemplar retrieval task replicated the results of Studies 1 and 2, with social category representations revealing levels of instability similar to those found with nonsocial categories. Once again, exemplar typicality affected stability, with more representative exemplars showing higher stability across time. Corroborating these results, within-participants reliability revealed more typical exemplars to be more stable over time. Once again, however, instability was not confined to the atypical exemplars: even the most typical exemplars showed some level of instability.
Study 4 Although we found a consistent picture across the three previous studies, it is possible that the degree of instability we obtained was, at least partially, due to measurement error or to participants’ deliberate monitoring of their responses. If participants felt that, for some reason, they should vary in the second session the responses they had given to identical instructions 3 or 4 weeks earlier, the level of instability we obtained might represent little more than an artifact derived from repeated measurement. Table 10 Within-Participant Stability by Social Category and Output Position Output position
Gypsy
Gay
African immigrant
Early Late
.56 .51
.71 .56
.64 .47
STEREOTYPES AS DYNAMIC KNOWLEDGE STRUCTURES
823
Our primary goal in the fourth study was to assess this possibility. We did so in two different ways. First, to show that stereotype instability is a result of context sensitivity and not simply a byproduct of measurement error, we manipulated the consistency of the context in which participants identified the traits associated with a given social stereotype. We did so by having participants perform an exemplar typicality rating task immediately before completing the stereotype trait selection task used in Study 1 both during Session 1 and Session 2. The rated exemplars were either equivalent or nonequivalent in typicality across sessions. If the attributes participants chose across sessions reflected only random variation, this manipulation should have no effect. In contrast, if stereotype fluidity reflects context sensitivity, variation in stereotype attribute choice would be driven by the context manipulation. Second, to assess the possibility that the variation in responses obtained across sessions in Studies 1–3 was due to participants’ deliberate attempts to change their responses, we compared their spontaneous responses with their attempts to accurately recall the choices they had made during the first session. Finally, we also wanted to assess the subjective or perceived stability of stereotypes to see whether our participants were aware of the fluidity in their social category representations or whether they perceive illusory consistency in their stereotypes (Ross & Conway, 1986).
We manipulated the equivalence of the exemplars rated for typicality across the two sessions. In the equivalent condition, the three exemplars rated in Session 2 were different from but equivalent in typicality to the ones judged in Session 1 (that is, participants saw either typical or atypical exemplars in both sessions). In the nonequivalent condition, the three exemplars rated in Session 2 were also different from, as well as nonequivalent in typicality to, those rated in Session 1 (that is, participants who first rated typical exemplars now were asked to rate atypical exemplars and vice versa). Stereotype attribute selection task. The stereotype attribute task always immediately followed the typicality rating task. As described in Study 1, participants had to choose from a list of 43 personality traits the five traits that best described each of the target groups. After choosing the traits, participants rated the centrality to the relevant stereotype of each of their chosen traits. Evaluations were made in a 9-point scale ranging from 1 (not at all central) to 9 (very central). All participants completed the attribute choice task immediately after the typicality rating task in Session 1. In Session 2, approximately 4 weeks later, half the participants repeated the attribute choice task and then estimated the number of traits that they had just chosen, which they had also chosen in the first session. These estimates served as a measure of subjective overlap. Memory task. The other half of the participants in Session 2 (all of whom had completed the stereotype attribute task in Session 1) were asked to reproduce the choices they had made in the first session as best they could.
Method
Results and Discussion
Participants and Design
Within-Participant Stability
Participants were 124 University of Lisbon students (71 female and 53 male) who volunteered for the study at the request of the researchers in return for partial course credit. The design of the study was a 2 (session: 1 and 2) ⫻ 2 (social category: gypsy and African immigrant) ⫻ 2 (exemplar: typical or atypical, rated in Session 1) ⫻ 2 (exemplar typicality: equivalent or nonequivalent across sessions) ⫻ 2 (task: stereotype attribute selection or recall in Session 2) mixed factorial design, with the first two factors being within-participants.
As in Study 1, we computed common-element correlations (overlap scores) from the choices of each participant across sessions (Bellezza, 1984a; McNemar, 1969). We then computed a 2 (session: 1 and 2) ⫻ 2 (social category: gypsy and African immigrant) ⫻ 2 (exemplar: typical or atypical, rated in Session 1) ⫻ 2 (exemplar typicality: equivalent or nonequivalent across sessions) ⫻ 2 (task: stereotype attribute selection or recall in session 2) mixed model ANOVA on these overlap scores. Only two significant main effects emerged. The first was an equivalence main effect, F(1, 113) ⫽ 5.27, p ⫽ .024, MSE ⫽ .044, such that participants who rated two sets of exemplars equivalent in typicality across sessions showed more stability in their stereotype attribute choices across session (M ⫽ .51) than participants who rated exemplars who differed in typicality across sessions (M ⫽ .45). The second significant main effect was a second session task effect, F(1, 113) ⫽ 54.00, p ⫽ .000, MSE ⫽ .044, showing that participants who performed the stereotype attribute selection task a second time were much more consistent (M ⫽ .59) than participants who were intentionally trying to reproduce their first session choices (M ⫽ .38). The superior consistency of participants who performed the stereotype attribute selection task a second time over those who tried to recall their choice may seem surprising, but such results are far from unprecedented. Repeating the selection task a second time may be considered as an implicit memory test of the choices made
Procedure All participants were tested twice as a group with the second session following the first session by 1 month. Participants were tested in small group sessions of up to 10 people. In Session 1, all students performed a typicality judgment task, completed the stereotype attribute selection task described in Study 1, and evaluated the centrality of each trait they selected. In Session 2, the same students performed a second typicality judgment task and then either repeated the stereotype attribute selection task or attempted to recall their responses from Session 1. Participants who repeated the trait selection task also estimated the number of traits chosen in Session 2 that they had also chosen in Session 1. Manipulation of context. To manipulate the context in which participants chose stereotype attributes, participants were first required to complete an exemplar typicality rating task. In both Session 1 and Session 2, participants first rated the typicality of three exemplars from the social category of gypsy and three from the social category of African immigrant. Participants received a booklet with each exemplar description typed on a single page in one set order. Instructions were as described in Study 2. Participants rated the typicality of each exemplar of each of the two groups on a 9-point rating scale ranging from 1 (extremely uncharacteristic of the group) to 9 (extremely characteristic of the group). The three-exemplar sets were either typical (two typical exemplars and one neutral exemplar) or atypical (two atypical exemplars and one neutral exemplar). Exemplars were selected from the pretest previously described in Study 2.4
4 A group of pretest participants generated descriptions of exemplars for each social category. Forty other participants then rated the typicality of each exemplar description on a 9-point rating scale. On the basis of these ratings, we then selected the most consensually agreed-on atypical, neutral, and typical exemplars for each category for use in the study.
824
GARCIA-MARQUES, SANTOS, AND MACKIE
the first time. As such, the comparison between participants who repeated the selection task or who try to recall the choices previously made can be equated with comparison between an implicit and an explicit memory test of the choices made in the first session after a 3 weeks interval. The amazing superiority of implicit over explicit memory tests over extended periods of time is one of the dissociative features of these two types of tests (E. R. Smith, 1990). For instance, Kolers (1976) showed that 1 year after some practice in reading inverted text, participants would read faster the inverted pages they had been presented with than new inverted text pages even though they were totally incapable of recognizing them. These results thus argue against taking variation in participants’ identification of stereotypic content across sessions as a function of either measurement error or deliberate changes in responding. On one hand, the equivalence main effect showed that variation in the stereotype attribution task reflected meaningful context sensitivity and not simply measurement error. On the other hand, the fact that participants trying to remember their first session choices were much less consistent than those repeating the stereotype attribute task casts severe doubt on the plausibility of alternative explanations that depend on participants’ explicit memory of their original responses.
Impact of Centrality on Stability As in Study 1, to explore the role of centrality on choice stability, we performed a median split on the centrality ratings of the traits chosen by each participant and computed separate overlap scores. We then computed a 2 (session) ⫻ 2 (social category) ⫻ 2 (Session 1 exemplar typicality) ⫻ 2 (exemplar typicality equivalence across sessions) ⫻ 2 (Session 2 task) ⫻ 2 (trait) mixed model ANOVA on the overlap scores. To avoid redundancy with the previous analysis, we report only effects involving centrality. Replicating Study 1, a main effect emerged for this factor, F(1, 83) ⫽ 65.45, p ⫽ .000, MSE ⫽ .119. Participants showed more overlap across sessions for central (M ⫽ .58) than for peripheral attributes (M ⫽ .29). Of more interest, two 2-way interactions qualified this main effect. The first was a significant Equivalence ⫻ Centrality interaction, F(1, 83) ⫽ 4.46, p ⫽ .038, MSE ⫽ .119. This effect was due to the fact that exemplar typicality equivalence across sessions had a much greater impact on the stability of central attributes chosen across sessions (Mequivalent ⫽ .66 vs. Mnonequivalent ⫽ .50) than it did on the stability of peripheral traits (Mequivalent ⫽ .29 vs. Mnonequivalent ⫽ .28). This finding suggests that fluidity in central compared with peripheral traits may reflect different processes. Whereas fluidity in central trait choices may reflect context sensitivity rather than measurement error, variation in peripheral trait choices may reflect the opposite. Future research should address this possibility. Moreover, note that context sensitivity is not always an obstacle to stereotype stability. For instance, when the context remained constant, stereotype stability across sessions was very high (M ⫽ .89). The second significant interaction was between second session task and centrality, F(1, 83) ⫽ 12.88, p ⫽ .001, MSE ⫽ .119. This effect was due to the fact that the stability of central traits was much higher (M ⫽ .79) than for peripheral traits (M ⫽ .38) for participants who repeated the stereotype attribute task than for
participants trying to reproduce their first session choices (Mcentral ⫽ .36 vs. Mperipheral ⫽ .21). This effect attests to the critical differences between naturally performing the stereotype task and performing it by trying to explicitly remember previous responses. In turn, this suggests that any within-participant fluidity that occurs in reported stereotype content is unlikely to have been influenced by explicit memory for earlier reactions.
Subjective Stereotype Stability Participants who performed the stereotype attribute task twice were also asked to estimate the number of traits chosen in the second session that they also chose in the first session. We used these estimates to compute “subjective” common-element correlations (subjective overlap scores). These subjective overlap scores could therefore be contrasted with “objective” overlap scores (on the basis of the actual chosen traits across sessions). To compare subjective and objective overlap scores, we computed a 2 (session) ⫻ 2 (social category) ⫻ 2 (Session 1 exemplar typicality) ⫻ 2 (exemplar typicality equivalence across sessions) ⫻ 2 (overlap) mixed model ANOVA. Again to avoid redundancy with previous analyses, we report only effects related to the overlap factor. A main effect emerged, F(1, 60) ⫽ 7.63, p ⫽ .008, MSE ⫽ .031, attesting to the fact that participants overestimated the degree of overlap (M ⫽ .65) in their responses across sessions relative to the actual degree of overlap (M ⫽ .59).5 This effect may help to explain why sometimes our intuitions regarding stereotype consistency are much stronger than our data. Whereas we may be inconsistency avoiders at heart, our cognitive system may be able to accommodate more variability than we give it credit for. Note that these results echo those found in the perceived attitude stability domain. Participants who are experimentally induced to significantly change their attitudes on critical issues report little or no attitude change at all (Bem & McConnell, 1970; Goethals & Reckman, 1973; Ross & Shulman, 1973). It may be the case that the illusion of stability is not incompatible with highly contextsensitive knowledge structures—it may even be a requirement of such structures. When we perceive our knowledge as instable or variable, we perceive it as having little validity and cease regarding it as a valuable guide for behavior (Bem, 1972; Kelley, 1973).
General Discussion The primary goal of these studies was to assess withinparticipant stability in the content and use of social category representations over time. Stability was assessed across sessions, 2 to 4 weeks apart, in a within-participant manner. We assumed natural variations in the idiosyncratic judgment contexts of our participants, such that particular moods, particular thoughts, recent experiences, and so forth would vary quite broadly across sessions. Such natural variability provided the appropriate conditions to test abstractionist intuitions concerning a high degree of stability in stereotyping against alternative positions. The results of the first three studies reported here provide considerable evidence of instability in the representation of both A noninterpretable four-way interaction also emerged, F(1, 60) ⫽ 4.94, p ⫽ .030, MSE ⫽ .015. 5
STEREOTYPES AS DYNAMIC KNOWLEDGE STRUCTURES
personally and culturally held social categories. First, we found low within-participant stability in the selection of traits as stereotypic of a social category (Study 1), in the typicality ratings of category members (Study 2), and in category exemplars retrieved (Study 3). Second, the level of within-participant instability we obtained was only partially accounted for by feature or exemplar centrality or importance. Instability greatly decreased for more central or important trait attributes (Study 1) and for more typical group members (Studies 2 and 3). However, even for more central attributes and typical exemplars, considerable instability was still present. Third, although accessible attributes (Study 1) and accessible exemplars (Study 3) were more stable across time, even they showed considerable instability. Accessibility and centrality, and accessibility and typicality, were both moderately correlated. The results of the fourth study are critical for three reasons. First, they provide data against the idea that fluidity in stereotype content across sessions is simply a byproduct of measurement error. Stereotype stability was a function of context constancy (particularly in the case of traits deemed central), suggesting that stereotype fluidity is likely to reflect sensitivity to the natural variation in the contexts in which stereotypes are constructed and used, rather than reflecting measurement error. Second, the results are equally discouraging for the notion that participants may consciously avoid making the same choices across sessions. The fact that our participants showed poor explicit memory for their first session choices suggests that explaining variation in stereotype construction as response monitoring or response editing is implausible. Finally, the results of Study 4 also suggest that within-participant stereotype consistency may be illusorily overestimated. To end this section, a final caveat is in order. Note that although we found considerable evidence for within-participants stereotype instability, we also found considerable evidence for betweensessions stability (see Study 1). We in no way claim that the question of stereotype stability should be addressed only from an intraindividual perspective. Both intraindividual and betweensample perspectives are important and complementary. We simply argue that stereotype research has addressed the question of stereotype stability in a way that pays too little attention to the intraindividual side of the coin. With our research, we attempted to contribute a more balanced approach.
825
Partial Retrieval and/or Variable Activation According to some exemplar approaches to categorization (e.g., Barsalou, Huttenlocher, & Lamberts, 1998; Kahneman & Miller, 1986; Nosofsky & Palmeri, 1997), knowledge structures are not represented by abstractions but instead by the whole class of stored exemplars. At any particular moment, and dependent on the demands of the cognitive task at hand, a subset of exemplars is retrieved so that a category judgment can be made or the processing of new exemplars facilitated. (Alternatively, exemplar retrieval may correspond to a continuous activation function, see Hintzman, 1986.) Context sensitivity is to be expected from such exemplarbased perspectives because category judgments and exemplar processing depend on the idiosyncratic nature of the small subset of exemplar retrieved. Models of attitudes as temporary constructs use a similar reasoning, contending that the stored information relevant to any given attitude is likely to be both extensive and contradictory and that at any given point in time, only a small portion of this information is retrieved and used to compute an attitude (Lord & Lepper, 1999; Wilson & Hodges, 1992). Partial retrieval thus accounts for stereotype instability. Although exemplar models would seem most compatible with partial retrieval, it is quite possible to envisage an abstraction-based model in which only part of the concept-relevant stored knowledge is retrieved depending on the situation (see Barsalou, 1999, 2002). Thus a partial retrieval temporary-abstractions model would also exhibit the required property of context sensitivity.
Situated Knowledge If people learn about categories and concepts in a succession of episodes and if category knowledge becomes grounded in situational knowledge, then it makes sense to assume that different situations activate different categorical knowledge (Yeh & Barsalou, 1996, 2002). For stereotypes, it is quite plausible to predict that the kind of stereotypic knowledge that is activated by the presence of a given gay person or gypsy should be quite different depending on the context in which information was gathered. Some exemplar models possess such a plasticity-inducing mechanism (e.g., Kruschke, 1992; Medin & Schaffer, 1978; Nosofsky, 1987), but again, abstraction-based models could integrate it (Barsalou, in press).
Compound Retrieval Cues Dynamic Alternatives to Enduring Abstractionism This pattern of results is clearly inconsistent with early abstractionist views—and perhaps with lay theories about stereotypes— but is quite consistent with more modern views of stereotype representation (exemplar, mixed models, or connectionist frameworks). In fact, despite the unique nature of social stereotypes, the general level of context sensitivity found in these studies is quite comparable with the results obtained from studies of common object concepts. Thus, although some of the content of the stereotypes studied here seemed relatively context independent, exhibiting greater accessibility and stability across time and context, a lot of our participants’ knowledge about the stereotypes was quite fluid. In this final section, we suggest some mechanisms that need to be integrated into views of social category representation to better account for their apparent fluidity.
The context dependent nature of categorical knowledge may derive from the compound nature of the retrieval cues that are used to retrieve such knowledge. According to global matching models (e.g., Hintzman, 1986), for example, self-assembled or context available cues are spontaneously integrated, forming compound retrieval cues that can be matched against stored memories for an output judgment (Dosher & Rosedale, 1989; McKoon & Ratcliff, 1992; Ratcliff & McKoon, 1988). Thus even if stereotypes are stable representations, the process of assembling or retrieving them in any given situation may be unreliable, in the sense that it is inherently dependent on the momentarily available situation (Medin & Ortony, 1989; E. R. Smith, 1989).
Source of Activation Confusion When a stereotype is constructed or retrieved, activation is a consequence of matching (attributes or group members more fre-
GARCIA-MARQUES, SANTOS, AND MACKIE
826
quently associated with other group cues become more activated). However, contextually activated attributes may also affect stereotype construction because we are usually aware only of the consequences of memory trace activation and not of its source (Ayers & Reder, 1998; Reder & Schunn, 1996). Thus, information that is not usually part of a stereotype may inadvertently be incorporated into one because of its momentary salience.
Summary All four of these proposed mechanisms describe processing mechanisms rather than representational formats. In that sense, they are compatible even with abstraction-based representations. And thus, it is quite possible to argue for the necessity of maintaining a role for abstractions in our conceptions of stereotypes but, at the same time, to acknowledge the situated and context sensitive nature of stereotypic representations. Like our data, these proposed cognitive mechanisms are incompatible, however, with assumptions of invariant retrieval of nonsituated stereotypic knowledge that dominated early social cognition-based research on stereotypes and that still heavily contaminate lay and intuitive theories of stereotypes. These four mechanisms are not necessarily incompatible and only further research can allow us to better articulate them and to choose among them. Our data are much more consistent with the idea that stereotypes, despite being widely shared and highly resilient, are nevertheless temporary constructs that are assembled in a flexible and context sensitive way to meet situational requirements and the perceiver’s goals. Across time and contexts, stereotypes may differ markedly, as a function of the variability of those times and contexts in which stereotypes are brought to mind. This view converges broadly with exemplar, mixed model, and connectionist perspectives on social categories, and also with recent positions taken in the cognitive literature (Barsalou, in press). Our findings suggest that considerable flexibility and fluidity in stereotype knowledge is possible. If we are in fact so capable of cognitive sensitivity and flexibility, why does the social information processor seem so consistency prone, avoiding dissonance, inconstancy, and incongruence? It may be that our apparent cognitive stability derives not from impoverished inputs (abstractions) to our cognitive system but from impoverished inputs constrained by our social environments. The world is a big and diverse place, but we choose to live in encapsulated noneventful social worlds where like-minded people live out similar lives. It is therefore possible that we have mistaken social consistency for cognitive consistency. If so, such an acknowledgement allows for a new perspective on stereotypes, one that recognizes the situated nature of social cognition and conceives information processing as a socially distributed network process (Clancey, 1997; Higgins, 2000; Wegner, 1995).
References Abelson, R. P. (1994). A personal perspective on social cognition. In P. G. Devine, D. L. Hamilton, & T. M. Ostrom, Social cognition: Impact on social psychology (pp. 15–39). San Diego, CA: Academic Press. Armstrong, S. L., Gleitman, L. R., & Gleitman, H. (1983). On what some concepts might not be. Cognition, 13, 263–308. Ashmore, R. D., & DelBoca, F. K. (1981). Conceptual approaches to stereotypes and stereotyping. In D. L. Hamilton (Ed.), Cognitive pro-
cesses in stereotyping and intergroup behavior (pp. 1–35). Hillsdale, NJ: Erlbaum. Ayers, M. S., & Reder, L. M. (1998). A theoretical review of the misinformation effect: Predictions from an activation-based memory model. Psychonomic Bulletin & Review, 5, 1–21. Babey, S. H., Queller, S., & Klein, S. B. (1998). The role of expectancy violating behavior in the representation of trait knowledge: A summaryplus-exception model of social memory. Social Cognition, 16, 287–339. Barsalou, L. W. (1982). Context-independent and context-dependent information in concepts. Memory and Cognition, 10, 82–93. Barsalou, L. W. (1983). Ad hoc categories. Memory and Cognition, 11, 211–227. Barsalou, L. W. (1985). Ideals, central tendency, and frequency of instantiation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 11, 629 – 654. Barsalou, L. W. (1987). The instability of graded structure: Implications for the nature of concepts. In U. Neisser (Ed.), Concepts reconsidered: The ecological and intellectual bases of categories (pp. 101–140). New York: Cambridge University Press. Barsalou, L. W. (1989). Intra-concept similarity and its implications for inter-concept similarity. In S. Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 76 –116). New York: Cambridge University Press. Barsalou, L. W. (1999). Perceptual symbols systems. Behavioral and Brain Sciences, 22, 577– 660. Barsalou, L. W. (2002). Being there conceptually: Simulating categories in preparation for situated action. In N. L. Stein, P. J. Bauer, & M. Rabinowitz (Eds.), Representation, memory, and development: Essays in honor of Jean Mandler (pp. 1–16). Mahwah, NJ: Erlbaum. Barsalou, L. W. (in press). Situated simulation in the human conceptual system. Language and Cognitive Processes. Barsalou, L. W., Huttenlocher, J., & Lamberts, K. (1998). Basing categorization on individuals and events. Cognitive Psychology, 36, 203–272. Barsalou, L. W., & Medin, D. L. (1986). Concepts: Static definitions or context-dependent representations? Cahiers de Psychologie Cognitive, 6, 187–202. Barsalou, L. W., & Sewell, D. R. (1984). Constructing representations of categories from different points of view. Emory Cognition Technical Report #2 (pp. 1– 66), Emory University, Atlanta, GA. Barsalou, L. W., Sewell, D. R., & Ballato, S. M. (1986). Assessing the stability of category representations with graded structure. Unpublished manuscript, Emory University, Atlanta, GA. Barsalou, L. W., Spindler, J. L., Sewell, D. R., Ballato, S. M., & Gendel, E. M. (1987). Assessing the stability of category representations with property generation. Unpublished manuscript, Emory University, Atlanta, GA. Bellezza, F. S. (1984a). Reliability of retrieval from semantic memory: Common categories. Bulletin of the Psychonomic Society, 22, 324 –326. Bellezza, F. S. (1984b). Reliability of retrieval from semantic memory: Noun meanings. Bulletin of the Psychonomic Society, 22, 377–380. Bellezza, F. S. (1984c). Reliability of retrieval from semantic memory: Information about people. Bulletin of the Psychonomic Society, 22, 511–513. Bem, D. J. (1972). Self-perception theory. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 6, pp. 1– 62). New York: Academic Press. Bem, D. J., & McConnell, H. K. (1970). Testing the self-perception explanation of dissonance phenomena: On the salience of premanipulated attitudes. Journal of Personality and Social Psychology, 14, 23–31. Bodenhausen, G. V., Schwarz, N., Bless, H., & Wanke, M. (1995). Effects of atypical exemplars in racial beliefs: Enlightened racism or generalized appraisals? Journal of Experimental Social Psychology, 31, 48 – 63. Brewer, M. B. (1988). A dual process model of impression formation. In
STEREOTYPES AS DYNAMIC KNOWLEDGE STRUCTURES R. Wyer & T. Srull (Eds.), Advances in social cognition (Vol. 1, pp. 1–36). Hillsdale, NJ: Erlbaum. Clancey, W. J. (1997). Situated cognition: On human knowledge and computer representations. New York: Cambridge University Press. Crocker, J., Fiske, S. T., & Taylor, S. E. (1984). Schematic bases of belief change. In J. R. Eiser (Ed.), Attitudinal judgment (pp. 197–226). New York: Springer-Verlag. Devine, P. G., & Elliot, A. J. (1995). Are racial stereotypes really fading? The Princeton trilogy revisited. Personality and Social Psychological Bulletin, 21, 1139 –1150. Dosher, B., & Rosedale, G. (1989). Integrating retrieval cues as a mechanism for priming in retrieval from memory. Journal of Experimental Psychology: General, 16, 519 –555. Dovidio, J. F., Brigham, J. C., Johnson, B. T., & Gaertner, S. L. (1996). Stereotyping, prejudice and discrimination: Another look. In C. N. Macrae, C. Stangor, & M. Hewstone (Eds.), Stereotypes and stereotyping (pp. 276 –319). New York: Guilford Press. 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., & 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 L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 23, pp. 1–74). San Diego, CA: Academic Press. Fiske, S. T., & Taylor, S. E. (1984). Social cognition. Reading, MA: Addison-Wesley. Garcia-Marques, L., & Mackie, D. M. (1999). The impact of stereotypeincongruent information on perceived group variability and stereotype change. Journal of Personality and Social Psychology, 77, 979 –990. Gilbert, G. M. (1951). Stereotype persistence and change among college students. Journal of Abnormal and Social Psychology, 46, 245–254. Goethals, G. R., & Reckman, R. F. (1973). The perception of consistency in attitudes. Journal of Experimental Social Psychology, 9, 491–501. Hamilton, D. L., & Sherman, J. W. (1994). Stereotypes. In R. S. Wyer, Jr., & T. K. Srull (Eds.), Handbook of social cognition (2nd ed., Vol. 2, pp. 1– 68). Hillsdale, NJ: Erlbaum. Hamilton, D. L., & Trolier, T. K. (1986). Stereotypes and stereotyping: An overview of the cognitive approach. In J. Dovidio, & S. L. Gaertner (Eds.), Prejudice, discrimination, and racism (pp. 127–163). New York: Academic Press. Hastie, R., & Park, B. (1986). The relationship between memory and judgment depends on whether the judgment task is memory-based or on-line. Psychological Review, 93, 258 –268. Higgins, E. T. (2000). Social cognition: Learning about what matters in the social world. European Journal of Social Psychology, 30, 3–39. Hintzman, D. L. (1986). “Schema abstraction” in a multiple-trace memory model. Psychological Review, 93, 411– 428. Judd, C. M., & Park, B. (1988). Out-group homogeneity: Judgments of variability at the individual and group levels. Journal of Personality and Social Psychology, 54, 778 –788. Kahneman, D., & Miller, D. T. (1986). Norm theory: Comparing reality to its alternatives. Psychological Review, 93, 136 –153. Karlins, M., Coffman, T. L., & Walters, G. (1969). On the fading of social stereotypes: Studies in three generations of college students. Journal of Personality and Social Psychology, 13, 1–16. Kashima, Y., Woolcock, J., & King, D. (1998). The dynamics of group impression formation: The tensor product model of exemplar-based social category learning. In S. Y. Read & L. C. Miller (Eds.), Connectionist models of social reasoning and social behavior (pp. 71–110). Mahwah, NJ: Erlbaum. Katz, D., & Braly, K. W. (1933). Racial stereotypes of 100 college students. Journal of Abnormal Social Psychology, 28, 280 –290.
827
Katz, D., & Braly, K. W. (1935). Racial prejudice and racial stereotypes. Journal of Abnormal Social Psychology, 30, 175–193. Kelley, H. H. (1973). The processes of causal attribution. American Psychologist, 28, 107–128. Kolers, P. A. (1976). Reading a year later. Journal of Experimental Psychology: Human Learning and Memory, 2, 554 –565. Kruschke, J. K. (1992). ALCOVE: An exemplar-based connectionist model of category learning. Psychological Review, 99, 22– 44. Linville, P. W., & Fischer, G. W. (1993). Exemplar and abstraction models of perceived group variability and stereotypicality. Social Cognition, 11, 92–125. Linville, P. W., Fischer, G. W., & Salovey, P. (1989). Perceived distributions of characteristics of in-group and out-group members: Empirical evidence and a computer simulation. Journal of Personality and Social Psychology, 57, 165–188. Lord, C. G., & Lepper, M. R. (1999). Attitude representation theory. In M. Zanna (Ed.), Advances in experimental social psychology (Vol. 31, pp. 265–343). New York: Academic Press. MacArthur, L. Z. (1982). Judging a book by its cover: A cognitive analysis of the relationship between physical appearance and stereotyping. In A. H. Hastorf & A. M. Isen (Eds.), Cognitive social psychology (pp. 149 –212). New York: Elsevier/Elsevier Science. Maurer, K. L., Park, B., & Rothbart, M. (1995). Subtyping versus subgrouping processes in stereotype representation. Journal of Personality and Social Psychology, 69, 812– 824. McCloskey, M. E., & Glucksberg, S. (1978). Natural categories: Well defined or fuzzy sets? Memory and Cognition, 6, 462– 472. McGuire, W. L. (1983). A contextualist theory of knowledge: Its implications for innovation and reform in psychological research. Advances in Experimental Social Psychology, 16, 1– 47. McKoon, G., & Ratcliff, R. (1992). Spreading activation versus compound cue accounts of priming: Mediated priming revisited. Journal of Experimental Psychology: Learning, Memory and Cognition, 18, 1155–1172. McNemar, Q. (1969). Psychological statistics (4th ed.). New York: Wiley. Medin, D. L., & Ortony, A. (1989). Psychological essentialism. In S. Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 179 –196). New York: Cambridge University Press. Medin, D. L., & Schaffer, M. M. (1978). A context theory of classification learning. Psychological Review, 85, 207–238. Medin, D. L., & Smith, E. E. (1984). Concepts and concept formation. Annual Review of Psychology, 35, 113–138. Murphy, G. L., & Medin, D. L. (1985). The role of theories in conceptual coherence. Psychological Review, 92, 289 –316. Nosofsky, R. M. (1987). Attention and learning processes in the identification and categorization of integral stimuli. Journal of Experimental Psychology: Learning, Memory and Cognition, 13, 87–109. Nosofsky, R. M., & Palmeri, T. J. (1997). An exemplar-based random walk model of speeded classification. Psychological Review, 104, 266 –300. Osherson, D. N., & Smith, E. E. (1981). On the adequacy of prototype theory as a theory of concepts. Cognition, 9, 35–58. Queller, S., & Smith, E. (2002). Subtyping versus bookkeeping in stereotype learning and change: Connectionist simulations and empirical findings. Journal of Personality and Social Psychology, 82, 300 –313. Ratcliff, R., & McKoon, G. (1988). A retrieval theory of priming. Psychological Review, 95, 385– 408. Reder, L. M., & Schunn, C. D. (1996). Metacognition does not imply awareness: Strategy choice is governed by implicit learning and memory. In L. M. Reder (Ed.), Implicit memory and metacognition (pp. 45–77). Mahwah, NJ: Erlbaum. Rosch, E. H., & Mervis, C. B. (1975). Family resemblances: Studies in the internal structure of categories. Cognitive Psychology, 7, 573– 605. Ross, M., & Conway, M. (1986). Remembering one’s own past: The construction of personal histories. In R. M. Sorrentino & E. T. Higgins
828
GARCIA-MARQUES, SANTOS, AND MACKIE
(Eds.), The handbook of motivation and cognition: Foundations of social behavior (122–144). New York: Guilford Press. Ross, M., & Shulman, R. F. (1973). Increasing the salience of initial attitudes: Dissonance versus self-perception theory. Journal of Personality and Social Psychology, 28, 138 –144. Roth, E. M., & Shoben, E. J. (1983). The effects of context on the structure of categories. Cognitive Psychology, 15, 346 –378. Rothbart, M., & John, O. P. (1985). Social categorization and behavior episodes: A cognitive analysis of the effects of intergroup contact. Journal of Social Issues, 41, 81–104. Rothbart, M., & John, O. P. (1993). Intergroup relations and stereotype change: A social-cognitive analysis and some longitudinal findings. In P. M. Sniderman, P. E. Tetlock, & E. G. Carmines (Eds.), Prejudice, politics and the American dilemma (pp. 32–59). Stanford, CA: Stanford University Press. Rothbart, M., & Lewis, S. (1988). Inferring category attributes from exemplar attributes: Geometric shapes and social categories. Journal of Personality and Social Psychology, 55, 861– 872. Rothbart, M., Sriram, N., & Davis-Stitt, C. (1996). The retrieval of typical and atypical category members. Journal of Experimental Social Psychology, 32, 309 –336. Santos, A. S. C. (2001). A instabilidade dos estereo´tipos: Experieˆncias em cognic¸a˜o social com uma metodologia de teste-reteste longitudinal. [Stereotype instability: Experiments in social cognition with a test–retest methodology]. Unpublished master’s thesis, Universidade de Lisboa, Lisboa, Portugal. Schmidt, F. L. (1996). Statistical significance testing and cumulative knowledge in psychology: Implications for the training of researchers. Psychological Methods, 1, 115–129. Sherman, J. W. (1996). Development and mental representation of stereotypes. Journal of Personality and Social Psychology, 70, 1126 –1142. Sia, T. L., Lord, C. G., Blessum, K. A., Thomas, J. C., & Lepper, M. R. (1999). Activation of exemplars in the process of assessing social category attitudes. Journal of Personality and Social Psychology, 76, 517–532. Smith, E. E., & Medin, D. L. (1981). Categories and concepts. Cambridge, MA: Harvard University Press. Smith, E. R. (1988). Category accessibility effects in a simulated exemplarbased memory. Journal of Experimental Social Psychology, 24, 448 – 463.
Smith, E. R. (1989). Procedural efficiency: General and specific components and effects on social judgment. Journal of Experimental Social Psychology, 25, 500 –523. Smith, E. R. (1990). Content and process specificity in the effects of prior experiences. In T. K. Srull & R. S. Wyer, Jr. (Eds.), Advances in social cognition (Vol. 3, pp. 1–59). Hillsdale, NJ: Erlbaum. Smith, E. R. (1998). Mental representations and memory. In D. T. Gilbert, S. T. Fiske, & G. Lindzey (Eds.), The handbook of social psychology (4th ed., Vol. 1, pp. 391– 445). New York: McGraw-Hill. Smith, E. R., & DeCoster, J. (1998). Knowledge acquisition, accessibility, and use in person perception and stereotyping: Simulation with a recurrent connectionist network. Journal of Personality and Social Psychology, 74, 21–35. Smith, E. R., & Mackie, D. M. (1995). Social psychology. New York: Worth Publishers. Smith, E. R., & Za´rate, M. A. (1992). Exemplar-based model of social judgments. Psychological Review, 99, 3–21. Snyder, M. (1981). On the self-perpetuating nature of social stereotypes. In D. L. Hamilton (Ed.), Cognitive processes in stereotyping and intergroup behavior (pp. 183–212). Hillsdale, NJ: Erlbaum. Taylor, S. E. (1981). A categorization approach to stereotyping. In D. L. Hamilton (Ed.), Cognitive processes in stereotyping and intergroup behavior (pp. 83–114). Hillsdale, NJ: Erlbaum. Van Rooy, D., Van Overwalle, F., Vanhoomissen, T., Labiouse, C., & French, R. (2003). A recurrent connectionist model of group biases. Psychological Review, 3, 536 –563. Wegner, D. M. (1995). A computer network model of human transactive memory. Social Cognition, 13, 319 –339. Wilson, T. D., & Hodges, S. D. (1992). Attitudes as temporary constructions. In L. L. Martin & A. Tesser (Eds.), The construction of social judgments (pp. 37– 65). Hillsdale, NJ: Erlbaum. Yeh, W., & Barsalou, L. W. (1996). The role of situation in concept learning. In G. W. Cottrell (Ed.), Proceedings of the eighteenth annual conference of the cognitive science society (pp. 469 – 474). Hillsdale, NJ: Erlbaum. Yeh, W., & Barsalou, L. W. (2002). The situated nature of concepts. Unpublished manuscript, Emory University, Atlanta, GA. Za´rate, M. A., & Smith, E. R. (1990). Person categorization and stereotyping. Social Cognition, 8, 161–185.
STEREOTYPES AS DYNAMIC KNOWLEDGE STRUCTURES
829
Appendix A Percentage of Choices of the Five Traits That Best Describe the Gypsy Group (Katz–Braly’s (1933) Task, Study 1) Descriptor Dishonest Trouble-maker Distrustful Ill-mannered Greedy Happy Ignorant Superstitious Disrespectful Show-off Poor Emotional Strong Ugly Rich Hardworking Conceited Lazy Aloof Unintelligent Shy Friendly Sad Naive Vulgar Passive Faithful Insensitive Honest Cultured Attractive Peaceful Dynamic Unfaithful Sensitive Discreet Fragile Unpretentious Polite Generous Sophisticated Respectful Intelligent
Cultural stereotype 78.30 76.00 54.30 45.70 28.30 28.30 26.10 26.10 23.90 19.60 15.20 13.00 10.90 8.70 6.50 6.50 6.50 6.50 4.30 4.30 4.30 4.30 2.20 2.20 2.20 2.20 2.20 2.20 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
82.60 82.60 39.10 34.80 13.00 19.60 32.60 26.10 30.40 19.60 8.70 8.70 8.70 6.50 6.50 10.90 10.90 8.70 4.30 4.30 4.30 2.20 2.20 0.00 8.70 2.20 0.00 0.00 2.20 0.00 0.00 0.00 2.20 2.20 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Descriptor Distrustful Trouble-maker Happy Superstitious Emotional Ill-mannered Dishonest Hardworking Conceited Ignorant Shy Strong Dynamic Greedy Disrespectful Rich Poor Aloof Friendly Faithful Show-off Sad Vulgar Honest Unfaithful Sensitive Ugly Lazy Unintelligent Naive Passive Cultured Peaceful Fragile Generous Intelligent Insensitive Attractive Discreet Unpretentious Polite Sophisticated Respectful
rwithin-items ⫽ .97 Note.
Individual stereotype 65.20 60.10 43.40 34.80 30.40 23.90 21.70 21.70 21.70 19.60 19.60 13.00 13.00 10.90 10.90 10.90 8.70 6.50 6.50 6.50 4.30 4.30 4.30 4.30 4.30 4.30 2.20 2.20 2.20 2.20 2.20 2.20 2.20 2.20 2.20 2.20 0.00 0.00 0.00 0.00 0.00 0.00 0.00
rwithin-items ⫽ .91
Items in bold represent the nine attributes most frequently mentioned by participants.
(Appendixes continue)
54.30 41.30 30.40 34.80 34.80 19.60 6.50 19.60 30.40 19.60 28.30 19.60 13.00 6.50 6.50 6.50 13.00 4.30 15.20 6.50 21.70 4.30 10.90 4.30 2.20 4.30 0.00 2.20 0.00 0.00 0.00 4.30 0.00 2.20 0.00 4.30 0.00 2.20 0.00 0.00 0.00 2.20 0.00
GARCIA-MARQUES, SANTOS, AND MACKIE
830
Appendix B Percentage of Choices of the Five Traits That Best Describe the Gay Group (Katz–Braly’s (1933) Task, Study 1) Descriptor Show-off Polite Disrespectful Conceited Emotional Unfaithful Sensitive Fragile Vulgar Peaceful Shy Unintelligent Friendly Insensitive Sophisticated Dishonest Ignorant Sad Happy Aloof Naive Discreet Respectful Trouble-maker Ill-mannered Superstitious Ugly Hardworking Passive Faithful Honest Cultured Attractive Unpretentious Distrustful Greedy Poor Strong Rich Lazy Dynamic Generous Intelligent
Cultural stereotype 58.70 47.80 43.50 37.00 34.80 30.40 30.40 26.10 19.60 19.60 17.40 15.20 13.00 8.70 8.70 6.50 6.50 6.50 4.30 4.30 4.30 4.30 4.30 2.20 2.20 2.20 2.20 2.20 2.20 2.20 2.20 2.20 2.20 2.20 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
60.90 39.10 45.70 32.60 30.40 21.70 41.30 32.60 8.70 17.40 15.20 13.00 8.70 6.50 13.00 6.50 4.30 2.20 6.50 4.30 10.90 6.50 2.20 4.30 0.00 0.00 6.50 2.20 4.30 4.30 2.20 2.20 2.20 4.30 4.30 0.00 0.00 0.00 2.20 2.20 0.00 0.00 2.20
rwithin-items ⫽ .96 Note.
Descriptor Sensitive Emotional Polite Peaceful Friendly Fragile Honest Shy Discreet Show-off Respectful Happy Sad Sophisticated Intelligent Conceited Cultured Attractive Generous Strong Faithful Unpretentious Hardworking Unfaithful Trouble-maker Distrustful Disrespectful Vulgar Dishonest Ill-mannered Greedy Ignorant Superstitious Poor Ugly Rich Lazy Aloof Unintelligent Naive Passive Insensitive Dynamic
Individual stereotype 76.10 41.30 41.30 34.80 30.40 28.30 26.10 19.60 19.60 15.20 15.20 13.00 13.00 13.00 13.00 10.90 10.90 10.90 10.90 8.70 6.50 6.50 4.30 4.30 2.20 2.20 2.20 2.20 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
65.20 47.80 43.50 39.10 21.70 10.90 10.90 37.00 21.70 15.20 13.00 6.50 8.70 6.50 6.50 15.20 15.20 8.70 10.90 2.20 17.40 4.30 4.30 2.20 0.00 6.50 4.30 6.50 0.00 0.00 2.20 0.00 0.00 0.00 0.00 0.00 0.00 2.20 0.00 2.20 0.00 2.20 0.00
rwithin-items ⫽ .92
Items in bold represent the nine attributes most frequently mentioned by participants.
STEREOTYPES AS DYNAMIC KNOWLEDGE STRUCTURES
831
Appendix C Percentage of Choices of the Five Traits That Best Describe the African Immigrant Group (Katz–Braly’s (1933) Task, Study 1) Descriptor Ignorant Poor Trouble-maker Hardworking Unintelligent Passive Ill-mannered Vulgar Dishonest Disrespectful Strong Ugly Lazy Naive Distrustful Superstitious Shy Happy Sad Peaceful Aloof Friendly Honest Discreet Greedy Emotional Conceited Unpretentious Respectful Intelligent Show-off Rich Faithful Insensitive Cultured Attractive Dynamic Unfaithful Sensitive Fragile Polite Generous Sophisticated
Cultural stereotype 63.00 56.50 39.10 37.00 28.30 26.10 21.70 21.70 19.60 19.60 19.60 19.60 17.40 15.20 10.90 10.90 10.90 8.70 8.70 8.70 6.50 6.50 6.50 4.30 2.20 2.20 2.20 2.20 2.20 2.20 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
52.20 45.70 41.30 41.30 32.60 28.30 15.20 21.70 21.70 19.60 32.60 19.60 19.60 13.00 23.90 6.50 6.50 2.20 6.50 2.20 0.00 4.30 4.30 2.20 2.20 2.20 0.00 4.30 4.30 0.00 10.90 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
rwithin-items ⫽ .94 Note.
Descriptor Hardworking Poor Strong Happy Ignorant Friendly Respectful Distrustful Sad Passive Honest Peaceful Shy Naive Trouble-maker Ill-mannered Vulgar Show-off Discreet Unpretentious Generous Dynamic Disrespectful Emotional Conceited Lazy Aloof Faithful Sensitive Fragile Greedy Superstitious Ugly Insensitive Intelligent Dishonest Rich Unintelligent Cultured Attractive Unfaithful Polite Sophisticated
Individual stereotype 56.50 52.20 32.60 28.30 28.30 23.90 23.90 19.60 19.60 19.60 17.40 17.40 13.00 13.00 10.90 10.90 10.90 8.70 8.70 8.70 8.70 6.50 4.30 4.30 4.30 4.30 4.30 4.30 4.30 4.30 2.20 2.20 2.20 2.20 2.20 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
56.50 50.00 32.60 21.70 23.90 21.70 6.50 17.40 10.90 6.50 13.00 17.40 15.20 13.00 17.40 10.90 10.90 6.50 10.90 13.00 8.70 10.90 6.50 17.40 8.70 4.30 0.00 0.00 8.70 4.30 0.00 10.90 4.30 2.20 4.30 2.20 0.00 0.00 0.00 0.00 0.00 2.20 0.00
rwithin-items ⫽ .92
Items in bold represent the nine attributes most frequently mentioned by participants.
Received January 26, 2005 Revision received February 7, 2006 Accepted February 7, 2006 䡲
Journal of Personality and Social Psychology 2006, Vol. 91, No. 5, 832– 844
Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.5.832
Terror Management and Attributions of Blame to Innocent Victims: Reconciling Compassionate and Defensive Responses Gilad Hirschberger Bar-Ilan University In this article, 4 studies test the hypothesis that reminders of personal death bias the normative attribution process and increase the motivation to blame severely injured, innocent victims. In Studies 1 and 2, primes of death led to greater attributions of blame to severely injured victims but did not significantly influence attributions of blame to either mildly injured victims or negatively portrayed others. In Study 3, primes of death led to greater attributions of blame to victims of circumstance but did not influence attributions of blame to victims who were explicitly responsible for their condition. In Study 4, innocent victims who were severely injured elicited more death-related cognitions than did victims who were responsible for their condition or who were only mildly injured. These findings indicate that the predictions of normative models of attribution may be moderated, and even overturned, when observers are reminded of their personal death such that defensive needs override rational inferential processes. Keywords: terror management, attribution, blame, just-world
1993). The reasons why people exhibit the tendency to flee from victims, and even more important, what psychological mechanisms enable them to justify their rejection, remain to be seen. One possibility that has been raised in the literature is that severe trauma disrupts observers’ illusion of invulnerability and exposes them to the possibility that such a calamity could also befall them (e.g., Janoff-Bulman & Yopyk, 2004; Novak & Lerner, 1968; Roessler & Bolton, 1978). Moreover, the encounter with the fragility and vulnerability of another may spark the unsettling awareness of personal mortality (e.g., Hirschberger, Florian, & Mikulincer, 2005; Livneh, 1985; Stangor & Crandall, 2000). The current research examines, from the perspective of terror management theory (TMT; e.g., Greenberg, Pyszczynski, & Solomon, 1997), whether the motivation to deny personal mortality stimulates the rejection of victims. Specifically, on the basis of attribution theory, the current research examines whether the awareness of personal death induces greater victim blaming that may provide a sense of moral justification that stands in the way of feeling callous or ruthless.
The relationship between fear of death and the blaming of victims has been a pivotal ingredient in much philosophical and literary thinking. In The Plague, for example, Camus (1948/1991) made use of a bubonic plague epidemic to raise questions about the rationalization of undeserved, capricious suffering. In particular, Father Paneloux, the religious authority who cannot accept a reality in which bad things happen to good people, admonishes his congregation: “Calamity has come on you, my brethren, and my brethren, you deserved it” (p. 94). The random, senseless killing of the plague seems to pose a threat to fundamental human beliefs and motivates an attempt to find justice in the pestilence: “[T]he just man need have no fear, but the evildoer has good cause to tremble” (p. 95). Thus, the plague becomes easier to digest—it has a reason and it will have an end once its purpose is completed. It seems that Camus’s (1948/1991) insight into this fearful and defensive aspect of human nature when confronted with the threat of death is also manifest in more mundane encounters with suffering others. The literature on reactions to victims suggests that the encounter with victims, especially those who have visible, physical injuries, often involves significant discomfort and the fleeting thought “I’m glad it’s not me.” However, this sense of relief may be accompanied by feelings of guilt for having a selfish rather than a compassionate response (e.g., Zola, 1984). To avoid the negative emotions ensuing from the dissonance between culturally prescribed values of compassion and the experience of personal distress, observers may opt to psychologically and behaviorally distance from victims (e.g., Bennett & Dunkel-Schetter, 1992; Pyszczynski, Greenberg, Solomon, Sideris, & Stubing,
Ambivalent Reactions to Victims Victims with severe and irreversible physical injuries pose a unique problem to nondisabled observers. On the one hand, these victims have experienced considerable misfortune in their lives and are therefore deserving of kindness and compassion from others. However, nondisabled observers often experience ambivalent feelings toward victims and seem hesitant to extend to them feelings of warmth and compassion. It is well documented that reactions toward unfortunate others are characterized by negative reactions such as ignoring, shunning, and distancing (e.g., Bennett & Dunkel-Schetter, 1992; Jones et al., 1984; Katz, 1981). Are observers oblivious to the suffering of others, or are they unaware of the impact their rejection has on victims? It does not seem likely that observers naively distance from victims without realizing the consequences of their actions. Rather, they may be
Parts of this research were completed during Gilad Hirschberger’s postdoctorate fellowship at the Institute of Personality and Social Research, University of California, Berkeley. Correspondence concerning this article should be addressed to Gilad Hirschberger, Department of Psychology, Bar-Ilan University, Ramat-Gan, 52900 Israel. E-mail:
[email protected] 832
TERROR MANAGEMENT AND ATTRIBUTION
motivated by a need to justify their behavior. As Crandall (2000) contended, “people, in the process of stigmatizing others, believe that the rejection, avoidance, and inferior treatment they dole out to stigmatized others are fair, appropriate, judicious—in other words, justified” (p. 126). These justification ideologies as Crandall (2000) described them are elaborate rationalizations that serve not only to create a physical and psychological chasm between observer and victim but also to provide explanations for why victims have earned their fate and deserve their suffering. Thus, the psychological benefits of victim blaming are twofold: First, the observer is shielded from the awareness that a similar fate may befall him or her, and second, the observer may reject and distance from the victim without experiencing any pangs of conscious. However, it would be an oversimplification to assert that observers always reject victims. In fact, much of the literature on reactions to victims has indicated that emotional and attitudinal ambivalence characterize reactions to victims with feelings ranging from compassion and care to disgust and aversion (e.g., Carver, Glass, & Katz, 1978; Katz, 1981; Scheier, Carver, Schulz, Glass, & Katz, 1978). According to Hafer and Be´gue (2005), observers are torn between feelings of perceived injustice, which elicit compassionate feelings, and between the need to believe in a just world, which leads to victim blaming. What are the conditions that determine whether an observer will respond with compassion or with rejection toward victims of misfortune? One possible answer to this question is offered by attribution theory, which has attempted to outline the evaluative processes that underlie attitudes toward misfortunate others.
Attributions of Blame Attribution theory is largely based on Kelly’s (1955) person-asscientist metaphor and posits that people are engaged in a rational process of investigation to better comprehend their physical and social environment. Although there are several different attribution theories, they are all concerned with understanding the causal structure of the social world (Weiner, 1992). Most attribution theorists realize that human behavior is not always rational, however the possible biases that may influence attributional process are often treated as exceptions to rational norms and not as central variables in the attributional process (Alicke, 2000). The rational decision stage models have delineated the process rational perceivers undergo when making judgments of blame (e.g., Shaver, 1970; Weiner, 1995). For example, according to the research of Weiner and his associates (e.g., Weiner, 1980; Weiner, Perry, & Magnusson, 1988), perceptions of control over the outcome moderate the effect of victim blaming, such that those who are perceived to be responsible for their condition elicit more blame, whereas those who are perceived to be innocent victims elicit more sympathy. However, according to Alicke (2000), these theories do not account for psychological processes that may lead to a deviation from rational evaluations. According to his culpable control model, when observers are motivationally biased they process information in a blame-validation mode. Under these conditions observers process the evidence relevant to the person’s responsibility in a biased manner by “exaggerating the actor’s volitional or causal control, by lowering their evidential standards for blame, or
833
by seeking information to support their blame attribution” (Alicke, 2000, p. 558). The culpable control model does not suggest that normative stage theories are invalid but rather that emotional and motivational biases deserve a more prominent place in these theories. Some earlier theories of blame have taken into account motivational biases that may influence observers’ judgment such as defensive attribution (e.g., Shaver, 1970) or the need to believe in a just world (Lerner & Miller, 1978). Accordingly, observers will display biased inferential processes when they feel personally threatened by the victim or when the victim threatens their belief system. Under these circumstances, blame may be attributed to a victim not on the basis of objective evidence but rather because of observers’ need to feel safe and protected. For example, defensive attribution research has shown that severity of injury is associated with greater attributions of blame (e.g., Burger, 1981; Robbennolt, 2000), and the just-world hypothesis predicts greater blame assignment to innocent victims (e.g., Lerner & Miller, 1978). It is noteworthy, that some of the predictions of the above motivational theories seem to be inconsistent with the predictions of normative stage models. For example, the defensive attribution hypothesis predicts that severity of outcome will be associated with greater attributions of blame. However, stage models such as Weiner’s (1995) theory contend that severity of outcome is associated with more sympathy to the victim. Similarly, just-world theory posits that innocent victims elicit more blame as they threaten the belief in a just world, whereas Weiner’s theory posits that innocent victims elicit less blame because of the straightforward and logical conclusion that an innocent victim is, in fact, innocent of any wrongdoing. The culpable control theory of blame (Alicke, 2000) offers a solution to the inconsistency of these models by suggesting that normative decision stage models, such as Weiner’s (1995) theory, may best explain observers’ judgments when observers are in a neutral emotional and motivational state. In this case, they are likely to make judgments on the basis of the objective evidence with a minimal involvement of factors that may bias the process. However, when observers are in a motivationally active state, the personal motivations of the observer may override the rational processes that rely on factual evidence, and, instead, personal motivations such as the need to feel safe or to maintain one’s beliefs may take precedence and distort inferential processes. Under these circumstances, the predictions of motivational theories, such as the defensive attribution hypothesis and the just-world theory, may best explain the attributional process. Although the culpable control model contributes to the understanding of blame attributions by combining normative blame theories with motivational theories, it is still not clear which are the factors that bias motivational processes and induce a blamevalidation mode. The current research proposes that observers’ fluctuating and seemingly inconsistent responses to victims may be a function of the degree to which they perceive a threat to their own physical integrity and a threat to their fundamental beliefs about justice. Accordingly, the theoretical and empirical framework of TMT (e.g., Greenberg et al., 1997) is used to examine whether brief reminders of personal mortality bias attributional processes toward physically injured victims.
HIRSCHBERGER
834 TMT
According to TMT (e.g., Greenberg et al., 1997; Solomon, Greenberg, & Pyszczynski, 1991), humans are caught in an intolerable paradox—they cherish life but are aware of the fact that life is transient and temporary. The paradox is manifested in the instinctual drive to sustain life that is frustrated by the realization that efforts to hold on to life are doomed, at some point, to fail. The inability to escape this predestined fate could potentially render humans helpless and consumed with terror. As there is no solution to the problem of death itself, humans have devised elaborate symbolic defense mechanisms that function to remove thoughts of death from conscious awareness. This denial of death, as Becker (1973) described it, is an ongoing dynamic process that functions to enable psychological equanimity and conscious oblivion. On the basis of these theoretical ideas, TMT postulates that two psychological constructs function as primary defense mechanisms against the anxiety associated with death awareness— cultural worldview validation and self-esteem enhancement. Embracing the cultural worldview or possessing high self-esteem is associated with more successful efforts at keeping death out of focal attention. Worldviews are symbolic social constructions of reality that are specific to a certain culture and may be threatened by a different conception of reality that is not compatible with it. For this reason, TMT has suggested that the encounter with a worldview-denying other may be particularly threatening and may lead to distancing, derogation, and even aggression against the other, especially when death is salient (Greenberg et al., 1997). Terror management defenses have been conceptualized as a dual process model wherein proximal and distal defenses take place in a temporal sequence (Pyszczynski, Greenberg, & Solomon, 1999). The initial and direct form of defense, the proximal defense, attends to conscious death concerns and attempts to remove conscious death-related thoughts from awareness. This may be achieved by actively suppressing death concerns (Arndt, Greenberg, Solomon, Pyszczynski, & Simon, 1997), by distraction (Greenberg, Pyszczynski, Solomon, Simon, & Breus, 1994), by shifting to an external focus of attention and avoiding selfreflective thought (Arndt, Greenberg, Simon, Pyszczynski, & Solomon, 1998), or by biasing inferential processes to deny one’s vulnerability (Greenberg, Arndt, Simon, Pyszczynski, & Solomon, 2000). The second line of defense, the distal defense, emerges only after participants are distracted from death-related thoughts and when these thoughts start to resurface but are not in focal attention (Arndt et al., 1997; Greenberg et al., 2000). Distal defenses are symbolic in nature and consist of attempts to embed oneself in a symbolic construct of meaning that offers death transcendence through literal and symbolic immortality (Pyszczynski et al., 1999). The defense of the cultural worldview has been identified as a primary distal terror management mechanism. Accordingly, terror management studies have consistently shown that when primed with death people exhibit a stronger identification with their worldview and respond more negatively to people or information that may compromise the veracity of the worldview. For example, studies have shown that when primed with death, Christian participants rated Christian targets more positively and Jewish targets more negatively (Greenberg et al.,
1990); American participants reacted more negatively to an antiAmerican essay writer (e.g., McGregor et al., 1998); German participants chose to sit further away from a Turkish confederate (Ochsmann & Mathy, 1994); and White American participants expressed more sympathy toward a White racist (Greenberg, Schimel, Martens, Pyszczynski, & Solomon, 2001). The effects of death primes on worldview validation are not limited to group relations but also affect the need to uphold central values and beliefs. Accordingly, recent studies have found that primes of death increase the need to believe in a just world and in a benevolent causal order of events in the social world (Landau et al., 2004, Studies 5–7).
Terror Management and Attribution The current research set out to understand the relationship between normative stage models of attribution of blame, such as Weiner’s (1995) theory, and between motivational accounts of attribution that represent a deviation from normative rational processes such as defensive attribution (e.g., Shaver, 1970) and the need to believe in a just world (e.g., Lerner & Miller, 1978). Biases to attributional processes may stem from the nature of the target, the characteristics or state of the observer, or the interaction between the two. The present study suggests that victims suffering from severe physical injuries may bias inferential processes in two different ways: either by increasing compassion toward victims or by increasing feelings of aversion and rejection toward them. The emotional, cognitive, and motivational state of the observer may moderate these different responses. On the basis of TMT, it is proposed that brief reminders of personal death may bias attributions of blame toward victims suffering from severe physical injuries because of the activation of proximal and distal terror management defenses. People with severe physical injuries pose a troubling reminder of the fragility and the vulnerability of the human body (e.g., Hirschberger et al., 2005). The encounter with a severely injured victim under deathsalient conditions elevates the awareness that one is physically fragile and susceptible to severe injury and death. This awareness may activate proximal terror management defenses that include the biasing of rational inferential processing to deny one’s vulnerability (Greenberg et al., 2000). The current research suggests that these biased inferential processes may take the form of greater attributions of blame toward a severely injured victim. Moreover, severely injured victims, primarily those who are not responsible for their condition, threaten the belief in a just world—a world in which people get their just deserts (e.g., Hafer & Be´gue, 2005). The belief in a just world is an important component of the cultural worldview, as it conveys that the world has order, logic, and meaning and that one is protected from randomness and happenstance (e.g., Landau et al., 2004). Thus, under death-salient conditions, observers are motivated to assign more blame to severely injured victims, especially those who are innocent, to restore the veracity of just-world beliefs. Overall, the current research posits that the defensive assignment of blame to victims involves both proximal and distal terror management mechanisms because a meaningful world is one in which people can anticipate, understand, and avoid real threats. However, a meaningful conception of a just world provides more than an illusory assurance of physical safety—it grants assurance that life
TERROR MANAGEMENT AND ATTRIBUTION
is meaningful, that other people will get their just deserts, and that one will survive in perpetuity beyond his or her physical demise. Some initial evidence supports the above conclusions and indicates that primes of death led men, but not women, to report less compassion toward members of their cultural in-group with physical disabilities, but primes of death did not influence their compassion toward members of their cultural out-group with physical disabilities (Hirschberger et al., 2005, Studies 1 and 2). Further, the exposure to in-group people with physical disabilities led men to experience elevated death-related cognitions and fears (Studies 3 and 4). Moreover, Landau and his colleagues (2004) have shown that primes of death led observers high in personal need for structure (PNS) to be more interested in discovering negative information about an innocent victim (Study 5) and that a positively portrayed innocent victim led participants high in PNS to experience elevated death-related cognitions (Study 6). These findings provide an empirical foundation for the contention of the current study that innocent victims pose a threat to terror management mechanisms and may bias the processes proposed by normative models of attribution (e.g., Weiner, 1995; Weiner et al., 1988). On this basis, it is hypothesized that under death-nonsalient conditions observers’ sense of personal vulnerability and the need to believe in a just world are low and they will judge a victim according to the logic of the evidence provided to them—an innocent victim will be judged less severely than a guilty victim, and a severely injured victim will be judged less severely than a mildly injured victim (e.g., Weiner et al., 1988). However, when death is salient observers will be motivated to attribute more blame to a severely injured innocent victim. The current research examines whether this effect will completely override the predictions of the rational model or whether death primes will only moderate the predictions of the rational model such that severely injured innocent victims will elicit a less sympathetic response than they do when death is not salient.
Study 1 The basic premise of the current research is that primes of death induce attributions of blame toward physically injured victims, especially if the injury is severe and irreversible. To examine this claim, Study 1 compares the impact of death primes on attributions of blame toward a victim with a permanent physical disability and toward a victim with a mild and reversible injury. Further, to control for the possibility that mortality salience may heighten attributions of blame toward any negatively portrayed other because of impatience or general negativity, a second control condition was included with a description of a clumsy person because previous research has indicated that negative attitudes exist toward clumsiness (Taylor & Mettee, 1971). Participants read a target description of a day in the life of the target person (disability, light injury, clumsy) that, in the case of the two injuries, offered no information about the causes of the condition (e.g., congenital, accident) or about victim responsibility. All three descriptions were identical aside from the defining characteristic of the target. Accordingly, it was hypothesized that primes of death would elicit greater attributions of blame only toward the target with a permanent and severe disability and not toward the other two targets.
835
Method Participants. One hundred forty-two Israeli undergraduate students from Bar-Ilan University (88 women, 51 men, and 3 participants who did not report their gender), ranging in age from 18 – 49 (Mdn ⫽ 23 years) participated in the study for course credit. Materials and procedure. The study was presented as a study of personality and social attitudes and was conducted in small groups (5–10 participants). In each group participants were randomly assigned to experimental conditions. Participants were given a packet of questionnaires and were asked to complete them at their own pace while making sure to follow the order of the questionnaires. The first questionnaire was a bogus personality inventory intended to disguise the goal of the study. Then participants were randomly assigned to one of two experimental conditions. In the mortality salience condition (N ⫽ 68), participants answered the following open-ended questions: “What do you think happens to you as you physically die and once you are physically dead?” and “Please briefly describe the emotions that the thought of your own death arouses in you.” In the pain salience condition (N ⫽ 74), participants received the same open-ended questions with all references to death replaced with “severe dental pain.” This procedure has been successfully used in numerous terror management studies (e.g., Florian & Mikulincer, 1997; Greenberg et al., 1990). Following the mortality salience induction, all participants completed a 19-item filler scale on leisure time activities. This scale was included as a distraction task because previous studies have shown that mortality salience effects occur after people have been distracted from thoughts of their own death (e.g., Greenberg et al., 1994). After completing the distracting task, participants were presented with a vignette describing an individual (“Dan”) who was portrayed as either clumsy or who had a physical disability or a mild injury. The target was described as being in the same age group as participants, and the story was depicted as a true story with all details accurate except for the target’s name, which had been changed to avoid identification. The vignette included a description of a typical day in the life of the target (e.g., “Dan likes to sit at cafe´s and watch people and often takes the bus and goes to Tel-Aviv”) and the difficulties the target faces on a daily basis (e.g., “picking up a fork that fell in a restaurant”). Next, participants completed a six-item Attribution of Blame Scale that measured the extent to which they attributed blame to the victim (“Dan is to blame for his condition”; “Dan could have done more to avoid his condition”; “The difficulties Dan experiences at a restaurant are his own fault”; “Dan is a victim of circumstances he has no control over”; “Dan is responsible for ending up like this”; “Dan experiences difficulties participating in many activities because of his bad luck”). Responses were made on a 7-point scale, ranging from 1 (strong disagreement) to 7 (strong agreement). A total attribution of blame score was computed by averaging responses on the six items (after reverse scoring two items). Higher scores indicated greater attribution of blame. In the present sample, the internal consistency of the scale was high with a Cronbach’s alpha of .86. After participants completed this scale, they filled out a short demographic sheet, were debriefed, and were thanked for their participation.
Results and Discussion A 2 ⫻ 2 analysis of variance (ANOVA) was conducted with prime condition (death, pain) and target characteristics (physical disability, mild injury, clumsy) as the factors. Attribution of blame scores served as the dependent measure. This analysis yielded a significant main effect of target characteristic, F(2, 136) ⫽ 73.42, p ⬍ .001. Planned contrasts were conducted by comparing the physical disability condition with the two control conditions as well as by comparing the two control conditions. This analysis revealed that, overall, participants assigned significantly less blame to the physically disabled target (M ⫽ 2.93, SD ⫽ .58) than
HIRSCHBERGER
836
to the mildly injured target (M ⫽ 4.43, SD ⫽ .98), t(139) ⫽ 8.56, p ⬍ .001, and the clumsy target (M ⫽ 4.77, SD ⫽ 1.13), t(139) ⫽ 10.53, p ⬍ .001, with a marginally significant difference between the clumsy target and the mildly injured target, t(139) ⫽1.65, p ⫽ .10. This main effect was moderated by a significant Prime Condition ⫻ Target Characteristic interaction, F(2, 136) ⫽ 3.31, p ⬍ .05. Tests for simple main effects revealed that when the target had a physical disability mortality salience led to greater attributions of blame compared to the control condition F(1, 136) ⫽ 9.88, p ⬍ .01. There were no significant differences between the mortality salience and control condition when the target was described as mildly injured F(1, 136) ⫽ .59, p ⫽ ns, or clumsy F(1, 136) ⫽ 1.11, p ⫽ ns. (See Table 1 for means and standard deviations.)1 Overall, these results suggest that targets with physical disabilities elicit the least amount of blame compared with targets with mild injuries or targets that are negatively portrayed. These findings support normative stage theories of attribution that posit that severe outcomes elicit more sympathy and less blame and that factors that are perceived to be under the targets’ control (i.e., clumsiness) elicit more blame (e.g., Weiner et al., 1988). However, death primes lead to significantly greater attributions of blame compared with the control condition only toward a target with a physical disability and not toward the other two targets. Thus, people with disabilities seem to elicit more sympathy and are generally assigned less blame for their condition. However, when observers are reminded of their own mortality, people with physical disabilities pose a threat to terror management mechanisms and consequently are more likely to be blamed for their condition. These findings support motivational theories of blame and indicate that the rational process suggested by normative theories may be biased when observers are reminded of their mortality and are exposed to a severely injured victim. One should note that although death primes moderated the impact of target characteristics on attributions of blame, they did not overturn the normative process such that people with disabilities elicited more blame than the other targets.
Study 2 In Study 2, we attempted to replicate the findings of Study 1, with the following three major differences: 1.
Study 2 was conducted on a sample of American undergraduate students lending the findings cross-cultural validity.
2.
In Study 2, target descriptions focused on the event that
Table 1 The Impact of Mortality Salience and Target Characteristics on Attributions of Blame Pain salience
Mortality salience
Target characteristic
M
SD
M
SD
Physical disability Mild injury Clumsiness
2.67 4.31 4.91
0.54 0.94 1.00
3.24 4.53 4.61
0.48 1.04 1.30
led to the injury (being hit by a car) and not on a general description of the target as in Study 1. Focusing on the injury increases confidence that it is physical disability that induces greater blame when death, and not negative target characteristics, is salient. Moreover, the use of a different scenario supports the contention that the results are due to the severity of injury and are not limited to a specific scenario. As in Study 1, target descriptions in Study 2 did not include information about victim responsibility for the accident. 3.
The mortality salience manipulation used in Study 2 is a novel method developed for that study and examines whether the results were specific to a certain priming method.
As in Study 1, the hypothesis of Study 2 was that primes of death will increase attributions of blame only toward a severely injured target and not toward a mildly injured target. On the basis of the findings of Study 1, it was also hypothesized that regardless of prime condition, physical disability will induce less blame than mild injury in accordance with normative models of attribution.
Method Participants. Eighty-seven American undergraduate students from the University of California, Berkeley (45 women, 41 men, and 1 participant who did not indicate his or her gender), ranging in age from 19 –36 (Mdn ⫽ 21) participated in the study for course credit. Materials and procedure. The study was presented as a study of personality and social attitudes and was conducted in small groups (5–10 participants). In each group, participants were randomly assigned to experimental conditions. Participants were given a packet of questionnaires and were asked to complete them at their own pace while making sure to follow the order of the questionnaires. The first questionnaire was a bogus personality inventory intended to disguise the goal of the study. Next, participants were assigned to either the mortality salience or the pain salience conditions. For the present study, a new, implicit, mortality salience prime was developed on the basis of Bargh, Chen, and Burrows’s (1996) word-scramble task. The task was presented to participants as an individual differences in cognition task, and the instructions read as follows: “The following sentences have been scrambled out of order. Please unscramble them in a way that makes sense.” The task consists of 23 sentences out of which 7 have a death-related meaning in the mortality salience condition (e.g., “I had a dream last night about dying”) or a pain-related meaning in the pain-salience condition (e.g., “I am not looking forward to my root canal”). The other 16 sentences were neutral in meaning and identical in both conditions (e.g., “I will wash my new car this weekend”). A pretest on 47 undergraduate students that did not participate in other parts of this research was conducted to examine whether this new mortality salience prime elicited thoughts of death. The participants were randomly divided into a mortality salience (N ⫽ 22) or pain salience (N ⫽ 25) condition and were asked to complete the word-scramble task. The dependent measure was the number of death-related completions in a word-stem completion task. Previous research has shown that this measure is a reliable indicator of the accessibility of death-related thoughts (e.g., Greenberg et al., 1994). An independent samples t test revealed that the death salience
1 In all four studies there were no significant main effects of gender and no significant interactions between gender and any of the other independent variables.
TERROR MANAGEMENT AND ATTRIBUTION condition elicited significantly more death-related completions (M ⫽ 1.73, SD ⫽ .7) than did the pain salience condition (M ⫽ 1.04, SD ⫽ .93), t(45) ⫽ 2.82, p ⬍ .01. Following the death prime, participants received a short description of a student from their university who crossed a street and got hit by a vehicle. Half of the students read a description that ended with the victim becoming severely disabled: “She was rushed to the hospital where it was determined that the injury was severe. She had fractures in several vertebrae of her upper spine and would remain confined to a wheelchair for the rest of her life.” The other half read a description that ended with the student being only mildly injured: “She was rushed to the hospital where it was determined that she was only mildly injured. Tests revealed that she had not hurt any vital organs and was expected to fully recover within a few weeks.” The scenarios did not include any information that revealed the responsibility of the victim for the accident. Participants were asked to complete a shortened four-item, English version of the Attribution of Blame Scale used in Study 1. This scale consisted of the four general attribution of blame items used in Study 1 and did not include two items that were specific to the scenario used in Study 1 (e.g., “The difficulties Dan experiences at a restaurant are his fault”). Responses were made on a 7-point scale, ranging from 1 (strong disagreement) to 7 (strong agreement). A total attribution of blame score was computed by averaging responses on the four items (after reverse scoring one item). Higher scores indicate greater attribution of blame. In the present sample, the internal consistency of the scale was acceptable with a Cronbach’s alpha of .72. Then, participants completed a short demographic sheet, were debriefed, and were thanked for their participation.
Results and Discussion A 2 ⫻ 2 ANOVA with mortality salience (death, pain) and outcome severity (severe injury, mild injury) as factors was conducted with attribution of blame as the dependent variable. This analysis revealed a significant main effect for scenario outcome, F(1, 83) ⫽ 7.7, p ⬍ .01, indicating that participants attributed more blame to the target with the mild outcome (M ⫽ 2.93, SD ⫽ 1.03) than to the target with the severe outcome (M ⫽ 2.32, SD ⫽ 1.13). The above main effect was moderated by a significant Mortality Salience ⫻ Outcome interaction, F(1, 83) ⫽ 8.26, p ⬍ .01. As can be seen in Table 2, tests for simple main effects indicated that when the accident resulted in a severe injury, mortality salience led to significantly greater attributions of blame compared with attributions in the pain salience condition, F(1, 83) ⫽ 5.22, p ⬍ .05. However, when the accident resulted in a mild injury, a marginally significant trend indicated that mortality salience led to lower attributions of blame than occurred in the control condition, F(1, 83) ⫽ 3.1, p ⫽ .08. To examine the hypothesis that under death-nonsalient conditions a severe outcome will elicit lower attributions of blame than will a mild outcome, we conducted additional tests for simple main
Table 2 The Impact of Mortality Salience and Scenario Outcome on Attributions of Blame Pain salience
837
effects on the mortality salience and the pain salience conditions. These analyses revealed that in the death-nonsalient condition more blame was attributed to the mildly injured victim compared with that attributed to the severely injured victim, F(1, 83) ⫽ 15.68, p ⫽ .001. No significant differences were found between these victims in the mortality salience condition (see Table 2 for means and standard deviations). Overall, the findings of Study 2 replicated the findings of Study 1 on a different cultural sample by using an implicit death prime and different scenario than in Study 1. The results of Study 2 indicate that a target with a severe injury elicits lower attributions of blame than does a target with a mild injury. However, death primes led to greater attributions of blame toward a target with a severe injury and not toward a target with a mild injury. These findings indicate that normative attributional processes are disrupted when observers are primed with death such that they are motivated to blame a severely injured victim, and to some extent may also be motivated to lower attributions of blame toward a mildly injured victim.
Study 3 Research has shown that the devaluation of victims is motivated by the need to believe that the world is a just and safe place, where people get what they deserve (e.g., Lerner & Simmons, 1966). When exposed to information that threatens this belief, observers are motivated to make attributions about people and circumstances that restore the veracity of the just-world belief (Godfrey & Lowe, 1975; Lerner & Miller, 1978). Ironically, an innocent victim may pose a greater threat to the belief in a just world than does a blameworthy victim, as he or she exposes observers to the reality of their own vulnerability and susceptibility to severe misfortune. A blameworthy victim, on the other hand, offers reassurance that one has control over one’s destiny and that serious misfortune can be avoided. From a terror management perspective, the threat posed by an innocent victim may be amplified when death is salient, as this information interferes with the need to deny the fragility of the human body and its susceptibility to death (e.g., Goldenberg, Pyszczynski, Greenberg, & Solomon, 2000) as well as the need to believe that the world is a just and fair place where everyone gets their just deserts (Landau et al., 2004). If terror management defenses function to provide an illusion of control over the uncontrollable reality of eventual death, innocent victims threaten these defenses and expose observers to the truth of their fragile, mortal nature. Accordingly, Study 3 examines whether the findings obtained in Studies 1 and 2—that death primes elicit attributions of blame toward severely injured and not moderately injured victims—will hold true only when the victim is clearly innocent of any wrongdoing.
Method
Mortality salience
Injury type
M
SD
M
SD
Severe Mild
1.92 3.19
0.94 1.08
2.67 2.65
1.27 0.94
Participants. One hundred sixty-nine Israeli undergraduate students (41 men, 128 women), ranging in age from 18 – 48 (Mdn ⫽ 22) participated in the study for course credit. Materials and procedure. The study was presented as a study of personality and social attitudes and was conducted in small groups (5–15 participants). In each group, participants were randomly assigned to ex-
HIRSCHBERGER
838
perimental conditions. Participants were given a packet of questionnaires and were asked to complete them at their own pace while making sure to follow the order of the questionnaires. The first questionnaire was a bogus personality inventory intended to disguise the study’s goal. Next, participants were assigned to either the mortality salience (N ⫽ 81) or the pain salience (N ⫽ 88) conditions, according to the same procedure used in Study 1. Following the distraction task, participants read a short description of an automobile accident that was formatted as a newspaper article to increase its credibility. These descriptions served to manipulate both outcome severity (severe, mild) and driver responsibility (careful driver, reckless driver). To manipulate driver responsibility, the driver (Yuval) was described as driving through an intersection on the way back home from work. In the careful driver condition, he was described as . . . crossing the intersection on a green light and driving within the speed limit, when suddenly a vehicle rushed toward him from his left side. The vehicle was speeding and ran a red light, and there was nothing Yuval could do to avoid the accident. In the reckless driver condition, Yuval was described as . . . crossing the intersection at high speed after the light had turned red, when suddenly another vehicle rushed toward him from his left side. The other vehicle had crossed on a green light and was not speeding, but it was too late. Nothing could be done to avoid the accident. To manipulate severity of outcome, an interview with the head of the hospital’s intensive care unit was described. In the mild outcome condition he described the victim (driver) in the following manner: He was very lucky and only mildly injured. He has some superficial wounds and a broken leg. We expect him to be released from the hospital in a few days, and within a few weeks he is expected to fully recover. In the severe outcome condition, the patient is described in the following manner: His spinal cord is damaged, and his condition is severe. We are doing everything we can to help, but he is likely to stay paralyzed from the neck down for the rest of his life. After reading the accident descriptions, participants completed the same 4-item scale used in Study 2 that assesses attribution of blame. In the present sample, the internal consistency of the scale was acceptable with a Cronbach’s alpha of .75. Then, participants filled out a short demographic sheet, were debriefed, and were thanked for their participation.
Results and Discussion A 2 ⫻ 2 ⫻ 2 ANOVA was conducted with mortality salience (death, pain), accident outcome (severe injury, mild injury), and driver’s responsibility (reckless, careful) as the factors. Attributions of blame served as the dependent variable. The analysis yielded a significant main effect of driver responsibility, indicating that significantly more blame was attributed to the reckless driver (M ⫽ 4.37, SD ⫽ .98) than to the careful driver (M ⫽ 2.91, SD ⫽ 1.05), F(1, 161) ⫽ 88.99, p ⬍ .001. This finding indicated that the manipulation of driver responsibility was successful. A marginally significant main effect of mortality salience, F(1, 161) ⫽ 2.77, p ⬍ .10, was also obtained indicating that primes of death led to slightly greater attributions of blame (M ⫽ 3.79, SD ⫽ 1.18) than did primes of physical pain (M ⫽ 3.49, SD ⫽ 1.3). These main
effects were moderated by a marginally significant Driver Responsibility ⫻ Accident Outcome interaction, F(1, 161) ⫽ 3.19, p ⫽ .08, and the expected three-way interaction between mortality salience, accident outcome, and driver responsibility, F(1, 161) ⫽ 4.38, p ⬍ .05. To examine the source of this significant interaction, two separate ANOVAs were conducted with mortality salience and accident outcome as the factors. The first was conducted on the mild injury condition and the second on the severe injury condition. The error term of the three-way interaction was used in each of these analyses. The ANOVA conducted on the mild injury condition yielded a significant main effect for driver responsibility, F(1, 161) ⫽ 67.8, p ⬍ .001, with the reckless driver eliciting more attributions of blame (M ⫽ 4.53, SD ⫽ .88) than the careful driver (M ⫽ 2.81, SD ⫽ 1.04). There were no other significant main effects and no significant interaction. However, the ANOVA conducted on the severe injury scenario yielded a significant main effect of mortality salience, F(1, 161) ⫽ 4.38, p ⬍ .05, indicating that attributions of blame were significantly greater in the mortality salience condition (M ⫽ 3.87, SD ⫽ .99) than in the control condition (M ⫽ 3.4, SD ⫽ 1.36). This finding replicates the results of Study 2. In addition, a significant main effect of driver responsibility was obtained, F(1, 161) ⫽ 28.18, p ⬍ .001, with the reckless driver eliciting more attributions of blame (M ⫽ 4.21, SD ⫽ 1.05) than the careful driver (M ⫽ 3.02, SD ⫽ 1.06). These main effects were moderated by a significant Mortality Salience ⫻ Driver Responsibility interaction, F(1, 161) ⫽ 5.13, p ⬍ .05. Tests for simple main effects revealed that when the victim drove recklessly, mortality salience did not lead to greater attributions of blame than did the pain salience condition, F(1, 161) ⫽ .02, p ⫽ ns. However, in line with the hypothesis, when the victim drove carefully, mortality salience led to greater attributions of blame compared with the pain salience condition, F(1, 161) ⫽ 9.15, p ⬍ .01. Additional tests for simple main effects were conducted to examine whether mortality salience in the careful driver condition led to more defensive blaming of a severely injured victim compared with blaming of a mildly injured victim. As expected, the analysis revealed that when death was salient, severely injured victims elicited greater attributions of blame than did mildly injured victims, F(1, 161) ⫽ 10.9, p ⫽ .001. No significant differences were found between these groups, or in the reckless driver condition, when death was not salient. (See Table 3 for means and standard deviations.) The results of Study 3 supported the hypotheses and indicated, as in Studies 1 and 2, that mortality salience increased attributions of blame toward severely injured victims. Study 3 adds to the findings of the previous two studies by indicating that this response takes place, in particular, toward innocent victims. Moreover, by specifying victim responsibility in Study 3, mortality salience not only moderated the normative attributional process as in Studies 1 and 2 but also produced a full defensive attribution response: Participants attributed more blame to the severely injured innocent victim under mortality salience conditions compared with blame attributions in both the control condition and the mildly injured victim condition. However, it should be noted that the normative pattern of results obtained in Studies 1 and 2, wherein severely injured victims elicited lower attributions of blame than did mildly injured victims, was not obtained in Study
TERROR MANAGEMENT AND ATTRIBUTION
Table 3 The Impact of Mortality Salience, Scenario Outcome, and Driver Responsibility on Attributions of Blame Pain salience Injury type
M
SD
Mortality salience M
SD
3.52b 2.76c
0.83 1.00
4.19a 4.62a
1.04 0.92
Careful driver Severe Mild
2.55c 2.85c
1.05 1.09
4.23a 4.44a
1.09 0.85
victims but that the greatest level of death-related cognitions would be elicited by severely injured victims who are innocent. Thus, the design of Study 4 was a 2 ⫻ 2 experimental design with target condition (severe, mild) and victim’s responsibility (innocent, guilty) as the factors. The accessibility of death-related cognitions served as the dependent variable.
Method
Reckless driver Severe Mild
839
Note. Means with different subscripts were significantly different at ␣ ⫽ ⬎.05.
3. It seems that when victim responsibility is specified, mortality salience increases attributions of blame to severely injured victims to the extent that they are greater than attributions of blame toward mildly injured victims, and thus the manipulation serves to override the normative response obtained in Studies 1 and 2.
Study 4 The first three studies indicated that primes of death increased attributions of blame toward a target with a severe and permanent physical injury, particularly if the target was an innocent victim. To further validate the hypothesis that severely injured innocent victims threaten terror management defenses, the current study examines whether the exposure to such victims elicits greater accessibility of death-related cognitions compared with both severely injured victims who are responsible for their condition and mildly injured victims. Previous TMT studies have examined the accessibility of deathrelated cognitions to measure the degree by which a certain construct or target threatens terror management mechanisms. Accordingly, studies have found that the physical aspects of sex elicit greater death-related cognitions among participants high in neuroticism (Goldenberg, Pyszczynski, McCoy, Greenberg, & Solomon, 1999) and that thoughts about relationship problems elicit higher death-related cognitions among people in a close relationship (Florian, Mikulincer, & Hirschberger, 2002). Recently, Landau et al. (2004, Study 6) showed that victims who are portrayed in a positive light elicit higher levels of death-related cognitions than victims portrayed in a negative light but only among observers high in personal need for closure. Similarly, Hirschberger et al. (2005) found that people with disabilities who were members of participants’ in-group elicited higher levels of death-related cognitions. The current study followed the logic of Studies 1–3 and examined whether only severely injured innocent victims threaten terror management defenses such that exposure to these victims would elicit greater death-related cognitions among observers. Accordingly, it was hypothesized that severely injured victims would elicit greater death-related cognitions than would mildly injured
Participants. One hundred forty-nine Israeli undergraduate students (36 men, 105 women, and 8 who did not indicate their gender), ranging in age from 19 –50 (Mdn ⫽ 23) participated in the study for course credit. Materials and procedure. The study was presented as a study of personality and social attitudes and was conducted in small groups (10 –20 participants). In each group, participants were randomly assigned to experimental conditions. Participants were given a packet of questionnaires and were asked to complete them at their own pace while making sure to follow the order of the questionnaires. The first questionnaire was a bogus personality inventory intended to disguise the study’s goal. Next, participants were randomly assigned to target outcome and victim responsibility conditions and received the same vignettes as in Study 3. After reading the target descriptions, the accessibility of death-related thoughts was assessed by a Hebrew version of the word-completion task, which had been constructed in English (e.g., Greenberg et al., 1994) and successfully used in Hebrew (e.g., Florian et al., 2002) on an Israeli sample. In our study, the task consisted of 19 Hebrew word fragments that participants were asked to complete with the first word that came to mind by filling in one missing letter. Eight of the 19 Hebrew fragments could be completed with either neutral or death-related Hebrew words. For example, participants saw the Hebrew fragment –VEL and could complete it with the Hebrew word HVEL (cord) or with the death-related EVEL (mourning). The possible death-related words were the Hebrew words for death, mourning, cadaver, grave, killing, dying, grief, and skeleton. The dependent measure was the number of death-related Hebrew words with which a participant completed the fragments. This score could range from 0 to 8. Then, participants completed a demographic sheet, were debriefed, and were thanked.
Results and Discussion To examine whether victim responsibility and outcome severity influenced the accessibility of death-related cognitions, a 2 ⫻ 2 ANOVA was conducted with outcome severity (severe injury, light injury) and victim responsibility (responsible, not responsible) as the factors. The accessibility of death-related cognitions served as the dependent variable. The analysis yielded a significant main effect of outcome severity, F(1, 145) ⫽ 9.2, p ⬍ .01, indicating that severely injured victims elicited significantly higher levels of death-related cognitions (M ⫽ 1.77, SD ⫽ 1.1) than did mildly injured victims (M ⫽ 1.27, SD ⫽ .89). This main effect was moderated by the expected Victim Responsibility ⫻ Outcome Severity interaction, F(1, 145) ⫽ 4.18, p ⬍ .05. Tests for simple main effects revealed that when the outcome was severe, innocent victims elicited significantly higher levels of death-thought accessibility than did accountable victims, F(1, 145) ⫽ 4.83, p ⬍ .05. However, when the outcome was mild there was no significant influence of target responsibility, F(1, 145) ⫽ .32, p ⫽ ns (see Table 4 for means and standard deviations). Additional tests for simple main effects indicated that outcome severity had a significant influence on death-related cognitions only when the victim was innocent, F(1, 145) ⫽ 12.19, p ⫽ .001. When the victim was
HIRSCHBERGER
840
Table 4 The Impact of Outcome Severity and Victim Responsibility on the Accessibility of Death-Related Thoughts Reckless driver
Careful driver
Injury type
M
SD
M
SD
Severe Mild
1.48 1.32
0.98 0.91
2.03 1.20
1.14 0.87
responsible for the accident, outcome severity had no significant influence on death-related cognitions, F(1, 145) ⫽ .52, p ⫽ ns. Overall, these findings support the hypotheses and add further validity to the results of Study 3 by indicating that severely injured victims, especially those who were innocent of any wrongdoing, elicited higher levels of death-related cognitions than did other victims.
General Discussion Reactions to misfortunate, suffering others are a window through which to observe the discord between acute defensive needs and basic human values. Although most cultural and religious teachings emphasize altruism and compassion, victims often experience derogation and rejection, only adding insult to their injury. The central premise of the current research was that the need to deny personal mortality motivates observers to relax their values of kindness and compassion in favor of self-protective needs. Two models of attribution were examined in the current research to explain the wavering and seemingly inconsistent reactions to victims described in the literature (e.g., Jones et al., 1984; Katz, 1981; Wright, 1983). The first model, the rational, normative model, suggested that (a) observers would have more positive reactions toward innocent victims compared with victims who are responsible for their fate; and (b) observers would have more positive reactions toward victims of severe misfortune compared with those experiencing mild consequences because a severe misfortune involves more suffering on behalf of the victim. The second model, the defensive model, predicted that under mortality salience conditions (a) innocent victims threaten terror management needs and consequently would elicit negative reactions in the form of greater attributions of blame; and (b) severe consequences are more personally threatening than mild, reversible consequences and thus would also elicit more attributions of blame from observers. Although these two models seem to contradict one another, the current series of studies yielded evidence that provides support for both. Specifically, the normative model was supported by the findings that when death was not salient (a) less blame was attributed to a victim suffering from severe consequences than to a victim suffering from mild consequences (Studies 1 and 2), and (b) less blame was attributed to innocent victims compared with victims responsible for their condition (Study 3). One should also note that in Study 2 mortality salience led to lower attributions of blame toward a mildly injured victim (at marginal significance) compared with the control condition. A similar direction was also observed in the careful driver condition of Study 3, wherein mildly
injured victims elicited slightly less blame in the mortality salience condition. However, there was no such direction in Study 1. Future research should examine the conditions under which death primes led to more positive responses toward victims that do not compromise terror management mechanisms (e.g., mildly injured victims). The defensive model was supported by the findings that (a) primes of death led to greater attributions of blame toward a victim suffering from a severe injury but not toward a victim suffering from a mild injury or toward a negatively portrayed other (Studies 1 and 2), (b) primes of death led to more defensive attribution toward an explicitly innocent victim suffering from a severe injury but not toward a victim who could be held accountable for a severe injury or any victim (innocent or accountable) incurring a mild injury (Study 3), (c) exposure to a severely injured innocent victim elicited the highest level of death-related cognitions compared with both mildly injured victims and victims accountable for their condition.
Death and Attitudinal Ambiguity The above findings support the hypothesis that death salience moderates the relationship between outcome severity, victim responsibility, and attributions of blame. Moreover, these findings concur with a long tradition in the rehabilitation and social psychology literatures suggesting that observers’ reactions to victims are characterized by emotional and attitudinal ambivalence (e.g., Hafer & Be´gue, 2005; Wright, 1983). However, previous research had not clearly specified the conditions that underlie emotional and attitudinal approach and avoidance. The major contribution of the current series of studies is in proposing a model that predicts when observers are likely to exhibit sympathetic and caring responses and when they are likely to go as far as to blame the victim. The results of the four studies reported here suggest that the ambivalence experienced toward victims may be based on motivated biases that distort the normative attribution process. Although severely injured and innocent victims usually elicit more sympathy and less blame (e.g., Weiner et al., 1988), death salience increases attributions of blame toward these victims. It should be noted that in Studies 1 and 2 mortality salience increased attributions of blame toward severely injured victims compared with the control condition, but it did not overturn the normative pattern of results such that severely injured victims elicited greater blame than did mildly injured victims. Only in Study 3, wherein victim responsibility was clearly specified, did mortality salience increase attributions of blame toward the severely injured innocent victim such that they were greater even compared with the mildly injured victim. The results of Study 4 further establish this conclusion and indicate that the severely injured innocent victims elicited the highest levels of death-related concerns. This pattern of results seems to suggest that only when both severity of outcome and victim responsibility are spelled-out does mortality salience increase attributions of blame to the extent that the predictions of the normative models of blame are overturned. In this case, the results seem to support the defensive attribution hypothesis (e.g., Robbennolt, 2000), which suggests that outcome severity will be associated with greater attributions of blame.
TERROR MANAGEMENT AND ATTRIBUTION
These findings appear to reflect the dilemma between compassion and rejection. By adding information on victim responsibility in Study 3 to the information on outcome severity in Studies 1 and 2, the threat is amplified and the need to defend the self overturns the rational process of feeling more compassion toward innocent victims. One should also note that the normative model of attribution was only partially supported in Study 3. Whereas reckless behavior elicited more blame than did careful behavior, mild outcomes did not elicit more blame than did severe outcomes as in Studies 1 and 2. It seems that the effect of mortality salience on attributions of blame toward the severely injured innocent victim was large enough to overturn the general tendency to attribute less blame to innocent victims. Moreover, the systematic progression of the first three studies clarifies the logic by which observers resolve the sympathy– blame dilemma when death is salient. In Study 1, participants read an ambiguous description of a target with a disability with no specific information about causes or consequences of the disability. In Study 2, more information was provided about the circumstances of the accident, yet the responsibility of the victim for the accident was left ambiguous. Only Study 3 provided explicit information about victim responsibility. However, in all three studies, death primes led to greater attributions of blame toward the victim. It seems that when death was salient observers were motivated to resolve the ambiguity surrounding the victim’s responsibility and tended to blame victims unless there were mitigating circumstances that made it clear that their (observers’) sense of personal safety was not compromised. A closer look at the targets in Studies 1, 2 and 3, who did not elicit defensive attribution when death was salient, completes this picture. In all three studies, the mildly injured victim did not elicit defensive attribution, because a mild injury does not pose a threat to terror management mechanisms. In Study 3, the explicitly responsible victim (reckless driver) did not elicit greater attributions of blame when death was salient, because knowing that a victim was responsible for his or her fate is consistent with the need to believe that people get their just deserts (e.g., Lerner, 1980) and removes the threat that one may be subject to severe misfortune caused by erratic circumstances.
Levels of Death Awareness and Reactions to Victims The relationship between death awareness and reaction to victims that emerges from the current research is in keeping with the findings of Hirschberger et al. (2005), who also revealed a similar relationship between death awareness and compassion toward victims in a series of four studies. In Hirschberger et al.’s research, death awareness was associated with less compassionate reactions to victims. Similarly, in the current research, death awareness was associated with more attribution of blame. One important difference, however, between the current research and Hirschberger et al. is that in the latter research similarity to victim and gender differences moderated the relationship between death primes and compassion to victims, such that the effects were observed only among men and only toward in-group people with disabilities. It is likely that gender differences in emotional expression, and particularly in the expression of compassion that were measured in Hirschberger et al.’s research, were responsible for the gender differences that were obtained in that research but were not obtained in the current research.
841
Similarly, in Landau et al.’s (2004) research death salience was associated with a greater interest in negative information about an innocent victim (Study 5), and a positively portrayed innocent victim elicited greater death-thought accessibility than a negatively portrayed innocent victim (Study 6). These findings are in line with the pattern of results of Study 4 in the current research. Landau et al.’s findings for positively portrayed victims paralleled the current findings on severely injured innocent victims. These results support the importance of just-world beliefs to terror management defenses. However, in Landau et al. this effect was only observed among people high in personal need for structure. The current research extends these findings by manipulating victim responsibility and outcome severity and by measuring a direct response, attribution of blame, to the victim.
The Illusion of Control and Existential Terror The realization that one is susceptible to severe misfortune as a result of random, meaningless causes over which one has little or no control can be extremely troubling for an organism motivated to deny death. The gravity of this psychological problem is most evident among victims of severe trauma, whose reaction is often characterized by feelings of terror, disillusionment, and meaninglessness (e.g., Janoff-Bulman, 1989, 1992). The experience of trauma shatters the fundamental illusion of safety, security, and control that most people are protected with, that which enables them to conduct their everyday lives with relative psychological equanimity. Victims of trauma are forced to come to terms with the unbearable randomness of injury and with the realization that the world is not safe and that humans are weak and fragile (JanoffBulman & Yopyk, 2004). It is not surprising, then, that observers witnessing the suffering of innocent victims react by attempting to find justice in the tragedy. The thought that severe misfortunes happen capriciously is intolerable, and observers are compelled to believe that the victim has somehow merited his or her fate. Indeed, a large body of research has documented the tendency to derogate victims’ personalities (e.g., Stokols & Schopler, 1973) or to ascribe prior misdeeds to victims in order to justify their suffering (e.g., Hafer, 2000; Lerner & Miller, 1978). According to Janoff-Bulman and Yopyk (2004), people are aware of the fact that terrible, capricious misfortunes may happen in the world, they just refuse to acknowledge that such misfortunes may happen in their world. On the basis of dual-process models that distinguish between explicit, rational modes of thinking and implicit, experiential modes of thinking (e.g., Epstein, 1998), they propose that although people may be consciously, rationally aware of the possibility of severe misfortune, at the experiential, gut level, they maintain a belief that they are safe and invulnerable. The above analysis is in line with a series of studies indicating that terror management defenses operate under experiential, implicit forms of cognitive processing (Simon et al., 1997). Thus, even though an observer rationally understands that a victim is innocent and that bad things may happen for no apparent reason, primes of death threaten the implicit belief that the world is fair and that one is safe. From a terror management standpoint, the belief in a just world is a fundamental component of death denial, as the uncontrollable and unpredictable nature of the world signifies the essence of human existential fears. Uncontrollable causes
842
HIRSCHBERGER
of misfortune pose a grave threat to a death-denying organism because they imply that anyone, including oneself, may be a victim of severe calamity and ultimately of death. Consequently, observers are compelled to resolve the incongruence between the implicit need to feel invulnerable and the explicit evidence that suggests innocent people may be victims of severe misfortune. When faced with the choice of denying one’s own invulnerability or denying a victim’s innocence, it appears that under death-salient conditions observers tend to choose the latter option. The denial of a reality wherein bad things may happen capriciously enables observers to resolve the dilemma between compassion and rejection. If victims are responsible for their fate then rejection becomes justifiable. As Becker (1973) contended “cultural illusion is a necessary ideology of self-justification” (p. 189)—the illusion of a just world is a self-deceptive belief that makes victim blaming reasonable.
Limitations Before reaching any final conclusions, one should keep in mind several limitations of the current research. First, there are various kinds of victims that may suffer from different types of misfortunes ranging from physical ailments to financial losses. The focus of the current research is on victims suffering from physical injuries. Future studies should examine whether the impact of death on defensive attribution is specific to physical forms of victimization or whether it would be obtained in other cases of victimization as well. Second, the current studies were conducted with college student samples. Although one advantage of this research is that both American and Israeli samples were recruited, lending the findings cross-cultural validity, future studies should examine other age and socioeconomic groups. Third, all four studies relied on self-report measures and on hypothetical scenarios with victims. To deal with this limitation the following steps were taken: (a) two different death primes were used to ensure that the effects were not specific to a certain prime; and (b) the scenarios in Studies 2, 3, and 4 were presented and formatted to look like newspaper articles to make them appear more real. Future studies should attempt to examine the impact of death on reactions to victims by using behavioral paradigms with real or confederate victims. Aside from the above methodological issues, there are also several conceptual issues regarding the concordance of the current findings with previous terror management research that merit attention. First, previous research has indicated that thoughts of death compared with thoughts of paralysis led participants who had been made to feel creative indicate more social consensus with their attitudes (Arndt, Greenberg, Solomon, Pyszczynski, & Schimel, 1999). At first sight, these findings seem to imply that thoughts of paralysis are not related to death anxiety and thus may undermine the central premises of the current research. However, the findings of Arndt et al. (1999) merely indicated that priming thoughts of personal death induce a stronger response than do primes of paralysis. Moreover, an extensive body of research conducted by Goldenberg and her colleagues (for a review see Goldenberg et al., 2000) and previous research conducted by Hirschberger et al. (2005) indicate that the physical, vulnerable nature of the human body elicits mortality concerns.
In the same vein, there may seem to be a logical inconsistency between the findings of Study 3 and Study 4. Namely, in Study 4 the severely injured innocent victim elicited the highest levels of death-related cognitions. Thus, one might argue that in Study 3 this victim should have induced higher blame attributions even in the pain salience condition. This argument may be extended to previous research that has indicated that relationship problems (Florian, Mikulincer, & Hirschberger, 2001), sex (Goldenberg et al., 1999), and physical disability (Hirschberger et al., 2005) also elicited increased levels of death-related cognitions. However, according to TMT, primes of personal death are unique and cannot be substituted by any of these other threats (e.g., Greenberg et al., 1997). It seems that to obtain a terror management response, the level of death accessibility must pass a certain threshold. This level is obtained with direct reminders of personal death and not with other constructs that are more indirectly related to death. Future research should examine the differences between direct mortality primes and other constructs that are associated with death-thought accessibility but that do not seem to have the same impact as mortality salience primes. Another avenue of research that may seem inconsistent with the current findings concerns the impact of death primes on the punishment of social transgressors. This research has consistently documented that primes of death led to more severe evaluations of transgressors and to the recommendation of harsher punishments (e.g., Florian & Mikulincer, 1997; Florian et al., 2001; Rosenblatt, Greenberg, Solomon, Pyszczynski, & Lyon, 1989). However, in Study 3 of the current research, mortality salience did not induce more attributions of blame toward a reckless driver. The central difference between the current study and previous research is the focus on victim rather than on perpetrator. It seems that to assuage threats to terror management, one must be assured that perpetrators will pay for their crimes and that victims have in some way merited their fate. A victim that is clearly guilty validates justworld beliefs, does not pose a threat to terror management mechanisms, and thus does not induce further blame when observers are primed with death. Ironically, innocent victims pose a threat to the terror management need to believe in a just world that is akin to a criminal getting away with a crime, and thus these victims, and not others, induce more blame when death is salient.
Conclusions The current research provides a comprehensive theoretical perspective that may settle the apparent disagreement between research traditions that have documented the tendency to blame victims and those that have revealed compassionate, caring responses toward them. Identifying terror management defenses as a moderator of reactions to victims may help explain how people, who may otherwise be kind, gentle, and honest, display behaviors that appear cold and callous. It seems that normative models of attribution, which are based on rationality and on predominant social values, are distorted when observers are confronted with reminders of their own death. In this case, a tension arises between normative values and defensive needs that may be resolved by asserting that an innocent victim is in fact guilty. This mechanism of moral disengagement (see Bandura, 1990) enables observers to psychologically detach from aversive reminders of personal death
TERROR MANAGEMENT AND ATTRIBUTION
without feeling that they are violating cultural norms about justice and compassion. The above analysis suggests that the distinction between defensive and nondefensive modes may be a useful way to conceptualize many complex human behaviors and seems to support Becker’s (1975) contention that “it is the disguise of panic that makes men live in ugliness and not the natural animal wallowing” (p. 169). The results of the current research place TMT as an effective framework that may provide an explanation for when and why people act inconsistently and, sometimes, in clear violation of normative thinking and moral reasoning.
References Alicke, M. D. (2000). Culpable control and the psychology of blame. Psychological Bulletin, 126, 556 –574. Arndt, J., Greenberg, J., Simon, L., Pyszczynski, T., & Solomon, S. (1998). Terror management and self-awareness: Evidence that mortality salience provokes avoidance of the self-focused state. Personality and Social Psychological Bulletin, 24, 1216 –1227. Arndt, J., Greenberg, J., Solomon, S., Pyszczynski, T., & Schimel, J. (1999). Creativity and terror management: Evidence that creative activity increases guilt and social projection following mortality salience. Journal of Personality and Social Psychology, 77, 19 –32. Arndt, J., Greenberg, J., Solomon, S., Pyszczynski, T., & Simon, L. (1997). Suppression, accessibility of death-related thoughts, and cultural worldview defense: Exploring the psychodynamics of terror management. Journal of Personality and Social Psychology, 73, 5–18. Bandura, A. (1990). Mechanisms of moral disengagement. In W. Reich (Ed.), Origins of terrorism: Psychologies, ideologies, theologies, states of mind (pp. 161–191). New York: Cambridge University Press. Bargh, J. A., Chen, M., & Burrows, L. (1996). Automaticity of social behavior: Direct effects of trait construct and stereotype priming on action. Journal of Personality and Social Psychology, 71, 230 –244. Becker, E. (1973). The denial of death. New York: Free Press. Becker, E. (1975). Escape from evil. New York: Free Press. Bennett, H. Y., & Dunkel-Schetter, C. (1992). Negative social reactions to victims: An overview of responses and their determinants. In L. Montada, S. Fillip, & M. J. Lerner (Eds.), Life crises and experiences of loss in adulthood (pp. 497–518). Mahwah, NJ: Erlbaum. Burger, J. M. (1981). Motivational biases in the attribution of responsibility for an accident: A meta-analysis of the defensive attribution hypothesis. Psychological Bulletin, 90, 496 –512. Camus, A. (1991). The plague. New York: Vintage Books. (Original work published 1948) Carver, C. S., Glass, D. C., & Katz, I. (1978). Favorable evaluations of blacks and the handicapped: Positive prejudice, unconscious denial, or social desirability? Journal of Applied Social Psychology, 8, 97–106. Crandall, C. S. (2000). Ideology and lay theories of stigma: The justification of stigmatization. In T. F. Heatherton, R. E. Kleck, M. R. Hebl, & J. G. Hull (Eds.), The social psychology of stigma (pp. 126 –152). New York: Guilford Press. Epstein, S. (1998). Cognitive– experiential self-theory. In D. F. Barone, M. Hersen, & V. B. Van Hasselt (Eds.), Advanced personality (pp. 35– 47). New York: Plenum Press. Florian, V., & Mikulincer, M. (1997). Fear of death and the judgment of social transgressions: A multidimensional test of terror management theory. Journal of Personality and Social Psychology, 73, 369 –380. Florian, V., Mikulincer, M., & Hirschberger, G. (2001). An existentialist view on mortality salience effects: Personal hardiness, death-thought accessibility, and cultural worldview defense. British Journal of Social Psychology, 40, 437– 453. Florian, V., Mikulincer, M., & Hirschberger, G. (2002). The anxiety-
843
buffering function of close relationships: Evidence that relationship commitment acts as a terror management mechanism. Journal of Personality and Social Psychology, 82, 527–542. Godfrey, B. W., & Lowe, C. A. (1975). Devaluation of innocent victims: An attribution analysis within the just-world paradigm. Journal of Personality and Social Psychology, 31, 944 –951. Goldenberg, J. L., Pyszczynski, T., Greenberg, J., & Solomon, S. (2000). Fleeing the body: A terror management perspective on the problem of human corporeality. Personality and Social Psychology Review, 4, 200 – 218. Goldenberg, J. L., Pyszczynski, T., McCoy, S. K., Greenberg, J., & Solomon, S. (1999). Death, sex, love, and neuroticism: Why is sex such a problem? Journal of Personality and Social Psychology, 77, 1173– 1187. Greenberg, J., Arndt, J., Simon, L., Pyszczynski, T., & Solomon, S. (2000). Proximal and distal defenses in response to reminders of one’s mortality: Evidence of a temporal sequence. Personality and Social Psychology Bulletin, 26, 91–99. Greenberg, J., Pyszczynski, T., & Solomon, S. (1997). Terror management theory of self-esteem and cultural worldviews: Empirical assessments and conceptual refinements. In P. M. Zanna (Ed.), Advances in experimental social psychology (Vol. 29, pp. 61–141). San Diego, CA: Academic Press. Greenberg, J., Pyszczynski, T., Solomon, S., Rosenblatt, A., Veeder, M., Kirkland, S., & Lyon, D. (1990). Evidence for terror management theory II: The effects of mortality salience on reactions to those who threaten or bolster the cultural worldview. Journal of Personality and Social Psychology, 58, 308 –318. Greenberg, J., Pyszczynski, T., Solomon, S., Simon, L., & Breus, M. (1994). The role of consciousness and accessibility of death-related thoughts in mortality salience effects. Journal of Personality and Social Psychology, 67, 627– 637. Greenberg, J., Schimel, J., Martens, A., Pyszczynski, T., & Solomon, S. (2001). Sympathy for the devil: Evidence that reminding whites of their mortality promotes more favorable reactions to white racists. Motivation and Emotion, 25, 113–133. Hafer, C. L. (2000). Do innocent victims threaten the belief in a just world? Evidence from a modified Stroop task. Journal of Personality and Social Psychology, 79, 165–173. Hafer, C. L., & Be´gue, L. (2005). Experimental research on just-world theory: Problems, developments, and future challenges. Psychological Bulletin, 131, 128 –167. Hirschberger, G., Florian, V., & Mikulincer, M. (2005). Fear and compassion: A terror management analysis of emotional reactions to physical disability. Rehabilitation Psychology, 50, 246 –257. Janoff-Bulman, R. (1989). Assumptive worlds and the stress of traumatic events: Applications of the schema construct. Social Cognition, 7, 113– 136. Janoff-Bulman, R. (1992). Shattered assumptions: Toward a new psychology of trauma. New York: Free Press. Janoff-Bulman, R., & Yopyk, D. J. (2004). Random outcomes and valued commitments: Existential dilemmas and the paradox of meaning. In J. Greenberg, S. L. Koole, & T. Pyszczynski (Eds.), Handbook of experimental existential psychology (pp. 122–140). New York: Guilford Press. Jones, E. E., Farina, A., Hastorf, A. H., Markus, H., Miller, D. T., & Scott, R. A. (1984). Social stigma: The psychology of marked relationships. New York: Freeman. Katz, I. (1981). Stigma: A social psychological analysis. Hillsdale, NJ: Erlbaum. Kelly, G. A. (1955). The psychology of personal constructs. New York: Norton. Landau, M. J., Johns, M., Greenberg, J., Pyszczynski, T., Martens, A., Goldenberg, J. L., & Solomon, S. (2004). A function of form: Terror
844
HIRSCHBERGER
management and structuring the social world. Journal of Personality and Social Psychology, 87, 190 –210. Lerner, M. J. (1980). The belief in a just world: A fundamental delusion. New York: Plenum Press. Lerner, M. J., & Miller, D. T. (1978). Just-world research and the attribution process: Looking back and ahead. Psychological Bulletin, 85, 1030 –1051. Lerner, M. J., & Simmons, C. H. (1966). Observers’ reactions to the “innocent victim”: Compassion or rejection? Journal of Personality and Social Psychology, 4, 203–210. Livneh, H. (1985). Death attitudes and their relationship to perceptions of physically disabled persons. Journal of Rehabilitation, 51, 38 – 41. McGregor, H. A., Lieberman, J. D., Greenberg, J., Solomon, S., Arndt, J., Simon, L., & Pyszczynski, T. (1998). Terror management and aggression: Evidence that mortality salience promotes aggression against worldview threatening others. Journal of Personality and Social Psychology, 74, 590 – 605. Novak, D. W., & Lerner, M. J. (1968). Rejection as a consequence of perceived similarity. Journal of Personality and Social Psychology, 9, 147–152. Ochsmann, R., & Mathy, M. (1994). Depreciating of and distancing from foreigners: Effects of mortality salience. Unpublished manuscript. Pyszczynski, T., Greenberg, J., & Solomon, S. (1999). A dual process model of defense against conscious and unconscious death-related thoughts: An extension of terror management theory. Psychological Review, 106, 835– 845. Pyszczynski, T., Greenberg, J., Solomon, S., Sideris, J., & Stubing, M. J. (1993). Emotional expression and the reduction of motivated cognitive bias: Evidence from cognitive dissonance and distancing from victims paradigm. Journal of Personality and Social Psychology, 64, 177–186. Robbennolt, J. K. (2000). Outcome severity and judgments of “responsibility”: A meta-analytic review. Journal of Applied Social Psychology, 30, 2575–2609. Roessler, R., & Bolton, B. (1978). Psychosocial adjustment to disability. Baltimore: University Park Press. Rosenblatt, A., Greenberg, J., Solomon, S., Pyszczynski, T., & Lyon, D. (1989). Evidence for terror management theory I: The effects of mortality salience on reactions to those who violate or uphold cultural values. Journal of Personality and Social Psychology, 57, 681– 690. Scheier, M. F., Carver, C. S., Schultz, R., Glass, D. C., & Katz, I. (1978). Sympathy, self-consciousness, and reactions to the stigmatized. Journal of Applied Social Psychology, 8, 270 –282.
Shaver, K. G. (1970). Defensive attribution: Effects of and relevance on the responsibility assigned for an accident. Journal of Personality and Social Psychology, 14, 101–113. Simon, L., Greenberg, J., Harmon-Jones, E., Solomon, S., Pyszczynski, T., Arndt, J., & Abend, T. (1997). Cognitive– experiential self-theory and terror management theory: Evidence that terror management occurs in the experiential system. Journal of Personality and Social Psychology, 72, 1132–1146. Solomon, S., Greenberg, J., & Pyszczynski, T. (1991). A terror management theory of social behavior: The psychological functions of selfesteem and cultural worldviews. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 24, pp. 93–159). New York: Academic Press. Stangor, C., & Crandall, C. S. (2000). Threat and the social construction of stigma. In T. F. Heatherton, R. E. Kleck, M. R. Hebl, & J. G. Hull (Eds.), The social psychology of stigma (pp. 62– 87). New York: Guilford Press. Stokols, D., & Schopler, J. (1973). Reactions to victims under conditions of situational detachment: The effect of responsibility, severity, and expected future interaction. Journal of Personality and Social Psychology, 25, 199 –209. Taylor, S. E., & Mettee, D. R. (1971). When similarity breeds contempt. Journal of Personality and Social Psychology, 20, 75– 81. Weiner, B. (1980). A cognitive (attribution)– emotion–action model of motivational behavior: An analysis of help giving. Journal of Personality and Social Psychology, 39, 186 –200. Weiner, B. (1992). Human motivation: Metaphors, theories, and research. Newbury Park, CA: Sage. Weiner, B. (1995). Judgments of responsibility: A foundation for a theory of social conduct. New York: Guilford Press. Weiner, B., Perry, R. P., & Magnusson, J. (1988). An attributional analysis of reactions to stigmas. Journal of Personality and Social Psychology, 55, 738 –748. Wright, B. A. (1983). Physical disability—A psychosocial approach. New York: Harper & Row. Zola, I. K. (1984). Communication barriers between “the able bodied” and “the handicapped.” In R. P. Marinelli & A. E. Dell Orto (Eds.), The psychological and social impact of physical disability (pp. 139 –147). New York: Springer.
Received April 22, 2005 Revision received February 17, 2006 Accepted February 21, 2006 䡲
Journal of Personality and Social Psychology 2006, Vol. 91, No. 5, 845– 856
Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.5.845
Transcending the “Here”: The Effect of Spatial Distance on Social Judgment Marlone D. Henderson, Kentaro Fujita, and Yaacov Trope
Nira Liberman Tel Aviv University
New York University Construal level theory proposes that increasing the reported spatial distance of events produces judgments that reflect abstract, schematic representations of the events. Across 4 experiments, the authors examined the impact of spatial distance on construal-dependent social judgments. Participants structured behavior into fewer, broader units (Study 1) and increasingly attributed behavior to enduring dispositions rather than situational constraints (Study 2) when the behavior was spatially distant rather than near. Participants reported that typical events were more likely and atypical events less likely when events were more spatially distant (Study 3). They were also less likely to extrapolate from specific cases that deviated from general trends when making predictions about more spatially distant events (Study 4). Implications for social judgment are discussed. Keywords: construal, spatial distance, psychological distance, extrapolation, unitization
the axis of the body (head–feet, front– back, left–right) and the three-dimensional space surrounding the body, whereas representations of spatially distant objects have been associated with mental constructions that are more global and schematized (e.g., Bryant & Tversky, 1999; Tversky, 2003). In representing spatially distant objects, individuals circumvent the need to encode all fine-grain metric values by relying instead on categorical information, which can lead to systematic biases and distortions in spatial distance judgments (e.g., Huttenlocher, Hedges, & Duncan, 1991; McNamara, 1986; Tversky, 1981). Despite the amount of work on spatial distance and mental representations, the question of how representations of near and distant events might affect social judgment remains largely unexplored. The present studies investigate the consequences of spatial distance for social judgment within the framework of construal level theory (CLT; Trope & Liberman, 2003). According to CLT, space is a dimension of psychological distance, along with time, social distance, and hypotheticality. Psychological distance is posited to affect the way individuals represent information such that psychologically distant events are represented more by their essential, general, and prototypical features (high-level construals) and psychologically near events are represented in terms of their incidental, specific, and unique features (low-level construals). CLT assumes that an association forms between psychological distance and level of construal and that this association is then overgeneralized, causing people to continue to form high-level construals for distant events and low-level construals for near events, even when information about events is completely known and reliable. Much of the research in support of CLT has focused on temporal distance from events (see Liberman, Trope, & Stephan, in press; Trope & Liberman, 2003) and has examined how temporal distance affects construal and construal-mediated choice and prediction. Temporal distance has been shown to affect a wide range of psychological phenomena, from person perception to selfregulation to interpersonal interactions (e.g., see Gilovich & Med-
How do individuals think and make judgments about events that take place in other neighborhoods, towns, states, continents, or planets? In other words, how does the perceived spatial distance of events from one’s immediate physical surrounding affect judgments and decisions about those events? Individuals frequently think about and make decisions regarding social events that are spatially near or distant. For example, a parent may make decisions about a child who is attending a nearby or faraway university. The present article examines how people’s responses to the same social event can change depending on whether it is believed to occur at a spatially near or distant location. A large body of work suggests that individuals’ understanding of spatially near versus distant objects is constructed through different sensory modalities and representational systems (e.g., McNamara, 1986; Tversky, 2003, 2005). In fact, spatial cognition research suggests that different areas of the brain might even be recruited to represent the same object at near and distant locations (e.g., Berti & Fassinetti, 2000; Halligan, Fink, Marshall, & Vallar, 2003). Representations of spatially near objects are dependent on
Marlone D. Henderson, Kentaro Fujita, and Yaacov Trope, Department of Psychology, New York University, Nira Liberman, Department of Psychology, Tel Aviv University, Tel Aviv, Israel. This research was supported by a National Science Foundation Graduate Student Fellowship to Marlone D. Henderson, a National Science Foundation Graduate Student Fellowship to Kentaro Fujita, National Institute of Mental Health Grant 1R01MH59030-01A1 to Yaacov Trope, and United States–Israel Binational Science Foundation Grant 2001057 to Nira Liberman and Yaacov Trope. We thank Ryan D. Coganow and Celia Gonzalez for their assistance with data collection. Special thanks also go to Ido Liviatan and Cheryl J. Wakslak for extensive discussion of ideas. Correspondence concerning this article should be addressed to Marlone D. Henderson, who is now at the Department of Psychology, University of Chicago, Chicago, IL 60637 or Kentaro Fujita, who is now at the Department of Psychology, Ohio State University, Columbus, OH 43210. E-mail:
[email protected] or
[email protected] 845
846
HENDERSON, FUJITA, TROPE, AND LIBERMAN
vec, 1995; Henderson, Trope, & Carnevale, in press; Keren & Roelofsma, 1995; Ross, 1989; Ross & Wilson, 2002; Sherman, Zehner, Johnson, & Hirt, 1983; Trope & Liberman, 2003; Wilson, Wheatley, Meyers, Gilbert, & Axsom, 2000; Zauberman & Lynch, 2005). In particular, research has consistently found that judgments and decisions for temporally distant rather than near events are determined more by higher level, schematic construals of events. For example, Nussbaum, Trope, and Liberman (2003) found that perceivers placed more weight on general, decontextualized characteristics (global trait concepts) and less weight on situation-specific states when predicting others’ behavior in a distant rather than a near future situation. In another set of studies, Liberman, Sagristano, and Trope (2002) found that individuals were more likely to expect distant rather than near future events to resemble the ideal case or prototype of an event’s category. For example, in one study they asked participants to list the events they expected to experience during either a good or bad day in either the near or distant future and had them rate the valence of each event. Liberman et al. found that the more distant future good and bad days were more uniformly positive and negative, respectively. CLT theorists posit that the same general principles that apply to temporal distance should also hold for other psychological distance dimensions, including spatial distance (Liberman et al., in press). Recently, Fujita, Henderson, Eng, Trope, and Liberman (2006) have extended research on psychological distance and construal to issues of spatial distance and mental representation of social events and objects. Specifically, they demonstrated that increasing the reported spatial distance of social events led individuals to represent the events more abstractly and globally. For example, in one study, they found that participants who imagined a spatially distant rather than near event preferred to identify actions associated with the event in terms of superordinate end states rather than subordinate means. In another study, they found that participants used more abstract language to describe an event that purportedly took place at a spatially distant rather than near location, suggesting a higher level of construal of the event (Semin & Fiedler, 1988). Although Fujita, Henderson, et al.’s (2006) findings offer support for CLT’s broader notion of psychological distance, whereby different distance dimensions are interrelated and similarly affect mental representation, the implications of spatial distance for social judgment and decision-making have yet to be explored. That is, their studies demonstrated the effects of spatial distance on level of construal but did not examine the implications of this effect for social judgment and prediction. In the present article, we extend the previous work in that direction. We also seek to extend CLT to novel paradigms of construal, prediction, and judgment that have not been examined with other dimensions of distance (Studies 1, 3, and 4). Our first study examined the effect of spatial distance on construal by using a construal task that has never been examined within the framework of CLT, namely, segmentation of continuous action (Newtson, 1973). If judgments about spatially distant events are based on more schematic, higher level construals of events, then continuous behavior should be segmented into fewer, broader action units (Study 1). In Studies 2– 4 we examined the effect of spatial distance on judgment and prediction in social situations in which a general, global view of the prediction problem suggests a different prediction than a more local, specific analysis of the same
problem. We expected that when predictions pertained to more spatially distant locations, they would derive more from the highlevel, global aspects of the problem and less from the low-level, local characteristics of the problem. In Study 2 we examined whether predictions of another’s behavior are based more on personal dispositions (e.g., attitudes) than on specific situational constraints when behavior purportedly occurs in a more distant location, an effect that has been demonstrated already with future temporal distance (Nussbaum et al., 2003). In Study 3 we treated the central tendency of a distribution as a prototypical, high-level feature, and a deviant case as a low-level feature. Accordingly, in this study we examined whether predictions about spatially distant outcomes would be more confident when they pertain to a typical event and less confident when they pertain to an atypical event. In Study 4 we viewed the general trend of a series of outcomes as a high-level construal of the prediction situation and a local deviation from the trend as a low-level construal of the same prediction situation. We predicted that extrapolations of a graph would be more in accord with the general trend and less in accord with the local deviation when they pertained to more spatially distant situations.
Study 1: Spatial Distance and Unitization Judgments Schemas represent knowledge about the prototypical features or aspects of objects and events (Rosch, 1975; Rosch & Llyod, 1978; Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976). Research has shown that schemas allow individuals to organize or “chunk” information about objects and events into consistent, meaningful, discernible units (Chase & Simon, 1973; Reitman, 1976). One domain that such researchers have been found to be important is the understanding of behavior (Cohen & Ebbesen, 1979; Lassiter, Geers, Apple, & Beers, 2000; Massad, Hubbard, & Newtson, 1979; Newtson, 1973; Newtson & Engquist, 1976; Wilder, 1978a; 1978b). Although behavior is continuous, perceivers must discern the relevant parts of action from the irrelevant to integrate the behavioral information with previously acquired knowledge. Perceivers with an activated schema have been found to adopt a more selective unitization strategy (dividing a behavior sequence into fewer meaningful actions) than perceivers without an activated schema (Cohen & Ebbesen, 1979; Markus, Smith, & Moreland, 1985). Markus et al. (1985), for example, asked individuals who had well-developed schemas in the domain of masculinity (masculine schematics) and individuals who lacked such schemas (masculine aschematic) to view two films that were either relevant or irrelevant to the domain of masculinity. They found that masculine schematics consistently divided the schema-relevant film into fewer, larger units than did masculine aschematic; no differences emerged for the schema-irrelevant film. In the current study, we tested the hypothesis that as spatial distance from an event is increased, individuals would form a high-level construal of actions that occurred during the event, resulting in fewer, broader units.
Method Participants. A sample of 86 individuals (54 women, 32 men) at New York University (NYU) participated in the study for $5 or in partial fulfillment of a course requirement. They were randomly assigned to condition, with equal numbers per condition. We included the sex of the participant as a factor for all of the analyses reported in this article and
SPATIAL DISTANCE AND SOCIAL JUDGMENT controlled for sex in all analyses; the pattern of results was unchanged in all cases. Thus, the sex of participants is not discussed further. Materials and procedures. Participants were tested in groups of 1 to 8 in a study ostensibly about reactions to a cartoon video. All instructions for the study took place on the computer. Participants were asked to imagine a scenario in which they secured a new position in a film production company. As part of their job, they were instructed to view a cartoon video that was currently in production. The cartoon was presented as a rough sketch of an event that took place during a camping trip for young teenagers. Participants learned that the principal characters in the video were drawn not as people but as shapes. They were told to imagine that the film depicted the action of three teenagers around a cabin at a well-known summer camp. The summer camp at which the events in the film took place was described as located “on the East Coast, about 3 miles from here” (spatially near condition) or “on the West Coast, about 3,000 miles from here” (spatially distant condition). Next, participants in both conditions were introduced to the unitization procedure. During a unitization procedure, participants typically view a film that contains a sequence of actions, and they are requested to push a button each time a meaningful unit of behavior occurs (e.g., see Newtson, 1973). The unitization task, which has been used in a variety of unitizing studies (e.g., Lassiter & Stone, 1984; Lassiter, Stone, & Rogers, 1988; Newtson, 1973), included the following instructions: The assignment your boss has given you is to watch the video of this East [West] Coast camping trip carefully, and to segment what you see into actions that seem natural and meaningful to you. While watching this cartoon video, you’ve been instructed to hit the space bar when, in your judgment, one meaningful action ends and another begins. There is no right or wrong way to do this; it’s up to you to decide whether or not an action seems natural and meaningful to you. Participants in all conditions then viewed and unitized a silent cartoon. Specifically, we showed participants the animated film developed by Heider and Simmel (1944).1 This film depicts two triangles and a circle that move against and around each other. Virtually all people (except for individuals with autism and Asperger’s syndrome) create a social plot for the stimuli in the film on the basis of movement of the shapes (Heider & Simmel, 1944; Klin, 2000). The key that participants pressed to mark a meaningful unit of action activated a computer program that recorded the number of actions identified (unitization rate). No instructions were given regarding the expected or appropriate size of participants’ units. All participants appeared able to perform the task. The film was shown on the computer and lasted 76 s. Heider and Simmel’s (1944) video was ideal for our purpose in this study because it depicts symbolic representations of behavior that can easily be framed as occurring in either a geographically close or remote location. More important, Heider and Simmel’s video contains the basic elements required for a unitization task, namely, streams of action, without the confound of potential differences in perceived similarity or group affiliation with the “social agents” performing the action. Following the film presentation, participants completed a series of questions that addressed other potential confounds. We measured participants’ difficulty in imagining the camping trip event using the following questions: “While watching the video, how difficult was it for you to picture it as a real life event?” “How easy was it to visualize the cartoon as a real life event?” (reverse scored). The answer scales ranged from 1 (not very difficult) to 7 (very difficult) and from 1 (not very easy) to 7 (very easy), respectively. The responses were averaged to form a single index of difficulty (r ⫽ .83). We also measured participants’ familiarity with summer camp in the specified spatial location (“How familiar are you with east [west] coast summer camps?”) and knowledge about summer camps in general (“How knowledgeable are you of summer camps in general?”). The answer scales ranged from 1 (not very familiar) to 7 (very familiar) and
847
from 1 (not very knowledgeable) to 7 (very knowledgeable), respectively. Finally, participants were debriefed and dismissed.
Results and Discussion We analyzed the mean number of key presses (i.e., the number of units into which the film was segmented). A smaller number of units indicates larger units. As predicted, participants who imagined the same event as spatially distant versus spatially near consistently divided the film into fewer units (10.09 vs. 14.41, Mann–Whitney U ⫽ 694.00, Z ⫽ 2.00, p ⬍ .05).2 This effect was not due to differences in difficulty of imagining, familiarity with, or knowledge about the social event. There was no significant difference between the two distance conditions in ease of imagining the event or in the reported amount of knowledge about it. Participants reported being more familiar with the social event in the spatially near location (M ⫽ 2.81) than in the spatially distant location (M ⫽ 1.86), t(84) ⫽ 2.16, p ⬍ .05. None of these variables, however, were significantly correlated with the number of segments (all rs ⱕ 1.0). Moreover, after adjusting for each of these as covariates, the results reported above remained significant, suggesting that they do not mediate the effect of spatial distance on social judgment. The results from this study offer the first evidence that psychological distance in general, and spatial distance in particular, affects the way people segment a stream of behavior. Increased spatial distance from an event leads to larger chunks of behavior during the event. As individuals construct broader conceptualizations of others’ behavior, they should be more likely to abstract the overarching, general purpose behind the actions of spatially distant others rather than simply identifying their behavior as discrete action (see Kudadjie-Gyamfi & Rachlin, 1995; Rachlin, 1995; see also Vallacher & Wegner, 1987). That is, forming larger behavioral chunks should facilitate construal in terms of looking for enduring, global dispositions and omitting more specific situational factors. One implication from these results then is that individuals will be more prone to abstract dispositions from actions of more spatially distant others. The next study tests this hypothesis.
Study 2: Spatial Distance and the Correspondence Bias In this study, we examined whether correspondent inferences— the tendency to make predictions about targets that are consistent with personal attitudes expressed in situationally constrained behavior—are stronger when people are drawing inferences from situationally constrained behavior that occurs in more spatially 1
We thank Sara Kiesler for making this video available on the Internet. A t test revealed the predicted difference between participants in the spatially near and spatially distant conditions, t(84) ⫽ 2.45, p ⫽ .02. However, as suggested by Markus, Smith, and Moreland (1985), a nonparametric statistic (based on the median) was used in reporting these results because the variance within the spatially near group was significantly larger than that within the spatially distant group. This difference in variance is consistent with the rest of our reasoning: Judgments about spatially distant events are based on more schematic, higher level construals of events, and thus participants with a spatially distant perspective are much more consistent in the type of units they produce. 2
HENDERSON, FUJITA, TROPE, AND LIBERMAN
distant locations. A considerable amount of person perception research has demonstrated a bias toward forming correspondent inferences (see Gilbert & Malone, 1995). In terms of CLT, this bias reflects a high-level construal of behavior in terms of abstract, decontextualized qualities rather than in terms of more concrete, situation-specific factors. Consistent with this framework, Nussbaum et al. (2003) demonstrated that the correspondence bias is stronger as the temporal distance from behavior is increased. The current study tested the hypothesis that increased spatial distance will also produce a greater correspondence bias. To test this hypothesis, we used a variation of the Jones and Harris (1967) attitude attribution paradigm. Participants read a situationally constrained or unconstrained essay arguing in favor of a controversial issue that was written by a spatially near or distant target. On the basis of this information, participants were asked to predict whether the writer would engage in behaviors that were congruent with the attitude expressed in the essay. These predictions served as a measure of correspondent attitude inferences. We hypothesized that situational constraints are more likely to attenuate correspondent attitude inferences about a spatially near writer than a spatially distant writer. The correspondence bias is thus more likely to be manifested when the reader believes the essay to be written in a distant location rather than a near location.
Method Participants. A sample of 60 individuals (39 women, 21 men) at NYU participated in the study for $10. They were randomly assigned to condition, with equal numbers in each condition. Materials and procedures. Participants were tested in groups of 1 to 8 in a purported study about essay reading. All instructions for this study and the remaining studies took place on paper. Participants read an essay arguing in favor of the institution of senior comprehensive exams at NYU, purportedly written by a NYU student for a writing class. They were told the writer was instructed either to write an essay that supported senior comprehensive examinations at NYU (constrained condition) or to express his or her view on senior comprehensive examinations at NYU (unconstrained condition). Half of the participants were told that the writer of the essay had written the essay while “in New York City, attending New York University’s Washington Square campus” (spatially near condition). The other half were told the writer had written the essay while “in Italy, attending New York University’s Study Abroad Program” (spatially distant condition). After reading the essay, participants predicted the writer’s attituderelated behaviors. The three behavior prediction questions were as follows: “While having a conversation with his or her friends, how likely is the writer (in Italy) to express views in favor of comprehensive examinations?” “How likely is the writer (in Italy) to express attitudes that favor comprehensive examinations if he or she were interviewed by NYU’s student newspaper?” and “If students had a chance to vote on the issue, how likely is the writer (in Italy) to vote in favor of comprehensive examinations?” Participants responded on a 7-point scale, anchored at 1 (not at all likely) and 7 (very likely). We created an index of essay-consistent (correspondent) attitude inference by averaging each participant’s responses to the three items (␣ ⫽ .83). As a check on the manipulation of situational constraints, participants answered the question “To what extent do you think the writer (in Italy) was forced to write about the view expressed in the essay?,” rating it on a scale that ranged from 1 (not at all forced) to 7 (completely forced). Participants also answered the question “How convincing is the essay about instituting comprehensive examinations?,” rating it on a scale that ranged from 1 (not at all convincing) to 7 (extremely convincing). After-
ward, participants were debriefed and thanked for their participation in the study.
Results and Discussion Manipulation checks. Participants’ perceptions of situational constraints were analyzed using a 2 (spatial distance: near vs. distant) ⫻ 2 (situational constraint: unconstrained vs. constrained) between-participants analysis of variance (ANOVA). As expected, we found a significant main effect of situational constraints, F(1, 56) ⫽ 10.19, p ⬍ .005, indicating that participants perceived the instructions given to the writer as more forceful in the constrained condition than in the unconstrained condition (Ms ⫽ 5.00 vs. 3.70, respectively). Neither spatial distance as a main effect (F ⬍ 1) nor spatial distance in interaction with situational constraints, F(1, 56) ⫽ 1.13, p ⫽ .29, affected these perceptions. We also analyzed participants’ perceptions of essay convincingness using a 2 (spatial distance) ⫻ 2 (situational constraint) between-participants ANOVA. Neither the main effects nor the interaction effect were significant (all Fs ⬍ 1). Attitude inference. The index of participants’ essay-consistent attitude inference was analyzed using a 2 (spatial distance: near vs. distant) ⫻ 2 (situational constraint: unconstrained vs. constrained) between-participants ANOVA. The results showed a main effect of situational constraint, F(1, 56) ⫽ 9.05, p ⬍ .005, with participants in the unconstrained condition making a stronger essayconsistent attitude inference (M ⫽ 5.81) than participants in the constrained condition (M ⫽ 4.95). Results also showed a main effect of spatial distance, F(1, 56) ⫽ 3.84, p ⫽ .06, with participants in the distant condition making a stronger essay-consistent attitude inference (M ⫽ 5.66) than participants in the near condition (M ⫽ 5.10). As expected, however, both of these main effects were qualified by the Spatial Distance ⫻ Situational Constraint interaction effect, F(1, 56) ⫽ 4.15, p ⬍ .05. As can be seen in Figure 1, specific comparisons revealed that there were no significant differences in the tendency to make essay-consistent attitude inferences among participants forming judgments about a spatially distant writer that was constrained (M ⫽ 5.52) versus unconstrained (M ⫽ 5.80), t ⬍ 1. It was only when participants formed
7
6 Attitude Inference
848
constrained
5
unconstrained
4
3 near
distant Spatial Distance
Figure 1. Essay-consistent attitude inference as a function of spatial distance and situational constraints (Study 2)
SPATIAL DISTANCE AND SOCIAL JUDGMENT
a judgment for a spatially near writer that they made a weaker essay-consistent attitude inference when the writer was constrained (M ⫽ 4.38) rather than unconstrained (M ⫽ 5.82), t(28) ⫽ 3.11, p ⬍ .005, d ⫽ 1.76. Furthermore, specific comparisons revealed that there were no significant differences in essayconsistent attitude inferences among participants who formed a judgment about an unconstrained writer who was spatially distant versus near (t ⬍1). In contrast, those forming judgments about constrained writers made essay-consistent attitude inferences to a greater degree when the writer was spatially distant rather than spatially near, t(28) ⫽ 2.43, p ⫽ .02, d ⫽ .92. These results support the present predictions regarding the effect of spatial distance from behavior on the correspondence bias. The attitudes of an essay writer are relatively general and decontextualized constructs that are a part of a high-level construal of writing an essay, whereas situational constraints are relatively concrete and contextual features that are a part of a low-level construal of writing the essay. We therefore predicted that spatial distance from an essay writer would lead participants to see the content of the essay as diagnostic of the writer’s attitude and give less weight to situational constraints on the essay writer. Supporting this prediction, the results show that participants were more likely to disregard the influence of situational constraints when drawing inferences about a spatially distant rather than near essay writer. It is important to note that the spatially near and distant groups did not differ in their perceptions of the situational constraints placed on the writer or in their perceptions of the convincingness of the writer’s essay, making it unlikely that spatial distance alters the fundamental perception of constrained behavior. Rather, we reason that the amount of spatial distance from a constrained target alters the extent to which people weigh such constraints when drawing an inference about the target. That is, the extent to which perceivers make correspondent inferences about a spatially close target should depend upon the amount of perceived constraint placed on the target’s behavior, whereas the extent to which perceivers make such inferences about a spatially distant target should be independent of perceived constraint. This would be manifest in an interaction between perceived constraint and distance condition, in which the degree of perceived constraint predicts spatially near attitude inferences but not spatially distant attitude inferences. Indeed, when we regressed participants’ attitude inferences on perceived constraint and its interaction with distance condition, results revealed that the interaction between distance condition and constraint condition no longer significantly predicted participants’ attitude inferences, whereas the interaction between perceived constraint and distance condition did so. Using simple slope analysis to illustrate the impact of perceived constraint on attitude inferences for the near and distant groups, we found that the attitude inferences that were made by participants in the distant group were unchanging as a function of perceived constraint, whereas the inferences that were made by those in the near group were less strong as perceived constraint increased. This supports our assertion that the situationally constrained essay was seen as more diagnostic of the corresponding attitude when the essay was purportedly written in a spatially distant location. That is, situational constraint, even when perceived, was not considered when drawing inferences from a distant behavior.
849
One potential alternative explanation for the present results might be that our participants held different stereotypes about NYU students in Italy versus New York in their endorsement of comprehensive examinations. That is, when the essay was unconstrained, participants might have taken the essay writer at his or her word, but when the essay was constrained, participants might have perceived the base rates in favor of comprehensive examinations to be greater on the spatially distant (Italy) campus. If the base rates were indeed perceived to be greater on the distant campus, that could explain why the correspondence bias was lower in the distant constrained group compared with the near constrained group. Although, unless the base rates are perceived to be 100% in favor of comprehensive examinations on the distant campus, on the basis of this alternative explanation one would still expect to find the correspondence bias to be lower in the distant constrained group than in the distant unconstrained group. The logic of this is that whereas at least some of the writers who are constrained to write in favor of comprehensive examinations would actually be opposed to them, none who freely choose to write in favor of them would be opposed to them. As reported above, however, the results do not support this interpretation, as we found no difference between the distant constrained and unconstrained groups in their likelihood to make an essay-consistent attitude inference. Moreover, results from a pilot study revealed that students do not in fact hold different stereotypes about NYU students on the different campuses in their endorsement of comprehensive examinations.3 The results from this study are consistent with the Nussbaum et al. (2003) findings that the correspondence bias is manifested more when predicting people’s distant future rather than near future behavior. Together, these findings suggest that judgments about spatially or temporally distant behavior are based on higher level construals of the behavior in terms of global dispositional qualities of the actor. The parallel in the effects of spatial and temporal distance on the correspondence bias, moreover, supports the more general notion that both types of distances constitute instances of psychological distance and that their similar effects on construal and prediction are attributable to this basic similarity between all types of psychological distance. One specific implication of these results might be that people may think that another’s true character is likely to be expressed in 3
Fifty-two participants were asked to estimate the likelihood that NYU students would endorse senior comprehensive examinations. Half of the participants were asked about students attending NYU’s New York campus (spatially near condition), whereas the other half were asked about students attending NYU’s Florence, Italy campus (spatially distant condition). Participants were specifically asked, “How likely are students in New York City, attending NYU’s Washington Square campus [in Florence, Italy attending NYU’s Study Abroad Program] to be in favor of instituting senior comprehensive examinations at NYU?” and “How likely are students in New York City, attending NYU’s Washington Square campus [in Florence, Italy attending NYU’s Study Abroad Program] to be opposed to instituting senior comprehensive examinations at NYU?” (reverse scored). Participants responded on a 7-point scale, anchored at 1 (not at all likely) and 7 (extremely likely). The responses were averaged to form a single index of endorsement (r ⫽ .82). As expected, participants in the spatially near condition (M ⫽ 2.60) did not differ significantly from participants in the spatially distant condition (M ⫽ 2.56) in their estimate of students’ endorsement of senior comprehensive examinations.
850
HENDERSON, FUJITA, TROPE, AND LIBERMAN
Table 1 Likelihood Ratings by Spatial Distance (Study 3) Near Typical event Number of inches of rainfall on campus has been 4.11 to 4.69 per month, with an average monthly rainfall of 4.40 Likelihood that the rainfall will be at least 4.11 Number of visits to health center per student has been .39 to .51, with an average of .45 Likelihood that the number of visits will be no greater than .50 visits Number of pages of photocopied documents per student per class has been 72 to 112, with an average of 92 Likelihood that students will receive no more than 109 pages Number of reported hours of sleep per student per night has been 6.3 to 7.1, with an average of 6.7 Likelihood that students will sleep more than 6.2 hr
Distant
M
SD
M
SD
7.00
1.69
7.20
1.32
5.92
1.77
6.48
1.90
4.63
1.88
5.96
1.95
5.96
2.24
6.72
1.65
Note. Numbers indicate the likelihood that events would happen the next semester (Fall 2005). Typicality of events was manipulated within-participant. Contact the authors for the exact wording of the scenarios.
distant places. This may underlie the romantic attitude about traveling. Indeed, many people plan vacations with significant others with the intent of discovering the true nature of themselves and those they travel with. Having this construal in mind, romantic partners, for example, may travel to remote places hoping to discover their partner’s true self, only to find that their partner is the same as in everyday life.
Study 3: Spatial Distance and Predictions for Typical and Atypical Events Higher level construals by definition impose one of a few possible alternative interpretations of information about events. Because inconsistent or irrelevant information about events is omitted or assimilated into such construals, judgments based on higher level construals are less ambiguous and more prototypical than judgments based on lower level construals. Therefore, when information is provided about the central tendency of an event’s category, predictions about whether spatially distant events will occur should be more affected by whether the events resemble the central tendency of the category than predictions about spatially near events. On the basis of this reasoning, we designed this study to test the hypothesis that as spatial distance from an event is increased, individuals will predict that typical events are more likely and atypical events are less likely.
Method Participants. A sample of 49 students (37 women, 11 men) at NYU participated in partial fulfillment of a course requirement; 1 participant failed to indicate his or her gender. They were randomly assigned to condition, with 24 participants in the spatially near condition and 25 participants in the spatially distant condition. Materials and procedures. Participants were tested in groups of 1 to 4 in a study ostensibly about NYU campuses. Participants were presented with eight scenarios describing the central tendency for various events’ categories (e.g., “Students report sleeping an average of 6.3 to 7.1 hours a night while studying at the NYU campus in Manhattan [Florence] during the fall semester, averaging 6.7 hours per night.”).4 The events were described as located at “the NYU campus in Manhattan” (spatially near condition) or “the NYU campus in Florence, Italy” (spatially distant condition). Half of the scenarios asked participants to estimate the likelihood that a typical event (i.e., an event that occurs within the stated range) would happen next year (see Table 1), while the other half asked partici-
pants to estimate the likelihood that an atypical event (i.e., an event that occurs outside of the stated range) would happen next year (see Table 2). In the example above, participants were asked to estimate the likelihood that “students will sleep greater than 6.2 hours per night while studying during the Fall 2005 semester at the NYU campus in Manhattan [Florence].” Participants responded on a 9-point scale, anchored at 1 (not at all likely) and 9 (extremely likely). The order of the scenarios was randomly determined and fixed across all participants. Afterward, participants were debriefed and dismissed.
Results and Discussion To test the effects of spatial distance on judgments that typical and atypical events would occur in the future, we averaged participants’ likelihood ratings together for each within-participant condition (scenarios that asked about typical vs. atypical events). Participants’ likelihood ratings were analyzed using a 2 (spatial distance: near vs. distant) ⫻ 2 (type of event: typical vs. atypical) repeated measures ANOVA, with the first factor a betweenparticipants variable and the last factor a within-participant variable. Although there was no main effect of spatial distance (F ⬍ 1), there was a significant main effect of type of event, F(1, 47) ⫽ 75.27, p ⬍ .001, with participants estimating a greater likelihood that typical (M⫽ 6.24) rather than atypical events (M⫽ 3.69) would occur. It is important to note that analyses did reveal a significant Spatial Distance ⫻ Type of Event interaction effect, F(1, 47) ⫽ 5.62, p ⬍ .05 (see Figure 2). Specific comparisons revealed, as expected, that the predicted likelihoods that typical events would occur were higher for spatially distant locations (M ⫽ 6.59) than spatially near locations (M ⫽ 5.88), t(47) ⫽ 2.13, p ⬍ .05, d ⫽ .62. Also as expected, the predicted likelihoods that atypical events would occur were lower for spatially distant locations (M ⫽ 3.36) than spatially near locations (M ⫽ 4.03), t(47) ⫽ 1.78, p ⫽ .08, d ⫽ .52.5 4
Materials can be obtained from the authors. We reran the analyses after standardizing the likelihood rating for each event so that scores reflected the degree to which the participants’ predicted likelihood for an activity deviated from the mean predicted likelihood for that activity by all participants. As expected, results still yielded a significant Spatial Distance ⫻ Type of Event interaction effect, F(1, 47) ⫽ 5.38, p ⬍ .05. 5
SPATIAL DISTANCE AND SOCIAL JUDGMENT
851
Table 2 Likelihood Ratings by Spatial Distance (Study 3) Near Atypical events Number of degrees in °F on campus has been 51 to 59, with an average of 55 Likelihood that the temperature will be no greater than 49°F Ratio of women to men on campus has been .47 to .52, with an average of .49 Likelihood that ratio will exceed .53 Number of hours of daylight on campus per day has been 13.8 to 14.6, with an average of 14.2 Likelihood that the length of daylight will be more than 14.7 Number of inches of snowfall on campus has been 1.9 to 2.7 per month, with an average of 2.3 Likelihood that the snowfall will be less than 1.9
Distant
M
SD
M
SD
3.25
2.23
3.48
1.78
5.25
2.11
3.92
1.68
3.29
2.18
2.48
1.96
4.33
2.09
3.56
1.53
Note. Numbers indicate the likelihood that events would happen the next semester (Fall 2005). Typicality of events was manipulated within-participant. Contact the authors for the exact wording of the scenarios.
One potential alternative explanation for the present findings might be that the base rates for what we presented as typical and atypical events were perceived to be different in the spatially near and distant conditions. For example, it is plausible that students believed that the base rate for the women-to-men ratio was smaller on NYU’s Florence campus and thus judged the likelihood of women outnumbering men on the Florence campus to be less likely. It is important to note, however, that we essentially provided base-rate information when we presented them with the central tendency information for atypical and typical events. This information was held constant across the two distance conditions. Nevertheless, it is still possible that the plausibility of this central tendency information differed between the two locations. To address this alternative explanation, we conducted a pilot study to examine whether students judged the different events described in Study 3 as more or less likely to happen in Italy versus New York. Thirty-one participants judged the average level of the events on NYU’s New York campus (spatially near condition), whereas another 31 participants judged the average level of the events on NYU’s Florence campus (spatially distant condition). For example, participants were asked, “How likely is it that the female to
Predicted Likelihood
7
6
atypical events typical events
5
male ratio of students during the fall semester at the NYU campus in Manhattan [Florence, Italy] is .49?” and “How likely is it that the average number of visits to the health center by each student during the fall semester at the NYU campus in Manhattan [Florence, Italy] is .45 visits?” (see Tables 1 and 2 for a complete list of events). Participants responded on a 7-point scale, anchored at 1 (not at all likely) and 7 (extremely likely). As expected, results revealed no systematic differences across the typical and atypical events between the spatially near and distant conditions; participants in the spatially near condition did not significantly differ from participants in the spatially distant condition in their estimated likelihoods for any of the events presented in Study 3. The present results add support to the idea that judgments about more spatially distant events are based on higher level construals of those events. We assumed that values that are representative of a distribution (those that are close to the mean) constitute a high-level construal of the distribution. We therefore hypothesized, and the results actually showed, that these values would be predicted with greater confidence for spatially remote locations than spatially proximal locations. Atypical events, on the other hand, were predicted to occur in proximal locations more than in spatially remote locations. These findings have implications for how people decide between devoting their limited resources toward preparing for likely and unlikely events. For example, individuals can be expected to make plans for events that take place at spatially distant locations that anticipate common or ordinary events, but they are not expected to plan for rare or unusual events (e.g., finance experts prepare more for unlikely market fluctuations in local markets). Although Murphy’s Law states, “If anything can go wrong, it will,” the results from this study suggest that individuals rely less on this adage when making predictions for spatially distant events.
Study 4: Spatial Distance and Extrapolations Based on Trends Versus Deviations
4
3 near
distant Spatial Distance
Figure 2. Predictions for events as a function of spatial distance and typicality of events (Study 3)
The previous study examined static situations in which the data participants based their predictions on did not show any trends or developments over time. In the present study we sought to extend our logic to dynamic situations, in which participants make inferences based on trends. Specifically, we examined the effect of spatial distance on people’s readiness to extrapolate from general trends as opposed to local information (deviations). Consider, for
HENDERSON, FUJITA, TROPE, AND LIBERMAN
Student Reported Satisfaction with Food Quality (1 to 9) 9 8
A
7 Satisfaction
example, an individual who has a history of behaving a certain way and then on one occasion deviates from her usual course of action. Does the perceived spatial distance from that person affect the weight that is placed on the deviant behavior when predicting how she will act in the future? More generally, when individuals have knowledge about the general history of any event and they encounter a specific case that is inconsistent with that history, how much weight are they likely to place on the specific case? In terms of CLT, general trends convey a high-level, abstract rule about how the future will manifest itself, whereas deviations from trends represent a low-level, concrete exception to the rule. Consequently, when information is provided about a general trend surrounding an event and a specific case that deviates from that trend, extrapolations for spatially distant rather than near events should be based more on the general trend than on the local deviation. In the current study, we presented participants with a series of graphs, each showing an upward or downward trend of cases charted over several years for various events related to an academic year (e.g., average photocopies per student per class). On each of the graphs, the final case that was presented for the last academic year always deviated from the overall trend of cases from the previous academic years. That is, the last and most recent case of a generally upward graph would deviate in the downward direction, and vice-versa for a generally downward graph (see Figure 3 for examples). Participants were asked to extrapolate whether the case for the next academic year would be consistent with the general trend or consistent with the specific case that deviated from the trend. We tested the hypothesis that as spatial distance from an event is increased, individuals would be more likely to extrapolate for the next academic year from a general trend relative to a specific case that deviates from the general trend.
6
B
5 4 3 2 1 1998
1999
2000
2001
2002
2003
2004
2005
2006
Academic Year
Average Hours Per Night of Sleep Reported by Students 8.5 8 7.5 Hours
852
A
7
B
6.5 6 5.5
Method Participants. A sample of 58 students (42 women, 16 men) at New York University participated for $10 or in partial fulfillment of a course requirement. They were randomly assigned to condition, with equal numbers in each condition. Materials and procedures. Participants were tested in groups of 1 to 10 in a study on predictions. Participants were presented with six graphs depicting information about several events that took place at NYU from 1999 to 2004 (e.g., “students’ reported satisfaction with food quality”).6 The events were described as occurring at “the NYU campus in Manhattan” (spatially near condition) or “the NYU campus in Florence, Italy” (spatially distant condition). Half of the graphs depicted an upward trend of cases for 1999 –2003, and the other half depicted a downward trend of cases for 1999 –2003. For the graphs that depicted an upward trend of cases, the last case for 2004 deviated downward. For the graphs that depicted a downward trend of cases, the last case for 2004 deviated upward (see Figure 3 for examples). Participants were asked to estimate the likelihood that cases for 2005 would go up from the previous year as well as the likelihood that cases would go down from the previous year. In the example above, participants were asked, “How likely will the graph continue to Point A?” and “How likely will the graph continue to Point B?” Participants responded on a 6-point scale, anchored at 1 (very unlikely) and 6 (very likely). We created two random orders of the graphs, included order as a factor for the analyses, and controlled for order in the analyses; the pattern of results was unchanged in all cases. Thus, the order of the graphs is not discussed further. Afterward, participants were debriefed and dismissed.
5 1998
1999
2000
2001
2002
2003
2004
2005
2006
Academic Year
Figure 3. Examples of graphs that were used to depict an upward or downward trend of cases for 1999 –2004 (Study 4)
Results and Discussion In this study, we asked participants to estimate whether the level of an event would continue up or down on a graph (see Figure 3 for examples). For every graph that depicted a general upward trend, we averaged across participants’ estimated likelihoods that the next point on the graph would go up, yielding an index of participants’ extrapolations based on upward trends. For these same generally upward graphs, we also averaged across participants’ estimated likelihoods that the next point would go down, yielding an index of participants’ extrapolations based on deviations from upward trends. The same was done for downward graphs. That is, for every generally downward graph, we averaged across participants’ estimated likelihoods that the next point on the 6
Materials can be obtained from the authors.
SPATIAL DISTANCE AND SOCIAL JUDGMENT
853
Table 3 Likelihood Ratings by Spatial Distance (Study 4) Near Trend Graph Upward trend Student satisfaction with food Average no. of photocopies Average class attendance Downward trend Medical office visits Acceptance rates Average hr per night of sleep
Distant Deviation
Trend
Deviation
M
SD
M
SD
M
SD
M
SD
3.62 3.48 3.52
1.01 1.43 1.09
4.03 4.24 4.03
0.78 0.79 0.82
4.21 3.62 3.93
1.11 1.21 0.88
3.21 3.93 3.72
1.32 1.13 1.10
3.52 4.59 4.00
0.99 1.05 1.07
4.07 3.00 3.45
0.80 1.22 0.87
4.00 3.97 4.28
1.10 1.40 0.92
3.41 3.31 3.24
1.15 1.20 0.95
Note. Numbers in the trend column indicate the likelihood that the graph would continue in direction suggested by the trend, and numbers under the deviation column indicate the likelihood that the graph would continue in the direction suggested by the case that deviated from the trend. The type of trend for the graphs was manipulated within-participant. Upward trends refer to graphs that depicted an incline in the rates from 1999 –2003 with a decline in the rate for 2004. Downward trends refer to graphs that depicted a decline in the rates from 1999 –2003 with an incline in the rate for 2004.
graph would go down (an index of participants’ extrapolations based on downward trends) and up (an index of participants’ extrapolations based on deviations from downward trends; see Table 3 for extrapolations for each graph). To test the effects of spatial distance on extrapolations, we analyzed participants’ extrapolations using a 2 (spatial distance: near vs. distant) ⫻ 2 (basis of extrapolations: trend vs. deviation) ⫻ 2 (type of trend: upward vs. downward) repeated measures ANOVA, with the first factor a between-participants variable and the last two factors within-participant variables. Results showed no main effects of spatial distance (F ⬍ 1) or the type of trend, F(1,56) ⫽ 2.25, p ⫽ .14. A main effect of the basis of extrapolations, F(1, 56) ⫽ 3.48, p ⫽ .07, did emerge, with participants evidencing a greater likelihood of extrapolating from general trends (M ⫽ 3.89) than from deviations from general trends (M ⫽ 3.64). Results also revealed a Basis of Extrapolations ⫻ Type of Trend effect, F(1, 56) ⫽ 15.24, p ⬍ .001. Specific comparisons revealed that although participants were more likely to extrapolate on the basis of downward trends (M ⫽ 4.06) than on deviations from downward trends (M ⫽ 3.41), t(57) ⫽ 3.68, p ⬍ .001, d ⫽ .97, they were equally likely to extrapolate based on upward trends (M ⫽ 3.73) and deviations from upward trends (M ⫽ 3.86), t ⬍ 1. It is important to note that, as expected, the Spatial Distance ⫻ Basis of Extrapolation effect also emerged, F(1, 56) ⫽ 3.97, p ⫽ .05, which was not qualified by the three-way interaction, F(1, 56) ⫽ 2.53, p ⫽ .12. In order to further examine the Spatial Distance ⫻ Basis of Extrapolation effect, we averaged across participants’ extrapolations on the basis of upward and downward trends, yielding an index of participants’ extrapolations based on general trends and across participants’ extrapolations on the basis of deviations from upward and downward trends, yielding an index of participants’ extrapolations based on deviations. As can be seen in Figure 4, specific comparisons revealed that participants in the spatially distant condition were less likely to extrapolate on the basis of deviations from general trends (M ⫽ 3.47) than were participants in the spatially near condition (M ⫽ 3.80), t(56) ⫽ 2.43, p ⬍ .05, d ⫽ .92; no differences emerged in their likelihood to extrapolate
on the basis of general trends, (M ⫽ 4.00 vs. M ⫽ 3.79), t(56) ⫽ 1.22, p ⫽ .23, d ⫽ .33. Results also showed that participants in the spatially distant condition were more likely to extrapolate on the basis of general trends than on deviations from general trends, t(28) ⫽ 2.47, p ⬍ .05, d ⫽ .94, whereas participants in the spatially near condition failed to show any difference in their extrapolations on the basis of general trends and deviations from general trends.7 Consistent with our hypothesis, these results demonstrate that individuals rely more on general trends and less on local deviations from those trends when making predictions about relatively distant locations. The results from this study complement the findings from the previous study, as both highlight how individuals are more likely to rely on global information (in this case, information about general trends) when making predictions for events that occur in spatially distant locations. Study 3 demonstrated people’s tendency to rely on global information when making predictions in static situations, whereas Study 4 demonstrated people’s tendency to rely on global information when making predictions in dynamic situations. An interesting implication of these studies is that when individuals (e.g., U.S.-affiliated stock brokers) make decisions (e.g., investments) based on information about spatially near events (e.g., the stock market on Wall Street) rather than distant events (e.g., the stock market in Tokyo), they will be more likely to exaggerate the significance of small departures from general data patterns. When individuals are faced with an unusual deviation (e.g., rejection of a paper) from how events have typically unfolded in the past (e.g., a positive trend in one’s career), the question is always whether this atypical case is a turning point or 7
We reran the analyses after standardizing the extrapolations for each graph so that extrapolations reflected the degree to which the participants’ extrapolations for a graph deviated from the mean extrapolation for that graph by all participants. As expected, results still yielded a significant Spatial Distance ⫻ Basis of Extrapolation interaction effect, F(1, 56) ⫽ 4.66, p ⬍ .05.
HENDERSON, FUJITA, TROPE, AND LIBERMAN
854 4.4
Predicted Likelihood
4.2 4 3.8
based on deviations based on trends
3.6 3.4 3.2 3 near
distant Spatial Distance
Figure 4. Extrapolations on the basis of general trends versus deviations from trends as a function of spatial distance (Study 4)
a fluctuation. The current findings have the potential to offer insights into how people respond to information that is unexpected or unforeseen given what they know about the past (e.g., freak weather events, spurts or collapses in financial market), as individuals appear to assign less weight to such information when forming judgments for spatially distant events. For example, these results suggest that as individuals (e.g., those with a history of infidelity) attempt to make drastic, personal changes (e.g., become more monogamous), those who are spatially distant rather than close who learn about such attempts are less likely to perceive any actual dispositional change (e.g., see Libby, Eibach, & Gilovich, 2005, for related findings).
General Discussion According to CLT, as individuals become more psychologically removed from events, their construal of events moves to a higher level. High-level construals structure the information about events in terms of abstract, global features that convey the essence of the event. As events become more psychologically close, individuals’ construal of events moves to a lower level. These construals are less structured and represent the available information about an event in terms of specific, local features. The spatial distance from events is thought to operate as a fundamental distance dimension through which individuals’ construal of events will move to higher or lower levels. In the current set of studies, we examined the extent to which increased spatial distance from events fostered greater reliance on high-level construals during judgment and prediction. Specifically, we showed that participants structured behaviors into broader units (Study 1) and increasingly attributed behavior to global dispositions rather than specific situational constraints (Study 2) when those behaviors were spatially distant rather than near. Moreover, we demonstrated that participants expected events that were prototypical and closer to the general case to be more likely when events were more spatially distant (Study 3). Finally, we demonstrated that participants were less likely to extrapolate from specific cases that deviated from general
trends when making predictions about more spatially distant events (Study 4). Together, these results support CLT’s assertion that judgments regarding experiences in distant locations are based more on global, abstract representations than judgments about the same experiences in near locations. It is important to address the possibility that participants in our studies may have perceived the near location as more self-relevant and, therefore, engaged in more effortful processing while forming judgments for spatially near experiences. One might argue, for example, that the greater unitization (Study 1) that was expressed for spatially near behavior required more effort. It is thus possible that high self-relevance might have enhanced the tendency to engage in more systematic, effortful (as opposed to heuristic, low-effort) processing (see Chen & Chaiken, 1999; Todorov, Chaiken, & Henderson, 2002) and that such a difference in processing explains the results obtained. However, several aspects of our results argue against this interpretation. First, in Study 1, no differences were found between conditions in the perceived difficulty of processing information about the social event. Second, in Studies 2 through 4, we took care to ensure that self-relevance was equal across the spatial distance conditions by holding participants’ school affiliation with the specified location constant across conditions. Third, regarding Study 3, there is no a priori reason to suspect that more effortful processing in the spatially near condition (if it even occurred) would reduce the utilization of information about the central tendency of a distribution. Last, the results from Study 4 cannot be accounted for by differences in effort as there is no a priori reason (similar to Study 3) to suspect that more systematic processing in the near condition would lead to more weight on instances that deviate from general trends when making predictions of future behavior. It is important to note that we do acknowledge that it is difficult to completely rule out this alternative explanation. However, given the consistency of results with CLT’s framework as well as the recent empirical work that has shown that experiences that occur at ostensibly spatially distant locations are construed at a higher level than spatially near experiences (Fujita, Henderson, et al., 2006), a construal level explanation seems to be the most consistent and parsimonious one available for the current findings. Nevertheless, it remains for future researchers to assess the exact role that construal level plays in the effects of spatial distance on social judgment. Our purpose in the present article was to demonstrate that several judgments that have been deemed as important for social reasoning and functioning are more likely to be based on highlevel construals for spatially distant rather than near events. Overall, the current results show that increasing the reported spatial distance of an event increases the impact of high-level information (i.e., central tendencies, general trends, dispositional characteristics) and decreases the impact of low-level information (i.e., incidental details, irregular outcomes, situationspecific task characteristics) on social judgment and decision-making.
Evaluation The implications of spatial construal for evaluation and choice also deserve attention. Previous CLT work has already demonstrated that the perceived value of objects and activities derives from their construal (Fujita, Trope, Liberman, & Levi-Sagi, 2006;
SPATIAL DISTANCE AND SOCIAL JUDGMENT
Liberman & Trope, 1998; Trope & Liberman, 2000). Individuals frequently form evaluations of objects and activities that differ in the valence associated with their high-level, central features and low-level, peripheral features. When the value of high-level features (e.g., quality of music bands) of an object or activity (e.g., music concert) is different from the value of its low-level features (e.g., quality of available food), then changing the level of representation of the target object or activity results in a corresponding change in its perceived value. By forming evaluations about spatially distant objects and activities that are based on high-level, central information and evaluations about spatially near objects and activities that are based on low-level, peripheral information, individuals’ evaluations can be expected to differ as a function of the amount of perceived spatial distance from objects and activities. Specifically, the findings from the present studies suggest that when individuals form evaluations and make choices for spatially distant rather than near objects and activities, they will give more weight to high-level concerns over low-level concerns when expressing their evaluations and preferences. Indeed, previous research has shown that individuals give more weight to their abstract ideals, values, and desirability concerns for temporally distant activities (Liberman & Trope, 1998; Liberman et al., in press; Pennington & Roese, 2003), suggesting that a similar focus will occur for spatially distant activities. This framework, for example, might account for why individuals (e.g., early European settlers) who plan to move to remote locations (e.g., the New World) often have visions of a utopia. Moreover, this framework might account for why organizations (e.g., the U.S. government, Al Qaeda) frequently make plans for distant locations (e.g., the Middle East, the United States) that center on establishing their preferred ideological systems.
Psychological Distance According to CLT (see Liberman, et al., in press; Trope & Liberman, 2003), space is just one of many psychological distance dimensions that influence the way individuals construe information when forming judgments about experiences. CLT posits that the same general principles that apply to spatial distance should also hold for other distance dimensions, including social distance. Future research should explore, for example, whether having individuals adopt a third-person rather than first-person perspective for another increases the impact of high-level information (e.g., typical behaviors, behaviors that convey a general pattern) and decreases the impact of low-level information (i.e., irregular behaviors, deviant behaviors) when forming judgments and evaluations about another. Furthermore, future research should also explore whether individuals put more weight on high-level information when forming judgments and evaluations of others and more weight on low-level information when forming judgments and evaluations of themselves. Indeed, Libby and Eibach (2004) and Fiedler, Semin, Finkenauer, and Berkel (1995) have reported effects that are consistent with these predictions.
Conclusion The four studies reported here suggest that increasing the reported spatial distance from events increases people’s propensity to rely on abstract, global, general information when forming
855
judgments and decisions in regard to those events. Unlike temporal distance, spatial distance from objects and events has received little, if any, attention in research on judgment and decisionmaking. As humans’ geographical and spatial horizons expand, it becomes increasingly important to study how humans transcend not only the “now” but also the “here.” We hope the present research is a step in this direction.
References Berti, A., & Fassinetti, F. (2000). When far becomes near: Remapping of space by tool use. Journal of Cognitive Neuroscience, 12, 415– 420. Bryant, D. J., & Tversky, B. (1999). Mental representations of spatial relations from diagrams and models. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25, 137–156. Chase, W. G., & Simon, H. A. (1973). Perception and chess. Cognitive Psychology, 4, 55– 81. Chen, S., & Chaiken, S. (1999). The heuristic-systematic model in its broader context. In S. Chaiken & Y. Trope (Eds.), Dual-process theories in social psychology (pp. 73–96). New York: Guilford Press. Cohen, C. E., & Ebbesen, E. B. (1979). Observational goals and schema activation: A theoretical framework for behavior perception. Journal of Experimental Social Psychology, 15, 305–329. Fiedler, K., Semin, G. R., Finkenauer, C., & Berkel, I. (1995). Actorobserver bias in close relationships: The role of self-knowledge and self-related language. Personality and Social Psychology Bulletin, 21, 525–538. Fujita, K., Henderson, M. D., Eng, J., Trope, Y., & Liberman, N. (2006). Spatial distance and mental construal of social events. Psychological Science, 17, 278 –282. Fujita, K., Trope, Y., Liberman, N., & Levi-Sagi, M. (2006). Construal levels and self-control. Journal of Personality and Social Psychology, 90, 351–367. Gilbert, D. T., & Malone, P. S. (1995). The correspondence bias. Psychological Bulletin, 117, 21–38. Gilovich, T., & Medvec, V. H. (1995). The experience of regret: What, when, and why. Psychological Review, 102, 379 –395. Halligan, P. W., Fink, G. R., Marshall, J. C., & Vallar, G. (2003). Spatial cognition: Evidence from visual neglect. Trends in Cognitive Science, 7, 125–133. Heider, F., & Simmel, M. (1944). An experimental study of apparent behavior. American Journal of Psychology, 57, 243–249. Henderson, M. D., Trope, Y., & Carnevale, P. J. (in press). Negotiation from a near and distant time perspective. Journal of Personality and Social Psychology. Huttenlocher, J., Hedges, L. V., & Duncan, S. (1991). Categories and particulars: Prototype effects in estimating spatial location. Psychological Review, 98, 352–376. Jones, E. E., & Harris, V. A. (1967). The attribution of attitudes. Journal of Experimental Social Psychology, 3, 1–24. Keren, G., & Roelofsma, P. (1995). Immediacy and certainty in intertemporal choice. Organizational Behavior & Human Decision Processes, 63, 287–297. Klin, A. (2000). Attributing social meaning to ambiguous visual stimuli in higher-functioning autism and Asperger syndrome: The social attribution task. Journal of Child Psychology and Psychiatry and Allied Disciplines, 41, 831– 846. Kudadjie-Gyamfi, E., & Rachlin, H. (1995). Temporal patterning in choice among delayed outcomes. Organizational Behavior & Human Decision Processes, 65, 61– 67. Lassiter, G. D., Geers, A. L., Apple, K. J., & Beers, M. J. (2000). Observational goals and behavior unitization: A reexamination. Journal of Experimental Social Psychology, 36, 649 – 659. Lassiter, G. D., & Stone, J. I. (1984). Affective consequences of variation
856
HENDERSON, FUJITA, TROPE, AND LIBERMAN
in behavior perception: When liking is in the level of analysis. Personality and Social Psychology Bulletin, 10, 253–259. Lassiter, G. D., Stone, J. I., & Rogers, S. L. (1988). Memorial consequences of variation in behavior perception. Journal of Experimental Social Psychology, 24, 222–239. Libby, L. K., & Eibach, R. P. (2004). Seeing meaning in action: Imagery perspective and action identification. Unpublished data. Libby, L. K., Eibach, R. P., & Gilovich, T. (2005). Here’s looking at me: The effect of memory perspective on assessments of personal change. Journal of Personality and Social Psychology, 88, 50 – 62. Liberman, N., Sagristano, M., & Trope, Y. (2002). The effect of temporal distance on level of mental construal. Journal of Experimental Social Psychology, 38, 523–534. Liberman, N., & Trope, Y. (1998). The role of feasibility and desirability considerations in near and distant future decisions: A test of temporal construal theory. Journal of Personality and Social Psychology, 75, 5–18. Liberman, N., Trope, Y., & Stephan, E. (in press). Psychological distance. In E. T. Higgins & A. W. Kruglanski (Eds.), Social psychology: Handbook of basic principles (Vol. 2). New York: Guilford Press. Markus, H., Smith, J., & Moreland, R. L. (1985). Role of the self-concept in the perception of others. Journal of Personality and Social Psychology, 48, 1494 –1512. Massad, C. M., Hubbard, M., & Newtson, D. (1979). Selective perception of events. Journal of Experimental Social Psychology, 15, 513–532. McNamara, T. (1986). Mental representations of spatial relations. Cognitive Psychology, 18, 87–121. Newtson, D. (1973). Attribution and the unit of perception of ongoing behavior. Journal of Personality and Social Psychology, 28, 28 –38. Newtson, D., & Engquist, G. (1976). The perceptual organization of ongoing behavior. Journal of Experimental Social Psychology, 12, 847– 862. Nussbaum, S., Trope, Y., & Liberman, N. (2003). Creeping dispositionism: The temporal dynamics of behavior prediction. Journal of Personality and Social Psychology, 84, 485– 497. Pennington, G. L., & Roese, N. J. (2003). Regulatory focus and temporal distance. Journal of Experimental Social Psychology, 39, 563–576. Rachlin, H. (1995). The value of temporal patterns in behavior. Current Directions in Psychological Science, 4, 188 –192. Reitman, J. (1976). Skilled perception in GO: Deducing memory structures from inter-response times. Cognitive Psychology, 8, 336 –356. Rosch, E. (1975). Cognitive representations of semantic categories. Journal of Experimental Psychology: General, 104, 192–223. Rosch, E., & Lloyd, B. B. (1978). Cognition and categorization. Hillsdale, NJ: Erlbaum. Rosch, E., Mervis, C., Gray, W., Johnson, D., & Boyes-Braem, P. (1976). Basic objects in natural categories. Cognitive Psychology, 8, 382– 439.
Ross, M. (1989). Relation of implicit theories to the construction of personal histories. Psychological Review, 96, 341–357. Ross, M., & Wilson, A. E. (2002). It feels like yesterday: Self-esteem, valence of personal past experiences, and judgments of subjective distance. Journal of Personality and Social Psychology, 82, 792– 803. Semin, G. R., & Fiedler, K. (1988). The cognitive functions of linguistic categories in describing persons: Social cognition and language. Journal of Personality and Social Psychology, 54, 558 –568. Sherman, S. J., Zehner, K. S., Johnson, J., & Hirt, E. R. (1983). Social explanation: The role of timing, set, and recall on subjective likelihood estimates. Journal of Personality and Social Psychology, 44, 1127– 1143. Todorov, A., Chaiken, S., & Henderson, M. D., (2002). The heuristicsystematic model of social information processing. In J. Dillard & M. Pfau (Eds.), The persuasion handbook: Developments in theory and practice (pp. 195–211). Thousand Oaks, CA: Sage. Trope, Y., & Liberman, N. (2000). Temporal construal and time-dependent changes in preference. Journal of Personality and Social Psychology, 79, 876 – 889. Trope. Y., & Liberman, N. (2003). Temporal construal. Psychological Review, 110, 403– 421. Tversky, B. (1981). Distortions in memory for maps. Cognitive Psychology, 13, 407– 433. Tversky, B. (2003). Structures of mental spaces: How people think about space. Environment and Behavior, 35, 66 – 80. Tversky, B. (2005). Functional significance of visuospatial representations. In P. Shah & A. Miyake (Eds.), Handbook of higher-level visuospatial thinking (pp. 1–70). Cambridge, England: Cambridge University Press. Vallacher, R. R., & Wegner, D. M. (1987). What do people think they’re doing? Action identification and human behavior. Psychological Review, 94, 3–15. Wilder, D. (1978a). Effects of predictability on units of perception and attribution. Personality and Social Psychology Bulletin, 4, 281–284. Wilder, D. (1978b). Predictability of behaviors, goals, and unit of perception. Personality and Social Psychology Bulletin, 4, 604 – 607. Wilson, T. D., Wheatley, T., Meyers, J. M., Gilbert, D. T., & Axsom, D. (2000). Focalism: A source of durability bias in affective forecasting. Journal of Personality and Social Psychology, 78, 821– 836. Zauberman, G., & Lynch, J. G., Jr. (2005). Resource slack and propensity to discount delayed investments of time versus money. Journal of Experimental Psychology: General, 134, 23–37.
Received November 2, 2005 Revision received February 20, 2006 Accepted March 6, 2006 䡲
INTERPERSONAL RELATIONS AND GROUP PROCESSES
The Costs and Benefits of Undoing Egocentric Responsibility Assessments in Groups Eugene M. Caruso
Nicholas Epley
Harvard University
University of Chicago
Max H. Bazerman Harvard Business School Individuals working in groups often egocentrically believe they have contributed more of the total work than is logically possible. Actively considering others’ contributions effectively reduces these egocentric assessments, but this research suggests that undoing egocentric biases in groups may have some unexpected costs. Four experiments demonstrate that members who contributed much to the group outcome are actually less satisfied and less interested in future collaborations after considering others’ contributions compared with those who contributed little. This was especially true in cooperative groups. Egocentric biases in responsibility allocation can create conflict, but this research suggests that undoing these biases can have some unfortunate consequences. Some members who look beyond their own perspective may not like what they see. Keywords: judgment and decision making, egocentrism, perspective taking, heuristics, biases
Unfortunately, research suggests that people too often claim credit like Banting and Macleod, rather than like Kahneman and Tversky. Indeed, people are notorious for claiming more responsibility in collective endeavors than they objectively deserve. In the classic demonstration (M. Ross & Sicoly, 1979), for example, married couples were asked to assess their responsibility for a variety of household activities, such as preparing breakfast, shopping, and making important decisions. When summed together, self-allocated responsibility exceeded 100%, indicating that at least one member of the couple was—perhaps sorely—mistaken. Similar results have been observed across domains as diverse as fund-raising (Zander, 1971), academics (M. Ross & Sicoly, 1979), and athletics (Brawley, 1984; Forsyth & Schlenker, 1977), just to name a few (for a review, see Leary & Forsyth, 1987). Such egocentric biases in responsibility allocations tend to occur, at least in part, because people focus too much on their own contributions and too little—if at all— on others’ contributions. As the opening example suggests, failing to credit others’ contributions by egocentrically focusing on one’s own can create considerable conflict among group members, even if they are not disagreeing about responsibility for a Nobel Prize (Babcock & Loewenstein, 1997; Forsyth, Berger, & Mitchell, 1981; Forsyth & Mitchell, 1979). These egocentric biases have been cited as one of the key instigators of dissatisfaction and conflict in groups (Bazerman & Neale, 1982; Neale & Bazerman, 1983; L. L. Thompson & Loewenstein, 1992). Reducing such egocentric biases by leading people to consider their collaborators’ contributions would therefore seem to be a simple strategy for minimizing the unhappiness, dissatisfaction, and conflict they produce.
Banting and Macleod won the 1923 Nobel Prize in Medicine for the discovery of insulin. Banting, so outraged at the credit given to Macleod, boycotted the ceremony in Stockholm and awarded half of his own prize money to a lab coworker. Macleod, who oversaw Banting’s experiments as director of the laboratory, conveniently failed to mention Banting in speeches about the research (Harris, 1946). Contrast this animosity to the 2002 Nobel Prize in Economics awarded to Daniel Kahneman, whose late collaborator and close friend Amos Tversky was ineligible to receive the distinction posthumously. Consistent with the collaborative nature of their relationship, Kahneman’s opening line in his award address emphasized the critical importance of Tversky’s efforts in their research together—a sentiment he has stressed many times before and since.
Eugene M. Caruso, Department of Psychology, Harvard University; Nicholas Epley, Graduate School of Business, University of Chicago; Max H. Bazerman, Harvard Business School. This research was supported by a Graduate Research Fellowship from the National Science Foundation to Eugene M. Caruso and by National Science Foundation Grant SES0241544 and a James S. Kemper Foundation Faculty Research Fund grant to Nicholas Epley. We thank Tara Abbatello, Dolly Chugh, Sean Darling-Hammond, Angela Kim, Sarah Murphy, Kristian Myrseth, Heather Omoregie, Dobromir Rahnev, Aram Seo, and Erin Rapien Whitchurch for assistance conducting these experiments. Correspondence concerning this article should be addressed to Nicholas Epley, University of Chicago, 5807 South Woodlawn Avenue, Chicago, IL 60637. E-mail:
[email protected] (Eugene M. Caruso),
[email protected] (Nicholas Epley), or
[email protected] (Max H. Bazerman).
Journal of Personality and Social Psychology, 2006, Vol. 91, No. 5, 857– 871 Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.5.857
857
858
CARUSO, EPLEY, AND BAZERMAN
However, removing an egocentric focus on one’s own contributions requires placing that focus on others’ contributions, and we suggest that whether egocentrism in social interaction exacerbates or diminishes conflict depends on what people see when they shine the spotlight on others’ contributions. To the extent that people believe they have done more work than their peers, considering others’ contributions by adopting their perspective may highlight how little others have contributed, thereby increasing perceptions of inequity and unfairness. We conducted four experiments to investigate the impact of perspective taking on egocentric allocations of responsibility and two indicators of group conflict— overall enjoyment and interest in future collaboration. We made several predictions. First, because people tend to focus too little on others’ contributions when assessing responsibility for group outcomes, we predicted that leading people to think about the actions of others would decrease self-allocations of responsibility. Second, we predicted a more complicated relationship between perspective taking and the two indicators of group conflict. Highlighting others’ contributions also highlights inequities in the contributions of group members that might have been overlooked. Doing so may lead group members who believe they contributed much (or members who actually did contribute much) to the group outcome to feel dissatisfied with others’ meager contributions and with the inequality in group members’ efforts. In contrast, group members who believed they contributed little (or members who actually did contribute little) may feel relatively more satisfied with others’ more substantial contributions and the extent to which they benefited from working with these group members. We therefore predicted that reducing egocentrism in groups through perspective taking would cause high credit claimers to report decreased enjoyment and interest in future collaborations, relative to their low credit-claiming counterparts. We based our predictions on the mechanisms that produce egocentric responsibility allocations and the importance of equity in groups, to which we now turn.
Egocentric Responsibility Allocations At least two mechanisms contribute to excessive responsibility claiming in groups: motivated reasoning and differential accessibility. In general, people are motivated to view themselves in a favorable light. Claiming more responsibility for positive group outcomes is obviously an effective strategy for improving and maintaining one’s self-image, and people commonly use it (Miller & Schlenker, 1985; M. Ross & Sicoly, 1979, Experiment 2; Schlenker & Miller, 1977). More relevant for the current studies, however, egocentric responsibility allocations are also produced by the differential accessibility of one’s own contributions relative to others’ contributions. People are always present (and usually attentive) for their own contributions, but are not necessarily present for others’ contributions. People are therefore more likely to both notice and recall their own contributions than others’ contributions (Schwarz et al., 1991; Tversky & Kahneman, 1973). Because noticing and remembering are both critical requirements for crediting contributions, people are likely to believe they have contributed more to a group project than others believe they contributed (M. Ross & Sicoly, 1979).1
Three empirical findings are consistent with this accessibility interpretation. First, people overestimate their contributions not only to positive group outcomes but also to negative group outcomes (Kruger & Gilovich, 1999; see also Caine & Schlenker, 1979; M. Ross & Sicoly, 1979). These findings suggest that a motivated desire to see oneself in a positive light is not sufficient to explain the tendency for exaggerated claims of contribution. Second, people naturally report considering information about themselves more than information about others when assigning responsibility (Brawley, 1984; S. C. Thompson & Kelly, 1981). This suggests that one’s own behavior is more accessible when allocating responsibility for collective endeavors and therefore more likely to be used in such judgments. Finally, varying the extent to which participants consider their own versus others’ contributions alters responsibility allocations. Increasing participants’ focus on their own contributions exacerbates the tendency to overestimate one’s contributions (Burger & Rodman, 1983; M. Ross & Sicoly, 1979), whereas increasing their focus on others’ contributions diminishes this tendency (Savitsky, Van Boven, Epley, & Wight, 2005). Even the most dispassionate group members, it appears, would conclude that they have contributed more than is warranted simply because their own contributions are so much easier to notice and recall than are others’ contributions. The impact of differential accessibility in responsibility allocations may be compounded by a related tendency for people to think of other group members as a collective rather than as individuals, even further masking their unique contributions (Savitsky et al., 2005). This tendency for people to pack the constituent elements of a category into a single unit is best seen in research on support theory (Rottenstreich & Tversky, 1997; Tversky & Koehler, 1994), which demonstrates that the perceived likelihood of an event is determined by the amount of support that can be generated in favor of a focal hypothesis relative to alternative hypotheses. Unpacking the constituent elements of a category— by describing them separately rather than collectively, for instance—increases the amount of support that can be generated in favor of a focal hypothesis and therefore increases its perceived likelihood. In one experiment, for example, people indicated that they were more likely to die from “heart disease, cancer, or other natural causes” than simply from “natural causes” (Tversky & Koehler, 1994). This existing research suggests that one effective way to reduce egocentric responsibility allocations is to increase the attention paid to other group members by asking people to unpack their collaborators, considering them as individuals rather than as the rest of the group. Consistent with this possibility, a series of experiments involving debate teams, MBA groups, and academic group projects found that participants asked to unpack (or think about) their collaborators as individuals claimed significantly less credit for the overall work than participants not encouraged to 1 In daily life, of course, motivated reasoning and differential accessibility can work in concert to produce egocentric responsibility allocations, as the desire to view oneself positively can influence the extent to which people search for accessible evidence consistent or inconsistent with this desire (Dawson, Gilovich, & Regan, 2002; Ditto & Lopez, 1992). Our point is not to disentangle these two mechanisms but to simply point out that either can produce egocentric assessments of responsibility.
UNDOING EGOCENTRISM
unpack their collaborators (Savitsky et al., 2005). Reducing egocentric allocations of responsibility by simply asking people to think about others’ contributions therefore seems like a logical way to help restore perceptions of fairness and reduce conflict over inequity in group interactions.
Egocentrism, Equity, and Group Well-Being People who overestimate their own importance may feel underappreciated or believe that others are trying to take advantage of them (Gilovich, Kruger, & Savitsky, 1999). In addition, those who appear to take more credit than they deserve for a group accomplishment are less well liked and thought to be less desirable collaborators (Forsyth et al., 1981). In fact, egocentrism is one of the key instigators of dissatisfaction and conflict in negotiations. Negotiators consistently overestimate the likelihood that a neutral arbitrator will agree with their egocentric assessments of fairness (Bazerman & Neale, 1982; Neale & Bazerman, 1983), and these biased perceptions predict negotiation impasse (L. L. Thompson & Loewenstein, 1992; Wade-Benzoni, Tenbrunsel, & Bazerman, 1996). These results have been replicated in a variety of negotiation contexts, regardless of the presence or absence of financial incentives for performance (Babcock, Loewenstein, Issacharoff, & Camerer, 1995; Camerer & Loewenstein, 1993; Loewenstein, Issacharoff, Camerer, & Babcock, 1993). If egocentric allocations of responsibility contribute to conflict and dissatisfaction within groups, and if focusing people on others’ contributions rather than on their own decreases egocentric allocations, then focusing people on others’ contributions should also reduce conflict and increase satisfaction. Although this argument is logically compelling, we suspect it is often wrong. Whether reducing egocentrism will decrease conflict and dissatisfaction within a group depends, we argue, on what people see when they look beyond their own perspective and into that of their collaborators. Almost all groups include natural variability in actual contributions, with some people doing more and others doing less. Reducing egocentric biases by strengthening the focus on others’ contributions is likely to highlight these differences in actual contributions that otherwise would have been relatively overlooked. Asking someone who contributes a great deal to consider others’ contributions might indeed decrease the relative importance of one’s own contribution but will also highlight the minimal contributions of individual others. This high credit claimer may now be more likely to feel like the division of labor was inequitable, to suspect that others were benefiting unfairly from his or her hard work, and to be less interested in continuing to work with this group in the future, compared with another group member who contributed less. Asking someone who contributes little, in contrast, will again decrease the relative importance of his or her own contributions but will also highlight the impressive efforts of others. A person who has benefited from others’ skills, abilities, and efforts may be more satisfied with his or her participation in the group and more interested in continuing with future collaborations, compared with someone who contributed more. Equity and fairness are paramount concerns in nearly all social relationships (Walster, Walster, & Berscheid, 1978), and reducing an egocentric focus on one’s own contributions may make violations of equity more salient than they
859
would otherwise be, with differential effects on measures of conflict in a group depending on whether one has contributed much or has contributed little. Notice that these predicted effects of reducing egocentrism could be independent of the group’s outcome. Indeed, the results from the hypothetical group project study described above made no mention whatsoever of the group outcome. Although people are more likely to claim responsibility for a group success than a group failure (e.g., Forsyth & Schlenker, 1977), our theory is less concerned with the differential impact of motivations or outcomes on attributions of responsibility than with the impact of decreasing egocentric biases in those attributions. Regardless of the group outcome, reducing a focus on one’s own contributions naturally calls attention to others’ contributions and highlights inequities among group members. Because such inequities are likely to produce relative dissatisfaction with group outcomes among those who contributed much but relative satisfaction with those who contributed little (e.g., Walster et al., 1978), we predicted relatively lower levels of satisfaction and desire for further collaboration among high credit claimers compared with low credit claimers. Ironically, such perspective taking may result in relative dissatisfaction among those who contributed much— precisely the people who would be most beneficial to the group’s future collaborations. Four experiments examined these hypotheses by investigating the impact of perspective taking in group endeavors on two proxies of group conflict, namely perceived enjoyment and interest in future collaborations. In each, participants working as part of a group were either asked to think carefully about the contributions of each individual group member or not before assessing responsibility or before assessing their satisfaction with the group. Consistent with previous research (Savitsky et al., 2005), we predicted that those led to consider their individual collaborators’ contributions would claim less responsibility than those not led to do so. More important, participants in each of our experiments also indicated their enjoyment with the group and their interest in continued collaboration. We predicted that explicitly leading participants to consider others’ contributions would result in relatively lower perceptions of enjoyment and desire for future collaboration among high credit claimers compared with low credit claimers. We did not predict any such relationship between credit claiming and enjoyment with the group or interest in future collaboration among those not explicitly led to consider others’ contributions. Although few people would choose to be more biased in their judgment rather than less, we suggest that reducing egocentric biases in collective endeavors may sometimes have important and unexpected costs.
Study 1: Authors Academic collaboration is the paradigmatic anecdote for egocentric responsibility allocations. The number and ambiguity of diverse tasks spread out over months or even years make accurate attributions virtually impossible. M. Ross and Sicoly (1979) discussed the problem of determining authorship as particularly relevant to their original investigation of egocentric biases, and suggestions for the appropriate way to overcome problems with authorship credit are a popular topic of discussion and debate (e.g., Fine & Kurdek, 1993; Goodyear, Crego, & Johnston, 1992; Zanna
CARUSO, EPLEY, AND BAZERMAN
860
& Darley, 2004). Given its prominence, we attempted to upgrade its anecdotal status by conducting research on authors of major academic journal articles. In addition, we examined our specific hypotheses about the impact of reducing egocentric biases by asking authors to indicate their satisfaction with the order of authorship and desire to collaborate in the future.
Method Participants. We selected manuscripts from five organizational behavior journals for this study: Academy of Management Journal (AMJ), Academy of Management Review (AMR), Administrative Science Quarterly (ASQ), Journal of Applied Psychology (JAP), and Organizational Behavior and Human Decision Processes (OBHDP). Articles with three to six authors,2 published between 1999 and 2001, were included in the investigation. E-mail addresses for all authors were available for 231 of the resulting 293 articles. If an author in this set had more than one article, all but one of those articles were randomly excluded to ensure that no author received the questionnaire more than once. Finally, all articles by two colleagues familiar with our hypotheses were excluded, leaving a sample of 145 articles with 484 unique authors. Procedure. Each participant received an individual e-mail with a link to an online questionnaire containing questions about the experience of writing the article with their author group. Participants were asked to complete the questionnaire in the next few days, and not to discuss their responses with anyone. All participants were assured, both in the e-mail invitation and on the first page of the online survey, that their responses would remain confidential and anonymous. We explained that their responses would be aggregated such that their identities would not be attached to the data for any analyses. Each article was randomly assigned to the self-focused (n ⫽ 108) or other-focused (n ⫽ 89) condition so that all authors of a particular article each received the same condition. Participants in the self-focused condition were simply asked, “Of the total work that your author group did on the article, what percent of the work do you feel you personally contributed?” Participants in the other-focused condition, in contrast, were first asked to write down the initials of their coauthors, and then told, For all authors of the paper, please take a few moments to think about the contributions that they made to the article. Go down the list one at a time and consider the work that each person prepared and the contributions they made based on their particular area of expertise. Following these instructions, respondents in the other-focused version indicated the percentage of work that each author (including themselves) contributed to this article. The order of the list of authors was held constant (“Author 1 contributed. . . ,” “Author 2 contributed. . . ,” etc.) to avoid confounding this manipulation with the order of self-allocations. Dependent measures. After reporting the amount of work contributed, participants were asked how interested they would be “in initiating a brand new line of research (independent of any current ongoing research projects) with this same author group,” how much they enjoyed “working with the group . . . compared to others with which you have worked,” how happy they were “right now with the order in which your name was listed among the authors of this paper,” and how happy they were “with the order in which your name was listed among the authors of this paper when the order was first decided.” Responses to these items were made on scales ranging from 1 (not at all) to 7 (very).
Results As might be expected, response rates were highest for first and second authors. Forty-six percent of first authors and 47% of
second authors completed the questionnaire, compared with 32% of third authors and 10% of fourth authors. Only 1 fifth author, and no sixth authors, responded. However, response rates by author order did not differ between conditions, 2(5, N ⫽ 197) ⫽ 4.42, ns. In addition, the average size of author groups did not differ between the self-focused (M ⫽ 3.42, SD ⫽ 0.73) and otherfocused conditions (M ⫽ 3.48, SD ⫽ 0.76), t(195) ⫽ 0.40, ns, nor did the mean author position or distribution of author numbers differ between these conditions (M ⫽ 2.05, SD ⫽ 0.96; M ⫽ 2.10, SD ⫽ 0.97, respectively), t(195) ⫽ 0.63, ns. Response rates may influence the generalizability of the following results to all compositions of author groups but do not influence the validity of comparisons between the two experimental conditions. In addition, gender had no significant effect on our dependent variables of interest in this or any other study in which it was measured and is therefore not reported further. Responsibility allocations. To assess responsibility allocations across different group sizes, we created an index of adjusted responsibility for the author by multiplying the self-report of work for each participant by the number of authors in his or her group (see also Savitsky et al., 2005). For instance, if a respondent claimed to have contributed 30% of the work in a four-author group, the adjusted responsibility for that author would be 120%. Note that we are making no claim that adjusted responsibilities greater than 100% necessarily represent an overestimation because author position is expected to be objectively related to this adjusted responsibility estimate. Rather, we are simply creating an index that allows for between-condition comparisons across different sizes of author groups.3 Our main interest in this research is not to replicate the well-established existence of egocentric biases in responsibility assessments but to investigate the consequence of reducing an egocentric focus on one’s contributions for group conflict. As predicted, authors who considered their coauthors’ contributions reported contributing less (M ⫽ 123.12%, SD ⫽ 50.55) than those who were not explicitly led to consider others’ contributions (M ⫽ 140.44%, SD ⫽ 65.70), t(195) ⫽ 2.04, p ⬍ .05, d ⫽ 0.294.4 2
We chose not to include groups of 2 authors because our analysis is unlikely to apply to such groups. If egocentric responsibility claims rest in part on considering the other group members collectively rather than individually (Savitsky et al., 2005), then such egocentrism should be largely eliminated when the rest of the group is only one other individual. That is, two-person groups are naturally unpacked. 3 Because we obtained responses from more than 1 person per author group, the analysis at the individual level does not account for statistical dependencies in the data. Because of confidentiality requirements, we could not identify individual respondents with their group in this study, making correction of this problem impossible. We return to this issue in Study 3, where we have independent observations at the group level. 4 Readers might be tempted to compare the overall index of claiming with a logical benchmark of 100%, but such a comparison is inappropriate given the higher response rates from first authors who likely were responsible for more credit than second or subsequent authors. Although we cannot be sure of the exact amount of egocentric responsibility claiming that occurred in this experiment, multiple replications in other research leave us little doubt that authors in the self-focused condition were exhibiting stronger egocentric biases than authors in the other-focused condition.
UNDOING EGOCENTRISM
Table 1 shows the mean amount of claiming by author order for the self-focused and other-focused conditions. To account for author order and group size, we also conducted a regression on raw self-allocated responsibility. This analysis predicted self-allocated responsibility from experimental condition and a set of dummy variables to control for seven possible combinations of group size (3 or 4 authors) and the participant’s author position (first, second, third, or fourth author), with the fourth author in a four-author article as a baseline.5 The results of this regression show a significant negative effect of considering others’ contributions on the percentage of work claimed (B ⫽ – 4.55), t(172) ⫽ –2.37, p ⬍ .02. This implies that explicitly considering others’ contributions, controlling for number of authors and authorship position, reduces self-allocated responsibility by an average of 4.5% on the raw responsibility allocation estimates. Interest in future collaboration. One might intuitively expect that enhancing others’ relative contributions would increase authors’ interest in future collaboration, but it did not. There was no significant difference in the desire to initiate a new line of research between authors in the self-focused (M ⫽ 4.74, SD ⫽ 1.91) and other-focused conditions (M ⫽ 5.09, SD ⫽ 2.03), t(195) ⫽ 1.26, ns. As predicted, however, the effect of considering others’ contributions did depend on the amount of responsibility authors claimed for themselves. In the other-focused condition, the more responsibility authors claimed, the less interested they were in future collaborations (r ⫽ –.40, p ⬍ .01). There was no significant relationship, however, in the self-focused condition (r ⫽ –.09, ns). These two correlations differ significantly from one another (z ⫽ 2.29, p ⬍ .025). To assess whether these correlational differences translated into mean differences among high and low credit claimers, we performed a median split on the adjusted responsibility index. In the other-focused condition, high credit claimers were less interested in future collaboration (M ⫽ 4.67, SD ⫽ 2.19) than were low credit claimers (M ⫽ 5.53, SD ⫽ 1.76), t(87) ⫽ 2.03, p ⬍ .05, d ⫽ 0.436. In the self-focused condition, high credit claimers (M ⫽ 4.64, SD ⫽ 1.90) did not differ significantly from low credit claimers (M ⫽ 4.90, SD ⫽ 1.94), t(106) ⫽ 0.710, ns, d ⫽ 0.141. Notice, however, that a median split on the responsibility index did not control for the participant’s author position, and the correlational analysis did not control for either the participant’s author position or the number of authors on a article. To control for both of these factors, we conducted a second regression examining the
Table 1 Mean Responsibility Claimed and Standard Deviations by Author Order and Condition (Study 1) Self-focused
Other-focused
Author order
n
M
SD
M
SD
1 2 3 4 5 6
66 68 47 15 1 0
59.74 38.17 29.96 21.67 10.00 —
19.87 12.70 12.66 9.83 — —
56.11 32.55 25.32 17.44 — —
16.00 7.51 6.56 5.25 — —
Note. Dashes indicate no data were available.
861
relationship between experimental condition (self-focused vs. other-focused), self-allocated responsibility, and the interaction between condition and self-allocations on the desire to initiate a new project. We again added a set of dummy variables to control for the seven possible combinations of the number of authors on the article and the participant’s author position. The results of this regression indicated that although experimental condition was positively related to the desire to initiate a new project (B ⫽ 1.29), t(170) ⫽ 1.85, p ⬍ .07, this effect was qualified by the interaction between the experimental condition and the level of claiming (B ⫽ ⫺0.03), t(170) ⫽ –1.75, p ⬍ .09. The impact of considering others’ contributions on the desire to initiate a new project was weaker (and even potentially negative) for those who claimed more. To see this more clearly, we ran separate regressions of experimental condition on desire to initiate a new project for the top quartile and the bottom quartile of claimers within each of the seven group size–author order combinations (using the same set of dummy variables as before). These regressions revealed that the impact of considering others’ contributions on the desire to initiate a new project was directionally negative and nonsignificant for the top quartile of credit claimers (B ⫽ ⫺0.72), t(30) ⫽ ⫺0.97, ns, but directionally positive and significant for the bottom quartile of credit claimers (B ⫽ 1.17), t(30) ⫽ 2.38, p ⬍ .025. Perceived enjoyment. We followed a similar analysis plan on the dependent variable of how much people enjoyed working with their author group. A similar pattern of correlations with selfallocations emerged for ratings of how much people enjoyed working with this group, with the other-focused condition more negative than the self-focused group (rs ⫽ – .49 and – .15, respectively; z ⫽ 2.38, p ⬍ .01). In addition, the top half of credit claimers (by a median split on the responsibility index) in the other-focused condition enjoyed their experience less (M ⫽ 5.48, SD ⫽ 1.50) than the bottom half of credit claimers did (M ⫽ 6.23, SD ⫽ 1.04), t(87) ⫽ 2.73, p ⬍ .01, d ⫽ 0.587, but no significant difference on perceived enjoyment emerged in the self-focused condition between high credit claimers (M ⫽ 5.42, SD ⫽ 1.46) and low credit claimers (M ⫽ 5.81, SD ⫽ 1.25), t(106) ⫽ 1.42, ns, d ⫽ 0.281. This pattern of results shows that both the desire to work with the group in the future and the enjoyment of working with the group decrease as work claimed for the self increases, but only among those who consider others’ contributions. Apparently, thinking about the other authors’ contributions decreases one’s overall evaluation of the group among those who feel they contributed more to the project, compared with those who feel they contributed less. Again, we conducted a regression examining the relationship of condition (self- versus other-focused), self-allocated responsibility, and the interaction between these two on how much people enjoyed working with their author group. We again added a set of dummy variables to control for the seven possible combinations of the number of authors on the article and the participant’s author 5 Because of the extremely low response rates from fifth and sixth authors, we dropped the 17 respondents from papers with 5 or 6 authors to simplify the regression analyses and reduce the number of dummy codes needed. Including the partial data from these groups does not alter the analysis in any meaningful way.
CARUSO, EPLEY, AND BAZERMAN
862
position. The results of this regression revealed that condition was positively related to how much people enjoyed working with their author group (B ⫽ 1.04), t(170) ⫽ 2.19, p ⬍ .05, and that this effect was qualified by the interaction with the level of credit claiming (B ⫽ ⫺0.02), t(170) ⫽ –2.22, p ⬍ .05, showing that the effect of self versus other focus was weaker (and even potentially negative) for those who claimed more. To see this more clearly, we again conducted separate regressions of condition (self-focused versus other-focused) on how much people enjoyed working with their author group for the top quartile and the bottom quartile of claimers within each of the seven group size–author order combinations (using the same set of dummy variables as before). These regressions revealed that the effect of condition on perceived enjoyment with the author group was marginally significant in the negative direction among the high credit claimers (B ⫽ ⫺0.99), t(30) ⫽ –1.89, p ⬍ .07, but positive and nonsignificant among the low credit claimers (B ⫽ 0.38), t(30) ⫽ 0.948, ns. Independent of group size and author order, considering others’ contributions did significantly increase authors’ happiness with the order in which their names appeared in the final author list. This was true both at the time the order was decided (other-focused, M ⫽ 6.52, SD ⫽ 1.01; self-focused, M ⫽ 6.13, SD ⫽ 1.48), t(195) ⫽ 2.10, p ⬍ .04, d ⫽ 0.302, and at the time of our experiment (other-focused, M ⫽ 6.47, SD ⫽ 1.10; self-focused, M ⫽ 5.97, SD ⫽ 1.66), t(195) ⫽ 2.44, p ⬍ .02, d ⫽ 0.351. Although this last effect is not central to the core theme of the present article, it is consistent with the logic that having a greater appreciation of the work of others reduces the likelihood of feeling that a higher status of authorship was deserved.
Discussion As predicted, considering others’ contributions significantly reduced the amount of responsibility authors claimed for themselves. Reducing a relatively egocentric focus on one’s own contributions, however, did not simultaneously lead to universally positive evaluations of the authors’ collaborative experiences. Leading participants to consider others’ contributions increased interest in future collaboration and reported enjoyment among those who believed they contributed relatively little to the project, relative to authors who believed they contributed more. Reminding people of how much more work they did than others in the group is exactly the violation of equity that would lead to dissatisfaction with a group project, whereas reminding participants how much they were helped by others’ contributions would likely increase satisfaction. That similar relationships did not arise in the self-focused condition suggests that these participants were primarily focused on their own contributions, consistent with the long line of research demonstrating egocentric biases in such responsibility allocations.
Study 2: Manipulating Perceived Contributions The results of Study 1 are consistent with our claim that reducing an egocentric focus on one’s own contributions reduces happiness with the group among those who report having done more than others, relative to those who report having done less. However, this study relied on participants’ reports of their contributions rather than manipulating perceived contributions directly and is
therefore open to a host of alternative interpretations that such correlational designs engender. Study 2 therefore tested our key hypotheses in a paradigm designed to manipulate people’s perceived contributions to a group. In particular, Study 2 used a well-replicated finding that the difficulty of retrieving information from memory can be used as a guide to its relative frequency, sometimes in spite of the amount of information retrieved. In one experiment, for instance, participants given the difficult task of listing 12 examples of their own assertive behavior rated themselves as less assertive than those given the easy task of listing only 6 examples (Schwarz et al., 1991). We adopted a similar approach in Study 2 by asking participants to think of either 3 or 10 personal contributions to a group project. We expected that those asked to recall only 3 contributions would believe that they contributed more to the group than those asked to recall a full 10 contributions. We therefore predicted that participants given the easy task of listing just 3 examples would show decreased enjoyment and desire for future collaboration after considering others’ contributions, compared with those given the difficult task of generating 10 examples. A related issue with the first study is that the reports of contributions in the self-focused and other-focused conditions were not necessarily equivalent. Our analysis assumes that relatively high credit claimers in the self-focused condition would be similarly high credit claimers in the other-focused condition. However, because self-allocated responsibility was not assessed until after the key experimental manipulations, this assumption remains empirically untested. To avoid this concern, we asked all participants in Study 2 the exact same question about how much they contributed before we introduced the self-focused versus other-focused manipulation. Study 2 was therefore not designed to test the consequence of considering others’ contributions on self-allocated responsibility but rather only to test the key relationship between these self-allocations and measures of group conflict.
Method Participants. Participants (N ⫽ 101) were approached in an undergraduate dining hall at Harvard University and completed the experiment in exchange for a candy bar. Procedure. Participants were handed a multipage questionnaire (adapted from Savitsky et al., 2005). On the first page, participants were asked to consider a project in which a group “worked together toward a common goal, and where the entire group as a whole was recognized for the outcome of the group project.” One participant, for instance, thought about a project in which “two classmates and I had to compose, memorize, and perform a dialogue for our Chinese class, and we were given a grade as a group.” On the next page, all participants were then asked to think about some specific contributions they personally made to the final group project. Approximately half of the participants (n ⫽ 50) were asked to list 10 specific contributions, whereas the other half (n ⫽ 51) were asked to list 3 such contributions. Participants were asked to pick a group project before learning how many contributions they needed to generate to rule out the possibility that participants in the two conditions might systematically report different types of group projects. On the following page, participants were asked to indicate what percentage of the group output they personally contributed. This virtually ensures that participants in both the self-focused and other-focused conditions were responding equivalently in their responsibility claims.
UNDOING EGOCENTRISM On the final page, participants in the other-focused condition were asked to think about the contributions of their other group members before proceeding. Specifically, they were asked to write the first name or initials of each of their fellow group members on blank lines we provided and then to take a moment to think back to specific things that each individual member contributed to their group. All participants then completed the same dependent measures of enjoyment with the group and desire to work with the group in the future that we used in Study 1. Finally, participants were asked to indicate how hard it was to generate (3 or 10) examples of personal contributions to the group on a scale ranging from ⫺5 (very difficult) to 5 (very easy).
Results Responsibility allocations. As intended, participants found it significantly easier to generate 3 contributions (M ⫽ 3.12, SD ⫽ 2.09) than 10 contributions (M ⫽ ⫺0.34, SD ⫽ 3.09), t(99) ⫽ 6.60, p ⬍ .0001, d ⫽ 1.33, with no differences between the self-focused and other-focused conditions (F ⬍ 1, ns). This ease translated into the predicted mean differences in reported contributions. Once again, an index of adjusted responsibility was created by multiplying participants’ responsibility allocations by their reported group size. Participants who were randomly assigned to the low-perceived-contribution condition (M ⫽ 135.32%, SD ⫽ 45.69) felt that they had contributed less to the group task than those in the high-perceived-contribution condition (M ⫽ 164.44%, SD ⫽ 88.10), t(99) ⫽ 2.08, p ⬍ .04, d ⫽ 0.418. Because the focusing manipulation happened after measuring self-allocated responsibility, we did not expect—nor did we find—any statistical difference in self-allocations between the self-focused (M ⫽ 153.39%, SD ⫽ 72.05) and other-focused (M ⫽ 146.58%, SD ⫽ 71.57) conditions (t ⬍ 1, ns) or any interaction on self-allocations between condition and the number of contributions requested (F ⬍ 1, ns). Interest in future collaboration. Recall we predicted that there would be no impact of listing contributions on interest in working with the group in the future or on recalled enjoyment among those in the self-focused condition, but that considering others’ contributions would produce lower evaluations on both measures among those asked to list few contributions compared with those asked to list many contributions. Indeed, a 2 (condition: self-focused versus other-focused) ⫻ 2 (contributions listed: 3 versus 10) analysis of variance on the desire to work with the group in the future yielded only this predicted significant interaction, F(1, 97) ⫽ 4.33, p ⬍ .04, 2p ⫽ .043. A similar significant interaction emerged on reported enjoyment for the group project, F(1, 97) ⫽ 4.25, p ⬍ .05, 2p ⫽ .042. We also examined the simple effects of our contribution condition on the two measures of satisfaction. Among other-focused participants, those in the high-contribution condition expressed less interest in working with the group again than those in the low-contribution condition, t(48) ⫽ 2.60, p ⬍ .015, d ⫽ 0.751, and also reported less enjoyment of the experience, t(47) ⫽ 2.61, p ⬍ .015, d ⫽ 0.761. There were no significant differences between high and low groups among the self-focused participants (ts ⬍ 1, ps ⬎ .70). All relevant means are presented in Table 2.
Discussion Study 2 makes several contributions to this research by helping to rule out alternative explanations for Study 1. First, Study 2
863
Table 2 Mean Ratings and Standard Deviations of Desire to Work With the Group Again in the Future and Enjoyment of the Project Among High- and Low-Perceived Contributors in the SelfFocused Versus Other-Focused Conditions (Study 2) Self-focused Contribution condition
M
Other-focused
SD
M
SD
5.38 4.08
1.95 1.57
5.65 4.58
1.72 1.14
Desire to work with group in future Low High
5.23 5.36
1.68 1.68
Enjoyment of project Low High
5.27 5.44
1.56 1.58
eliminates the possibility of any systematic difference between those who may have naturally reported (and actually contributed) more or less to the group project by manipulating perceived contributions across participants rather than by simply measuring perceived contributions. Compared with those led to feel they contributed relatively little, those led to feel that they contributed relatively more showed a decrease in desire to work with the group again, and decreased enjoyment of the project, when they were led to think about other group members’ contributions just before reporting their satisfaction with the group. Second, the introduction of the perspective-taking manipulation after the reports of contribution rules out the possibility that the results of Study 1 were driven solely by some quirk in the manipulation before the responsibility judgments. That is, it is possible that those who claimed to have contributed much in the selffocused conditions would not have made similar claims in the other-focused conditions of our first study. This manipulation itself may have differentially affected the reports of people at different levels of claiming, rendering our comparisons of high and low claimers across the two conditions untenable. The current study eliminates this problem by capturing claims of fairness in an identical manner in both conditions and shows that considering others’ contributions at the time of responsibility judgments is not necessary to obtain the effects on group satisfaction. Finally, Study 2 expands on our earlier findings by moving beyond correlational data and demonstrating the same pattern of results experimentally. The significant interactions on desire to work with the group again and on enjoyment with the project establish a more direct link between reducing egocentric biases and indices of group conflict. The significant simple effects among only the other-focused participants reinforce the reliability of these results. Although considering others’ contributions may uniformly decrease egocentric biases in responsibility allocations, it does not have a uniform effect on indices of group conflict. Whether this form of perspective taking helps or hinders group satisfaction depends critically on the perceived contributions of the perspective takers themselves.
Study 3: Manipulating Actual Contributions The studies reported so far both suggest that reducing an egocentric focus has different effects among those who believe (or are
864
CARUSO, EPLEY, AND BAZERMAN
led to believe) they have contributed relatively more or relatively less than their other group members. Although Study 2 helps rule out explanations associated with natural differences between high and low claimers by manipulating perceived contributions, no study so far has manipulated actual contributions. Without controlling actual contributions to the group, the previous experiments cannot definitively demonstrate that those who have actually contributed more to a group will be less happy than those who contributed less when taking the perspective of others. We have suggested that this occurs because considering others’ contributions highlights how much they have done compared with oneself, producing a relatively negative assessment when one has contributed more than others but a relatively positive assessment when one has contributed less. For example, imagine that you worked on a project in a group of four people, and that you did a considerable amount— 40% of the total work. Without giving much more thought to it, you may conclude that doing 40% is a reasonable share of the total work. But thinking harder about others’ individual contributions highlights how much less each of the other individuals did than you—say, 15%, 20%, and 25%. An assessment of one’s own work in isolation should therefore lead to more favorable assessment of the group and an increased interest in working with the group in the future, compared with thinking more specifically about others’ contributions. In contrast, if you contributed less than others, say 15%, considering others’ contributions would highlight how much you benefited from others’ efforts and lead to a more favorable assessment, compared with those who contributed little but did not explicitly consider others’ contributions. In fact, this is precisely what participants reported when we conducted an experiment using this very example. In this experiment, undergraduates were approached in public places on the Harvard University campus, and they were asked to complete a brief questionnaire in exchange for a candy bar. The survey asked participants to imagine that they had been part of a group of four students that completed a class project. Half the participants were told that they had contributed 15% of the total work, whereas the other half were told that they had completed 40% of the total work. Furthermore, some participants in each of these groups were given no explicit comparison with the other group members. Others were told how much each of their individual group members contributed in an effort to make their contributions relative to others—and the corresponding departures from equity—more salient. The low contributors (15%) were told that the other members of their group had contributed 20%, 30%, and 35%, whereas the high contributors (40%) were told that the other members had contributed 15%, 20%, and 25%. All participants were then asked to indicate how much they thought they contributed compared with others on a scale ranging from 1 (none of the work) to 6 (all of the work), and how much they would want to work with this group again in the future on a scale ranging from 1 (not at all) to 6 (very much). Results revealed the predicted interaction on desire for future collaboration, such that low contributors who had no explicit comparison available wanted to work with the group relatively less (M ⫽ 3.45, SD ⫽ 1.45) than low contributors who had the explicit comparison available (M ⫽ 4.07, SD ⫽ 1.41), but that high contributors with no comparison (M ⫽ 2.91, SD ⫽ 1.21) wanted to work with the group relatively more than high contributors with an explicit comparison available (M ⫽ 2.55, SD ⫽ 0.91), F(1, 113) ⫽
4.38, p ⬍ .04, 2p ⫽ .037. Furthermore, the correlation between how much people thought they contributed and how much they wanted to work with the group again was more negative among participants who had the explicit comparisons available (r ⫽ –.48, p ⬍ .001) than among participants with no such comparison available (r ⫽ –.13, ns; z ⫽ – 2.11, p ⬍ .04), suggesting that the explicit comparisons exacerbated the tendency for relatively high credit claimers to feel less enthusiastic about future collaboration with the group, compared with relatively low credit claimers. Of course, this experiment involves purely hypothetical contributions and is therefore only suggestive. Study 3 was designed to make a more substantial contribution, as well as overcome a number of shortcomings of the earlier studies, by randomly assigning participants to contribute a relatively high or relatively low amount of work to an actual group project. In addition, many of the contributions considered in the previous experiments have been somewhat ambiguous. Successful research collaborations, for instance, require a mix of knowledge, insight, and effort. It is therefore difficult to calibrate claims of responsibility with some objective standard. Study 3 therefore used an objective and quantifiable measure of contribution to the group. Finally, Study 3 allowed for analyses of complete groups rather than the partial groups in Studies 1 and 2. These complete groups allow us to address concerns about the statistical nonindependence of some of the data in our previous experiments. In particular, participants in Study 3 were asked to write a group essay, with some members of the group asked to write more sentences than other members. We predicted that those who actually contributed much to the group essay would report less enjoyment and interest in working with the group in the future when explicitly asked to consider others’ contributions, relative to participants who contributed less. This paradigm allows us to investigate the impact of reducing an egocentric focus on one’s actual contributions on measures of group conflict.
Method Participants. One hundred thirty-six participants from an existing participant pool of Boston-area residents reported to a computer laboratory. Participants were paid $20 for this 90-min study. Procedure. Participants were randomly assigned to condition by having them select a card from a bowl. Participants were seated in front of individual computers and told that they were participating in a study about virtual workgroups. Because many groups today comprise people in different locations, work on group projects is often completed independently by members of the group. Therefore, participants were told they would be writing an essay jointly with three other group members, but that they would not be interacting or discussing their work as they wrote. Participants first filled out some background questions on the computer. These questions asked about their undergraduate major, the number of essays they had written in college, how much they enjoyed writing essays, and included a series of analogies and anagrams for them to complete. These questions were ostensibly designed to determine the type and amount of work that each individual would be assigned to contribute. In reality, participants were randomly assigned to write either a high (15) or low (5) number of sentences in the final essay. After completing the background questionnaire, participants received eight pages of background material from the World Health Organization Web site (http://www.who.int) about the current state of the HIV/AIDS epidemic to use in their essay and a detailed outline that explained what topics their essay should cover. Participants were given 12 min to read
UNDOING EGOCENTRISM through the material and were then assembled in groups of 4 in separate rooms that each contained a single computer. Participants were assigned to one of four roles (Alpha, Beta, Kappa, or Delta). Alpha and Kappa were each responsible for writing 15 sentences, whereas Beta and Delta wrote 5 sentences each (although the exact number of sentences was never made explicit to the participants). The experimenter informed the participants that they would be writing sentences sequentially and that they were not to talk to one another during the essay writing. The experimenter then called out a role and the number of sentences to be written by that participant. For instance, the experimenter began by saying, “To begin Section 1, Alpha should write the first two sentences.” Participants had approximately 1 min for every sentence they had to write before the experimenter moved on to the next instruction. Roles were called out in a predetermined random order so that participants could not anticipate who would be responsible for writing next. Once the essay was finished, participants reported back to their individual computers to complete the final dependent measures. Individuals in the self-focused groups (n ⫽ 17 groups) simply reported the percentage of the overall essay that they contributed, whereas those in the other-focused groups (n ⫽ 17 groups) thought about each of their other group members before reporting how much each person—including themselves— contributed. These responsibility estimates were followed by the same questions about desire to work with the group on a new project and enjoyment of the group task. In addition, participants were asked to rate the quality of the sentences that they had written compared with those of their other group members on a scale from ⫺3 (much worse) to 3 (much better). A manipulation check asked participants to estimate the percentage of the total sentences in the essay that they had personally written. Finally, we asked participants to rate how happy they were with the division of labor in their group on a scale from 1 (not at all) to 7 (very). This allowed us to test our claim that considering others’ contributions influences reported enjoyment with the group and interest in future collaboration because it highlights relative inequities in responsibility among the group members.
Results We conducted all analyses at the level of the group, treating high versus low contributors as a within-groups variable and focusing condition as a between-groups variable. Responsibility allocations. Participants successfully perceived the actual contribution manipulation. Those assigned to write a low number of sentences reported contributing less (M ⫽ 13.85%, SD ⫽ 6.68) than those assigned to write a high number of sentences (M ⫽ 32.09%, SD ⫽ 10.99), t(33) ⫽ 10.06, p ⬍ .0001, d ⫽ 3.50. More important, considering others’ contributions influenced reported contributions to the essay. Participants in the otherfocused condition claimed to have contributed less to the group essay (M ⫽ 20.98%, SD ⫽ 3.54) than participants in the selffocused condition (M ⫽ 25.62%, SD ⫽ 5.68), t(33) ⫽ 2.86, p ⬍ .008, d ⫽ 1.00. There was no interaction between the focusing manipulation and the high–low contribution manipulation (F ⬍ 1, ns). Although considering others’ contributions reduced selfallocated responsibility, it is interesting to note that summed selfallocations of all 4 group members did not exceed 100%. This may not be especially surprising to readers, however, because this paradigm that allowed complete control over actual contributions, also avoided many of the key attributes that produce egocentric biases in daily life. In particular, all participants’ contributions were easily noticed by all group members (although they might not have been equally credited), contributions were relatively unambiguous, and overall satisfaction with the group product was rel-
865
atively low. Ratings for the quality of the final essay (M ⫽ 3.43, SD ⫽ 1.63) were not above the midpoint on the scale; in fact, they were significantly below the midpoint, t(34) ⫽ 3.80, p ⬍ .001. Research suggests that making contributions harder to notice, more ambiguous, and more desirable would increase participants’ tendency to claim more credit for themselves than is logically possible (e.g., Dunning, Leuenberger, & Sherman, 1995; Miller & Schlenker, 1985; Savitsky et al., 2005). Interest in future collaboration. Of course, the purpose of this study was not to provide yet another demonstration of the oftdocumented egocentric biases in responsibility allocation in group endeavors, but rather to investigate the consequences of reducing an egocentric focus on one’s own contributions. As in the previous studies, participants’ actual contributions influenced their evaluations of their group. There was a marginally significant interaction between condition and amount of contribution on desire to work with the group again in the future, F(1, 32) ⫽ 2.72, p ⫽ .109, 2p ⫽ .078. As seen in Table 3, high contributors in the self-focused groups (M ⫽ 4.21, SD ⫽ 1.72) were directionally more likely to want to work with the group again compared with low contributors (M ⫽ 3.56, SD ⫽ 0.92), whereas the opposite pattern was found in the other-focused groups, in which high contributors (M ⫽ 3.74, SD ⫽ 1.38) desired future collaboration less than low contributors (M ⫽ 4.44, SD ⫽ 1.65). Neither of these simple effects was significant (Fs ⬍ 1.2, ps ⬎ .20). Perceived enjoyment. A much stronger effect emerged, however, for ratings of enjoyment with the group. The interaction between condition and amount of contribution on enjoyment was significant, F(1, 32) ⫽ 6.98, p ⬍ .015, 2p ⫽ .179. Table 3 shows that high contributors in the self-focused groups (M ⫽ 4.15, SD ⫽ 1.54) reported enjoying their group experience directionally more than low claimers (M ⫽ 3.65, SD ⫽ 1.16); as expected, this simple difference was nonsignificant, F ⬍ 1, p ⬎ .35. But once again, the opposite pattern was found in the other-focused groups, in which high contributors (M ⫽ 3.65, SD ⫽ 1.50) enjoyed the project less than low contributors (M ⫽ 5.12, SD ⫽ 1.10), F(1, 32) ⫽ 7.78, p ⬍ .01, 2p ⫽ .196. A composite measure of these two indices of group conflict (r ⫽ .85) revealed the predicted significant interaction, F(1, 32) ⫽ 4.85, p ⬍ .035, 2p ⫽ .132. In addition, simple effects tests on this composite measure revealed the predicted
Table 3 Mean Ratings and Standard Deviations of Desire to Work With the Group Again in the Future and Enjoyment of the Project Among High and Low Contributors in the Self-Focused Versus Other-Focused Conditions (Study 3) Self-focused Contribution condition
M
Other-focused
SD
M
SD
4.44 3.74
1.65 1.38
5.12 3.65
1.10 1.50
Desire to work with group in future Low High
3.56 4.21
0.92 1.72
Enjoyment of project Low High
3.65 4.15
1.16 1.54
866
CARUSO, EPLEY, AND BAZERMAN
results. The simple effect of focus among high versus low contributors was significant among the other-focused groups, F(1, 32) ⫽ 4.24, p ⬍ .05, 2p ⫽ .117, but not among the self-focused groups, F(1, 32) ⫽ 1.12, p ⬎ .25, 2p ⫽ .034. Happiness with division of labor. To explore a possible mechanism for the results in this study, we asked participants how happy they were with the division of labor in their group. Otherfocused participants who had contributed much were significantly less happy with the division of labor (M ⫽ 3.76, SD ⫽ 1.92) compared with other-focused participants who had contributed little (M ⫽ 5.65, SD ⫽ 1.54), F(1, 32) ⫽ 17.26, p ⬍ .001, 2p ⫽ .350, but there was no significant difference in happiness between high (M ⫽ 4.00, SD ⫽ 1.91) and low (M ⫽ 4.35, SD ⫽ 1.97) contributors in the self-focused condition, F(1, 32) ⬍ 1, p ⬎ .40, 2p ⫽ .019. The interaction between contribution and focus was significant, F(1, 32) ⫽ 5.70, p ⬍ .025, 2p ⫽ .151. To test whether this measure of happiness with the division of labor mediated the effect of self- versus other-focus on our dependent measures, we tested for moderated mediation using the procedure outlined by Baron and Kenny (1986). Essentially, we predicted that happiness with the division of the work would mediate the effect of high versus low contributors in the otherfocused, but not self-focused, condition. To test this prediction, we treated contribution condition, the focusing condition, and their interaction as independent variables, happiness with the division of labor as the mediator, and the composite measure of group conflict as the dependent variable. As noted earlier, the interaction of these variables affected happiness with the division of work (t ⫽ ⫺2.39, p ⬍ .025), and happiness with the division of work in turn affected the composite measure of group conflict (t ⫽ 5.27, p ⬍ .001). When the mediator was added to the model, happiness with the division of work remained a significant predictor of group conflict (t ⫽ 4.66, p ⬍ .001), but the interaction term became marginally significant (t ⫽ ⫺1.77, p ⬍ .08). This drop was itself significant (z ⫽ ⫺2.15, p ⬍ .04). Separate mediational analyses confirmed that happiness with the division of labor significantly mediated the effect of contribution amount on group satisfaction for the otherfocused groups (z ⫽ ⫺2.58, p ⬍ 01), but not the self-focused (z ⫽ ⫺.73, p ⬎ .45) groups. Such results provide suggestive evidence that happiness, or the lack thereof, with the division of labor at least partially accounts for the difference in group satisfaction ratings among those who consider their group members’ contributions.
Discussion Study 3 replicates and extends earlier findings by directly manipulating actual, quantifiable contributions among complete groups. Just moments after completing a task, those who were led to reflect on how much more than others they contributed reported enjoying the experience less than those who reflected on how much less than others they contributed, whereas no such difference was found among those who were not specifically focused on their relative contributions. The random assignment of high and low contributors rules out alternative explanations about characteristics specific to those who actually tend to contribute more or less to a group enterprise. The use of complete groups in Study 3 also rules
out the possibility that our earlier results were produced solely by an artifact of selection bias in response rates or other issues associated with the statistical interdependence of responses from incomplete groups. Although we took great pains to assign participants to contribute many or few sentences, and although the manipulation was successfully reflected in participant self-reports, it is conceivable that number of sentences was not as objective a measure of contribution as we may have hoped. For example, those who wrote only a few sentences may have actually written sentences of a much better quality; hence, the amount of total work claimed could reflect some combination of quality and quantity. To test this possibility, we included one last dependent measure in the procedure that asked the participants themselves to rate the quality of their own sentences relative to those of their fellow group members on a scale ranging from 0 (much worse) to 10 (much better). Participants who contributed little did not think their sentences were any better or worse than those who contributed much, and there was no interaction between focus condition and level of contribution on reported quality of one’s own sentences (F ⬍ 1, ns).
Study 4: Moderation by Competition Considering others’ contributions in a group endeavor can have potentially deleterious effects on at least some group members’ well-being. Studies 1–3 demonstrated these results in contexts involving relatively cooperative groups, in which rewards are given to the group as a whole and individuals are not singled out for special recognition of their performance. When rewards are given to all group members in equal measure, violations of equity in contributions may be seen as unacceptable and upsetting. Loafers in these cooperative groups should therefore be particularly unwanted because of the clear inequity between effort and rewards. Not all groups, however, share this kind of cooperative reward structure. Many groups are competitive in nature, in which the greatest rewards are given to the group member who contributed the most. In these competitive groups, claiming more responsibility is synonymous with success or victory, and feeling responsible for the majority of the group’s accomplishments may potentially translate to greater enjoyment and desire to continue working with this group in the future. After all, loafers within a competitive group will actually increase the rewards given to a person who contributes a great deal. The impact of undoing egocentrism may therefore depend on the competitive versus cooperative nature of the group. We examined this potentially important moderator of our earlier results in Study 4 by asking participants to recall a competitive or cooperative group of which they were a part, and to either consider the other group members’ contributions or not. We predicted that considering others’ contributions would reduce enjoyment and interest in further collaboration among those who contributed much compared with those who contributed little in cooperative groups, but not necessarily influence either measure among those who contributed much compared with those who contributed little in competitive groups.
UNDOING EGOCENTRISM
Method Participants. Seventy participants composed mainly of college students in the Boston area were recruited from an existing participant pool. Participants completed the experiment in exchange for $5. Procedure. Participants arrived at a computer laboratory and completed an unrelated study. Following this task, participants were handed a questionnaire packet for the current study similar to that used in Study 2. The first sheet of this packet asked participants to think of a recent time when they participated in a group project, with between two and five other people, that was now finished. Approximately half of the participants were asked to describe a cooperative group and the other half a competitive group. Participants in the cooperative group condition recalled a project in which they “worked together as a group toward a common goal, and in which the entire group as a whole was recognized for the outcome of the project,” such as “a project from a job you have held where the team as a whole was recognized or rewarded for its efforts.” For instance, 1 participant reported a project in which her “string quartet practiced together as a group to prepare for an upcoming concert”; another described a group that “worked together to design, print, and distribute organ donation awareness pamphlets and cards on campus.” Participants in the competitive group condition, in contrast, were asked to recall a group project in which they “worked together as a group toward a common goal, but where the individual members of the group were recognized separately for their contribution to the outcome of the project,” such as “a project from a job you have held where you were working as part of a group but individually competing with your group members for a raise, bonus, or the affection of the boss.” For instance, 1 participant described a competitive project in which “our track relay team was competing to win the state race, but we were competing against each other for individual scholarships”; another thought about “a history project in school where a group of five people gave a presentation on a historical period, and each person had his/her own aspect to report on and get graded on”; and another mentioned being “part of the fencing team, composed of four people competing for three slots – the overall objective was to win as many meets as possible, but my individual objective was to hold on to my slot as #2 on the team.” After writing a short description of the project and indicating how many people were in their group, participants randomly assigned to the selffocused condition (n ⫽ 36) indicated how much work they had contributed to the group project. In contrast, participants randomly assigned to the other-focused condition (n ⫽ 34) were asked to write down the first name or initials of all other group members, to think about each member’s specific contributions to the group, and to place a check mark next to each name once they had done so. On the following page, participants indicated how much each member, including themselves, had contributed. All participants in the other-focused condition received the same instructions, but half these participants listed their own contributions first, whereas half listed their own contributions last. The order in which their own contributions were listed, however, had no significant influence on the amount of work claimed (t ⬍ 1, ns) and is therefore not discussed further. Finally, all participants indicated how interested they would be in working with this group again in the future and how much they enjoyed working in the group, both on scales ranging from 1 (not at all) to 7 (very much). Participants also indicated how cooperative or competitive they felt the group was on an 11-point scale, ranging from –5 (very competitive) to 5 (very cooperative), and how well they knew their fellow group members before participating in the project on an 11-point scale, ranging from –5 (not at all) to 5 (very well).
Results As intended, participants in the cooperative condition (M ⫽ 2.73, SD ⫽ 2.34) rated their groups as more cooperative than those in the competitive condition (M ⫽ 0.94, SD ⫽ 2.92), t(68) ⫽ 2.85,
867
p ⬍ .006, d ⫽ 0.692. There were no significant differences between conditions in enjoyment with the group, t(68) ⫽ 1.45, ns, d ⫽ 0.352, or how well participants knew other group members, t(68) ⫽ 1.02, ns, d ⫽ 0.248. Unexpectedly, we found that the other-focused condition (M ⫽ 4.75, SD ⫽ 1.03) thought of groups with more members than the self-focused condition (M ⫽ 4.26, SD ⫽ 0.96), t(68) ⫽ 2.04, p ⬍ .05, d ⫽ ⫺0.683; group size was thus used as a covariate in the subsequent analyses. Responsibility allocations. To create an index of adjusted responsibility, as in Studies 1 and 2, we multiplied participants’ responsibility allocations by their reported group size. Once again, leading participants to consider their collaborators reduced egocentric responsibility allocations. Participants in the other-focused condition (M ⫽ 125.44%, SD ⫽ 47.42) claimed to be responsible for less of the overall work than participants in the self-focused condition (M ⫽ 153.71%, SD ⫽ 76.73), F(1, 67) ⫽ 5.44, p ⬍ .025, 2p ⫽ .075. There was no difference between the competitive and cooperative groups in the total amount of work claimed (t ⬍ 1, ns). Because participants were randomly assigned to conditions, we expected no overall difference in claiming between the cooperative and competitive groups. Interest in future collaboration. For the cooperative group condition, we expected that participants’ desire to work with their group in the future would replicate those of the cooperative groups used in the previous studies. Consistent with this prediction, the correlation between self-allocated responsibility and desire for future work was more negative (r ⫽ – .73) in the other-focused condition than in the self-focused condition (r ⫽ – .36), albeit this difference was only marginally significant (z ⫽ 1.56, p ⬍ .06). Participants in the competitive group condition, in contrast, showed the opposite pattern. These same correlations were positive in the other-focused condition (r ⫽ .32) and slightly negative in the self-focused condition (r ⫽ –.10). As predicted, the overall 2 (self-focused versus other-focused) ⫻ 2 (competitive versus cooperative group) interaction on these correlations was significant (z ⫽ 3.47, p ⬍ .001). To investigate mean differences for our key claims in the cooperative condition, we performed a median split on the adjusted responsibility index and then tested the difference between high and low claimers on desire for future collaboration. The top half of claimers in the other-focused– cooperative condition were less interested in future collaboration (M ⫽ 3.50, SD ⫽ 2.13) than were the bottom half of credit claimers (M ⫽ 5.62, SD ⫽ 1.38), F(1, 32) ⫽ 7.89, p ⬍ .01, 2p ⫽ .198. In the self-focused– cooperative condition, however, high credit claimers (M ⫽ 4.55, SD ⫽ 1.43) did not differ significantly from low credit claimers (M ⫽ 5.86, SD ⫽ 1.60), F(1, 32) ⫽ 2.54, p ⬎ .12, 2p ⫽ .074. Although showing the opposite pattern, neither of these simple effects was significant among participants in the competitive conditions (Fs ⬍ 1, ps ⬎ .45), suggesting that focusing on others’ contributions is only detrimental to high credit claimers’ desire to collaborate again when they are working in cooperative group contexts. Perceived enjoyment. Ratings of enjoyment showed a somewhat similar but weaker pattern. As before, the correlation between self-allocated responsibility and enjoyment was more negative in the other-focused (r ⫽ –.33) than in the self-focused condition (r ⫽ – .22), but this difference was not significant (z ⫽ 0.48, ns). These correlations did not reverse, however, in the
CARUSO, EPLEY, AND BAZERMAN
868
competitive groups condition between those in the self-focused (r ⫽ –.06) and other-focused (r ⫽ –.07) conditions. The median split on the adjusted index also showed consistent results. Relatively high claimers in the other-focused– cooperative condition were marginally less interested in future collaboration (M ⫽ 4.13, SD ⫽ 2.25) than were relatively low claimers (M ⫽ 5.37, SD ⫽ 1.50), F(1, 32) ⫽ 2.88, p ⬍ .10, 2p ⫽ .082. In the self-focused– cooperative condition, high claimers (M ⫽ 5.56, SD ⫽ 1.17) did not differ significantly from low claimers (M ⫽ 6.29, SD ⫽ 1.22; F ⬍ 1, p ⬎ .35, 2p ⫽ .026). Again, neither of these simple effects was significant among participants in competitive conditions (Fs ⬍ 1, ps ⬎ .50). As in Study 3, we created a composite measure of these two indices of group conflict (r ⫽ .82). As expected, the simple effect of low versus high claiming on this composite was significant in the other-focused– cooperative condition, F(1, 32) ⫽ 5.74, p ⬍ .025, 2p ⫽ .152, but not in the self-focused– cooperative condition, F(1, 32) ⫽ 1.79, p ⬎ .15, 2p ⫽ .053.
Discussion Study 4 again confirms that, regardless of the type of group endeavor, considering others’ contributions reduces selfallocations of responsibility to group projects. However, this experiment also suggests that the potential negative effects of perspective taking within groups may be moderated by the cooperative versus competitive nature of the group. In cooperative groups, higher claimers were less likely than lower claimers to want to collaborate in the future once they had thought about the contributions of their other group members. In competitive groups, however, higher claimers were not less willing to work with the group again after considering the contributions of their fellow group members. These results are consistent with Studies 1–3, all of which investigated groups designed to be cooperating together toward a common goal and common group output. When individuals feel they have done more than what they should have had to do in these groups, they may feel that the group is taking advantage of them. Such feelings could result in their unwillingness to collaborate with the same group of people in the future. When people think about their contributions to competitive groups, on the other hand, those who feel they have contributed more may have been more successful (the relay racer who actually wins the individual track scholarship, for instance). Among these people, the more they think about the others against whom they were competing, the more likely they may want to compete with those people again. Given their past success in the group and the credit they likely received, they should not necessarily be less inclined to desire future interactions with the same group. Study 4 demonstrates that the type of group project can moderate the effect of reducing an egocentric focus on one’s contributions on the desire for future interactions with the group.
General Discussion Group members often appear to have their heads stuck in the proverbial sand when allocating responsibility for collective endeavors. Across a wide variety of domains, people tend to claim more responsibility for group outcomes than others would likely
give them credit for because they tend to egocentrically focus on their own contributions. This egocentric bias was reduced in three experiments by simply asking participants, before allocating responsibility, to think about their collaborators’ contributions, and in one study by manipulating the ease with which one’s own contributions came to mind. However, considering others’ contributions in cooperative groups consistently decreased enjoyment and interest in future collaboration among those who felt they contributed relatively more and among those who actually contributed relatively more than others, suggesting that leaving one’s head in the sand may sometimes be an effective strategy for maintaining group cohesion and happiness. The reverse pattern was found in the competitive groups of Study 4, however, suggesting that the practical implications of egocentric biases in group endeavors is more complicated than one might expect. Although these experiments shed important light on the consequences of undoing egocentric biases in social interaction, and perhaps on the consequences of going beyond one’s own egocentric perspective more generally, these experiments are clearly the start of a research program rather than the end of one. These experiments did not, for instance, clearly identify the key reasons why considering others’ contributions produced our observed results. Study 3 provided at least some support for our predicted mediator of happiness with the division of labor, but variance remained to be explained in that analysis. It is possible that there are other important mediators that we have yet to consider, such as overall liking for other group members, the relative power or status of the perspective taker within the group, or perhaps the nature of the group’s outcome. Our results may be especially likely to occur when those who consider their other group members naturally dislike those members, are in a position of weak power, or have worked toward an unsatisfying or unsuccessful outcome. In addition, the relationship between effort and group rewards may also moderate the impact of undoing egocentrism in group interaction. Some groups are rewarded as individuals based on the effort they contributed, such as the author groups in Study 1 who are listed in order of their presumed effort and overall contributions. Other groups, however, receive equal rewards regardless of the amount of effort they personally contribute, such as in the essay groups in Study 3 and the cooperative groups in Study 4. Although we found that undoing egocentrism can lead those who contributed much to feel less satisfied with their group than those who contributed little in both kinds of groups, we suspect that the impact of undoing egocentrism may be larger in groups that are rewarded as a whole than in groups that are rewarded as individuals. People who contributed much may feel that they have been duly rewarded for their efforts in groups in which individual members are rewarded but may feel that others who contributed less have received undue credit in groups that get rewarded as a whole. These potential moderators of our results highlight the difficulty in finding one single approach that is likely to reap the benefits of reducing egocentric biases that appear to create conflict within groups, without making those who contribute the most the least happy or interested in working with the group in the future. One simple possibility for an alternative debiasing technique, however, might be to have participants consider only their other group members’ contributions, rather than considering both their own and others’ contributions as participants did in all of the experi-
UNDOING EGOCENTRISM
ments reported here. Of course, one’s own contributions are likely to be highly salient regardless of whether one is asked to consider them explicitly or not, but explicitly considering one’s own contributions in light of others’ may particularly highlight the lack of equity that leads to relative dissatisfaction compared with those who contribute little to the group. What these experiments did do, however, is begin a program of research looking carefully at both the positive and negative consequences of undoing a particularly pernicious bias in human judgment, namely egocentrism (Epley, Caruso, & Bazerman, 2006). In general, psychologists interested in human judgment and decision making have focused their empirical attention on identifying errors and biases in human judgment for both their practical and theoretical importance. This research tradition has provided an impressive corpus of knowledge, much of which demonstrates that people’s interpretations of events are largely determined by their own unique perspectives on those events. For example, people tend to view themselves and their futures more positively than is both logically and realistically possible (e.g., Brown, 1986; Epley & Dunning, 2000; Kunda, 1990; Taylor, 1989; Weinstein, 1980). People also tend to overestimate the extent to which others will share their attitudes, emotions, and knowledge (Keysar, 1994; Nickerson, 1999; L. Ross & Ward, 1996; Van Boven, Dunning, & Loewenstein, 2000); the extent to which others are focused on them and their behavior (Fenigstein, 1984; Gilovich, Medvec, & Savitsky, 2000); and the speed with which they will complete important projects (Buehler, Griffin, & MacDonald, 1997; Buehler, Griffin, & Ross, 1994). A moderately common view among interested psychologists is that these egocentric or egoistic biases are largely adaptive (e.g., Gigerenzer, Todd, & the ABC Research Group, 1999; Taylor, 1989). These illusions, the story goes, contribute to psychological well-being and protect an individual’s positive sense of self (Taylor & Brown, 1988). As a result, these positive illusions increase personal commitment, enhance persistence at difficult tasks, and facilitate coping with aversive and uncontrollable events. Positive illusions also allow people in their everyday lives to maintain cognitive consistency, belief in a just and meaningful world, and a sense of personal control and efficacy necessary to take beneficial risks (Greenwald, 1980). Some have even gone as far as to advocate the selection of salespeople on the basis of the magnitude of their positive illusion, or “learned optimism” (Seligman, 1990). The logic is that unrealistically high levels of optimism bolster salesforce persistence. Although each of these findings may be true in some specific situations (e.g., severe health conditions), and although positive illusions may prove beneficial in helping people cope with tragic events, they can also create harm. People regularly invest their life savings in new businesses that have little chance of success. Employees falsely assume that they are irreplaceable and find that their ultimatums to the boss are met with a quick firing. Other researchers caution that positive illusions are likely to have a negative impact on learning and on the quality of decision making, personnel decisions, and responses to organizational crises (“the hole in the ozone layer isn’t that big”), and can contribute to conflict and discontent (Brodt, 1990; Kramer, Newton, & Pommerenke, 1993; Tyler & Hastie, 1991). And more relevant to the specific focus of the current studies, positive illusions lead organizational members to claim an inappropriately large proportion of
869
the credit for positive outcomes, to overestimate their value to the organization, and to set objectives that have little chance of success. Despite this lengthy list of helpful and harmful effects of positive illusions, the empirical science of understanding the conditions under which these illusions help versus hurt has lagged far behind. Our research provides systematic evidence, under controlled experimentation, that the cooperative versus competitive nature of a group may be one such critical condition, and we feel that future experiments may reveal other important factors about the nature of the group, its task, or its outcome that may influence the effects of perspective taking on members’ satisfaction with the group. Perhaps most important, however, the current article highlights the specific and complicated relationship between judgmental biases and psychological or behavioral outcomes. Public discourse about the functionality of mental operations is often quite simplistic, and assumes that judgmental biases are either beneficial or they are not. This research demonstrates that reducing egocentric biases in collaborative groups may be harmful for happiness and future collaborations among some participants (i.e., high credit claimers), but helpful for others (i.e., low credit claimers). Although few people would wish to be more biased than less, reducing egocentric biases in group contexts may not be the panacea for conflict and impasse that much existing research suggests. It is important to bear in mind that reducing egocentric biases not only diminishes people’s focus on themselves but also can increase their focus on others. Practitioners are therefore advised to remember that the benefits of removing egocentric blinders may depend on what people are able to see once they do so. Although reducing egocentric biases certainly has its benefits, these experiments demonstrate that they may have some unexpected—and undesirable— costs as well. Those who encourage group members to move beyond their own egocentric perspective and consider the efforts of their coworkers may wish to pause and consider what group members will see when they actually do so.
References Babcock, L., & Loewenstein, G. (1997). Explaining bargaining impasse: The role of self-serving biases. Journal of Economic Perspectives, 11, 109 –126. Babcock, L., Loewenstein, G., Issacharoff, S., & Camerer, C. (1995). Biased judgments of fairness in bargaining. American Economic Review, 85, 1337–1343. Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173–1182. Bazerman, M. H., & Neale, M. A. (1982). Improving negotiation effectiveness under final offer arbitration: The role of selection and training. Journal of Applied Psychology, 67, 543–548. Brawley, L. R. (1984). Unintentional egocentric biases in attributions. Journal of Sport Psychology, 6, 264 –278. Brodt, S. E. (1990). Cognitive illusions and personnel management decisions. In C. L. Cooper & I. T. Robertson (Eds.), International review of industrial and organizational psychology (Vol. 5, pp. 229 –279). New York: Wiley. Brown, J. D. (1986). Evaluations of self and others: Self-enhancement biases in social judgments. Social Cognition, 4, 353–376. Buehler, R., Griffin, D., & MacDonald, H. (1997). The role of motivated
870
CARUSO, EPLEY, AND BAZERMAN
reasoning in optimistic time predictions. Personality and Social Psychology Bulletin, 23, 238 –247. Buehler, R., Griffin, D., & Ross, M. (1994). Exploring the “planning fallacy”: Why people underestimate their task completion times. Journal of Personality and Social Psychology, 67, 366 –381. Burger, J. M., & Rodman, J. L. (1983). Attributions of responsibility for group tasks: The egocentric bias and the actor– observer difference. Journal of Personality and Social Psychology, 45, 1232–1242. Caine, B. T., & Schlenker, B. R. (1979). Role position and group performance as determinants of egotistical perceptions in cooperative groups. Journal of Psychology, 101, 149 –156. Camerer, C., & Loewenstein, G. (1993). Information, fairness, and efficiency in bargaining. In B. A. Mellers & J. Baron (Eds.), Psychological perspectives on justice (pp. 155–181). Boston: Cambridge University Press. Dawson, E., Gilovich, T., & Regan, D. T. (2002). Motivated reasoning and performance on the Wason selection task. Personality and Social Psychology Bulletin, 28, 1379 –1387. Ditto, P. H., & Lopez, D. F. (1992). Motivated skepticism: Use of differential decision criteria for preferred and nonpreferred conclusions. Journal of Personality and Social Psychology, 63, 568 –584. Dunning, D., Leuenberger, A., & Sherman, D. A. (1995). A new look at motivated inference: Are self-serving theories of success a product of motivational forces? Journal of Personality and Social Psychology, 59, 58 – 68. Epley, N., Caruso, E. M., & Bazerman, M. H. (2006). When perspective taking increases taking: Reactive egoism in social interaction. Journal of Personality and Social Psychology, 91, 872– 889. Epley, N., & Dunning, D. (2000). Feeling “holier than thou”: Are selfserving assessments produced by errors in self or social prediction? Journal of Personality and Social Psychology, 79, 861– 875. Fenigstein, A. (1984). Self-consciousness and the overperception of self as a target. Journal of Personality and Social Psychology, 47, 860 – 870. Fine, M. A., & Kurdek, L. A. (1993). Reflections on determining authorship credit and authorship order on faculty–student collaborations. American Psychologist, 48, 1141–1147. Forsyth, D. R., Berger, R. E., & Mitchell, T. (1981). The effects of self-serving vs. other-serving claims of responsibility on attraction and attribution in groups. Social Psychology Quarterly, 44, 56 – 64. Forsyth, D. R., & Mitchell, T. (1979). Reactions to others’ egocentric claims of responsibility. Journal of Psychology, 103, 281–285. Forsyth, D. R., & Schlenker, B. R. (1977). Attributional egocentrism following performance of a competitive task. Journal of Social Psychology, 102, 215–222. Gigerenzer, G., Todd, P. M., & the ABC Research Group. (1999). Simple heuristics that make us smart. New York: Oxford University Press. Gilovich, T., Kruger, J., & Savitsky, K. (1999). Everyday egocentrism and everyday interpersonal problems. In R. Kowalski & M. Leary (Eds.), The social psychology of emotional and behavioral problems: Interfaces of social and clinical psychology (pp. 69 –95). Washington, DC: American Psychological Association. Gilovich, T., Medvec, V. H., & Savitsky, K. (2000). The spotlight effect in social judgment: An egocentric bias in estimates of the salience of one’s own actions and appearance. Journal of Personality and Social Psychology, 78, 211–222. Goodyear, R. K., Crego, C. A., & Johnston, M. W. (1992). Ethical issues in the supervision of student research: A study of critical incidents. Professional Psychology: Research and Practice, 23, 203–210. Greenwald, A. G. (1980). The totalitarian ego: Fabrication and revision of personal history. American Psychologist, 35, 603– 618. Harris, S. (1946). Banting’s miracle: The story of the discovery of insulin. Philadelphia: Lippincott. Keysar, B. (1994). The illusory transparency of intention: Linguistic perspective taking in text. Cognitive Psychology, 23, 165–208.
Kramer, R. M., Newton, E., & Pommerenke, P. L. (1993). Selfenhancement biases and negotiator judgment: Effects of self-esteem and mood. Organizational Behavior and Human Decision Processes, 56, 110 –133. Kruger, J., & Gilovich, T. (1999). “Naı¨ve cynicism” in everyday theories of responsibility assessment: On biased assumptions of bias. Journal of Personality and Social Psychology, 76, 743–753. Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108, 480 – 498. Leary, M. R., & Forsyth, D. R. (1987). Attributions of responsibility for collective endeavors. In C. Hendrick (Ed.), Review of personality and social psychology (Vol. 8, pp. 167–188). Newbury Park, CA: Sage. Loewenstein, G., Issacharoff, S., Camerer, C., & Babcock, L. (1993). Self-serving assessments of fairness and pretrial bargaining. Journal of Legal Studies, 22, 135–159. Miller, R. S., & Schlenker, B. R. (1985). Egotism in group members: Public and private attributions of responsibility for group performance. Social Psychology Quarterly, 48, 85– 89. Neale, M. A., & Bazerman, M. H. (1983). The role of perspective-taking ability in negotiating under different forms of arbitration. Industrial and Labor Relations Review, 36, 378 –388. Nickerson, R. S. (1999). How we know—and sometimes misjudge—what others know: Imputing one’s own knowledge to others. Psychological Bulletin, 125, 737–759. Ross, L., & Ward, A. (1996). Naive realism: Implications for social conflict and misunderstanding. In T. Brown, E. Reed, & E. Turiel (Eds.), Values and knowledge (pp. 103–135). Hillsdale, NJ: Erlbaum. Ross, M., & Sicoly, F. (1979). Egocentric biases in availability and attribution. Journal of Personality and Social Psychology, 37, 322– 336. Rottenstreich, Y., & Tversky, A. (1997). Unpacking, repacking, and anchoring: Advances in support theory. Psychological Review, 104, 406 – 415. Savitsky, K., Van Boven, L., Epley, N., & Wight, W. (2005). The unpacking effect in responsibility allocations for group tasks. Journal of Experimental Social Psychology, 41, 447– 457. Schlenker, B. R., & Miller, R. S. (1977). Egocentrism in groups: Selfserving biases or logical information processing? Journal of Personality and Social Psychology, 35, 755–764. Schwarz, N., Bless, H., Strack, F., Klumpp, G., Rittenauer-Schatka, H., & Simons, A. (1991). Ease of retrieval as information: Another look at the availability heuristic. Journal of Personality and Social Psychology, 61, 195–202. Seligman, M. E. P. (1990). Learned optimism. New York: Knopf. Taylor, S. E. (1989). Positive illusions: Creative self-deception and the healthy mind. New York: Basic Books. Taylor, S. E., & Brown, J. D. (1988). Illusion and well-being: A social psychological perspective on mental health. Psychological Bulletin, 103, 193–210. Thompson, L. L., & Loewenstein, G. F. (1992). Egocentric interpretations of fairness and interpersonal conflict. Organizational Behavior and Human Decision Processes, 51, 176 –197. Thompson, S. C., & Kelly, H. H. (1981). Judgments of responsibility for activities in close relationships. Journal of Personality and Social Psychology, 41, 469 – 477. Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5, 207–232. Tversky, A., & Koehler, D. J. (1994). Support theory: A nonextensional representation of subjective probability. Psychological Review, 101, 547–567. Tyler, T., & Hastie, R. (1991). The social consequences of cognitive illusions. In M. H. Bazerman, R. J. Lewicki, & B. Sheppard (Eds.),
UNDOING EGOCENTRISM Handbook of negotiation research: Research on negotiation in organizations (Vol. 3, pp. 69 –98). Greenwich, CT: JAI Press. Van Boven, L., Dunning, D., & Loewenstein, G. (2000). Egocentric empathy gaps between owners and buyers: Misperceptions of the endowment effect. Journal of Personality and Social Psychology, 79, 66 –76. Wade-Benzoni, K. A., Tenbrunsel, A. E., & Bazerman, M. H. (1996). Egocentric interpretations of fairness in asymmetric, environmental social dilemmas: Explaining harvesting behavior and the role of communication. Organizational Behavior and Human Decision Processes, 67, 111–126. Walster, E., Walster, G. W., & Berscheid, E. (1978). Equity: Theory and research. Boston: Allyn & Bacon.
871
Weinstein, N. D. (1980). Unrealistic optimism about future life events. Journal of Personality and Social Psychology, 39, 806 – 820. Zander, A. (1971). Motives and goals in groups. New York: Academic Press. Zanna, M. P., & Darley, J. M. (2004). Mentoring: Managing the faculty– graduate student relationship. In J. M. Darley, M. P. Zanna, & H. L. Roediger, III (Eds.), The compleat academic: A career guide (2nd ed., pp. 117–131). Washington, DC: American Psychological Association.
Received May 20, 2005 Revision received January 22, 2006 Accepted February 26, 2006 䡲
New Editors Appointed, 2008 –2013 The Publications and Communications Board of the American Psychological Association announces the appointment of six new editors for 6-year terms beginning in 2008. As of January 1, 2007, manuscripts should be directed as follows: • Behavioral Neuroscience (www.apa.org/journals/bne), Ann E. Kelley, PhD, Department of Psychiatry, University of Wisconsin–Madison Medical School, 6001 Research Park Boulevard, Madison, WI 53719. • Journal of Experimental Psychology: Applied (www.apa.org/journals/xap), Wendy A. Rogers, PhD, School of Psychology, Georgia Institute of Technology, 654 Cherry Street, Atlanta, GA 30332-0170. • Journal of Experimental Psychology: General (www.apa.org/journals/xge), Fernanda Ferreira, PhD, The School of Philosophy Psychology and Language Sciences, The University of Edinburgh, 7 George Square, Edinburgh EH8 9JZ, United Kingdom. • Neuropsychology (www.apa.org/journals/neu), Stephen M. Rao, PhD, Division of Neuropsychology, Medical School of Wisconsin, 8701 West Watertown Plank Road, Medical Education Building, Room M4530, Milwaukee, WI 53226. • Psychological Methods (www.apa.org/journals/met), Scott E. Maxwell, PhD, Department of Psychology, University of Notre Dame, Notre Dame, IN 46556. • Psychology and Aging (www.apa.org/journals/pag), Fredda Blanchard-Fields, PhD, School of Psychology, Georgia Institute of Technology, 654 Cherry Street, Atlanta, GA 30332-0170. Electronic manuscript submission. As of January 1, 2007, manuscripts should be submitted electronically via the journal’s Manuscript Submission Portal (see the Web site listed above with each journal title). Manuscript submission patterns make the precise date of completion of the 2007 volumes uncertain. Current editors, John F. Disterhoft, PhD, Phillip L. Ackerman, PhD, D. Stephen Lindsay, PhD, James T. Becker, PhD, Stephen G. West, PhD, and Rose T. Zacks, PhD, respectively, will receive and consider manuscripts through December 31, 2006. Should 2007 volumes be completed before that date, manuscripts will be redirected to the new editors for consideration in 2008 volumes.
UNDOING EGOCENTRISM Handbook of negotiation research: Research on negotiation in organizations (Vol. 3, pp. 69 –98). Greenwich, CT: JAI Press. Van Boven, L., Dunning, D., & Loewenstein, G. (2000). Egocentric empathy gaps between owners and buyers: Misperceptions of the endowment effect. Journal of Personality and Social Psychology, 79, 66 –76. Wade-Benzoni, K. A., Tenbrunsel, A. E., & Bazerman, M. H. (1996). Egocentric interpretations of fairness in asymmetric, environmental social dilemmas: Explaining harvesting behavior and the role of communication. Organizational Behavior and Human Decision Processes, 67, 111–126. Walster, E., Walster, G. W., & Berscheid, E. (1978). Equity: Theory and research. Boston: Allyn & Bacon.
871
Weinstein, N. D. (1980). Unrealistic optimism about future life events. Journal of Personality and Social Psychology, 39, 806 – 820. Zander, A. (1971). Motives and goals in groups. New York: Academic Press. Zanna, M. P., & Darley, J. M. (2004). Mentoring: Managing the faculty– graduate student relationship. In J. M. Darley, M. P. Zanna, & H. L. Roediger, III (Eds.), The compleat academic: A career guide (2nd ed., pp. 117–131). Washington, DC: American Psychological Association.
Received May 20, 2005 Revision received January 22, 2006 Accepted February 26, 2006 䡲
New Editors Appointed, 2008 –2013 The Publications and Communications Board of the American Psychological Association announces the appointment of six new editors for 6-year terms beginning in 2008. As of January 1, 2007, manuscripts should be directed as follows: • Behavioral Neuroscience (www.apa.org/journals/bne), Ann E. Kelley, PhD, Department of Psychiatry, University of Wisconsin–Madison Medical School, 6001 Research Park Boulevard, Madison, WI 53719. • Journal of Experimental Psychology: Applied (www.apa.org/journals/xap), Wendy A. Rogers, PhD, School of Psychology, Georgia Institute of Technology, 654 Cherry Street, Atlanta, GA 30332-0170. • Journal of Experimental Psychology: General (www.apa.org/journals/xge), Fernanda Ferreira, PhD, The School of Philosophy Psychology and Language Sciences, The University of Edinburgh, 7 George Square, Edinburgh EH8 9JZ, United Kingdom. • Neuropsychology (www.apa.org/journals/neu), Stephen M. Rao, PhD, Division of Neuropsychology, Medical School of Wisconsin, 8701 West Watertown Plank Road, Medical Education Building, Room M4530, Milwaukee, WI 53226. • Psychological Methods (www.apa.org/journals/met), Scott E. Maxwell, PhD, Department of Psychology, University of Notre Dame, Notre Dame, IN 46556. • Psychology and Aging (www.apa.org/journals/pag), Fredda Blanchard-Fields, PhD, School of Psychology, Georgia Institute of Technology, 654 Cherry Street, Atlanta, GA 30332-0170. Electronic manuscript submission. As of January 1, 2007, manuscripts should be submitted electronically via the journal’s Manuscript Submission Portal (see the Web site listed above with each journal title). Manuscript submission patterns make the precise date of completion of the 2007 volumes uncertain. Current editors, John F. Disterhoft, PhD, Phillip L. Ackerman, PhD, D. Stephen Lindsay, PhD, James T. Becker, PhD, Stephen G. West, PhD, and Rose T. Zacks, PhD, respectively, will receive and consider manuscripts through December 31, 2006. Should 2007 volumes be completed before that date, manuscripts will be redirected to the new editors for consideration in 2008 volumes.
Journal of Personality and Social Psychology 2006, Vol. 91, No. 5, 872– 889
Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.5.872
When Perspective Taking Increases Taking: Reactive Egoism in Social Interaction Nicholas Epley
Eugene M. Caruso
University of Chicago
Harvard University
Max H. Bazerman Harvard Business School Group members often reason egocentrically, believing that they deserve more than their fair share of group resources. Leading people to consider other members’ thoughts and perspectives can reduce these egocentric (self-centered) judgments such that people claim that it is fair for them to take less; however, the consideration of others’ thoughts and perspectives actually increases egoistic (selfish) behavior such that people actually take more of available resources. A series of experiments demonstrates this pattern in competitive contexts in which considering others’ perspectives activates egoistic theories of their likely behavior, leading people to counter by behaving more egoistically themselves. This reactive egoism is attenuated in cooperative contexts. Discussion focuses on the implications of reactive egoism in social interaction and on strategies for alleviating its potentially deleterious effects. Keywords: reactive egoism, egocentrism, perspective taking, negotiation, conflict
interpersonal conflict (Babcock & Loewenstein, 1997; Bazerman & Neale, 1982; Messick, 1995). When such problems with psychological vision arise, a seemingly simple and effective strategy for determining the optimal solution would be to actively consider an opposing side’s perspective. Labor union representatives, for instance, would seem well advised to consider management’s likely concerns before going to the negotiating table. So too would environmentalists seem wise to consider constraints faced by industry representatives in order to propose changes that are likely to be implemented. And indeed, reducing egocentric biases by considering another person’s perspective has been found to have a variety of beneficial effects in social interaction. Considering others’ perspectives, for instance, increases the likelihood of helping another person in need (Batson, 1994), reduces the use of stereotypes when forming impressions (Galinsky & Moskowitz, 2000), increases negotiation effectiveness (Neale & Bazerman, 1983), and diminishes a variety of problematic egocentric biases in judgment (Savitsky, Van Boven, Epley, & Wight, 2005; Wade-Benzoni, Tenbrunsel, & Bazerman, 1996). It is therefore of little surprise that actively considering the other side’s perspective is considered to be a critical component of successful conflict resolution (Paese & Yonker, 2001). In this article, however, we suggest that the consequences of considering others’ perspectives in social interactions may be more complicated than these positive results might suggest. Although reducing an egocentric focus on one’s own concerns and interests by considering others’ thoughts and perspectives may make an optimal solution more readily accessible, we suggest it can ironically lead people to behave in an even less optimal fashion. The reason involves the thoughts, attitudes, and likely behaviors that are activated when people shift their focus from their own concerns and interests to consider the concerns and interests of others. In particular, considering others’ concerns and interests may high-
People in the midst of disagreements often fail to see “eye to eye.” Plaintiffs consistently request larger damage awards than defendants are willing to give. Environmentalists consistently demand more extensive changes to industrial practices than industry representatives believe are reasonable. And labor unions predictably argue that management undervalues their efforts, whereas management predictably argues that increasing salaries would reduce profitability and hasten bankruptcy. In just one of many recent examples, players from the National Hockey League were “locked out” of play by team owners in the summer of 2004, and the subsequent season was eventually canceled, at least in part, because players rejected a proposed salary cap that they considered to be patently unfair but that owners claimed was necessary to run a sustainable business. Divergent interests can lead to divergent perspectives, and failing to understand an opposing side’s perspective can lead to egocentric assessments of fairness and can create
Nicholas Epley, Graduate School of Business, University of Chicago; Eugene M. Caruso, Department of Psychology, Harvard University; Max H. Bazerman, Harvard Business School, Harvard University. This research was supported by Grant SES0241544 from the National Science Foundation and by the James S. Kemper Foundation Faculty Research Fund awarded to Nicholas Epley and by a Graduate Research Fellowship from the National Science Foundation awarded to Eugene M. Caruso. We thank Tara Abbatello, Kristin Boyd, Daniel Carroll, Dolly Chugh, Leif Holtzman, Nick Josefowitz, Angela Kim, Adriana Luciano, Celene Menschel, Sarah Murphy, Kristian Myrseth, Heather Omoregie, Dobromir Rahnev, Aram Seo, Ashley Siler, Bree Tse, Bev Whelan, Erin Rapien Whitchurch, and Joe Whitchurch for assistance conducting these experiments. Correspondence concerning this article should be addressed to Nicholas Epley, University of Chicago, 5807 South Woodlawn Avenue, Chicago, IL 60637. E-mail:
[email protected] 872
REACTIVE EGOISM
light self-interested motives in others’ perceptions or likely behavior. A person selling a house, for instance, who considers a buyer’s perspective might now be more focused on a buyer’s interest in obtaining a low selling price. Although this seller may now recognize that his or her house may be worth less than originally thought, he or she may also now feel compelled to demand an even higher selling price to balance out a presumably low offer from the other side in order to obtain a fair offer. Or an oil executive who considers environmentalists’ concerns may recognize the benefits of making environmentally friendly changes but may actually be even less willing to offer these changes in order to combat the presumably extreme demands that the environmental lobby is likely to make. Or team owners who try to predict the salary demands of their players may concede that the athletes deserve a pay increase but the owners may actually suggest a pay decrease to counteract the exaggerated demands they come to expect from the players’ union. We suggest in this article that reducing an egocentric focus on one’s own concerns and interests by considering others’ perspectives in social interaction may reduce egocentric biases in judgment, but it may also lead to reactive egoism in behavior (i.e., egoistic or self-serving behavior in reaction to the presumably egoistic behavior of others). Notice that our predictions about the consequences of perspective taking focus solely, at this point, on what psychologists commonly refer to as cognitive perspective taking. Cognitive perspective taking involves intuiting, as accurately as possible, another person’s thoughts, feelings, attitudes, interests, or concerns in a particular situation (Chartrand & Bargh, 1999; Davis, 1983; Epley, Savitsky, & Gilovich, 2002; Mead, 1934; Savitsky, Epley, & Gilovich, 2001; Stotland, 1969). Although very common in daily life, this form of perspective taking is also very specific and should not be confused with other meanings of the term such as empathizing with another person’s distress (Batson, 1994), simulating one’s own reactions in another person’s situation (imagineself perspective; Stotland, 1969; see also Galinsky & Moskowitz, 2000), or visually “seeing” a scene from another person’s perspective (Piaget, 1932/1965). Whether the predictions we make about the consequences of cognitive perspective taking generalize to these other forms of perspective taking is an issue we consider further in the General Discussion. Notice also that our predictions about the consequences of perspective taking involve a critical interaction between judgment and behavior. In particular, we predict that considering others’ perspectives should decrease egocentric or self-centered biases in judgments of what is objectively fair but should increase egoistic or self-interested behavior compared with those who remain more egocentrically focused and do not consider others’ perspectives. Homeowners who consider a buyer’s perspective may believe their house is not worth as much as they would have thought but will ask for more nonetheless. An oil executive may see the wisdom in an environmentally friendly policy but may offer fewer environmentally friendly changes. And a team owner may privately value a star player’s skills more after taking the player’s perspective but may offer a smaller salary at the negotiating table. Understanding this interaction between judgment and behavior therefore requires two separate explanations, one detailing why considering others’ perspectives should reduce egocentric biases in judgment and another detailing why considering others’ perspectives should in-
873
crease egoism in behavior. We offer each of these explanations in turn.
Egocentrism in Judgment Some people occasionally report out-of-body experiences, but life for most people is very much an in-body enterprise. People perceive the world directly through their own sensory organs and interpret those perceptions by using schemas and expectations that are firmly planted in their own brains. This means that one’s own unique perspective on the world is immediately and easily available, whereas others’ perspectives must be deliberately inferred. Experience is faster and more reliable than inference, meaning that a person’s egocentric perspective generally serves as a default in judgment that must be deliberately corrected or adjusted when necessary. Because such correction processes are notoriously insufficient (Epley & Gilovich, 2004; Gilbert, 2002; Tversky & Kahneman, 1974), many social judgments are egocentrically biased. For instance, people tend to overestimate the extent to which others share their knowledge, preferences, and attitudes (Keysar & Bly, 1995; Keysar, Ginzel, & Bazerman, 1995; Nickerson, 1999; L. Ross, Greene, & House, 1977), overestimate the extent to which others attend to their behavior and appearance (Gilovich, Medvec, & Savitsky, 2000; Gilovich, Savitsky, & Medvec, 1998), and overestimate the extent to which others’ impressions of them will match their self-assessments (Epley et al., 2002; Kenny & DePaulo, 1993; Savitsky et al., 2001). More relevant for the current research, people’s assessments of fairness in social interactions tend to be egocentrically biased as well (Babcock & Loewenstein, 1997; Paese & Yonker, 2001). People tend to believe that they deserve more credit for collaborative endeavors than is logically (and statistically) possible (Leary & Forsyth, 1987; M. Ross & Sicoly, 1979) and that they also deserve a larger share of available resources than others believe is fair (Babcock & Loewenstein, 1997). In one study, for instance, participants believed that they should be paid nearly $5 more, on average, than their partner for identical work (Messick & Sentis, 1983). In another, plaintiffs in a mock court case believed it was fair for them to receive approximately twice as much in damages, on average, than defendants believed was fair (Loewenstein, Issacharoff, Camerer, & Babcock, 1993). What is more, these participants believed that their egocentric biases would be shared by the trial judge as well (Babcock, Loewenstein, Issacharoff, & Camerer, 1995; Loewenstein et al., 1993). These egocentric perceptions of fairness can then create conflict between individuals and groups. In the mock court case just described, the magnitude of egocentric biases between plaintiffs and defendants significantly predicted an impasse that required third-party adjudication. Similar negative influences of egocentric biases have been reported elsewhere (Thompson & Loewenstein, 1992; WadeBenzoni et al., 1996). Such egocentric biases in fairness and resource allocation appear to arise from both motivational and cognitive sources. Motivationally speaking, people are likely to seek evidence consistent with preferred or self-interested outcomes and to evaluate evidence inconsistent with these preferred outcomes more critically than evidence consistent with preferred outcomes (Dawson, Gilovich, & Regan, 2002; Ditto & Lopez, 1992; Ditto, Scepansky,
EPLEY, CARUSO, AND BAZERMAN
874
Munro, Apanovitch, & Lockhart, 1998; Lord, Ross, & Lepper, 1979). More relevant for the current research, however, are cognitive mechanisms that render evidence supporting self-serving assessments of fairness more readily accessible than evidence supporting self-defeating assessments. People are more likely to notice and to attend to their own contributions to group endeavors than to others’ contributions, making a self-serving allocation of rewards seem justifiable (M. Ross & Sicoly, 1979; Thompson & Kelly, 1981). So too are people likely to be more focused on their own interests and concerns than on others’ interests and concerns, leading them to notice and to attend to information that supports their interests more than information that supports others’ interests (Lord, Ross, & Lepper, 1979), to weight such supportive evidence more heavily in judgment (Babcock et al., 1995; Messick & Sentis, 1979), and to recall more easily such supportive evidence in memory (Thompson & Loewenstein, 1992). One reason people believe it is fairer to satisfy their own interests than others’ interests is because they do not notice, attend to, or care about others’ interests as much as their own. These results make it clear why increasing the focus on others’ contributions and interests may highlight information that an individual might have naturally overlooked and might reduce self-serving allocations of responsibility and judgments of fairness (Savitsky et al., 2005).
Reactive Egoism in Behavior If failing to consider others’ perspectives in social interaction creates egocentric biases and conflict, and if considering others’ perspectives can reduce egocentric biases that apparently create such conflict, then it is more of a logical syllogism than an empirical hypothesis that considering others’ perspectives should reduce interpersonal conflict. Although this analysis is perfectly logical, it is not perfectly psychological for three reasons. First, people’s judgments need not match their behavior. People eat Twinkies when they know they should not, spend money when they know they should be saving, and go on vacations when they know they should be working. In one of many empirical examples of this divergence between attitudes and behavior, undergraduates in one experiment universally expressed anger at the lack of adequate student housing on campus, but only those personally affected actually took action to fix the problem (Regan & Fazio, 1977). Similarly, knowing what is fair in a social interaction or negotiation and behaving fairly are two very different things (Batson, Kobrynowicz, Dinnerstein, Kampf, & Wilson, 1997; Batson, Thompson, & Chen, 2002). Altering people’s egocentric assessments of fairness does not therefore necessitate an analogous change in behavior. Second, much of social interaction is strategic, aimed at achieving some underlying goal. This is especially true in the competitive contexts that lead to divergences in interests and create conflict like those mentioned in the opening paragraph. Altering people’s otherwise egocentric focus on their own concerns and interests may make a purely egocentric allocation of resources, for instance, seem unfair, but it is unlikely to alter the underlying interests and concerns on which these egocentric perceptions were based. As a result, behavior may be less likely to change in competitive or strategic interactions than might perceptions of fairness, making a
disconnect between judgment and behavior especially likely in these contexts. Finally, and perhaps most importantly, reducing egocentric biases in social interaction by shifting the focus to others’ perspectives means that people’s behavior is now likely to be determined in large part by what they see when they look into the mind of those people. Several empirical findings suggest that such perceptions are unlikely to be especially flattering or conducive to cooperative behavior. Chief among them is that people tend to believe they are fairer and more moral than are others (Allison, Messick, & Goethals, 1989; Epley & Dunning, 2000; Messick, Bloom, Boldizar, & Samuelson, 1985). Considering what another person is likely to believe is a fair resolution to a dispute will almost certainly highlight a resolution that appears anything but fair. But not only do people believe they are fairer than others in a relative sense, research has suggested that people may believe that others are patently unfair and self-interested in an absolute sense. People tend to assume that others’ behavior is guided by their self-interest with relatively little concern about fairness or justice (Kramer, 1994), a belief that would not be especially problematic except that it often exaggerates the impact of self-interest on others’ attitudes and thoughts (Miller, 1999; Miller & Ratner, 1998). In one study, for instance, spouses were asked to allocate responsibility for a series of positive as well as negative marital tasks (e.g., resolving conflicts and causing arguments, respectively), and to anticipate how their spouse would allocate responsibility for these same tasks (Kruger & Gilovich, 1999). Spouses actually allocated responsibility in a self-centered way, claiming more responsibility for both positive and negative tasks (see also M. Ross & Sicoly, 1979). Spouses believed their partners, however, would allocate responsibility in a self-serving way by claiming more responsibility for positive tasks but less responsibility for negative tasks. People’s judgments of the external world may or may not be distorted by their self-interest, but it is clear that they are not as distorted by self-interest as others believe them to be. Notice how this naive cynicism (Kruger & Gilovich, 1999) can work against benefits otherwise gleaned from reducing egocentric biases in social interaction. Looking into the minds of others may highlight cynical and self-interested motivations that would have been otherwise overlooked if people were egocentrically focused on their own interests and concerns. To counteract these presumably selfish motivations in others, people may react by behaving even more selfishly in return. This hypothesis is consistent with the reciprocity norm observed in bargaining negotiations (Esser & Komorita, 1975), in which the behavior of one person often leads to similar (if not identical) behavior in another. In addition, predictions about another’s behavior can influence how one decides to act. For instance, in one prisoner’s dilemma game, those who defected anticipated approximately four times as much defection from others compared with those who cooperated (Dawes, McTavish, & Shaklee, 1977). What’s more, people tend to fear that they will be exploited by the selfish behavior of others if they behave charitably and are therefore motivated to behave in line with their own self-interest (Kelley & Stahelski, 1970; Messe´ & Sivacek, 1979). When people expect others to behave selfishly, such reactive egoism could heighten rather than diminish conflict and disagreement in social interaction, producing exactly the op-
REACTIVE EGOISM
posite results that adopting others’ perspectives was designed to produce.
Current Research We therefore predicted that considering others’ perspectives in social interactions—namely competitive social interactions— would decrease egocentric biases in perceptions of fairness and resource allocations but would ironically increase egoistic behavior in those interactions compared with people who did not consider others’ perspectives. We tested these hypotheses in four separate experiments involving both simulated (Experiments 1, 2, 3, and 5) as well as actual (Experiment 4) resource allocation negotiations. We predicted that reactive egoism would be caused by the increasingly cynical thoughts that were activated upon considering others’ perspectives, and we tested this hypothesis directly in Experiments 2 and 4. Finally, we investigated the extent to which reactive egoism in social interaction would be moderated by the competitive versus cooperative nature of the interaction. Cynical thoughts about others’ behaviors should be especially likely to come to mind in competitive environments when others’ interests are opposed to one’s own but should be less likely in cooperative environments where interests align. We tested this situational moderator in Experiments 4 and 5.
Experiment 1: Fish Commons dilemmas provide textbook contexts for observing egocentric biases and conflict in social interaction. Such dilemmas involve multiple parties making decisions that affect all of their final outcomes. Commons dilemmas are defined by a context in which each party has a choice to cooperate or defect (or to select a behavior on a continuum between the two) in which all parties would benefit through mutual cooperation but individuals would profit through defection. In real life, people often fail to solve these dilemmas optimally. Too few behave in a manner that is best for the group as a whole, and the fixed pool of resources is quickly depleted. Such a case happened with commercial fishermen in the North Atlantic in the 1980s who overharvested the once abundant cod until it was on the brink of collapse and the fishery was closed by government regulators (and it is still closed to this day). We modeled Experiment 1 after this very incident by adapting the Shark Harvesting and Resource Conservation (SHARC) case (Wade-Benzoni et al., 1996), a negotiation exercise based on the North Atlantic fishery collapse. In this simulated negotiation exercise, participants were assigned to represent one of four fishing industries and eventually to indicate the percentage of the overall fish stock that was fair for them to take as well as the percentage of the stock that they would actually take. Some groups completed these measures after considering other group members’ perceptions of fairness, whereas others did not consider others’ perceptions. We predicted that those who considered other group members’ perceptions of fairness would report that it was fair for them to take less of the overall fishing stock but would actually take more of the stock compared with participants who did not consider the other members’ perceptions.
875 Method
Participants Harvard University undergraduates (N ⫽ 160) participated in this experiment as a class exercise in an introductory social psychology course. The course featured 2 lectures and 1 small discussion section per week. There were a total of 13 discussion sections in this course, each consisting of 14 to 18 students.
Procedure Students within each section were divided into groups of 4, and participants within each group were randomly assigned to one of the four following association representative roles: “Large Commercial Fishers Association,” “Small Commercial Fishers Association,” “Recreational Competition Fishers Association,” and “Recreational Tours Association.” Two days before their section meeting, students received an e-mail containing a general overview of the SHARC case and their confidential role-specific information and instructions. Students were instructed to read through the materials and to prepare to interact with classmates representing the other three roles in the simulated conference, but they did not complete any dependent measures at this time. These materials informed participants that they would be participating in a conference as a spokesperson for one of the four fishing associations mentioned above, all of which depend on shark fishing for income. The goal of the conference was to address the overharvesting crisis by determining the amount that each association of fishers should reduce their current harvesting level in order to preserve the fish stocks. The commercial fishers harvest more shark than the recreational fishers and are also less concerned with the future harvest of any particular species as switching to a different type of fish is relatively easy for them. The dilemma is therefore asymmetric in that each organization contributes a different amount to the current overharvesting problem and depends on the future health of the resource to a different extent, thereby creating different interests and concerns among the associations. In trying to reach agreement about harvest reductions, participants from each association wanted to (a) maximize current profit and (b) avoid depleting the number of sharks left that would jeopardize future harvests. Participants received a formula to enable profit calculations on the basis of the combined harvesting levels of all groups, and they learned that each of the associations was currently harvesting at their maximum capacity such that the overall harvest would have to be cut in half to maintain a sustainable population (from a current total harvest of 5,000 metric tons to a total of 2,500 metric tons). Participants were told that all relevant concerns were accounted for in the information and profit equations that they received. Exact details of the profit calculations and equations participants received are available from the authors. Experimental conditions. On arrival to the section meeting, participants were asked to review the materials they received over e-mail and to complete the first dependent measure in the experiment: estimates of fair harvesting levels. All participants indicated what they believed to be fair harvesting levels before being separated into individual groups for their discussion with the other association representatives. The way in which participants completed this dependent measure served as our key independent variable. Participants in approximately half of the groups (i.e., the self-focused condition) were simply asked, “Of the total harvest taken, what do you think is a fair percentage for your group to harvest?” Participants in the remaining groups (i.e., the other-focused condition) first considered the perspectives of the other group members before indicating what was fair for them to harvest. In particular, participants in these groups were first told, “Please take a minute to think about the other groups. As you can imagine, they may have different priorities than you do and are likely to view this situation from a different perspective. Thinking about the other
EPLEY, CARUSO, AND BAZERMAN
876
groups, what amount will each of them indicate is fair for them to harvest?” These participants were then asked, “Now, what do you think are fair harvesting levels for each group, including your own?” The order of the list of associations was held constant throughout all versions to minimize order effects. Assignment to the self-focused or other-focused conditions was done at the level of the course section to ease presentation of the stimulus materials in the large group setting. Random assignment to the self-focused versus other-focused groups was done at the level of the group, whereas assignment to each of the four roles was done randomly within each group. The simulated conference. Students from each role were randomly assigned to groups of 4 when they arrived for their section meeting such that one representative from each of the four roles was included in each group. Because each group needed to have representation from just one of the roles, only complete groups of 4 were included in the analysis. Each group was sent to a separate room with instructions to discuss how much each association would harvest. The first dependent measure form (which they had already completed) was collected from participants as they entered the room. Participants were told that they could follow any format they wished in discussing the issues and the potential solutions but that they were not allowed to make any binding commitments to specific harvesting levels (as happens in many real-life negotiations). After 25 min of discussion, the experimenter entered the room and distributed the second set of materials. Participants received role-specific final instructions, including the final two dependent measures. Participants left the discussion rooms and moved into a large lecture hall to ensure privacy and confidentiality when filling out the final forms. Final dependent measures. The final instructions reminded the spokespersons of the interests and goals of their associations. Participants once again indicated what they believed to be fair harvesting levels on a form similar to the one they completed before discussion. Participants in both groups then recorded their actual harvesting level for the next year. Participants handed in their forms and were told that the researchers would perform the relevant calculations and would share aggregate results with the class during the next lecture (which they did as promised).
Results
Table 1 Perceptions of Fair Allocations and Actual Behavior Among Groups in the Self-Focused and Other-Focused Conditions (Experiment 1) Fair to harvest (%)
Actual harvest
Focus condition
Before discussion
After discussion
Tons
% of total
Self Other
133 113
111 116
2,846 3,620
56.9 72.4
Note. Percentages listed in the first two columns are the summed percentage of the total harvest participants believed was fair for their association to take. Figures reported in the third column are the metric tons of the fishing stock that groups, on average, would actually harvest. The percentages of total reported in the far right column are out of the total metric tons available for harvest (5,000) rather than out of the total metric tons available for a sustainable harvest (2,500).
condition) found that egocentric assessments of fairness were stronger before discussion than after discussion (Wade-Benzoni et al., 1996). We replicated that result here. Groups in the selffocused condition claimed it was fair for them to take 133% of the harvest before discussion but only 111% after discussion, t(19) ⫽ 5.54, p ⬍ .001. Groups in the other-focused condition claimed it was fair for them to take 113% before discussion and 116% after discussion, t(19) ⬍ 1, ns. These fairness estimates after discussion did not differ between conditions, t(38) ⬍ 1, ns, suggesting that just thinking about others produces similar effects on fairness judgments as does actually talking to them in this situation. The repeated measures interaction for fairness estimates before and after discussion across conditions was significant, F(1, 38) ⫽ 21.26, p ⬍ .001.
Fairness Estimates Before Discussion
Actual Behavior
Participants in both conditions estimated the percentage of the available harvest that they thought was fair for their group to take. The fairness estimates for each of the four roles were summed within each group to create a measure of fairness for the group as a whole. As can be seen in Table 1, groups in the self-focused condition claimed it was fair for them to harvest significantly more (133%) than groups in the other-focused condition (113%), t(38) ⫽ 3.04, p ⬍ .005. Getting participants to think about their fellow group members by considering their perceptions of fairness caused a significant reduction in egocentric assessments of fairness. This manipulation did not eliminate egocentric assessments of fairness, however. The 113% claimed fair in the other-focused condition was still significantly higher than the normative baseline of 100%, t(19) ⫽ 3.48, p ⬍ .005.
We predicted that considering others’ perspectives would decrease egocentric judgments of fairness but would actually increase egoistic behavior. Results confirmed this prediction. Despite claiming that they deserved to take less before discussion, groups in the other-focused condition actually ended up taking more of the available harvest than did groups in the self-focused condition, t(38) ⫽ 5.57, p ⬍ .001. Other-focused groups took 72.4% of the available harvest compared with the 56.9% taken by the self-focused groups. This translated into lower profits for groups in the other-focused condition (M ⫽ $62.4 million) than for groups in the self-focused condition (M ⫽ $68.6 million). Although self-focused groups consumed less than did other-focused groups, it is worth noting that both of these figures are significantly larger than the 50% figure required for a sustainable harvest, ts(19) ⫽ 12.35 and 3.29, respectively (both ps ⬍ .01). This pattern of behavior, coupled with the fairness judgments, is consistent with our predictions. Considering others’ perspectives led groups to report that it was fair for them to take less of the overall harvest than groups in the self-focused condition before the group discussion, but groups in the other-focused condition actually took more of the overall harvest than groups in the selffocused condition. To test for the statistical significance of this interaction, we standardized both fairness judgments and harvest
Fairness Estimates After Discussion After the group discussion, all participants again estimated the percentage of the available harvest that they thought was fair for their group to take. This discussion session allows representatives from the different associations to talk with one another and share their thoughts on the harvesting situation. The original demonstration of this simulation (with conditions similar to the self-focused
REACTIVE EGOISM
amounts and submitted them to a 2 (condition: self-focused vs. other-focused) ⫻ 2 (measure: fairness judgments vs. harvest behavior) analysis of variance (ANOVA) with repeated measures on the second variable. The predicted interaction was indeed significant, F(1, 38) ⫽ 43.48, p ⬍ .001.
Discussion These results confirmed all of our major predictions. Egocentric biases in judgments of fairness were reduced by leading participants to consider the perspectives of their other group members, but this reduction did not lead to an analogous change in behavior. Although these participants reported that it was fair for them to take less of the shared resource, they actually ended up taking more of that resource compared with participants whose egocentric biases in judgment remained intact. It is worth noting that the self-focused and the other-focused groups did not differ in their stated judgments of fairness after discussion; hence, the consideration of others’ concerns and interests before discussion seems to be responsible for the observed differences in subsequent egoistic behavior rather than being simply a drop in egocentric allocations of fairness. This suggests that people’s actions in this context are guided more strongly by their beliefs about how others are likely to behave rather than by what they believe is objectively fair for themselves to take. This result is consistent with our hypothesis that considering others’ concerns and interests activates beliefs about others’ behavior, which, in turn, lead people to behave more selfishly themselves. This result also suggests that simply being made aware of the reality constraints involved in the situation by attending to everyone’s needs and thinking about it more carefully is insufficient to account for the complete pattern of results we observed in the other-focused conditions. Looking into the minds of others and intuiting their thoughts and likely behavior increased egoistic behavior, but simply talking with others did not. We interpret these behavioral results as a form of reactive egoism. Considering others’ perspectives highlighted, we believe, self-interested motives and likely actions on the part of other group members. Believing that others would behave selfishly led participants to behave more selfishly as well, even though they indicated indirectly that such behavior was unfair. Such cynical thoughts were unlikely to be considered, or processed as fully, when participants were not explicitly led to consider others’ perspectives. Experiment 1, however, was intended as a demonstration of our main phenomena rather than as a test of its underlying mechanism. To test this mechanism more directly, we conducted a follow-up simulation with a new group of 50 participants, each of whom imagined being a representative of the “Recreational Competition Fishers Association.” These participants did not engage in the actual exercise with other participants but instead simply read all of the role-relevant materials from the SHARC exercise and reported what they believed was fair for their association to harvest and then to indicate what they would actually harvest. When finished, all participants also indicated the amount they believed each of the other fishing associations would actually harvest. Via the same procedures as were used in Experiment 1, half of the participants considered others’ perspectives before completing these measures, whereas the other half did not.
877
As in Experiment 1, participants who considered others’ perspectives claimed it was fair for them to take less of the harvest (M ⫽ 13.52%) than participants who did not (M ⫽ 25.00%), t(48) ⫽ 6.60, p ⬍ .001, but these participants actually decided to take more of the harvest (M ⫽ 732.70 metric tons) than did participants who did not consider others’ perspectives (M ⫽ 629.2 metric tons), t(48) ⫽ ⫺2.18, p ⬍ .05. This again produced a significant Condition ⫻ Measure interaction, F(1, 48) ⫽ 49.23, p ⬍ .001. More important, participants who considered others’ perspectives also reported that others would take significantly more of the overall harvest (M ⫽ 3303.90 total metric tons) than participants who did not consider others’ perspectives (M ⫽ 2656.30 total metric tons), t(48) ⫽ ⫺3.53, p ⬍ .001. Beliefs about what others would take, however, were only marginally correlated with what participants indicated they would take (r ⫽ .25, p ⬍ .08). These results are therefore suggestive only of the mechanism underlying Experiment 1. We designed Experiment 2, in part, to investigate that mechanism more fully.
Experiment 2: Grants Experiment 2 was patterned after a generic resource allocation conflict in which members of different groups compete for a fixed pool of resources. In this experiment, undergraduates were asked to imagine that the dean of their university had received an anonymous donation to improve student life and that this gift was to be distributed among the student residence houses (dorms). Participants imagined that they were elected as a representative of their house and indicated the amount of money they believed was fair for their own group to receive as well as the amount they would actually request from the dean. Half of the participants did so after considering the perspectives of other house representatives, whereas the other half did not consider others’ perspectives. We predicted that those who considered others’ perspectives would claim that they deserved less money from the dean but would actually request more money compared with participants who did not consider others’ perspectives. When finished with these measures, participants also indicated the amount they expected other representatives would request from the dean. We predicted that those who considered others’ perspectives would indicate that other representatives would request more money than participants who did not consider others’ perspectives and that this difference would mediate the difference in participants’ own requests.
Method Participants Two hundred ninety-three Harvard University undergraduates completed an online questionnaire. As compensation, 2 participants were randomly chosen to win $250 each.
Procedure Participants were sent an e-mail with a link to an online survey. The first question randomly assigned them to either the self-focused (n ⫽ 141) or the other-focused (n ⫽ 152) conditions. The next screen asked participants to consider their current living situation in college. At the end of the 1st year at Harvard University, all
EPLEY, CARUSO, AND BAZERMAN
878
Figure 1. Mediational analysis for Experiment 2. Coefficients are standardized betas. The coefficient in parentheses is the direct relationship between the independent variable (condition) and the dependent variable (own request from the dean), controlling for the proposed mediator (others’ request from the dean). *p ⬍ .05. **p ⬍ .01.
students are randomly assigned to live in one of 12 “houses” (clusters of dormitories) on campus. They were asked to think about their house and imagine that the dean of the university had received an anonymous donation of $100,000 to be used “to support and improve student life in the undergraduate houses.” The dean would allocate the funds on the basis of both the need of the houses (to support infrastructure, social activities, etc.) as well as the strength of the proposal submitted by the students in each house. Consistent with Experiment 1, the self-focused group next reported what they thought was a fair percentage of the grant for their house to receive. The other-focused group, in contrast, first took a moment to think carefully about the needs of each house, after which they reported what they thought was a fair percentage for each house to receive (including their own). Following the fairness estimates, participants in both groups were told to imagine that they had volunteered to act as a representative for their house to put together the proposal for the dean. Everything about the proposal was said to be complete except for the amount of money they were going to request from the dean. Following these instructions, participants indicated what amount of the available $100,000 they would request from the dean in their proposal. Just prior to making this estimate, those in the other-focused group were once again asked to consider the perspectives and the likely behavior of the student representatives from the other houses. Finally, participants were asked a series of questions to measure their beliefs about the other representatives’ likely behavior. Specifically, they were asked to indicate (a) how much money (on average) they thought the other house representatives would request in their proposals, (b) how fair they thought the requests from the other houses would be (0 ⫽ not at all fair, 10 ⫽ very fair), and (c) to what extent they thought the other houses would attempt to gain more funding by exaggerating their own need (0 ⫽ not at all, 10 ⫽ very much).
Results Fairness Versus Requests As in Experiment 1, considering others’ interests and concerns reduced egocentric assessments of fairness. Participants in the other-focused condition claimed it was fair for them to receive less (M ⫽ 10.4%) than participants in the self-focused condition (M ⫽ 11.9%), t(291) ⫽ 2.18, p ⬍ .03. However, when asked how much they would actually request from the dean, those in the otherfocused condition actually requested more (M ⫽ $14,368) than those in the self-focused condition (M ⫽ $12,236), t(291) ⫽ ⫺2.21, p ⬍ .03. After standardizing these two dependent variables so they could be directly compared, a 2 (condition: self-focused vs. other-focused) ⫻ 2 (measure: fairness versus request) ANOVA
with repeated measures on the second variable revealed that this interaction pattern was significant, F(1, 291) ⫽ 14.41, p ⬍ .001.
Predictions About Others We propose that considering others’ perspectives in these competitive contexts activated cynical thoughts about others’ likely behavior, leading people to react more egoistically in return. Participants’ predictions about the likely behavior of others support this assertion. As expected, those in the other-focused condition thought that representatives of the other houses would request more from the dean (M ⫽ $14,790) than those in the self-focused condition (M ⫽ $12,130), t(284) ⫽ ⫺2.32, p ⬍ .025.1 Similarly, other-focused participants suspected that representatives from the other houses would be more likely to exaggerate their need (M ⫽ 6.80) than self-focused participants (M ⫽ 6.18), t(291) ⫽ ⫺2.78, p ⬍ .01. The focusing manipulation did not, however, have a significant effect on how fair participants thought the proposals of the other houses would be (Mself-focused ⫽ 5.40, Mother-focused ⫽ 5.18), t(291) ⫽ 0.97, ns. To examine whether beliefs about others’ likely selfish behavior mediated the impact of condition on participants’ own behavior, we followed the mediational procedure outlined by Baron and Kenny (1986). Condition (self-focused or other-focused) served as the independent variable in this analysis, amount participants requested from the dean served as the dependent variable, and estimates of what other house representatives would request served as the mediator. As can be seen in Figure 1, condition had a significant effect on estimates of how much participants would request as well as on how much others would request. When both the independent variable (condition) and the mediator (estimates of others’ requests) were entered simultaneously into a regression predicting participants’ own request, the effect of condition was no longer significant, but estimates of others’ requests were a significant predictor of actual requests. The effect of condition decreased significantly with the addition of estimates about others to the model (Sobel z ⫽ 2.29, p ⬍ .025). 1
Degrees of freedom vary for this analysis because 7 people failed to complete this question.
REACTIVE EGOISM
Discussion These results again confirm our key predictions and expand on the demonstration in Experiment 1 in two ways. First, Experiment 2 provided a novel demonstration and expanded on the generalizability of Experiment 1 by using a resource allocation task rather than a commons dilemma. Second, Experiment 2 provided evidence consistent with our proposed mechanism such that participants who considered others’ perspectives expected others to be more self-serving and egoistic than participants who did not consider others’ perspectives. This perceived egoism from others led people to react by behaving more egoistically themselves and to request more of the pool of fixed resources, even though they reported that they deserved considerably less than they were requesting. Looking into the minds of others can lead people to reduce self-serving assessments, but it can also ironically lead people to behave more self-servingly than they would have otherwise. Instead of removing the barriers to conflict resolution that egocentric biases appear to have put in place, considering others’ perspectives may actually make matters worse by activating thoughts about others’ self-interested motivations and likely behavior (Miller, 1999). In situations with diverging perspectives, reducing an egocentric focus on one’s own interests may not be the most prudent course of action. There is, however, one notable weakness of Experiment 2. Although mediational analyses support our proposed mechanism of reactive egoism, such results are simply correlational and cannot rule out the opposite causal model—that considering others’ perspectives somehow alters people’s own behavior and in turn influences people’s predictions about others’ behavior via egocentric projection. Indeed, reversing the mediator and the dependent variable in the mediational analysis also provides evidence for significant mediation (z ⫽ 2.18, p ⬍ .05). This is not very surprising as the amount participants requested themselves was highly correlated with what they believed others would request (r ⫽ .67, p ⬍ .001). To be sure, a large amount of existing data has suggested that people do indeed use their own thoughts and behaviors as a guide to others’ thoughts and behaviors (e.g., Ames, 2004; Epley, Keysar, Van Boven, & Gilovich, 2004; Krueger & Clement, 1994; L. Ross et al., 1977), but we believe that there are two reasons to doubt that this reverse causal path explains the pattern of data we observed. The first reason is theoretical. The account we have offered in which perspective taking highlights cynical thoughts about others provides a parsimonious and a priori explanation of both the decrease in egocentric biases in fairness estimates as well as the increase in egoistic behavior. It is harder to see how the reverse causal path could as parsimoniously account for this interaction between judgments of fairness and behavior. Our inability to identify a compelling alternative does not, of course, mean that such an alternative does not exist. The second reason is empirical. Although the correlation in Experiment 2 between the amount that participants requested and the amount they believed others would request was generally very strong, it was significantly stronger in the other-focused condition (r ⫽ .70) than in the self-focused condition (r ⫽ .52, z ⫽ 2.49, p ⬍ .05). This difference in correlations is consistent (albeit not exclusively so) with our hypothesis that participants’ own egoistic behavior in the other-focused condition was a reaction to the
879
presumed behavior of others rather than simply a projection of their own behavior onto others. The reverse causal story in which participants’ own behavior is projected onto others would not predict a difference between these correlations. Notice also that the results of Experiment 1 are inconsistent with this reverse causal path. Egocentric biases in fairness were identical after discussion for both participants who considered others’ perspectives and those who did not, but their behavior differed. This divergence between judgment and behavior is consistent with our predictions, but it is inconsistent with an account based solely on egocentric projection from one’s own presumed or intended behavior to others’ presumed or intended behavior. These reasons are consistent with our theoretical account, but they are only suggestive and may also be consistent with other alternative interpretations of Experiments 1 and 2. Stronger support for our proposed mechanism rather than this alternative account would therefore come from an experiment that directly manipulates the proposed mediator rather than one that simply measures it, which pits these two accounts directly against one another. Experiment 3 does exactly that.
Experiment 3: Grants, Take Two We propose that the reactive egoism observed in Experiments 1 and 2 stemmed from participants’ beliefs about others’ selfish behavior (activated by adopting others’ perspectives), rather than from participants’ own selfish behavior influencing their beliefs about others’ behavior. To test between these two accounts, we designed an experiment in which either we manipulated participants’ predictions about their own behavior and then measured their predictions about others’ behavior, or we manipulated participants’ predictions about others’ behavior and measured participants’ predictions about their own behavior. In particular, participants in Experiment 3 considered the same resource allocation dilemma used in Experiment 2. One group of participants (those in the manipulate-others condition) first indicated either the highest or lowest plausible amount of money that they believed the other house representatives would request, on average, and then indicated the amount they would actually request. The other group of participants (those in the manipulate-self condition) did precisely the opposite—they first indicated either the highest or lowest plausible amount they would request from the dean and then indicated the amount they believed the other house representatives would request, on average. The account we have offered predicts that manipulating predictions about others’ behavior will have a stronger effect on participants’ predictions about their own behavior than vice versa. The alternative account based on egocentric projection, in contrast, would predict that manipulating predictions about one’s own behavior would have a stronger effect on predictions of others’ behavior than vice versa.
Method Participants One hundred thirty-six Harvard University undergraduates completed an online questionnaire. As compensation, 1 participant was chosen at random to win $250. This participant pool was the same as that used in Experiment 2, so 5 participants who also participated in Experiment 2 were excluded from the following analyses.
EPLEY, CARUSO, AND BAZERMAN
880 Procedure
The materials were identical to those used in Experiment 2, but the procedure varied in several important ways. After reading the experimental materials, participants in the manipulate-others condition were reminded that the dean had a limited amount of funds to distribute and that everyone was therefore recommended to request either the highest (in the highrequest condition) or the lowest (in the low-request condition) amount they believed it was plausible to request. Participants then indicated how much, on average, they thought the other representatives would request from the dean and finally indicated how much they would actually request. Participants in the manipulate-self condition were given the same information but first indicated how much they would actually request and then indicated the amount they believed other representatives, on average, would request.
Results and Discussion To test our key hypothesis, participants’ responses were submitted to a 2 (condition: manipulate-self vs. manipulate-others) ⫻ 2 (request: high vs. low) ⫻ 2 (target predicted: self vs. other) ANOVA with repeated measures on the last factor. This produced a main effect of request, F(1, 132) ⫽ 16.56, p ⬍ .001, qualified by the predicted three-way interaction, F(1, 132) ⫽ 6.39, p ⫽ .01. As can be seen in Table 2, participants in the manipulate-others condition matched others’ presumed requests quite closely, with no significant difference between predictions of others and requests for self between the high- and low-request conditions; interaction: F(1, 68) ⫽ .61, p ⫽ .44. Participants who indicated the highest amount others would request asked for significantly more from the dean themselves (M ⫽ $18,291) than those who indicated the lowest amount others would request (M ⫽ $9,267), t(68) ⫽ 2.69, p ⬍ .01. In the manipulate-self condition, in contrast, a clear discrepancy in what people themselves would request versus what they thought others would request emerged; interaction: F(1, 64) ⫽ 5.86, p ⬍ .05. Although participants themselves requested more in the high-request condition (M ⫽ $20,364) than in the low-request condition (M ⫽ $9,656), t(64) ⫽ ⫺2.95, p ⬍ .01, there was no significant difference between these two conditions in the amount participants believed that others would request (Ms ⫽ $16,016 and $13,828, respectively), t(64) ⫽ .85, p ⫽ .40. Finally, the predicted requests of self and others were strongly correlated in this experiment. But as with Experiment 2, this correlation was significantly stronger in the manipulate-others condition (r ⫽ .76, p ⬍ .001) than in the manipulate-self condition (r ⫽ .41, p ⫽ .001, z ⫽ 3.20, p ⫽ .001).
Table 2 Predicted Amount Requested in the High- and Low-Request Conditions for Self and Others Among Participants in the Manipulate-Self Versus Manipulate-Others Conditions (Experiment 3) Manipulate-self
Manipulate-others
Request condition
Self amount
Others’ amount
Self amount
Others’ amount
High Low Difference
$20,364 $ 9,656 $10,708
$16,016 $13,828 $ 2,178
$18,291 $ 9,267 $ 9,025
$19,000 $11,528 $ 7,476
These results are consistent with our theoretical account and are inconsistent with the reverse causal path, but they are not as tightly linked to the empirical findings of Experiment 2 as we would have liked to eliminate all concerns about this alternative interpretation in that experiment. To provide such tightly linked support, we conducted one additional follow-up experiment by using 141 Harvard university undergraduates who again completed an online questionnaire for a chance to win a $250 lottery. These participants completed a procedure similar to that in Experiment 3; but instead of indicating the highest or lowest plausible values they or others would request, these participants were given the average amount participants thought others would request in the other-focused and self-focused conditions in Experiment 2. In particular, participants in the manipulate-others condition were told that after considering the relevant facts, the other representatives, on average, had decided to request either $14,367 (in the high-request condition) or $12,236 (in the low-request condition) from the dean. Participants then indicated how much they would actually request. Participants in the manipulate-self condition, in contrast, were told to imagine that after carefully considering the relevant facts that they had decided to request either $14,367 (in the high-request condition) or $12,236 (in the low-request condition). Participants in the manipulate-self condition then indicated how much they believed others, on average, would request from the dean. As in Experiment 3, manipulating how much participants believed others would request significantly affected how much participants themselves requested from the dean, with those in the high-request condition asking for significantly more (M ⫽ $17,338) than those in the low-request condition (M ⫽ $13,703), t(75) ⫽ 2.76, p ⬍ .01. Manipulating what participants themselves requested, in contrast, had no significant influence on how much they believed that others, on average, would request in the high(M ⫽ $13,957) versus in the low-request conditions (M ⫽ $13,716), t(62) ⫽ 0.16, p ⫽ .86. The interaction between these two effects approached significance, F(1, 137) ⫽ 3.02, p ⫽ .08. Although the mediational analysis in Experiment 2 is ambiguous with respect to the causal impact of considering others’ behavior on influencing one’s own behavior (i.e., the causal connection in Experiment 3), in this additional follow-up experiment it is crystal clear. Both experiments demonstrate that considering others’ likely behavior had a significant effect on one’s own predicted behavior, whereas considering one’s own behavior had no significant effect on predictions of others’ behavior. This experiment is clearly consistent with our proposed mechanism underlying reactive egoism and is inconsistent with the alternative model based on the reverse causal path in which participants’ own behavior influences their predictions of others’ behavior.
Experiment 4: Chocolate Chips One implication of our account of reactive egoism in social interaction is that such reactive behavior should occur only when the consideration of another’s perspective highlights divergent self-interests. Such divergences are the sine qua non of competitive group contexts, but not all group contexts are competitive. Groups often work together in a cooperative fashion to achieve a mutually shared goal. Self-interests in such cooperative contexts converge, rather than diverge, and looking into the minds of others in these contexts should not uncover the diverging self-interests that
REACTIVE EGOISM
emerge in the competitive contexts we have studied thus far. If behavior toward other group members is mediated by what people see when they look into the minds of those group members, then the reactive egoism we observed in Experiments 1, 2, and 3 should be moderated by the competitive versus the cooperative nature of the group. In competitive contexts, considering others’ perspectives should increase egoistic behavior. In cooperative contexts, considering others’ perspectives should highlight shared interests and therefore should not increase egoistic behavior. We tested these hypotheses in Experiment 4 by designing an actual commons dilemma in the laboratory. Instead of using simulated behavior as in Experiment 1, Experiment 4 measured actual behavior by having participants take from a fixed pool of resources we supplied to them. In particular, participants in Experiment 4 were asked to bake chocolate chip cookies in the laboratory. The key feature of this experiment was that participants were given only a small amount of premium chocolate chips for the entire group to use in their cookies. The amount of chips (by weight) taken by each participant—analogous to the amount of fish taken in Experiment 1—served as our key measure of behavior. As in the earlier experiments, half of the participants considered the perspectives of their other group members before indicating the amount of the fixed resource it was fair for them to take, whereas the other half did not. In addition, half of the groups competed against each other, whereas the other half cooperated with one another. This was accomplished by manipulating the payoff structure of the experiment. Competitive groups were told that the person who made the best cookies in the group would be eligible for a monetary award, whereas cooperative groups were told that the group who made the best cookies would be eligible for a monetary award as a group. We predicted that considering others’ perspectives would decrease egocentric perceptions of fairness regardless of the cooperative or competitive nature of the group but that it would increase egoistic behavior only in the competitive contexts. We therefore predicted a significant interaction between judgment and behavior among those in the competitive groups. Because cooperative contexts should highlight shared rather than divergent interests, we did not predict the same interaction between judgment and behavior among those in the cooperative condition.
Method Participants Ninety-four Harvard University undergraduates participated in this study in exchange for $15 or partial course credit.
Procedure Participants were greeted by an experimenter in a waiting area immediately upon arrival to the laboratory and were then led to an individual testing room. They were told that they would be baking chocolate chip cookies in this experiment and that they would have to share a limited amount of ingredients with 5 other group members. They were informed that there were just enough ingredients for everyone to bake with their given recipes. All were told that their cookies would be tasted and rated by an independent judge. Participants in the cooperative conditions were told that they would receive a team score based on the average ratings of the cookies that each team member baked such that their team score depended on everyone’s
881
performance. They were further notified that their group of 6 was competing against other groups of 6 and that the team with the highest team score would win $100. Participants in the competitive conditions, on the other hand, were told that they were competing against the other members of their own group for the best individual score such that the person in the group of 6 with the highest score would be entered into a lottery for $100. All participants were told that they had been randomly selected to be the first member of their group to choose ingredients. Because many experimental sessions occurred back-to-back, all participants were told that “the group before you is just finishing up” to reduce any suspicion associated with seeing another participant who was already in the midst of baking. They were also told that they might see the next person in their group and were instructed not to talk to any other participant they might see in the kitchen or the common area. After hearing the procedure, participants completed a set of measures that reminded them that there were two types of chocolate chips available for them to use in their cookies: premium Godiva chocolate chips and generic CVS-brand diet chips.2 Those in the self-focused conditions were asked to report what percentage of each type of chips they thought was fair for them to take. As in the previous experiments, those in the other-focused groups were first asked to think about the situation from the perspective of their other group members. They then reported what they believed was fair for each of their other group members to take, as well as what they thought was fair for themselves to take. Once this form was completed and returned, the experimenter entered the room to provide participants with a recipe for Nestle´ Toll House chocolate chip cookies as well as two bowls containing the relevant ingredients for the participants to mix. Participants were told that they needed to go to the kitchen to see how the oven worked and to collect the remaining ingredients necessary for baking. There were two bowls of chocolate chips in the kitchen; one was clearly labeled Godiva and the other was labeled CVS. There were half as many chips in the bowl of Godiva compared with the bowl of CVS to highlight the relative scarcity of the premium ingredient. There were, in fact, just enough total chips for six batches of cookies. After a brief description of how to set the oven temperature, the experimenter reminded the participant that he or she was randomly chosen to select ingredients first. Participants were left alone in the kitchen to choose their chips and then returned to their individual lab room to finish preparing their cookies. While participants were alone preparing their cookies, the experimenter entered the kitchen to weigh the two bowls of chips to obtain a measure of how much of each variety was taken by each participant. When they finished preparing their dough, participants reported back to the kitchen with four cookies on a baking sheet and placed them in the oven. While waiting for their cookies to bake, participants completed another set of dependent measures. Specifically, participants were asked to indicate the percentage of chips that their fellow group members would have taken had they been first to select ingredients. They were then asked to rate the Godiva chips compared with the CVS chips “in terms of how valuable they were to the quality of your cookies” on an 11-point scale (0 ⫽ CVS much more valuable, 10 ⫽ Godiva much more valuable), followed by one final measure of others’ likely behavior that asked participants to rate how “cooperative” (working together) compared with how “competitive” (competing against each other) they felt their group members were likely to be (0 ⫽ very competitive, 10 ⫽ very cooperative). When the baking was finished, one of the four cookies was saved and later rated for quality by a “judge” (blind to experimental condition and hypothesis) on a 7-point scale.
2
CVS is a large pharmacy retailer that carries a number of generic, low-cost brands of common food items, whereas Godiva is a well-known manufacturer of premium quality chocolate. The difference between these two brands, and their obvious desirability, was well-known to participants.
EPLEY, CARUSO, AND BAZERMAN
882
After 38 participants completed this procedure successfully, the 39th participant inadvertently burned his cookies, setting off the fire alarm and forcing the evacuation of the 15-floor building and all of the residents therein. Due to public safety concerns and additional ire from colleagues, we were unable to continue baking cookies following this incident. As a result, actual ratings of the cookies were not included in the following analyses. All subsequent participants completed the same procedure up to the point of baking, at which point they completed the final set of dependent variables and were then given the option of taking their dough home to bake for themselves. Because this procedural change came at the very end of the experiment and did not alter participants’ preceding experience in any way, it is not surprising that there were no significant differences in the results obtained before and after this procedural change in the following analyses. At the end of the experiment, participants were debriefed and probed for suspicion about the nature of the study. They were informed about the cooperative– competitive manipulation and were told that, in fact, the names of 2 participants would be drawn at random for the $100 prizes.
Results Judgment Versus Behavior All participants were asked to estimate the percentage of the available Godiva chips that they thought was fair for them to take and then were allowed to actually take some amount of Godiva chips to use in their baking. As in the preceding experiments, we predicted a significant interaction between judgment and behavior among participants in the competitive condition. Among participants in the competitive condition, those in the other-focused condition claimed it was fair for them to take fewer Godiva chips (M ⫽ 26.01%) than did those in the self-focused condition (M ⫽ 31.01%), but those in the other-focused condition actually took more chips (M ⫽ 3.01 oz.) than did those in the self-focused condition (M ⫽ 2.55 oz.). A 2 (condition: self-focused vs. otherfocused) ⫻ 2 (measure: fairness vs. behavior) ANOVA with repeated measures on the standardized scores for the second factor indicated that this interaction was significant, F(1, 45) ⫽ 4.73, p ⬍ .05, although neither of these simple effects was significant, ts(45) ⫽ 1.49 and ⫺1.57, respectively ( ps ⬎ .20). Among participants in the cooperative condition, however, only a main effect of perspective taking on judgment and behavior emerged, F(1,45) ⫽ 5.47, p ⬍ .05, with no significant interaction (F ⬍ 1, ns). As in the competitive condition, those in the otherfocused condition reported it was fair for them to take fewer Godiva chips (M ⫽ 20.33%) than those in the self-focused condition (M ⫽ 29.49%), but these other-focused participants also took fewer chips (M ⫽ 1.62) than did self-focused participants (M ⫽ 2.26). Follow-up contrasts revealed that both of these simple effects were significant, ts(45) ⫽ 3.10 and 2.47, respectively (both ps ⬍ .05). Overall, these results indicate predictably divergent effects of perspective taking on judgment versus behavior (see Table 3). For participants’ judgments of fairness, only a significant main effect of perspective-taking condition emerged, F(1, 90) ⫽ 4.47, p ⬍ .05, with no main effect of cooperative versus competitive condition nor any significant interaction (Fs ⬍ 1.3). Those in the otherfocused condition reported that it was fair for them to take fewer Godiva chips (M ⫽ 23.34%) than did participants in the selffocused condition (M ⫽ 30.20%), t(92) ⫽ 2.12, p ⬍ .05. For participants’ actual behavior, however, there was a significant
Table 3 Percentage of Resource Participants Perceived Was Fair for Them to Take and the Amount Actually Taken by Participants in the Self-Focused Versus Other-Focused Conditions in the Competitive Versus Cooperative Group Contexts (Experiment 4) Competitive
Cooperative
Focus condition
Fair to take (%)
Amount taken
Fair to take (%)
Amount taken
Self Other
31.01 26.01
2.55 3.01
29.49 20.33
2.26 1.62
Note. The percentage of Godiva chips participants perceived was fair for them to take across conditions is reported in percentages, whereas the amount of Godiva chips actually taken is reported in ounces.
interaction between our two experimental conditions, F(1, 90) ⫽ 4.02, p ⬍ .05. The decrease in the amount that participants in the other-focused condition claimed was fair for them to take was matched by a drop in the amount actually taken among those in the cooperative condition but not among those in the competitive condition.
Ratings About Others We again predicted that participants’ behavior—in this case, the amount of Godiva chips they actually took—would be mediated by their beliefs about others’ likely behavior. We obtained two such measures: (a) the amount of Godiva chips participants believed each of the others would take if they had gone first and (b) participants’ subjective ratings of how cooperative versus competitive they expected their other group members to be. The average of the first measure was significantly correlated with the second (r ⫽ ⫺.36, p ⬍ .001) and showed the same pattern of responses, so we reverse scored the subjective rating of how cooperative participants expected others to be such that higher numbers indicated more competitive behavior, standardized responses for both measures, and then collapsed them into a single composite to ease presentation. A 2 (condition: self-focused vs. other-focused) ⫻ 2 (context: cooperative vs. competitive) ANOVA on this standardized composite revealed a significant main effect of context, F(1, 90) ⫽ 16.18, p ⬍ .001, qualified by the predicted interaction, F(1, 90) ⫽ 5.47, p ⬍ .01. Other-focused participants expected others to be significantly more competitive in the competitive condition (M ⫽ .57) than in the cooperative condition (M ⫽ ⫺.54), t(47) ⫽ 5.17, p ⬍ .001, but self-focused participants showed no difference in how they expected others to behave in the competitive (M ⫽ .05) versus the cooperative conditions (M ⫽ ⫺.09; t ⬍ 1). These results are consistent with our prediction that considering others’ perspectives will highlight motives and likely behaviors that people would have otherwise overlooked.
Mediational Analysis We predicted that these beliefs about others’ likely behavior would mediate the effect of perspective taking in cooperative and competitive groups on one’s own behavior. Because we expected (and found) that perspective taking had different effects in coop-
REACTIVE EGOISM
erative and competitive groups, we were essentially predicting that thoughts about others’ likely behavior would mediate the interaction (or moderating effect) of our perspective taking and cooperativeness manipulations on actual behavior. As such, we followed the procedure for mediated moderation outlined in Baron and Kenny (1986). We treated perspective taking, cooperative– competitive context, and their interaction as the independent variables, predictions about others’ competitive behavior as the mediator, and the actual amount of Godiva taken as the dependent variable. As can be seen in Figure 2, the interaction of perspective taking and cooperativeness had a significant effect on the dependent measure (amount of Godiva taken) as well as on the proposed mediator (predictions of others’ competitive behavior), and the proposed mediator significantly predicted the dependent measure. Finally, when the mediator was added to the overall model, the interaction term was significantly reduced (z ⫽ 2.02, p ⬍ .05), and it became nonsignificant (t ⫽ ⫺.17, ns). The proposed mediator, however, remained significant in this complete model (t ⫽ 9.68, p ⬍ .001). This pattern of results suggests that predictions about others’ cooperative versus competitive behavior mediated the interaction of perspective taking and cooperative– competitive context on the amount of Godiva actually taken by participants. As in Experiment 2, however, the reverse causal path also yielded a significant mediational path (z ⫽ 2.11, p ⬍ .05), but the interaction term in this analysis also remained significant, (t ⫽ 2.73, p ⬍ .01), suggesting only partial mediation for this reverse causal path. These mediational analyses are again only suggestive for the causal direction of the underlying mechanism because the mediator and key dependent measures were so highly correlated (r ⫽ .70, p ⬍ .001), and we again defer to the results of Experiment 3 for a more definitive demonstration of our proposed model.
Discussion These results are consistent with our predictions that the competitive versus cooperative nature of a group is an important moderator of the impact of considering others’ perspectives on social interaction. In particular, these results suggest that the competitive versus cooperative nature of a group can moderate the impact of perspective taking on behavior.
883
Competitive contexts are generally defined by competing interests among group members, and the consideration of others’ perspectives in the competitive condition of this experiment decreased egocentric biases but directionally increased egoistic behavior. Cooperative contexts, in contrast, are generally defined by shared interests among group members. Considering others’ perspectives in the cooperative condition of this experiment reduced egocentric biases just as in the competitive condition, but it also decreased egoistic behavior, unlike in the competitive condition. In fact, participants in cooperative groups who considered others’ perspectives tended to behave less egoistically than did participants who did not consider others’ perspectives, behavior that was in line with their judgments of fairness. Additional analyses again suggested that these behavioral effects were mediated by participants’ beliefs about others’ likely behavior. Considering others’ perspectives led participants in competitive contexts to believe that other group members would have been more selfish and competitive, leading participants to react by behaving more selfishly in return. This reactive egoism did not occur in the cooperative conditions, it appears, because such cynical thoughts simply were not activated in this context.
Experiment 5: Coalitions Although consistent with our hypotheses, all of the experiments presented thus far are open to an alternative interpretation. In particular, instead of influencing people’s beliefs about others’ behavior, it is possible that considering others’ perspectives simply led people to think harder about their behavior and the reality constraints of the situation, thereby increasing the likelihood of making a rational response. In Experiment 4, for instance, it is in one’s own best interests (at least in the short term) to compete in the competitive context and to cooperate in the cooperative context, and it is possible that adopting others’ perspectives simply made this rational response more transparent and therefore more likely. A clear test between this alternative interpretation and the one we have offered would require manipulating the competitive versus cooperative group context while holding the actual reward structure and reality constraints of these groups constant. Experiment 5 did exactly this by simply manipulating the framing (or
Figure 2. Mediational analysis for Experiment 4. Coefficients are standardized betas. The independent variable in this analysis (between-conditions interaction) is the interaction between self-focused versus other-focused conditions and the competitive versus cooperative nature of the experimental incentives. The coefficient in parentheses is the direct relationship between the independent variable (between-conditions interaction) and the dependent variable (amount of Godiva taken), controlling for the proposed mediator (others’ predicted competitive behavior). *p ⬍ .05. **p ⬍ .01.
EPLEY, CARUSO, AND BAZERMAN
884
description) of a novel negotiation exercise. All participants experienced the same exercise, but half were told they were playing the “cooperative alliance game,” whereas the other half were told they were playing the “strategic competition game.” This experiment was inspired by earlier findings that people play more competitively in a prisoner’s dilemma game when it is described as the “Wall Street game” than when the game is described as the “community game” (Liberman, Samuels, & Ross, 2004). The particular negotiation exercise in Experiment 5, however, is novel. Although objectively identical in terms of their structural payoff and therefore in their rational response, changing the way the game is described is likely to change participants’ beliefs about the normative responses of other participants and to alter participants’ own behavior in turn. A significant framing effect in this experiment would not rule out the possibility that considering others’ perspectives increases the likelihood of identifying the most rational response in a social interaction (as it would require affirming the null hypothesis), but it would rule in the importance of others’ presumed behavior as a determinant of reactive egoism.
Method Participants engaged in a 4-person negotiation in which they represented companies involved in either the development or the distribution of fuel cells. The goal was for participants to maximize their profits, and to do so they needed to form strategic alliances with other members of the group (ideally with all members of the group). Creating an alliance generated financial value but came at a cost of effort to each of the participants that reduced their profits. Effort was not explicitly monitored, however, and individuals could profit by entering into an alliance and then expending less effort than they promised. This procedure was designed to mimic real-world strategic alliances, which often fail in part because individual parties do not contribute sufficient resources to support the endeavor. Participants who formed alliances indicated what percentage of the total profit they believed was fair for them to receive, which served as our measure of perceived fairness. The effort participants actually chose to expend in the exercise (at a cost to their profits) served as our behavioral measure of selfishness. Approximately half of the participants were told they were participating in the strategic competition game, whereas the other participants were told that they were participating in the cooperative alliance game. We expected to observe reactive egoism in the competition game but not in the alliance game. That is, we expected that participants who considered others’ perspectives in the strategic competition game would say it was fair for them to receive less of the overall profits (indicating that they would expend more effort) but would actually expend less effort than participants who did not consider others’ perspectives. We did not expect this reactive pattern among participants in the cooperative alliance game, but, we instead expected, as in Experiment 4, that those who considered others’ perspectives would claim they deserved less of the overall profit and would demonstrate behavior consistent with that assessment (i.e., they would expend more effort to the alliance).
Participants One hundred twenty-four participants from an existing participant pool of Boston-area residents participated in exchange for $15 and the chance to win an additional $25 on the basis of their performance.
Procedure Participants reported to a computer laboratory in groups of up to 36 and drew a number from a bowl that randomly assigned them to their experi-
mental conditions (described in the following paragraphs). The numbers on the cards also corresponded to the hypothetical company that the individual would be representing in the experiment: Alpha, Beta, Cappa, or Delta. Participants were seated in front of individual computers, provided with written instructions about the experiment, and listened to a presentation from the experimenter clarifying the rules of the game. The instructions explained that participants would be involved in a simulated negotiation exercise, with each individual representing one of four firms in the emerging fuel cell market. Two firms, Alpha and Cappa, were fuel cell developers; they were working to develop and refine the basic fuel cell design. Firms Beta and Delta were cell distributors; they were responsible for connecting fuel cells to the residential market. All four firms were described as having important patents or offering unique benefits in the industry. Ostensibly, as the result of the interconnected nature of this industry, each firm in this exercise was worthless alone but potentially valuable in an alliance with other firms. The goal of the simulation therefore was to have fuel developers form strategic alliances with cell distributors. No firm was allowed to participate in more than one alliance, and three-way alliances were not allowed. If the firms managed to form an alliance, the alliance would create financial value that would need to be divided among the members of the alliance. Such value, however, came at a cost. Being in an alliance required a firm to contribute resources to support the alliance, meaning that each of the firms would also incur some costs. The precise value and cost for all possible alliance types were each made explicit in the instructions, the exact details of which are available from the authors. The general structure of these values and costs was specified such that firms that did not form an alliance created no value but also incurred no costs. Two-way alliances between one cell developer and one cell distributor created some value and incurred some costs, and four-way alliances had the potential to create more value than any of the two-way alliances. If the four parties agreed to a four-way alliance, the value of that alliance would depend on the combined effort that the four firms contributed independently to support the alliance. That is, each firm chose which of three levels of effort, or cost, they would forgo in order to make the alliance as a whole more profitable. Participants knew they would be making this decision privately and anonymously. In this sense, the game was similar in structure to the social dilemmas faced by participants in Experiments 1 and 4. As an incentive, each participant received one “lottery ticket” for every $1 million earned in this simulation, and a computer selected one winning ticket from each of the four firms to receive an additional $25 in the experiment. All participants thus had the same financial incentive to maximize their individual profits. After hearing the instructions, participants filled out a form on the computer asking them to indicate what percentage of the total alliance value they thought would be fair for them to receive, should they enter into a four-way alliance. Consistent with our previous studies, those in the self-focused condition simply reported the amount they thought was fair for their own firm to receive, whereas those in the other-focused condition were first asked to think about the situation from the perspective of the other firms and then to report what they thought was fair for each firm to receive (including their own). As mentioned earlier, we manipulated the cooperative versus competitive nature of the experiment simply by changing the name of the game, while keeping the structure identical across all participants. Those in the cooperative condition were told they would be working with each other in the cooperative alliance game, whereas those in the competitive condition were told they would be competing with each other in the strategic competition game. The name of the game was displayed on the top of the instruction sheets and on the top of each form presented on the computer. When finished with their fairness estimates, participants gathered in groups of 4 in small conference rooms to discuss, negotiate, and decide on any alliances. Participants had 20 min for discussion, after which they had to be prepared to state which alliance, if any, they had agreed upon. Firms
REACTIVE EGOISM entering an alliance needed to submit an alliance agreement form that specified which type of alliance was reached and how the ultimate value of the alliance would be distributed among the four firms. After each firm signed the form, the experimenter entered the values into a computer that performed the relevant profit calculations. Participants then returned to their computers to complete the final dependent measures privately. The first of these was the level of effort they chose to contribute to support their alliance. Any firm entering a four-way alliance specified a high, medium, or low level of effort. Finally, participants indicated how cooperative they thought their group interaction was on an 11-point scale ranging from 0 (very competitive) to 10 (very cooperative). When finished, a computer program calculated the individual profits for each firm and chose the $25 prize winners on the basis of the results. Participants were then debriefed, paid, and dismissed.
Results and Discussion Judgment Versus Behavior Before actually agreeing on any alliances, participants were asked to imagine they were successful in forming a four-way alliance and to estimate the percentage of the resulting profits they thought was fair for them to receive. Following the group interaction, participants had to decide what amount of effort to contribute to support their alliance (all groups agreed on a four-way alliance). As increased effort corresponded to less selfish behavior (sacrificing more individual profit for the good of the group), we expected lower levels of effort in the other-focused competitive groups. We coded a low level of effort as 1, a medium level as 2, and a high level as 3. Once again, asking participants in competitive groups to consider the situation from the perspective of the other firms diminished egocentric perceptions of fairness but actually increased egoistic behavior. Other-focused participants in competitive groups claimed that it was fair for them to receive marginally less of the profits (M ⫽ 27.0%) than their self-focused counterparts (M ⫽ 31.1%), t(66) ⫽ 1.74, p ⬍ .09, but otherfocused participants actually contributed less effort (M ⫽ 2.34) than did self-focused participants (M ⫽ 2.78), t(66) ⫽ 2.68, p ⬍ .01. A 2 (condition: self-focused vs. other-focused) ⫻ 2 (measure: fairness vs. behavior) ANOVA with repeated measures on the standardized scores for the second factor indicated that this predicted interaction was significant, F(1, 66) ⫽ 8.67, p ⬍ .01. No such interaction, however, emerged for participants in the cooperative condition (F ⬍ 1, ns). Other-focused participants in the cooperative condition claimed they deserved nonsignificantly less of the profits (M ⫽ 28.9%) than did self-focused participants (M ⫽ 32.2%), t(53) ⫽ 1.37, p ⬎ .20, and showed no significant difference in the amount of effort they actually contributed (Ms ⫽ 2.81 and 2.79, respectively). Overall, the results reveal the consistent pattern found in the previous studies. For judgments of fairness, participants in the other-focused condition indicated that it was fair for them to take less of the overall profit (M ⫽ 28.0%) than participants in the self-focused condition (M ⫽ 31.5%), t(121) ⫽ 2.09, p ⬍ .05.3 The amount claimed fair to take did not differ between those in the cooperative (M ⫽ 30.3%) and the competitive (M ⫽ 29.2%) conditions, t(121) ⬍ 1, ns, and there was no interaction between the perspective-taking and cooperative– competitive groups, F(1, 119) ⬍ 1, ns. For actual behavior, an ANOVA revealed a significant Condition ⫻ Framing interaction, F(1, 120) ⫽ 4.67, p ⬍ .05.
885
As noted earlier, inspection of the means revealed that participants in the self-focused cooperative (M ⫽ 2.79), self-focused competitive (M ⫽ 2.78), and other-focused cooperative (M ⫽ 2.81) groups all contributed more effort than those in the other-focused competitive group (M ⫽ 2.34).
Ratings of Competition–Cooperation Participants also rated how cooperative versus competitive their group interaction was, and these ratings showed effects similar to those seen in Experiment 4. Participants in the other-focused condition rated their interaction as more competitive in the competitive frame condition (M ⫽ 6.17) than in the cooperative frame condition (M ⫽ 7.88), t(62) ⫽ 2.79, p ⬍ .01. There was no significant difference, however, in the self-focused condition between those in the competitive frame condition (M ⫽ 7.29) and those in the cooperative frame condition (M ⫽ 7.47; t ⬍ 1). This interaction was significant, F(1, 120) ⫽ 4.04, p ⬍ .05. A mediational analysis with this measure of competition– cooperation as a mediator of the interaction between our two experimental conditions on participants’ own behavior was nonsigificant (z ⫽ 1.57, p ⫽ .12), as was the analysis reversing the mediator and the dependent measure (z ⫽ 1.62, p ⫽ .10). Of course, the implication of this measure is more ambiguous than the result from Experiment 4 because participants completed this measure after a real interaction rather than simply imagining or predicting how others would behave as in the previous experiments. This measure is, however, at the very least consistent with those of the previous experiments and again suggests that looking into the minds of others may highlight motivations and interests that an egocentric focus on the self might otherwise overlook. Overall, the results of Experiment 5 confirm our main predictions that altering the description of the participants’ exercise would alter their behavior, despite holding constant the actual structure and incentive system of the game. Those who believed they were playing the strategic competition game displayed the patterns of reactive egoism observed in the preceding experiments. Those who considered others’ interests and concerns demonstrated weaker egocentric biases in perceptions of fairness but actually behaved more selfishly compared with participants who did not consider others’ perspectives. Those who believed they were playing the cooperative alliance game, in contrast, did not show this pattern of reactive egoism, again demonstrating how the competitive versus cooperative context can alter the impact of perspective taking in social interaction. These results are not open to the alternative interpretation that perspective taking simply increased the tendency to notice, and therefore to choose, the most rational response or to attend more carefully to the reality constraints of the situation. The rational response and reality constraints remained constant across all experimental conditions, and yet the predicted pattern of behavior emerged.
General Discussion It is unlikely that a blind person would ever argue with a sighted person about the color of a painting, the shape of a cloud, or the 3
One participant who failed to fill in this question was excluded from analyses of fairness estimates.
886
EPLEY, CARUSO, AND BAZERMAN
extent to which one’s shirt matches one’s pants. People’s eyes are universally recognized conduits for their visual experience, and the reasons for divergent visual perceptions are therefore relatively easy to identify and appreciate. Harder to identify and appreciate, however, are the ways in which psychological factors such as preexisting expectations, attitudes, or self-interest can also influence people’s perceptions of the external world. Two equally sighted people with divergent interests and beliefs may look at the very same stimulus—from a court settlement to a labor contract to an aggressive play in a sporting event—and “see” very different things (Hastorf & Cantril, 1954; L. Ross & Ward, 1995, 1996). Unlike divergent perceptions between the sighted and the blind, however, a failure to identify and appreciate the reasons for divergences in psychological perspectives can lead to heated conflict between individuals, between groups, and between nations (Pronin, Puccio, & Ross, 2002). One commonly offered strategy for alleviating these divergences in psychological perspectives is for people to actively consider the other side’s thoughts, interests, and concerns in the hopes that doing so will overcome egocentric biases in judgment. In the present experiments, we investigated the consequence of such perspective taking on people’s judgments and behaviors in a series of social interactions. We observed two consistent results. First, considering others’ perspectives did indeed diminish egocentric assessments of fairness. Those who considered what other group members would think is a fair allocation reported that it was fair for them to take a smaller percentage of fixed resources than those who did not consider others’ perspectives. This occurred in a wide variety of social interactions, with resources ranging from fish (Experiment 1) to grant money (Experiments 2 and 3) to chocolate chips (Experiment 4) to corporate profits (Experiment 5).4 Our second consistent finding was that these reductions in egocentric assessments of fairness were not consistently matched by reductions in egoistic or self-serving behaviors. Instead, each of the experiments demonstrated a reliable tendency for reactive egoism in competitive groups. Although those who considered others’ perspectives claimed it was fair for them to take less of a fixed resource than those who did not consider others’ perspectives, these participants actually took more of those very resources when given the opportunity to do so. This occurred, as Experiments 2 and 4 suggested and as Experiment 3 demonstrated directly, because considering others’ perspectives led participants to believe that others would behave more selfishly. Egocentric assessments of fairness are obvious and well-documented sources of conflict between individuals and within groups, but reducing those egocentric biases does not necessarily reduce conflict in behavior. Sometimes, in fact, considering others’ perspective could make matters worse. In addition to demonstrating this interaction between judgment and behavior, and obtaining evidence to support our proposed mechanism of reactive egoism, Experiments 4 and 5 also demonstrated that the competitive versus cooperative nature of a group is an important moderator of the impact of perspective taking on behavior. In competitive groups defined by divergent interests and goals, the consideration of others’ perspectives leads to reactive egoism. In cooperative groups defined by shared interests and goals, however, perspective taking reduces egoistic behavior. Looking into the mind of a competitor highlights self-interested
motives and leads people to behave more self-interestedly in return. Looking into the mind of a cooperative collaborator, however, highlights shared interests and leads to more cooperative behavior in return. The impact of perspective taking on behavior among individuals or within groups, then, depends critically on what people see when they look into the minds of others. These results suggest that one of the keys to harnessing the benefits of perspective taking in groups without incurring the costs of reactive egoism is to highlight the shared interests between otherwise competitive group members. We did this explicitly in Experiment 5 simply by changing the description of the group task to make cooperation a salient goal, while keeping the competitive structure of the group interaction constant. The importance of highlighting superordinate goals is not, of course, a new revelation for social psychologists (Sherif, 1958), but it does provide greater insight into exactly how and why a focus on superordinate goals is likely to reduce conflict in social interactions. Research suggests that one of the main benefits of perspective taking is to coordinate social goals and thereby create social bonds (Galinsky, Ku, & Wang, 2005). Highlighting the potential for coordination between group members may be sometimes necessary to ensure that perspective taking produces these beneficial outcomes. It is unlikely, however, that the competitive versus the cooperative nature of the group is the only important moderator of reactive egoism, and we believe there are at least two additional moderators that seem particularly promising for future research to pursue. The first promising moderator is the specific procedure involved in perspective taking. As we mentioned earlier, perspective taking does not refer to a specific set of mental operations but encompasses a broad range of procedures and instantiations (for a review see Galinsky et al., 2005). In our experiments, perspective taking was an entirely cognitive enterprise in which participants considered the likely thoughts and actions of other group members. Given that people tend to think they are more fair than others (Messick et al., 1985; Messick & Sentis, 1979) and tend to see others with divergent interests as being more extreme than they 4 It is worth noting that these results are interesting in their own right, as manipulations that may appear conceptually similar to our perspectivetaking manipulation have proven ineffective in reducing egocentric assessments of fairness in past research (Babcock et al., 1995; Lord et al., 1979). Most relevant is an experiment in which participants in a mock court trial were asked to generate the most convincing arguments they possibly could for their opponent’s side (Babcock et al., 1995). Instead of diminishing egocentric assessments of fair settlements in this case between opposing sides, as we found here, this manipulation actually produced a marginally significant increase in egocentric assessments. We suspect the difference between Babcock et al. (1995) and our own studies is that participants in the former study were explicitly asked to generate compelling arguments for the opposing side’s case, whereas participants in our experiments were simply asked to think about what others would think was a fair resolution. Generating arguments is a relatively effortful process, and any difficulty in generating these arguments would likely lead people to conclude that the opposing side’s arguments are not especially compelling (Schwarz, 1998). Simply thinking about what others would think is fair, in contrast, requires relatively little mental effort and readily activated relevant information that people would have otherwise overlooked. The manipulation used in our research is more analogous to what has been called unpacking in past research (Tversky & Koehler, 1994) and is a manipulation that has proven successful in reducing egocentric biases (Savitsky et al., 2005).
REACTIVE EGOISM
actually are (Robinson, Keltner, Ward, & Ross, 1995), this kind of perspective taking may have been particularly likely to activate cynical thoughts about others’ motives in competitive contexts. It is possible that alternate forms of perspective taking may be less likely to activate the cynical thoughts that produce reactive egoism. For instance, Batson & Moran (1999; see also Batson & Ahmad, 2001) have reported that empathizing with the unpleasant circumstances of another person can lead to increased cooperation in a prisoner’s dilemma game. In this experiment, some participants were asked to empathize with a woman who had just described the break up of a romantic relationship, whereas others remained objective while reading this description. Those who empathized with the negative emotions of this woman were more likely to cooperate with her in a subsequent prisoner’s dilemma game than those who did not. This may have occurred because people were simply reluctant to inflict more negative affect by defecting on someone currently experiencing such pain, but it may also have occurred because the particular manner in which one adopts another’s perspective moderates one’s reactions toward them. Indeed, we have reported elsewhere (Caruso, Epley, & Bazerman, 2006) the results of a hypothetical prisoner’s dilemma game in which those who considered others’ thoughts and likely behavior—akin to the perspective-taking manipulation used in the studies reported here—were significantly more likely to defect (60.0%) than participants who did not consider others’ thoughts and likely behavior (27.5%). This occurred because participants who considered others’ thoughts before deciding what to do believed that others were significantly more likely to defect than those who only considered others’ likely behavior after making their own decision. We have found similar results in a simulated trust game (Berg, Dickhaut, & McCabe, 1995), in which participants imagined that they personally had $10 in Round 1 and could give some amount of it to their partner. The amount given would be tripled, and the partner would then have the opportunity to return any amount of the tripled money to the participant. Those who first considered how the other player was likely to behave passed significantly less money to their partner in Round 1 (M ⫽ $3.00) than those who did not (M ⫽ $4.77). To the extent that empathizing with another person’s situation reduces cynical or self-interested thoughts about others, the negative consequences of perspective taking that we document here might be minimized or eliminated. Similarly, asking people to imagine what they would think if they were in another person’s role—akin to Stotland’s (1969) imagine-self condition—might also diminish the cynical thoughts that created reactive egoism in the present experiments (Batson, Early, & Salvarani, 1997). This would occur, however, only if people actually believed that they personally would not behave egoistically if they were placed in another person’s situation. Although people consistently believe they are fairer than others (Messick et al., 1985), it remains an empirical question whether people would imagine being selfless if placed in another role. The second promising moderator is the specific identity of the target of perspective taking. Recall that the key mechanism underlying the consequences of perspective taking is the thoughts that people are led to consider when they look into the mind of another person. Participants in the experiments reported here were asked to consider the perspective of relatively unknown targets. But targets
887
known to be selfish would likely produce even more reactive egoism, and targets known to be selfless would produce even less reactive egoism compared with these unknown targets. Even in the absence of direct knowledge about the selfishness of the target, people often have preexisting beliefs about certain classes of other people that might influence their intuitions about their motives. Because people tend to trust in-groups and distrust out-groups (e.g., Levine & Campbell, 1972; Tajfel, Billig, Bundy, & Flament, 1971), we could predict that people would expect their friends or other in-group members to behave less selfishly than their enemies or other out-group members such that reactive egoism would be more likely to occur when taking the perspective of the latter groups than the former. Finally, we believe this research has interesting implications for the impact of self-interest (or egoism) on social interaction in general and for conflict resolution in particular. For social interaction, these results join a growing body of research investigating the actual versus assumed impact of self-interest on judgment and behavior (Epley & Dunning, 2000; Kruger & Gilovich, 1999; Miller, 1999; Miller & Ratner, 1998; Ratner & Miller, 2001) and suggest that the assumed impact of self-interest on others’ behavior can influence the apparent impact of self-interest on people’s own behavior. In our experiments, perspective taking increased selfish or egoistic behavior because it led participants to expect selfish or egoistic behavior from their other group members. Participants behaved more selfishly, it appears, because of their theories that others would behave selfishly rather than because they were explicitly acting as self-interested agents. These results are consistent with Miller’s (1999) suggestion that self-interest may influence behavior by operating as a descriptive social norm rather than (or perhaps in addition to) operating as a core social motive. It is interesting that research suggests that people tend to overestimate the impact of self-interest on others’ attitudes (Miller & Ratner, 1998) and judgments (Kruger & Gilovich, 1999), meaning that the reactive egoism we observed in our experiments may have been something of an overreaction. Egocentrically attending to one’s own concerns, interests, and perspective in social interaction can create its own set of problems, but undoing that egocentric focus can create quite another. For conflict resolution in particular, these results imply that the intuitive appeal of reducing an egocentric focus on one’s own interests and concerns may produce some potentially negative consequences. It is virtually axiomatic that considering others’ perspectives is desirable in negotiations and conflict situations (Neale & Bazerman, 1983; Paese & Yonker, 2001), but our results place an important caveat on this general sentiment. Sometimes considering others’ perspectives can increase the very egoistic behavior that perspective taking was designed to reduce. Care should be taken when suggesting that people should look beyond their own perspective, as those who look into the minds of others may not like what they see.
References Allison, S. T., Messick, D. M., & Goethals, G. R. (1989). On being better but not smarter than others: The Muhammad Ali effect. Social Cognition, 7, 275–295. Ames, D. (2004). Strategies for social inference: A similarity contingency model of projection and stereotyping in attribute prevalence estimates. Journal of Personality and Social Psychology, 87, 340 –353.
888
EPLEY, CARUSO, AND BAZERMAN
Babcock, L., & Loewenstein, G. (1997). Explaining bargaining impasse: The role of self-serving biases. Journal of Economic Perspectives, 11, 109 –126. Babcock, L., Loewenstein, G., Issacharoff, S., & Camerer, C. (1995). Biased judgments of fairness in bargaining. American Economic Review, 85, 1337–1343. Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173–1182. Batson, C. D. (1994). Prosocial motivation: Why do we help others? In A. Tesser (Ed.), Advanced social psychology (pp. 333–381). Boston: McGraw-Hill. Batson, C. D., & Ahmad, N. (2001). Empathy-induced altruism in a prisoner’s dilemma II: What if the target of empathy has defected? European Journal of Social Psychology, 31, 25–36. Batson, C. D., Early, S., & Salvarani, G. (1997). Perspective taking: Imagining how another feels versus imagining how you would feel. Personality and Social Psychology Bulletin, 7, 751–758. Batson, C. D., Kobrynowicz, D., Dinnerstein, J. L., Kampf, H. C., & Wilson, A. D. (1997). In a very different voice: Unmasking moral hypocrisy. Journal of Personality and Social Psychology, 72, 1335– 1348. Batson, C. D., & Moran, T. (1999). Empathy-induced altruism in a prisoner’s dilemma. European Journal of Social Psychology, 29, 909 –924. Batson, C. D., Thompson, E. R., & Chen, H. (2002). Moral hypocrisy: Addressing some alternatives. Journal of Personality and Social Psychology, 83, 330 –339. Bazerman, M. H., & Neale, M. A. (1982). Improving negotiation effectiveness under final offer arbitration: The role of selection and training. Journal of Applied Psychology, 67, 543–548. Berg, J., Dickhaut, J., & McCabe, K. (1995). Trust, reciprocity, and social history. Games and Economic Behavior, 10, 122–142. Caruso, E. M., Epley, N., & Bazerman, M. H. (2006). Reactive egoism and unwarranted distrust. Unpublished manuscript, Harvard University. Chartrand, T. L., & Bargh, J. A. (1999). The chameleon effect: The perception– behavior link and social interaction. Journal of Personality and Social Psychology, 76, 893–910. Davis, M. H. (1983). Measuring individual differences in empathy: Evidence for a multidimensional approach. Journal of Personality and Social Psychology, 44, 113–126. Dawes, R. M., McTavish, J., & Shaklee, H. (1977). Behavior, communication, and assumptions about other people’s behavior in a commons dilemma situation. Journal of Personality and Social Psychology, 35, 1–11. Dawson, E., Gilovich, T., Regan, D. T. (2002). Motivated reasoning and performance on the Wason selection task. Personality and Social Psychology Bulletin, 28, 1379 –1387. Ditto, P. H., & Lopez, D. F. (1992). Motivated skepticism: The use of differential decision criteria for preferred and nonpreferred conclusions. Journal of Personality and Social Psychology, 63, 568 –584. Ditto, P. H., Scepansky, J. A., Munro, G. D., Apanovitch, A. M., & Lockhart, L. K. (1998). Motivated sensitivity to preference-inconsistent information. Journal of Personality and Social Psychology, 75, 53– 69. Epley, N., & Dunning, D. (2000). Feeling “holier than thou”: Are selfserving assessments produced by errors in self or social psychology? Journal of Personality and Social Psychology, 79, 861– 875. Epley, N., & Gilovich, T. (2004). Are adjustments insufficient? Personality and Social Psychology Bulletin, 30, 447– 460. Epley, N., Keysar, B., Van Boven, L., & Gilovich, T. (2004). Perspective taking as egocentric anchoring and adjustment. Journal of Personality and Social Psychology, 87, 327–339. Epley, N., Savitsky, K., & Gilovich, T. (2002). Empathy neglect: Recon-
ciling the spotlight effect and the correspondence bias. Journal of Personality and Social Psychology, 83, 300 –312. Esser, J. K., & Komorita, S. S. (1975). Reciprocity and concession making in bargaining. Journal of Personality and Social Psychology, 31, 864 – 872. Galinsky, A. D., Ku, G., & Wang, C. S. (2005). Perspective-taking and self– other overlap: Fostering social bonds and facilitating social coordination. Group Processes & Intergroup Relations, 8, 109 –124. Galinsky, A. D., & Moskowitz, G. B. (2000). Perspective taking: Decreasing stereotype expression, stereotype accessibility, and in-group favoritism. Journal of Personality and Social Psychology, 78, 708 –724. Gilbert, D. T. (2002). Inferential correction. In T. Gilovich, D. Griffin, & D. Kahneman (Eds.), Heuristics and biases: The psychology of intuitive judgment (pp. 167–184). Cambridge, England: Cambridge University Press. Gilovich, T., Medvec, V. H., & Savitsky, K. (2000). The spotlight effect in social judgment: An egocentric bias in estimates of the salience of one’s own actions and appearance. Journal of Personality and Social Psychology, 78, 211–222. Gilovich, T., Savitsky, K., & Medvec, V. H. (1998). The illusion of transparency: Biased assessments of others’ ability to read our emotional states. Journal of Personality and Social Psychology, 75, 332–346. Hastorf, A. H., & Cantril, H. (1954). They saw a game: A case study. Journal of Abnormal and Social Psychology, 49, 129 –134. Kelley, H. H., & Stahelski, A. J. (1970). Social interaction basis of cooperators’ and competitors’ beliefs about others. Journal of Personality and Social Psychology, 16, 66 –91. Kenny, D. A., & DePaulo, B. M. (1993). Do people know how others view them? An empirical and theoretical account. Psychological Bulletin, 114, 145–161. Keysar, B., & Bly, B. (1995). Intuitions of the transparency of idioms: Can one keep a secret by spilling the beans? Journal of Memory and Language, 34, 89 –109. Keysar, B., Ginzel, L. E., & Bazerman, M. H. (1995). States of affairs and states of mind: The effect of knowledge of beliefs. Organizational Behavior and Human Decision Processes, 64, 283–293. Kramer, R. M. (1994). The sinister attribution error: Paranoid cognition and collective distrust in organizations. Motivation and Emotion, 18, 199 –230. Krueger, J., & Clement, R. (1994). The truly false consensus effect: An ineradicable and egocentric bias in social perception. Journal of Personality and Social Psychology, 65, 596 – 610. Kruger, J., & Gilovich, T. (1999). “Naive cynicism” in everyday theories of responsibility assessment: On biased assumptions of bias. Journal of Personality and Social Psychology, 76, 743–753. Leary, M. R., & Forsyth, D. R. (1987). Attributions of responsibility for collective endeavors. In C. Hendrick (Ed.), Review of personality and social psychology: Group processes. (Vol. 8, pp. 167–188). Newbury Park, CA: Sage. Levine, R. A., & Campbell, D. T. (1972). Ethnocentrism: Theories of conflict, ethnic attitudes, and group behaviour. New York: Wiley. Liberman, V., Samuels, S. M., & Ross, L. (2004). The name of the game: Predictive power of reputations versus situational labels in determining prisoner’s dilemma game moves. Personality and Social Psychology Bulletin, 9, 1175–1185. Loewenstein, G., Issacharoff, S., Camerer, C., & Babcock, L. (1993). Self-serving assessments of fairness and pretrial bargaining. Journal of Legal Studies, 22, 135–159. Lord, C. G., Ross, L., & Lepper, M. R. (1979). Biased assimilation and attitude polarization: The effects of prior theories on subsequently considered evidence. Journal of Personality and Social Psychology, 37, 2098 –2109. Mead, G. H. (1934). Mind, self, and society. Chicago: University of Chicago Press.
REACTIVE EGOISM Messe´, L. A., & Sivacek, J. M. (1979). Predictions of others’ responses in a mixed-motive game: Self-justification or false consensus? Journal of Personality and Social Psychology, 37, 602– 607. Messick, D. M. (1995). Equality, fairness, and social conflict. Social Justice Research, 8, 153–173. Messick, D. M., Bloom, S., Boldizar, J. P., & Samuelson, C. D. (1985). Why we are fairer than others. Journal of Experimental Social Psychology, 21, 480 –500. Messick, D. M., & Sentis, K. P. (1979). Fairness and preference. Journal of Experimental Social Psychology, 15, 418 – 434. Messick, D. M., & Sentis, K. P. (1983). Fairness, preference, and fairness biases. In D. M. Messick & K. S. Cook (Eds.), Equity theory: Psychological and sociological perspectives (pp. 61– 64). New York: Praeger. Miller, D. T. (1999). The norm of self-interest. American Psychologist, 54, 1053–1060. Miller, D. T., & Ratner, R. K. (1998). The disparity between the actual and assumed power of self-interest. Journal of Personality and Social Psychology, 74, 53– 62. Neale, M. A., & Bazerman, M. H. (1983). The effects of perspective taking ability under alternate forms of arbitration on the negotiation process. Industrial and Labor Relations Review, 36, 378 –388. Nickerson, R. S. (1999). How we know—and sometimes misjudge—what others know: Imputing one’s own knowledge to others. Psychological Bulletin, 125, 737–759. Paese, P. W., & Yonker, R. D. (2001). Toward a better understanding of egocentric fairness judgments in negotiation. The International Journal of Conflict Management, 12, 114 –131. Piaget, J. (1965). The moral judgment of the child (M. Gabain, Trans.). New York: Free Press. (Original work published 1932) Pronin, E., Puccio, C., & Ross, L. (2002). Understanding misunderstanding: Social psychological perspectives. In T. Gilovich, D. W. Griffin, & D. Kahneman (Eds.), Heuristics and biases: The psychology of intuitive judgment (pp. 636 – 665). Cambridge, England: Cambridge University Press. Ratner, R. K., & Miller, D. T. (2001). The norm of self-interest and its effects on social action. Journal of Personality and Social Psychology, 81, 5–16. Regan, D. R., & Fazio, R. (1977). On the consistency between attitudes and behavior: Look to the method of attitude formation. Journal of Experimental Social Psychology, 13, 28 – 45. Robinson, R., Keltner, D., Ward, A., & Ross, L. (1995). Actual versus assumed differences in construal: “Naı¨ve realism” in intergroup perceptions and conflict. Journal of Personality and Social Psychology, 68, 404 – 417. Ross, L., Greene, D., & House, P. (1977). The false consensus effect: An egocentric bias in social perception and attribution processes. Journal of Experimental Social Psychology, 13, 279 –301.
889
Ross, L., & Ward, A. (1995). Psychological barriers to dispute resolution. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 27, pp. 255–304). San Diego, CA: Academic Press. Ross, L., & Ward, A. (1996). Naive realism in everyday life: Implications for social conflict and misunderstanding. In T. Brown, E. Reed, & E. Turiel (Eds.), Values and knowledge (pp. 103–135). Hillsdale, NJ: Erlbaum. Ross, M., & Sicoly, F. (1979). Egocentric biases in availability and attribution. Journal of Personality and Social Psychology, 37, 322–336. Savitsky, K., Epley, N., & Gilovich, T. (2001). Do others judge us as harshly as we think? Overestimating the impact of our failures, shortcomings, and mishaps. Journal of Personality and Social Psychology, 81, 44 –56. Savitsky, K., Van Boven, L., Epley, N., & Wight, W. (2005). The unpacking effect in responsibility allocations for group tasks. Journal of Experimental Social Psychology, 41, 447– 457. Schwarz, N. (1998). Accessible content and accessibility experiences: The interplay of declarative and experiential information in judgment. Personality and Social Psychology Review, 2, 87–99. Sherif, M. (1958). Superordinate goals in the reduction of intergroup conflict. American Journal of Sociology, 63, 349 –356. Stotland, E. (1969). Exploratory studies in empathy. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 4, pp. 271–313). San Diego, CA: Academic Press. Tajfel, H., Billig, M. G., Bundy, R. F., & Flament, C. (1971). Social categorization and intergroup behaviour. European Journal of Psychology, 1, 149 –177. Thompson, L., & Loewenstein, G. (1992). Egocentric interpretations of fairness and interpersonal conflict. Organizational Behavior and Human Decision Processes, 51, 176 –197. Thompson, S. C., & Kelly, H. H. (1981). Judgments of responsibility for activities in close relationships. Journal of Personality and Social Psychology, 41, 469 – 477. Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185, 1124 –1130. Tversky, A., & Koehler, D. J. (1994). Support theory: A nonextensional representation of subjective probability. Psychological Review, 101, 547–567. Wade-Benzoni, K. A., Tenbrunsel, A. E., & Bazerman, M. H. (1996). Egocentric interpretations of fairness in asymmetric, environmental social dilemmas: Explaining harvesting behavior and the role of communication. Organizational Behavior and Human Decision Processes, 67, 111–126.
Received August 8, 2005 Revision received February 15, 2006 Accepted February 26, 2006 䡲
Journal of Personality and Social Psychology 2006, Vol. 91, No. 5, 890 –903
Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.5.890
Cultural Affordances and Emotional Experience: Socially Engaging and Disengaging Emotions in Japan and the United States Shinobu Kitayama
Batja Mesquita
University of Michigan
Wake Forest University
Mayumi Karasawa Tokyo Woman’s Christian University The authors hypothesized that whereas Japanese culture encourages socially engaging emotions (e.g., friendly feelings and guilt), North American culture fosters socially disengaging emotions (e.g., pride and anger). In two cross-cultural studies, the authors measured engaging and disengaging emotions repeatedly over different social situations and found support for this hypothesis. As predicted, Japanese showed a pervasive tendency to reportedly experience engaging emotions more strongly than they experienced disengaging emotions, but Americans showed a reversed tendency. Moreover, as also predicted, Japanese subjective well-being (i.e., the experience of general positive feelings) was more closely associated with the experience of engaging positive emotions than with that of disengaging emotions. Americans tended to show the reversed pattern. The established cultural differences in the patterns of emotion suggest the consistent and systematic cultural shaping of emotion over time. Keywords: culture, emotion, self
appraising the situation as one in which his or her own personal goals are being unfairly blocked by someone. Similarly, to experience guilt means that the person is construing the situation as one in which he or she has failed to repay certain obligations to someone else or has failed to perform an expected duty to the person. What themes are likely to be highlighted in a given situation depends in part on the likelihood of the themes to be activated in memory and made available in the construction of emotional experience (Frijda, Mesquita, Sonnemans, & Van Goozen, 1991; Lazarus, 1991, 1999; Mesquita, 2003; Schachter & Singer, 1962). For example, a certain aversive situation may make someone feel different emotions, such as anger and guilt, depending on the theme that is made accessible in the situation. If the theme of unfair goal blockage comes to mind, then anger will most likely be the dominant emotion, but if the theme of failed repayment is more accessible, then the person will instead be more likely to experience guilt. It is even possible that both themes are simultaneously available, producing certain “mixed” or “ambivalent” feelings. Systematic cross-cultural comparison can provide a potent means by which to test the foregoing analysis on emotion. Culture comprises a large number of symbolic resources, such as lay theories, schemas, images, and icons, that are distributed in a given group of people (Chiu & Hong, 2005; Kitayama, Duffy, & Uchida, in press; Sperber, 1996). Culture, therefore, may be expected to constantly supply meanings and thus modify the accessibility of different emotion themes. That is, meanings and practices of different cultural contexts may encourage certain themes over others and, as a consequence, may give rise to systematic cultural variation in emotional experience. Drawing on this reasoning, the present work seeks to examine Japanese and American self-reports of emotions across different social situations.
Virtually all emotions are intensely meaningful (Lutz, 1988) in the sense that each of the emotions captures global thematic features of the attendant situation. These features have been referred to as core relational themes (Lazarus, 1991, 1999) or simply as appraisals (e.g., Ellsworth, 1994; Frijda, 1986; Mesquita & Ellsworth, 2001; Mesquita & Leu, in press). By means of these themes, emotions simplify a complex social situation and reconstitute it in a single brush that is intrinsically meaningful to the person who experiences them. For example, emotions such as pride, friendly feelings, anger, and guilt all reflect meaningful themes that go beyond mere positive or negative evaluations. These themes describe the way individuals perceive their relationship to the surrounding environment (Ellsworth, 1994; Kitayama, Karasawa, & Mesquita, 2004; Mesquita & Ellsworth, 2001). For example, although pride and friendly feelings are both positive, they are associated with very different themes that may respectively be characterized as personal achievement and social harmony. Likewise, anger and guilt are similar in their unpleasantness, but they are linked to profoundly different themes that can be described as unfair goal interference and failure of repayment. Thus, to experience anger implies, among others, that the person is
Shinobu Kitayama, Department of Psychology, University of Michigan; Batja Mesquita, Department of Psychology, Wake Forest University, Mayumi Karasawa, Department of Communication, Tokyo Woman’s Christian University. We thank Nick Bowman, Keiko Ishii, Jinkyung Na, Yu Niiya, Yukiko Uchida, and Ayse Uskul for helpful comments on an earlier version of this article. We also thank Rich Gonzalez for his advice on statistics. Correspondence concerning this article should be addressed to Shinobu Kitayama, Department of Psychology, University of Michigan, 503 Church Street, Ann Arbor, MI 48109-1043. E-mail:
[email protected] 890
CULTURE AND EMOTION
Social Engagement and Disengagement Independence and interdependence constitute two major sets of social and cultural tasks and associated ideas that are ubiquitous in all societies and cultures. These tasks and ideas can supply many themes for emotions (Kitayama, Markus, & Kurokawa, 2000; Kitayama, Markus, & Matsumoto, 1995; Markus & Kitayama, 1991b). Specifically, some themes, such as social harmony and failed repayment, are premised on the presence of an interdependent self that seeks a harmonious relationship of some kind. When relational harmony is realized, the theme of social harmony may arise. In contrast, when harmony has been temporarily disrupted and a person seeks to restore it, the theme of failed repayment may be highlighted. Themes that are derived from social interdependence and relationally embedded nature of self are referred to as socially engaging. Emotions are considered socially engaging when they are about these engaging themes. All engaging emotions, especially those that are positive, may thus be expected to be both deriving from and affirming the interdependence of self. In contrast, some other themes, such as personal achievement and unfair infringement of personal goals and desires, are premised on the presence of an independent, autonomous self that seeks to pursue such goals and desires. Whereas the theme of personal achievement may arise from success on an important task, the theme of unfair goal interference may become salient when the person faces certain obstacles that are intentionally held against his or her goal pursuit. Themes that are grounded in independence and autonomy of self and its separateness from others in a relationship may be said to be socially disengaging. All disengaging emotions, especially those that are positive, may be expected to be both deriving from and affirming the independence of self. Evidence suggests that individuals spontaneously recognize the engaging and disengaging orientation of different emotions. Kitayama, Markus, and Negishi (1989, as cited in Markus & Kitayama, 1991a) asked Japanese participants to rate many pairs of emotions for their perceived similarities. Likewise, Kitayama et al., (2000) asked both Japanese and American participants to rate how frequently they experienced a number of different emotions. A matrix of correlations among the emotions was used as a measure of psychological similarities among them. In both cases, a multidimensional scaling analysis performed on the similarity data yielded a dimension of social orientation in addition to the dimension of pleasantness. Thus, a set of emotions, both positive (e.g., friendly feelings and respect) and negative (e.g., guilt and shame) constituted the engagement end of the social orientation dimension, with the opposite, disengagement end defined by a different set of both positive emotions (e.g., pride and feelings of superiority) and negative emotions (e.g., anger and frustration).1 These findings suggest that both Japanese and Americans spontaneously categorize emotions in terms of both pleasantness and social orientation. This evidence sets the stage for a next step of inquiry, namely, to examine the possibility that cultural variation in emotional experience may be accounted for by these two dimensions of emotions.2
Cultural Affordances and Emotional Experience Cultural variation in emotional experience may be anticipated on the basis of the idea that although tasks, concerns, and goals
891
related to independence and those pertaining to interdependence are all potentially available in all cultures, the distribution of these symbolic resources is uneven. Markus and Kitayama (1991b) have suggested that in North America, ideas related to independence and disengagement of the self (e.g., personal achievement, goal pursuit, free choice, and personal rights) are highlighted and elaborated by means of corresponding daily tasks, routines, and lay theories. North American culture may then be described as independent. In contrast, in Asia, including Japanese culture, ideas of interdependence and engagement (e.g., social harmony, duty to groups, adjustment and fitting in, and sympathy) are highlighted and salient. Asian cultures may therefore be said to be interdependent. Of course, not all people in any given cultural context explicitly approve the themes or values that are salient and sanctioned in that context. Nor do they always succeed in attaining such values. For example, many Americans explicitly oppose the hegemonic value of independence (Oyserman, Coon, & Kemmelmeier, 2002). Likewise, people may not always act on any given cultural value and belief. Nevertheless, depending on the pervasiveness of tasks and concerns that highlight either independence and disengagement or interdependence and engagement, cultural environments can carry the potential of evoking very different sets of emotions and other psychological responses. This potential has been called cultural affordances (Kitayama & Markus, 1999). The notion of cultural affordance entails the following two testable predictions.
Intensity of Experiencing Engaging and Disengaging Emotions The first prediction concerns a consequence of the cultural affordances for engaging and disengaging emotional themes. Cultural affordances result from a biased pool of symbolic resources of culture that are brought to bear on the construction of concrete daily situations (Kitayama, Markus, Matsumoto, & Norasakkunkit, 1997; Morling, Kitayama, & Miyamoto, 2002). These symbolic resources are used to define general classes of events and episodes that are available in a given cultural context (e.g., school, taking an exam, business, success, failure, and the like). But more importantly, they profoundly influence more subtle, yet powerful nuances and psychological meanings (e.g., pride, shame, obligation, honor, and the like) that are added to the lived experience of such events and episodes. These meanings may often be highly idio1
These findings also suggest that the most dominant meaning of pride is “pride in the self” in both languages. Pride can, of course, be produced by achievement of ingroup members. But this more interpersonal or social pride must be linguistically marked at least in both English and Japanese (e.g., pride in my friend). 2 Emotion words have diverse meanings, and, moreover, no single word is likely to have the exact translation equivalent in another language (Wierzbicka, 1994). Nevertheless, our previous studies used multidimensional scaling analyses to show that underlying core appraisals, such as pleasantness and social orientation, are still common across cultures. This fact has enabled us to draw systematic cross-cultural comparisons. Note, however, that these core appraisals alone can by no means cover the entire range of meanings different emotion words and emotion concepts can carry. Hence, our analysis implies no one-to-one correspondence between the core appraisals and emotion words or concepts.
892
KITAYAMA, MESQUITA, AND KARASAWA
syncratic and hardly predictable in any specific instances, and, yet, because they are derived or fostered by a pool of symbolic resources that are available in a given culture, they may be systematically biased over many situations and episodes in accordance with the specific, historically crafted and accumulated contents of this pool. In this way, individual experiences may be collectively constructed through sociohistorical processes (Kitayama & Markus, 1999; Kitayama et al., 1997). For example, “getting an A in an important course” is likely to be both positive and disengaging (causing one to feel proud) in all cultures. Nevertheless, certain engaging themes, icons, and ideas, such as the smiling faces of parents who would be very happy to know about their daughter’s performance in the course, may be more available in interdependent cultural contexts than in independent cultural contexts. If so, then a subtle yet distinctly engaging element or nuance may be added to the experience for those in interdependent contexts, but such addition of engaging nuances may be relatively unlikely for those in independent contexts. This example illustrates how subtle the effect of cultural affordances can sometimes be. It also suggests, however, that such subtle effect can be very powerful over time because it is highly recurrent and present all the time, insofar as all individuals must necessarily be drawing on the pool of symbolic resources of their own culture. Accordingly, we expected that, over time and across many different situations, Asian, interdependent cultures would make socially engaging themes salient and thus would facilitate the corresponding engaging emotions such as friendly feelings and guilt. Similarly, we expected that American, independent culture would, over time and across different situations, make salient socially disengaging themes and would thus foster the corresponding disengaging emotions such as pride and anger.3 Recently, Kitayama and colleagues (2000) investigated emotional experience as a function of the social orientation of emotions. The researchers asked both Japanese and American participants to report how frequently they experienced both positive and negative emotions that were either engaging or disengaging. As predicted, Japanese reported that they experienced engaging emotions, both positive and negative, more frequently than disengaging emotions, both positive and negative. In contrast, a pattern shown by Americans belied the prediction. Overall, American participants reported that they experienced positive engaging emotions such as friendly feelings and close feelings much more frequently than the remaining three categories of emotions. It is possible that the reported frequency of emotional experience was skewed in a positive, desirable direction by memory distortion and reporting bias. Oishi (2002) showed that such a memory distortion is quite pervasive when Americans make a frequency judgment of emotional experience. Specifically, positive engaging emotions may be especially desirable emotions to experience, and, as a consequence, Americans may have sought to see themselves in a positive light by selectively remembering the experience of these emotions and/or by simply overreporting it. Past research indicates that Japanese do not show any strong tendency toward positive self-images (Heine, Lehman, Markus, & Kitayama, 1999; Kitayama & Markus, 1999). The Kitayama et al. (2000) evidence implies that in order to carry out a valid test of the present prediction, it is important to use a much more episodic task that minimizes any memory bias or reporting bias. Moreover, the cultural affordances for both engage-
ment for Japanese and disengagement for Americans are expected to be pervasive in the respective cultural contexts primarily because they are highly consistent and, thus, cumulative over numerous situations that recur in daily life (Abelson, 1985). We therefore anticipated that the hypothesized effect of cultural affordances would be best captured when emotional experience was examined over a number of different situations. With the foregoing considerations in mind, in Study 1, we had participants remember a very concrete episode on each of many days in which they felt strong emotions and to report how intensely they experienced different emotions in that episode. Likewise, in Study 2, participants were to remember concrete episodes that were fitted to many different situation types and then to report the emotions they experienced in these episodes.
Predictors of Well-Being Our second prediction is derived from the idea that emotions that vary with regard to their social orientation affirm different forms of self. Whereas positive engaging emotions affirm the interdependent self, positive disengaging emotions affirm the independent self. Moreover, past research has shown that Asians are chronically motivated toward social harmony and other related goals of interdependence, but Americans are chronically motivated toward personal control and other related goals of independence (Kitayama, Karasawa, Curhan, Ryff, & Markus, 2006; Markus & Kitayama, 1991b). We therefore expected that socially engaging positive emotions, such as friendly feelings and feelings of respect, would best predict subjective well-being in Japanese groups. In contrast, among Americans, we predicted that socially disengaging positive feelings, such as pride and self-esteem, would best predict subjective well-being. Past evidence is consistent with the present predictions. For example, in the aforementioned study by Kitayama and colleagues (2000), for Japanese the reported frequency of experiencing happiness was better predicted by the reported frequency of experiencing positive engaging emotions (e.g., friendly feelings) than by the reported frequency of experiencing positive disengaging emotions (e.g., pride), but this pattern was reversed for Americans. Some other studies have shown analogous patterns by using personality measures of engagement and disengagement. For example, Kwan, Bond, and Singelis (1997) have shown that among Americans, subjective well-being is predicted by self-esteem (disengagement) more strongly than by relationship harmony (engagement), but among Hong Kong Chinese, subjective well-being was equally predicted by both. Both Kang, Shaver, Min, and Jin (2003) and Uchida, Kitayama, Mesquita, Reyes, and Morling (2004) have reported consistent findings with different measures of social engagement. Furthermore, Kitayama and colleagues (2006) have used a much wider range of measures of well-being and observed the same pattern with adult, noncollege student samples. Overall, 3 This argument amounts to the hypothesis that culture “primes” different emotion themes. Unlike social cognition researchers who focus on effects of specific priming stimuli such as I versus we (Brewer & Gardner, 1996) or cultural icons such as Great Walls or Marilyn Monroe (Hong, Morris, Chiu, & Benet-Martinez, 2000), our emphasis is on the entire pool of icons, lay theories, and other symbolic resources that are unevenly distributed across different cultural contexts.
CULTURE AND EMOTION
893
Table 1 Emotions in Each of the Four Examined Emotion Types and reliabilities in Studies 1 and 2 Reliability Study 1 Emotion type Positive engaging emotions
Positive disengaging emotions
Negative engaging emotions
Negative disengaging emotions
Emotion Friendly feelings Close feelingsa Respect Sympathy Proud Superior Top of the worlda Self-esteemb Guilt Indebted Ashamed Afraid of causing trouble on anotherb Sulky feelings Frustration Angry
Study 2
JPN
U.S.
JPN
U.S.
.83
.55
.86
.67
.78
.78
.72
.73
.80
.71
.82
.67
.83
.83
.79
.76
Note. JPN ⫽ Japan; U.S. ⫽ United States. a Not included in Study 1. b Not included in Study 2.
the studies summarized here show that both engagement and disengagement are associated with subjective well-being and happiness, but the relative significance of the two factors vary across cultures, with disengagement more important for Americans and engagement more important for Asians (see also Mesquita et al., 2005, for relevant evidence). One significant shortcoming of the present literature stems from the fact that all of the existent studies examine correlations across individuals, showing culturally divergent association patterns between individual propensities toward either engagement or disengagement and individual propensities toward happiness and wellbeing. It is yet to be determined whether analogous patterns of association can be observed within each individual. In order to examine whether individuals would in fact feel happier when they experience either engagement or disengagement, we examined associations between engagement/disengagement and happiness within each individual participant.
To maximize ecological and cultural validity, we sought to test our predictions with naturally occurring emotional events in Study 1. We used a diary method to test both Japanese and Americans in their native cultural contexts.
(not experienced it at all) to 5 (experienced it very strongly). The questionnaire was printed on the backside of a prestamped postcard in Japan. It was printed on a sheet of paper, which was to be enclosed in a prestamped envelope in the United States. In both cases, the participants were asked to send the questionnaire to the psychology office every day. Only those data that were postmarked on the same or the next day were analyzed. Three Japanese male participants and 2 American male participants who failed to return the questionnaire on more than 7 days were excluded from the analysis. Materials. All emotion terms were selected, and their translation equivalents developed, through a series of discussions among Japanese– English bilinguals (see Kitayama et al., 2000, for details). Drawing on our previous work (Kitayama et al., 2000, 1995), the list of emotions contained four theoretically derived types of emotion terms, which were defined by their position on the pleasantness (positive and negative) and the social orientation (engaging and disengaging) dimensions.4 Whereas engaging emotions both result from and foster social engagement of the self, disengaging emotions both result from and foster social disengagement of the self. The emotion terms used are listed in Table 1. The scales for the four types of emotions had reasonable reliabilities in both studies (see Table 1 for Cronbach’s alphas). In addition to these four emotion scales, defined by pleasantness and social orientation, we included several emotion terms to indicate well-being or general positive emotions (happy, elation, relaxation, and calmness) and negative well-being or general negative emotions (unhappy, sadness, fear, depression, boredom, and disgust), respectively. Common among the terms for general positive and general negative emotions was that they did not specify any particular social orientation.
Method
Results and Discussion
Participants and procedure. Thirty-eight Japanese college students at a Japanese university (20 men and 17 women, and 1 whose gender is unknown) and 49 American college students at a U.S. university (29 men and 20 women) participated in the study in exchange for course credits. Participants were asked to remember “the most emotional episode of the day” at the end of each of 14 consecutive days. They were asked to briefly describe the episode and then to report how strongly they experienced each of 27 emotions. The ratings were made on a 6-point scale ranging from 0
Emotional experience. We anticipated that whereas engaging emotions would be more strongly experienced by Japanese than by Americans, disengaging emotions would be more strongly expe-
Study 1: Reported Intensity of Experiencing Emotions in Naturally Occurring Situations
4 Four additional emotions were included (excited, sleepy, feeling like babied [amae in Japanese], and afraid of causing trouble to someone) in Study 1.
KITAYAMA, MESQUITA, AND KARASAWA
894
rienced by Americans than by Japanese. This prediction implies a significant interaction between culture and the social orientation of emotion. In testing this prediction, however, it is important to distinguish positive from negative emotional situations because our predictions on cultural differences in social orientation would be expected to hold primarily for the focal themes in the attendant situations (Mesquita & Ellsworth, 2001; Mesquita & Frijda, 1992)—namely, the themes that constitute the most dominant meaning of the situation. Thus, we expected our prediction of cultural differences in social orientation to emerge for positive emotions in the positive situations and negative emotions in the negative situations. We did not have any predictions with regard to the emotions that were inconsistent in pleasantness with the situation’s main theme, except that they would be of relatively low intensity. We divided all the reported situations into two sets, containing positive and negative events, respectively. We first computed the mean intensity ratings for general positive emotions and general negative emotions. The scales for general positive and general negative emotions had reasonable reliability (positive emotions: ␣s ⫽ .92 and .88, for Japanese and Americans, respectively; negative emotions: ␣s ⫽ .80 and .78, for Japanese and Americans, respectively). We then classified as positive the situations for which the average of the general positive emotions was greater than that for general negative emotions and as negative the situations for which the reverse was the case. All respondents reported both negative and positive emotional events. Mean intensity ratings of the four types of emotions that varied in pleasantness and social orientation were computed for each Japanese
participant and submitted to an analysis of variance (ANOVA), with two between-subjects variables (culture and gender) and three within-subjects variables (emotion pleasantness, emotion social orientation, and situation pleasantness). Four participants with missing cells (3 Americans and 1 Japanese) were excluded, as was another Japanese participant whose gender information was missing. So the analysis was performed on 33 Japanese and 44 Americans. The pertinent means are shown in Figure 1. As expected, we observed a significant interaction between culture and social orientation only for the emotions that were matched in pleasantness to the attendant situations. This observation is underscored by a significant interaction involving culture, social orientation of the emotion, emotion pleasantness, and situation pleasantness, F(1, 73) ⫽ 19.70, p ⬍ .0001. The results of separate ANOVAs performed on the four conditions that were marked by both emotion pleasantness and situation pleasantness are summarized in Table 2. When the emotions and the situations were matched in pleasantness, the interaction between culture and social orientation of the emotions was highly significant. When both the emotions and the situations were positive, Japanese reportedly experienced the engaging emotions (e.g., friendly feelings) more strongly than the disengaging emotions (e.g., pride), t(73) ⫽ 4.41, p ⬍ .001. In contrast, Americans showed a reliable reversal, experiencing the disengaging emotions (e.g., pride) reportedly more strongly than the engaging emotions (e.g., friendly feelings), t(73) ⫽ 2.60, p ⬍ .01. Comparing between cultures, Americans reported a greater intensity of experiencing disengaging emotions than did Japanese, Am ericans
4.2
Negative Situations
Positive Situations
4 3.8 3.6 3.4 Intensity ofexperience
3.2 3 2.8 2.6 2.4 2.2 2 1.8 1.6 1.4 1.2 1 Engaging
Disengaging Positive
Engaging
Disengaging
Engaging
N egative
Disengaging P ositive
Engaging
Disengaging
N egative
Type ofEm otions
Figure 1. Reported intensity of experiencing positive and negative emotions that are either engaging or disengaging in positive and negative situations in Study 1. The bar attached to each mean indicates the standard error of that mean.
CULTURE AND EMOTION
895
Table 2 F Tests From Analyses of Variance Performed on Four Conditions That Varied in the Pleasantness of Emotions and the Pleasantness of the Situations (in Study 1) Positive situation
Negative situation
Variable
Positive emotion
Negative emotion
Positive emotion
Negative emotion
Culture Gender Culture ⫻ Gender Social orientation (SO) SO ⫻ Culture SO ⫻ Gender SO ⫻ Culture ⫻ Gender
14.31*** 1.23 ⬍1 ⬍1 20.69*** ⬍1 1.55
1.29 ⬍1 ⬍1 6.17* 8.31** 2.15 2.27
11.25** ⬍1 1.11 ⬍1 ⬍1 ⬍1 ⬍1
2.78 1.89 ⬍1 56.32*** 21.70*** 7.74* 2.82
* p ⬍ .05.
** p ⬍ .001.
*** p ⬍ .0001.
t(73) ⫽ 5.58, p ⬍ .01, but this difference disappeared for engaging emotions, t(73) ⫽ 1.45, ns. When both the emotions and the situations were unpleasant, the disengaging emotions (e.g., anger) were reportedly experienced more strongly in both cultural groups than the engaging emotions (e.g., guilt), but, as expected, this tendency was reliably more pronounced for Americans than for Japanese. The disengaging emotions (e.g., anger) were reportedly more strongly experienced by Americans than by Japanese, t(73) ⫽ 3.78, p ⬍ .01, but this cultural difference was reversed, albeit nonsignificantly, for the engaging emotions (e.g., guilt), t(73) ⫽ 1.37, ns. When emotions and situations were not matched in pleasantness, the intensity of emotions was low; moreover, as predicted, the Culture ⫻ Social Orientation interaction was much weaker. This interaction was statistically significant for negative emotions in positive situations. Disengaging positive emotions were reportedly experienced more strongly by Americans than by Japanese, t(73) ⫽ 2.16, p ⬍ .05. No such difference was present for engaging emotions. As can be seen in Figure 1, however, the interaction pattern was very weak. For positive emotions in negative situations, the Culture ⫻ Social Orientation interaction was trivial. In addition, for negative emotions in negative situations, the Gender ⫻ Social Orientation interaction was significant. Although disengaging emotions were more strongly experienced than engaging emotions, this effect was more pronounced for women (Ms ⫽ 3.56 vs. 2.30) than for men (Ms ⫽ 3.61 vs. 2.56). If anything, the pattern goes against the common assumption that women are more interdependent than men (Cross & Madson, 1997). Note that in both cultures, negative disengaging emotions such as anger and frustration were reportedly experienced much more strongly than negative engaging emotions such as guilt and shame. We suspect that the disengaging negative emotions are more arousing than their engaging counterparts, and, as a consequence, the intensity of the former was experienced to be greater than that of the latter (Sonnemans & Frijda, 1994). Predictors of well-being. In order to determine when individuals would feel happy or unhappy, we carried out an analysis in two steps. First, for each participant, the ratings of general positive (or negative) emotions over all the 14 episodes were regressed on the corresponding ratings of engaging positive (or negative) emotions and those of disengaging positive (or negative) emotions. The respective regression coefficients were then averaged over all the
participants in the two cultures. These two steps were simultaneously estimated with an algorithm provided by the hierarchical linear models (HLM) analysis (Raudenbush & Bryk, 2002). The pertinent regression coefficients are summarized in Table 3. Preliminary analysis showed no significant gender effect, so this variable was dropped. The regression coefficient for engaging positive emotions (e.g., friendly feelings) was positive in both cultures, indicating that individuals felt happier when they found themselves connected with close others. As predicted, however, the effect of engaging emotions (e.g., friendly feelings) was stronger for Japanese than for Americans, 2(1) ⫽ 7.39, p ⬍ .01. In contrast, the effect of disengaging emotions (e.g., pride) was much stronger for Americans than for Japanese, 2(1) ⫽ 21.13, p ⬍ .0001. Within each country, the relative magnitude of the effect of the two types of positive emotions conformed to our predictions. In Japan, the effect of engaging emotion was significantly larger than the effect of disengaging emotion, 2(1) ⫽ 11.65, p ⬍ .0001. In contrast, in the United States, the effect of disengaging emotion was no different from the effect of engaging emotion, 2(1) ⫽ 1.16, p ⬎ .25. Because positive situations are likely to induce all positive emotions and negative situations are likely to induce all negative emotions, one might expect that engaging positive emotion would be positively correlated with disengaging positive emotion. This in fact was the case. When this correlation was computed for each of the participants, it was substantially positive, with means of .54
Table 3 Regression Coefficients That Predict General Positive Emotion as a Function of Engaging Positive Emotion (e.g., Friendly Feelings) and Disengaging Positive Emotion (e.g., Self-Esteem) Within Each Individual (in Study 1) Predicting general positive emotion (e.g., happiness)
Predicting general negative emotion (e.g., unhappiness)
Culture
Engaging emotion
Disengaging emotion
Engaging emotion
Disengaging emotion
Japanese American
0.68 0.50
0.27 0.60
0.21 0.15
0.59 0.50
896
KITAYAMA, MESQUITA, AND KARASAWA
and .47 and standard deviations of .23 and .27 for Japanese and Americans, respectively. Because multicollinearity can introduce considerable noise in the estimation of regression coefficients (Neter, Kutnerm, Wassermand, & Nachtsheim, 1996), we ran another analysis that is theoretically equivalent and, yet, that is not susceptible to the multicolleanearity problem. For each participant, we first subtracted the mean intensity of engaging positive emotions from the mean intensity of disengaging positive emotions. This difference score was used to predict the mean intensity of general positive emotions. As predicted, and replicating the first regression analysis, the effect of country was highly significant, with the average standardized regression coefficient for the difference score being significantly greater for Americans than for Japanese (Ms ⫽ .27 and ⫺.37 for Americans and Japanese, respectively), 2(1) ⫽ 19.15, p ⬍ .0001. As predicted, the American coefficient was significantly positive, 2(1) ⫽ 15.06, p ⬍ .0001, indicating that Americans feel happier when they experience more disengaging (rather than engaging) positive emotions. In contrast, the Japanese coefficient was significantly negative, 2(1) ⫽ 28.67, p ⬍ .0001, meaning that Japanese feel happier when they experience more engaging (rather than disengaging) positive emotions.5 Next, we carried out a comparable analysis for negative emotions. As can be seen in Table 3, unhappiness was more reliably predicted by disengaging negative emotion than by engaging negative emotion. This was the case for both Japanese and Americans, 2(1) ⫽ 102.34, p ⬍ .0001 and 2(1) ⫽ 42.09, p ⬍ .0001, respectively. Moreover, the effect of both engaging emotion and the effect of disengaging emotion tended to be larger for Japanese than for Americans, but only the latter effect was statistically significant, 2(1) ⫽ 6.20, p ⬍ .02. The same conclusion was obtained in an additional analysis in which a difference score between engaging negative emotion and disengaging negative emotion was used to predict general negative emotion. Correlation between two indices of emotional experience. We used two primary indices of emotional experience in the present study (i.e., intensity and predictor of happiness). In order to explore whether the two indices of emotional experience may be correlated, we first obtained a single measure for each index. We start with the relative magnitude of experiencing disengaging (vs. engaging) emotions. For each participant, the average intensity rating for positive engaging emotions in positive situations was subtracted from the average intensity rating for positive disengaging emotions in the same positive situations. Likewise, the average intensity rating for negative engaging emotions in negative situations was subtracted from the average intensity rating for negative disengaging emotions in the same negative situations. These two relative intensities were averaged to yield a single measure of the relative intensity of experiencing disengaging (rather than engaging) emotions. As may be predicted, the index was positive for Americans and negative for Japanese, and, moreover, the cross-cultural difference was quite substantial (Ms ⫽ .37 vs. ⫺.13), t(78) ⫽ 5.01, p ⬍ .0001. The effect size was large, as indicated by Cohen’s d ⫽ 1.16 (Cohen, 1988). Next, we computed for each participant a standardized regression coefficient that predicts happiness as a function of disengaging (rather than engaging) positive emotions. As may be predicted, the coefficient was positive for Americans and negative for Japanese, and, moreover, the cross-cultural difference was substantial
(Ms ⫽ .13 vs. ⫺.22), t(79) ⫽ 4.38, p ⬍ .0001. The effect size was large, as indicated by Cohen’s d ⫽ 0.99. Finally, we computed the correlation between the two indices. Both of them showed substantial variations, and, yet, the correlation was no different from zero either in Japan or in the United States (rs ⫽ .29 and ⫺.14, ps ⬎ .10, for Japanese and Americans, respectively).
Study 2: Reported Intensity of Experiencing Emotions in a Preselected Set of Situations Although Study 1 confirmed our predictions in naturally occurring, ecologically and culturally valid settings, this same feature of the study made it difficult to exclude one alternative interpretation. That is to say, the cultural differences in emotional response may have happened because very different sets of emotional events were recalled in the two cultures. For example, Japanese may have experienced more engaging emotions than did Americans because they were more likely to be in a situation that involved a meaningful social relationship. If, however, emotional responses were different because culturally accessible emotional themes were different, then the findings from Study 1 should be replicated even when respondents reported their emotional responses to the same set of social situations. Study 2 was designed to address this issue.
Method Participants and procedure. Fifty-five Japanese (20 men and 35 women) and 46 Americans (23 men and 23 women) participated in a study on “emotional experiences in daily life.” The Japanese participants, all undergraduates at a Japanese university, received either course credit or 1000 yen (U.S. $8) for their participation. American participants, all Caucasian American undergraduates who were temporarily studying at a Japanese university, received 1000 yen for their participation. All participants were tested in small groups. Respondents were asked to remember the latest event of each of 22 different event categories and to report the extent to which they experienced each of 25 emotions in the situation.6 A 6-point rating scale that ranged from 0 (not experienced it at all ) to 5 (experienced it very strongly) was used. Materials. We adopted 22 types of situations from earlier work by Reyes (1997; see Table 4 for a list of the situation types) because they spanned a wide variety of daily emotion elicitors. Some situations involved social relations, others involved study and work, and still others had to do with daily hassles and bodily conditions of the self. Furthermore, a priori the 22 situations seemed to include about as many positive as negative events. Inadvertently, one of the situations included in the Japanese set (“saw somebody I like”) was missing in the American set, whereas one situation included in the American set (“saw the person I have a crush on”) was missing in the Japanese set.
Results and Discussion Intensity of emotional experience. In order to control for the effect of the pleasantness of situations, we first divided the 21 5 The beta from this analysis was highly correlated with the difference between the beta for engaging emotion and the one for disengaging emotion from the first analysis (r ⫽ .84). 6 Three of the emotion terms were related to “amae”—a Japanese indigenous emotion glossed as desire for dependency (Doi, 1971). These terms were not analyzed.
CULTURE AND EMOTION
897
Table 4 Emotional-Experiences-in-Daily-Life Situations Used in Study 2 Japan Situation
Happiness
U.S.
Unhappiness
Happiness
Unhappiness
0.33 0.30 0.40 0.81 0.57 0.56 0.51 0.40
2.72 2.85 2.69 2.49 1.69 2.78 2.82 1.54
0.29 0.20 0.14 0.31 1.01 0.29 0.21 0.44
1.80 1.35 1.52 2.20 1.98 1.76 1.48 1.68
0.92 0.98 0.88 0.37 0.77 0.46 0.77 0.75
1.12 1.07 0.93 1.29 1.08 1.64 1.23 1.25
1.37 1.86 2.24 2.01 1.14
1.03 1.73 1.48 1.58 0.69
0.84 0.26 1.01 1.26 0.80
— 0.74
1.54 —
1.27 —
Positive situation Positive interaction with friends Something good happened to a family member Good interaction with family members Heard a comment about my appearance Watched TV or listened to music Read a novel or book Participated in a sports activity Class got canceled
3.58 2.36 3.38 1.54 2.99 2.46 2.65 2.92
Negative situation Caught up in a traffic jam Thought about my appearance Overloaded with work (e.g., schoolwork) Getting ill or injured Problem with a family member Difficulty understanding a lecture Took an exam Studied for an exam
0.73 1.07 0.85 0.60 0.59 0.50 0.94 0.72 Ambivalent situation
Late for an appointment Didn’t get enough sleep Argument or problem with friends Something bad happened to a family member Skipped a class
0.50 0.51 0.36 0.42 1.30 Negative situation
Saw the person I have a crush on Saw somebody I like
— 3.18
Note. The 23 situations are classified into different types, depending on the relative reported intensity of experiencing happiness-versus unhappiness-related emotions. U.S. ⫽ United States; dashes indicate data were not obtained or are not reported.
common situations into two categories, one of positive and one of negative events. Positive events were operationalized as events that, in both cultures, elicited more general positive emotions than general negative emotions, whereas negative events were those events that, in both cultures, elicited more general negative than general positive emotions. Specifically, we calculated for each situation the average intensity of experiencing general positive emotions (␣s ⫽ .90 and .85 for Japanese and Americans, respectively) and that of experiencing general negative emotions (␣s ⫽ .80 and .74 for Japanese and Americans, respectively), after which we determined for each situation the ratio of general positive and general negative emotions for each culture separately (see Table 4). Of the 21 situations, 16 were successfully classified as being either predominantly positive or predominantly negative for both cultures. The remaining 5 situations were excluded from the analysis on intensity of emotional experience. We performed an ANOVA in which subjects were used as a random variable. For each participant, mean emotion intensities were computed for the four types of emotions and for positive and negative situations separately. Type of emotion is characterized by
both social orientation (engaging vs. disengaging) and pleasantness (positive vs. negative). The mean intensity ratings for each of the four emotion types were submitted to an ANOVA, with two between-subjects variables (culture and gender) and three withinsubjects variables (emotion pleasantness, emotion social orientation, and situation pleasantness). This ANOVA assessed the generalizability of effects relative to the variability across different subjects. We also sought to determine the generalizability of the effects relative to the variability across different situations with another ANOVA wherein situations served as a random variable. For each of the situations that were either positive or negative, mean emotion intensities were computed for the four types of emotions calculated separately for the four groups of participants that differed in culture and gender. The mean intensity ratings were then submitted to an ANOVA, with one between-situation factor (situation pleasantness) and four within-situation factors (emotion social orientation, emotion pleasantness, participant culture, and participant gender). This ANOVA assessed the generalizability of the effects across different situations.
KITAYAMA, MESQUITA, AND KARASAWA
898
Pertinent means by culture are illustrated in Figure 2. As can be seen, the predicted interaction between culture and social orientation of emotions is evident for those cases in which situation pleasantness is matched with emotion pleasantness, namely, for both positive emotions that were experienced in positive situations and negative emotions that were experienced in negative situations. In the remaining two cases, no such interaction is evident. This observation is underscored by an interaction that involved culture, social orientation of the emotion, emotion pleasantness, and situation pleasantness. This interaction was significant in both the between-subject analysis, F(1, 97) ⫽ 41.33, p ⬍ .0001, and the between-situation analysis, F(1, 14) ⫽ 5.15, p ⬍ .05. We subsequently conducted four separate sets of ANOVAs for the four possible combinations of pleasantness of emotions and pleasantness of situations. Again, two separate ANOVAs assessed the generalizability of the effects across both subjects and situations. The results of these analyses are summarized in Table 5. For the positive emotions in the positive situations, the predicted Culture ⫻ Social Orientation interaction was significant in the between-subjects analysis and also reached marginal significance in the between-situations analysis. Both Japanese and Americans reportedly experienced the engaging emotions (e.g., friendly feelings) more strongly than the disengaging emotions (e.g., pride). As predicted, however, the reliable Culture ⫻ Social Engagement interaction indicates that this tendency was reliably stronger for Japanese than for Americans. From a different angle, the disengaging emotions (e.g., pride) were reportedly experienced more strongly by Americans than by Japanese, t(97) ⫽ 2.79, p ⬍ .001, but this difference vanished for the engaging emotions (e.g.,
Japanese
friendly feelings; t ⬍ 1). For the negative emotions in the negative situations, the Culture ⫻ Social Orientation interaction proved to be significant in both the between-subjects analysis and the between-situations analysis. Both Japanese and Americans reportedly experienced the disengaging emotions (e.g., anger) more strongly than the engaging emotions (e.g., guilt). As predicted, however, this tendency was reliably more pronounced for Americans than for Japanese. In the remaining two cases in which the pleasantness of the emotions did not match with the pleasantness of the situations, the Culture ⫻ Emotion Social Orientation interaction was negligible. As can also be seen in Table 5, there were some sporadic significant interactions involving gender. Two of them (Social Orientation ⫻ Gender interaction for the positive emotions in both positive and negative situations) achieved statistical significance in both the between-subjects analysis and the between-situations analysis. These interactions are in accordance with the idea that women are more interdependent than men (Cross & Madson, 1997). That is, in the positive situations, although both men and women reportedly experienced engaging positive emotions more strongly than disengaging positive emotions in the positive situations, this effect was more pronounced for women (Ms ⫽ 2.81 and 2.16) than for men (Ms ⫽ 2.68 and 2.27). In the negative situations, whereas men reportedly experienced disengaging positive emotions more strongly than engaging positive emotions (Ms ⫽ 1.62 vs. 1.50), women reportedly experienced engaging emotions more than disengaging emotions (Ms ⫽ 1.67 vs. 1.56). Caution is warranted, however, because the comparable finding was absent in Study 1. In fact, Study 1 showed one interaction involving gender
Am ericans
3.2
Negative Situations
Positive Situations
3 2.8
Intensity ofexperience
2.6 2.4 2.2 2 1.8 1.6 1.4 1.2 1 Engaging
Disengaging Positive
Engaging
Disengaging
Engaging
N egative
Disengaging Positive
Engaging
Disengaging
N egative
Type ofEm otions
Figure 2. Reported intensity of experiencing positive and negative emotions that are either engaging or disengaging in positive and negative situations in Study 2.
CULTURE AND EMOTION
899
Table 5 F Tests From Analyses of Variance Performed on Four Conditions That Varied in the Valence of Emotions and the Valence of Situations (in Study 2) Positive situation Positive emotion
Negative situation
Negative emotion
Positive emotion
Negative emotion
Variable
Between subjects
Between situations
Between subjects
Between situations
Between subjects
Between situations
Between subjects
Between situations
Culture Gender Culture ⫻ Gender Social orientation (SO) SO ⫻ Culture SO ⫻ Gender SO ⫻ Culture ⫻ Gender
1.73 ⬍1 2.11 85.38*** 25.30*** 4.48* ⬍1
4.42† ⬍1 4.47† 12.65** 4.02† 5.47† ⬍1
1.92 ⬍1 ⬍1 2.79 ⬍1 ⬍1 ⬍1
⬍1 ⬍1 1.09 1.69 ⬍1 ⬍1 ⬍1
7.56* ⬍1 ⬍1 ⬍1 1.07 6.04* ⬍1
17.11** ⬍1 2.54 ⬍1 ⬍1 15.04*** ⬍1
⬍1 1.06 1.21 175.78*** 17.37*** ⬍1 ⬍1
⬍1 3.60† 4.75† 33.42*** 11.06* ⬍1 ⬍1
† p ⬍ .10.
* p ⬍ .05.
** p ⬍ .001.
*** p ⬍ .0001.
that contradicted this pattern. Evidently, the gender effects are weaker and less consistent than the culture effects. Predictors of well-being. In this analysis, all the 22 situations were used.7 Over the diverse array of situations, we predicted that Americans would be happiest when they succeeded in tasks of independence, whereas Japanese would be happiest when they succeeded in tasks of interdependence. Preliminary analysis showed no significant gender effect, so this variable was dropped. An HLM analysis performed on the measures of emotional experience provided strong support for this prediction. As shown in Table 6, the degree of experiencing general positive emotions (e.g., happiness) was significantly increased as a function of the degree of experiencing both engaging and disengaging positive emotions (e.g., friendly feelings and pride, respectively). As predicted, however, the effect of engaging emotions (e.g., friendly feelings) was stronger for Japanese than for Americans, 2(1) ⫽ 9.23, p ⬍ .05, but the effect of disengaging emotions was stronger for Americans than for Japanese, 2(1) ⫽ 30.00, p ⬍ .0001. Within each culture, the relative magnitude of the effects of the two types of positive emotions provided strong support for our analysis. Thus, in Japan, the effect of engaging emotions was much stronger than the effect of disengaging emotions, 2(1) ⫽ 17.13, p ⬍ .0001, whereas in the United States, the effect of disengaging emotions was reliably stronger than the effect of engaging emotions, 2(1) ⫽ 8.20, p ⬍ .05.
Table 6 Regression Coefficients That Predict General Positive Emotion as a Function of Engaging Positive Emotion (e.g., Friendly Feelings) and Disengaging Positive Emotion (e.g., Self-Esteem) Over 22 Situations Within Each Individual (in Study 2) Predicting general positive emotion (e.g., happiness)
Predicting general negative emotion (e.g., unhappiness)
Culture
Engaging emotion
Disengaging emotion
Engaging emotion
Disengaging emotion
Japanese American
0.61 0.41
0.30 0.65
0.24 0.20
0.56 0.40
We also used a difference score between the intensity rating for positive disengaging emotions and the intensity rating for positive engaging emotions to predict happiness. This analysis eliminated the multicollinearity problem associated with the first analysis. We replicated the first analysis and found that the average standardized regression coefficient for Americans was significantly positive (M ⫽ .14), 2(1) ⫽ 10.93, p ⬍ .002, indicating that Americans feel happy primarily when they feel disengaging (rather than engaging) positive emotions. In contrast, the standardized regression coefficient for Japanese was significantly negative (M ⫽ ⫺.44), 2(1) ⫽ 154.05, p ⬍ .0001, indicating that Japanese feel happy primarily when they feel engaging (rather than disengaging) positive emotions. Remember 21 of the 22 situations were common across the two cultures. For each of the 21 situations, we tested the predicted cross-cultural difference by predicting happiness as a function of both engaging and disengaging positive emotions in each situation. The pattern was very consistent across situations, with the predicted cross-cultural difference evident in 19 of the 21 situations. We also sought to predict the experience of general negative emotions (e.g., unhappiness) as a function of negative engaging and negative disengaging emotions. As in Study 1, the effect of disengaging negative emotions (e.g., anger) was stronger than the effect of engaging negative emotions (e.g., guilt) for both Japanese and Americans, 2(1) ⫽ 51.06, p ⬍ .0001. The conclusion remained the same when a difference score between disengaging negative emotion and engaging negative emotion was used to predict happiness. Correlation between two indices of emotional experience. As in Study 1, we first computed, for each participant, a single index of the relative magnitude of experiencing disengaging (vs. engaging) emotions. As predicted, this index took a positive value for Americans and a negative value for Japanese (Ms ⫽ .42 and ⫺.13), t(99) ⫽ 7.28, p ⬍ .0001. Moreover, the cross-cultural difference was quite large, with Cohen’s d ⫽ 1.46. Next, also for each participant, we computed a standardized regression coefficient predicting happiness as a function of disengaging (rather than 7
Results were no different when the situation that was unique to each country was excluded.
KITAYAMA, MESQUITA, AND KARASAWA
900
engaging) positive emotions. The average coefficient, as predicted, was positive for Americans and negative for Japanese (Ms ⫽ .05 and ⫺.37), t(99) ⫽ 7.34, p ⬍ .0001. Moreover, the cross-cultural difference was large, with Cohen’s d ⫽ 1.46. Finally, we correlated the two indices. Notably, the correlation was significantly positive both in Japan and in the United States (rs ⫽ .37 and .58, ps ⬍ .05, for Japanese and Americans, respectively). The convergence between the relative intensity measure and the happiness measure here is in sharp contrast with the finding in Study 1, in which we found no such convergence.
General Discussion Culture and Experience of Emotion Past research on culture and emotion has primarily focused on the potential for emotions (Mesquita, Frijda, & Scherer, 1997), that is, the emotions that people are capable of having in principle. Thus, much cross-cultural research on emotions has aimed at finding similarities in the aptitude for certain aspects of emotions, such as the facial or vocal expressions and the physiological response patterns associated with particular emotions. In fact, this research has yielded overwhelming evidence for similarity with regard to the emotional responses that people in different cultures are capable of having and expressing (Ekman, 1994; Scherer, 1997; Scherer, Schorr, & Johnstone, 2001; Scherer, Wallbott, & Summerfield, 1986), although some careful comparisons have also yielded culture specificity even at this level (e.g., Menon & Shweder, 1994; Mesquita, 2001). The present study takes the cross-cultural research on emotions beyond the potential for emotions to the actual experience of emotions (Mesquita, 2003). The most important contribution of the present work, then, was to demonstrate that there exists reliable and systematic cross-cultural variation in emotional experience. We focused on emotions that are distinguished on the dimensions of pleasantness and social orientation. Drawing on our earlier work, we proposed that, across cultures, emotional experience is organized according to these dimensions, such that one can crossculturally distinguish between positive and negative emotions and, moreover, between socially engaging and disengaging emotions. We expected that the meanings and practices of independent cultural contexts encourage themes of independence such as personal achievement and goal blockage, and the meanings and practices of interdependent cultural contexts make themes of connectedness or the lack thereof salient. As a consequence, we predicted systematic cultural variation in emotional experience. In particular, we expected that interdependent cultural contexts would activate engaging emotions more, whereas independent cultural contexts would activate socially disengaging emotions more. We conducted two studies, comparing the emotions experienced in Japanese, interdependent contexts with those in North American, independent contexts.
pendent cultures, we found systematic tendencies for Japanese to experience engaging emotions more strongly than Americans do and, conversely, for Americans to experience disengaging emotions more strongly than Japanese do. These findings were consistent across the two studies that adopted somewhat different methodologies. Notably, both studies included a relatively large number of different emotional episodes (7–22) per respondent. The data thus reflect cultural patterns of emotional responses across different people as well as situations, rather than merely representing onetime measurements of emotional feelings in response to one particular event. The present data are thus consistent with the notion that cultural affordances for engaging or disengaging emotions are consistently present across different situations over all individuals engaging in the respective cultural contexts and, thus, highly pervasive even though their effect sometimes may not be readily detectable in a single person’s response to a single, specific situation or episode. We believe that the present methods that assess emotions across a number of situations are much more suited to reveal systematic effects of culture than more common methods in the emotion research that focus on a very small number of situations because these effects are mediated by cultural affordances that are subtly, but consistently present in many ordinary daily practices and pervasive ideas, beliefs, and assumptions.
Predictors of Happiness Moreover, in accordance with the notion that positive engaging themes are affirming of interdependence of self, we found that well-being is best predicted by positive emotions that are engaging among Japanese. We also hypothesized that positive disengaging themes are affirming of independence of self. In support of this idea, we found in Study 2 that positive disengaging emotions best predicted well-being of Americans, although this pattern was weak in Study 1. These findings indicate that Japanese and Americans are chronically motivated toward the very different goals of interdependence (e.g., social harmony) and independence (e.g., personal control), respectively (Kitayama et al., 2006). An overarching conclusion, then, is that across cultures, both positive engagement and positive disengagement can promote well-being. Yet, in Japan, those individuals who are embedded in close, relatively harmonious relations and thus are likely to experience friendly feelings, respect, and the like tend to enjoy more well-being; however, in the United States, social interdependence may be less important for well-being than is standing on one’s own feet, striving for personal achievement, and maintaining high selfesteem. Our findings are consistent with several past studies that make this general point (e.g., Kang et al., 2003; Kitayama et al., 2006, 2000; Kwan et al., 1997). Yet, the present findings are one of the first in the literature that demonstrates this consistency with ecologically valid daily experiences of emotions. Mesquita and colleagues have recently amassed evidence that is convergent with the evidence described here (Mesquita & Karasawa, 2002; Mesquita et al., 2005).
Intensity of Emotional Experience In support of the idea that socially engaging themes such as social harmony and failed repayment are prevalent in interdependent cultures but socially disengaging themes such as personal achievement and right infringement are more prevalent in inde-
Cultural Affordances and the Two Effects of Social Orientation We have argued that people differentially experience engaging and disengaging emotions because of the cultural affordances they
CULTURE AND EMOTION
are exposed to, with American contexts affording disengaging emotions and Japanese contexts affording engaging emotions. We have also argued that either engaging or disengaging positive emotions are more closely linked to happiness and subjective well-being because of culturally divergent motivational propensities, with Americans oriented toward independence and Japanese oriented toward interdependence. One significant question that must be addressed in future work concerns the connection between the two hypotheses tested in the present work. As a first approximation, we might posit that cultural affordances for engaging and disengaging social orientations are instrumental in nurturing and fostering relatively chronic motivational tendencies toward the corresponding goals of interdependence and independence (Kitayama et al., 1997). That is, people who are chronically exposed to cultural affordances for disengaging emotions may eventually acquire motivational tendencies toward independence, whereas those chronically exposed to cultural affordances for engaging emotions may eventually acquire motivational tendencies toward interdependence. If so, then the relative intensity of emotional experience and the predictor of happiness may be expected to be closely related to one another. That is, our two indices of emotional experience should converge. This is exactly what we found in Study 2. It is of interest, however, that we did not find any such correlation in Study 1. We suspect that this was because in Study 1, participants recalled only a very specific type of social episode, namely, the most emotional experience of the day. People’s motivational tendencies (as revealed in the happiness measure) may be influenced not only by cultural affordances existing in this type of situation (as indicated by the relative intensity measure) but also by cultural affordances associated with myriad other types of situations. It stands to reason that each person’s motivational tendencies are best predicted by the entire pool of cultural affordances he or she is exposed to over many different types of situations. It would follow that the relative intensity measure that is based on a wide array of situations (see Study 2) is a better proxy for the entire pool of cultural affordances than the corresponding measure that is based only on one type of situation (see Study 1). This may account for the fact that a highly reliable correlation between the relative intensity measure and the happiness measure was observed in Study 2 but not in Study 1. Although speculative, this analysis suggests that the notion of cultural affordances is useful in furthering researchers’ understanding of both origins and underlying mechanisms for a variety of cross-culturally divergent psychological effects.
Toward an Implicit Measure of Self-Orientation The present results may be brought to bear on the measurement of orientation of the self as independent or interdependent. In order to capture individual differences on this dimension, researchers have traditionally relied on several different attitudinal scales (e.g., Singelis, 1994). All these scales probe one’s independence or interdependence in terms of explicit self-reports (“In general, I make my own decisions” for independence and “When my opinion is in conflict with that of another person’s, I often accept the other opinion” for interdependence). Although these attitude scales have been used in numerous studies (Oyserman et al., 2002), the validity of these measures have recently been called into question. Some
901
researchers have pointed out certain methodological artifacts such as acquiescence response bias (Schimmack, Oishi, & Diener, 2005) and reference group effect (Heine, Lehman, Peng, & Greenholtz, 2002). Some others have argued that people’s behavioral propensities toward independence or interdependence may be implicit; that is, they may be rarely self-reflective, deliberate, or even conscious. If there is only a limited conscious access to such behavioral propensities (Kitayama, 2002; see also Nisbett & Wilson, 1977), then explicit self-report may be ill-suited as a tool for measuring them. The present work used the notion of independent and interdependent self to predict some systematic cross-cultural variations in the pattern of emotional responses. Yet, given the results reported in this article, we wonder whether the converse of the present logic may prove to be equally fruitful in future work. That is, it may be possible to use the pattern of emotion responses as a means for implicit assessment of independence and interdependence. One obvious measure is the extent to which engaging or disengaging positive emotions are linked to happiness. Furthermore, at least if emotional experience is assessed over a wide variety of social situations (as in Study 2), it may be reasonable to use the relative intensity of experiencing engaging versus disengaging emotions. Individuals may be said to be more independent (or interdependent) if their happiness is predicted more by disengaging positive (or engaging positive) emotions, if they experience disengaging (or engaging) emotions more, or both.8 Study 2 provided initial evidence for the convergent validity of these measures. Future research should directly assess the predictive validity of these implicit measures of independence and interdependence with respect to a variety of cross-cultural differences in cognition, such as analytic versus holistic thought (e.g., Nisbett, Peng, Choi, & Norenzayan, 2001), and motivation, such as self-enhancement and self-improvement (e.g., Heine et al., 1999). We suspect that many of such behavioral or online responses may be better predicted by implicit measures of independence and interdependence than by their explicit counterparts, such as the one developed by Singelis (1994).
Uncovering Social Orientation of Emotions The present work has provided yet another piece of evidence for the dimension of social orientation as central in analyzing emotional experience. Nevertheless, this dimension has never been acknowledged in appraisal theories of emotion (Scherer et al., 2001). This apparent neglect may be because the appraisal theories have their set of dimensions that overlap with social engagement and disengagement. To be more specific, whereas social engagement may often be conflated with submissiveness, externality in control, and lower self-control, social disengagement may have a lot in common with dominance, internality in control, and higher self-control. This means, for example, that a state of social relationship that can be described and experienced as “harmonious,” “respectful,”, and thus “socially engaging” can alternatively be described and experienced as “submissive,” “externally con8 Yet another implicit measure that is promising is the spontaneous accessibility of personal versus relational attributes in a 20-statement test (Balcetis, Dunning, & Miller, 2006).
902
KITAYAMA, MESQUITA, AND KARASAWA
trolled,” and “losing self-control.” Likewise, a state of social relationship that can be described and experienced as “separated,” “independent,” and “ego centered” can also be described and experienced as “in control,” “self-enhancing,” and “dominating.” It then appears that the appraisal dimensions assumed in the contemporary literature on emotion provide alternative ways of conceptualizing social engagement and disengagement. Like ambiguous figures, any given social situation can be multiply construed. However, it seems important to stay experientially close to the experiencer’s own perspective, as it is this perspective—not the researchers’—that will shape the nature of emotional experience and subsequent behaviors. We suggest that especially in cultures that emphasize the connectedness between people, a situation may rather be captured as one of engagement than as one of low control, submissiveness, and the like. We believe that exploring such perspectival effects on emotions will be a next significant challenge in future research on culture and emotion.
Limitations A few limitations of the present work should be acknowledged. First, we only used verbal report of emotions. Emotions, however, can be unconscious or, even if not, may not always be amenable to verbal articulation. Future work should examine nonverbal measures of emotion to determine whether the pattern of cultural variation can be reliably identified with such measures as well. The second limitation concerns the notion of social orientation itself. One major feature of negative emotions that vary in this regard involves motivational tendencies. Thus, whereas engaging negative emotions, such as guilt and shame, are considered to entail a strong motivational tendency to repair damaged social harmony and to restore interdependence, disengaging negative emotions, such as anger and frustration, are considered to entail a strong motivational tendency to repair damaged independence and to restore the sense of autonomy. These motivational tendencies have so far been rarely addressed in the emotion literature, and the present research is no exception. Future research should focus on the link between emotional and motivational processes. This new focus would allow emotion researchers to explicitly link the emotion literature to other related areas of personality and social psychology, such as altruism, intrinsic motivation, and aggression. Third, we relied on the notion of cultural affordances to make the predictions that were tested in this work. Although we did not directly assess the affordances themselves, such assessment has been successfully attempted in the domains of self-esteem (Kitayama et al., 1997), control (Morling et al., 2002), and cognitive style (Miyamoto, Nisbett, & Masuda, 2006). Future work should identify in detail the nature of cultural affordances for both engagement and disengagement. Finally, present psychological research on culture, including the studies described here, is based largely on college student samples. Future work on emotion and culture should cover the entire range of the life cycle (Kitayama et al., 2006). Moreover, with initial evidence for both regional variations (Kitayama, Ishii, Imada, Takemura, & Ramaswamy, in press) and variations as a function of social class (Schooler, in press; Snibbe & Markus, 2005), it will be increasingly important to pay careful attention to similarities and differences within any single macroscopic culture. By so
doing, it will be possible to better understand the process by which culture influences and shape one’s emotional experience.
References Abelson, R. P. (1985). A variance explanation paradox: When a little is a lot. Psychological Bulletin, 112, 128 –132. Balcetis, E., Dunning, D., & Miller, R. L. (2006). Do collectivists “know themselves” better than individualists? Cross-cultural studies of the “holier than thou” phenomenon. Unpublished manuscript, Cornell University. Brewer, M. B., & Gardner, W. L. (1996). Who is this “we”? Levels of collective identity and self representations. Journal of Personality and Social Psychology, 71, 83–93. Chiu, C., & Hong, Y. (2005). Cultural competence: Dynamic processes. In A. Elliot & C. S. Dweck (Eds.), Handbook of motivation and competence (pp. 489 –505). New York: Guilford Press. Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Erlbaum. Cross, S. E., & Madson, L. (1997). Models of the self: Self-construals and gender. Psychological Bulletin, 122, 5–37. Doi, D. T. (1971). The anatomy of dependence (2nd ed.). Tokyo: Kodansha. Ekman, P. (1994). Strong evidence for universals in facial expressions: A reply to Russell’s mistaken critique. Psychological Bulletin, 115, 268 – 287. Ellsworth, P. C. (1994). Sense, culture, and sensibility. In S. Kitayama & H. R. Markus (Eds.), Emotion and culture: Empirical studies of mutual influence (pp. 23–50). Washington, DC: American Psychological Association. Frijda, N. H. (1986). The emotions. Cambridge, England: Cambridge University Press. Frijda, N. H., Mesquita, B., Sonnemans, J., & Van Goozen, S. (1991). The duration of affective phenomena or emotions, sentiments, and passions. International Review of Studies on Emotion, 1, 187–225. Heine, S. J., Lehman, D. R., Markus, H. R., & Kitayama, S. (1999). Is there a universal need for positive self-regard? Psychological Review, 106, 766 –794. Heine, S. J., Lehman, D. R., Peng, K., & Greenholtz, J. (2002). What’s wrong with cross-cultural comparisons of subjective Likert scales? The reference-group effect. Journal of Personality and Social Psychology, 82, 903–918. Hong, Y.-Y., Morris, M. W., Chiu, C.-Y., & Benet-Martinez, V. (2000). Multicultural minds. American Psychologist, 55, 709 –720. Kang, S., Shaver, P. R., Min, K., & Jin, H. (2003). Culture-specific patterns in the prediction of life satisfaction: Roles of emotion, relationship quality, and self-esteem. Personality and Social Psychology Bulletin, 29, 1596 –1608. Kitayama, S. (2002). Culture and basic psychological processes—Toward a system view of culture: Comment on Oyserman et al. (2002). Psychological Bulletin, 128, 89 –96. Kitayama, S., Duffy, S., & Uchida, Y. (in press). Self as cultural mode of being. In S. Kitayama & D. Cohen (Eds.), Handbook of cultural psychology. New York: Guilford Press. Kitayama, S., Ishii, K., Imada, T., Takemura, K., & Ramaswamy, J. (in press). Voluntary settlement and the spirit of independence: Evidence from Japan’s “Northern Frontier.” Journal of Personality and Social Psychology. Kitayama, S., Karasawa, M., Curhan, K., Ryff, C. D., & Markus, H. R. (2006). Independence, interdependence, and well-being: Divergent patterns in the United States and Japan. Unpublished manuscript, University of Michigan. Kitayama, S., Karasawa, M., & Mesquita, B. (2004). Collective and personal processes in regulating emotions: Emotion and self in Japan and
CULTURE AND EMOTION the U.S. In P. Philippot & R. S. Feldman (Eds.), The regulation of emotion (pp. 251–273). Hillsdale, NJ: Erlbaum. Kitayama, S., & Markus, H. R. (1999). Yin and yang of the Japanese self: The cultural psychology of personality coherence. In D. Cervone & Y. Shoda (Eds.), The coherence of personality: Social cognitive bases of personality consistency, variability, and organization. (pp. 242–302). New York: Guilford Press. Kitayama, S., Markus, H. R., & Kurokawa, M. (2000). Culture, emotion, and well-being: Good feelings in Japan and the United States. Cognition & Emotion, 14, 93–124. Kitayama, S., Markus, H. R., & Matsumoto, H. (1995). Culture, self, and emotion: A cultural perspective on “self-conscious” emotions. In J. P. Tangney & K. W. Fisher (Eds.), Self-conscious emotions: The psychology of shame, guilt, embarrassment, and pride (pp. 439 – 464). New York: Guilford Press. Kitayama, S., Markus, H. R., Matsumoto, H., & Norasakkunkit, V. (1997). Individual and collective processes in the construction of the self: Self-enhancement in the United States and self-criticism in Japan. Journal of Personality and Social Psychology, 72, 1245–1267. Kwan, V. S. Y., Bond, M. H., & Singelis, T. M. (1997). Pancultural explanations for life satisfaction: Adding relationship harmony to selfesteem. Journal of Personality and Social Psychology, 73, 1038 –1051. Lazarus, R. S. (1991). Emotion and adaptation. New York: Oxford University Press. Lazarus, R. S. (1999). Stress and emotion: A new synthesis. New York: Springer Publishing Company. Lutz, C. (1988). Unnatural emotions: Everyday sentiments on a Micronesian atool and their challenge to Western theory. Chicago: University of Chicago Press. Markus, H. R., & Kitayama, S. (1991a). Cultural variation of the self. In G. R. Goethals & J. Strauss (Eds.), Multidiscriplinary perspectives on the self (pp. 18 – 48). New York: Springer-Verlag. Markus, H. R., & Kitayama, S. (1991b). Culture and the self: Implications for cognition, emotion, and motivation. Psychological Review, 98, 224 –253. Menon, U., & Shweder, R. A. (1994). Kali’s tongue: Cultural psychology and the power of shame in Orissa, India. In S. Kitayama & H. R. Markus (Eds.), Emotion and culture: Empirical studies of mutual influence (pp. 241–284). Washington, DC: American Psychological Association. Mesquita, B. (2001). Emotions in collectivist and individualist contexts. Journal of Personality and Social Psychology, 80, 68 –74. Mesquita, B. (2003). Emotions as dynamic cultural phenomena. In R. Davidson, H. Goldsmith, & K. R. Scherer (Eds.), The handbook of the affective sciences (pp. 871– 890). New York: Oxford University Press. Mesquita, B., & Ellsworth, P. (2001). The role of culture in appraisal. In K. R. Scherer, A. Schorr, & T. Johnstone (Eds.), Appraisal processes in emotion: Theory, methods, research (pp. 233–248). New York: Oxford University Press. Mesquita, B., & Frijda, N. H. (1992). Cultural variations in emotions: A review. Psychological Bulletin, 112, 179 –204. Mesquita, B., Frijda, N. H., & Scherer, K. R. (1997). Culture and emotion. In P. Dasen & T. S. Saraswathi (Eds.), Handbook of cross-cultural psychology (Vol. 2, pp. 255–297). Boston: Allyn & Bacon. Mesquita, B., & Karasawa, M. (2002). Different emotional lives. Cognition & Emotion, 16, 127–141. Mesquita, B., Karasawa, M., Haire, A., Izumi, K., Hayashi, A., Idzelis, M., et al. (2005). Emotions as culture-specific ways of relating: A comparison between Japanese and American groups. Unpublished manuscript, Wake Forest University, Winston-Salem, NC. Mesquita, B., & Leu, J. (in press). The cultural psychology of emotion. In S. Kitayama & D. Cohen (Eds.), Handbook of cultural psychology. New York: Guilford Press.
903
Miyamoto, Y., Nisbett, R. E., & Masuda, T. (2006). Culture and the physical environment: Holistic versus analytic perceptual affordances. Psychological Science, 17, 113–117. Morling, B., Kitayama, S., & Miyamoto, Y. (2002). Cultural practices emphasize influence in the United States and adjustment in Japan. Personality and Social Psychology Bulletin, 28, 311–323. Neter, J., Kutnerm, M. H., Wassermand, W., & Nachtsheim, C. J. (1996). Applied linear statistical modes (4th ed.). Chicago: Irwin. Nisbett, R. E., Peng, K., Choi, I., & Norenzayan, A. (2001). Culture and systems of thought: Holistic versus analytic cognition. Psychological Review, 108, 291–310. Nisbett, R. E., & Wilson, D. (1977). Telling more than we can know: Verbal reports on mental processes. Psychological Review, 84, 231–253. Oishi, S. (2002). The experiencing and remembering of well-being: A cross-cultural analysis. Personality and Social Psychology Bulletin, 28, 1398 –1406. Oyserman, D., Coon, H. M., & Kemmelmeier, M. (2002). Rethinking individualism and collectivism: Evaluation of theoretical assumptions and meta-analyses. Psychological Bulletin, 128, 3–72. Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and data analysis methods (2nd ed.). Newbury Park, CA: Sage. Reyes, J. A. S. (1997). Personality, life events, and appraisal as determinants of emotions: A test of an integrated model in a non-Western culture. Unpublished dissertation, Washington State University. Schachter, S., & Singer, J. (1962). Cognitive, social and physiological determinants of emotional state. Psychological Review, 69, 379 –399. Scherer, K. R. (1997). The role of culture in emotion-antecedent appraisal. Journal of Personality and Social Psychology, 73, 902–922. Scherer, K. R., Schorr, A., & Johnstone, T. (2001). Appraisal processes in emotion: Theory, methods, research. New York: Oxford University Press. Scherer, K. R., Wallbott, H. G., & Summerfield, A. B. (Eds.). (1986). Experiencing emotion: A cross-cultural study. Cambridge, England: Cambridge University Press. Schimmack, U., Oishi, S., & Diener, E. (2005). Individualism: A valid and important dimension of cultural differences between nations. Personality & Social Psychology Review, 9, 17–31. Schooler, C. (in press). Culture and social structure: The relevance of social structure to cultural psychology. In S. Kitayama & D. Cohen (Eds.), Handbook of cultural psychology. New York: Guilford Press. Singelis, T. M. (1994). The measurement of independent and interdependent self-construals. Personality and Social Psychology Bulletin, 20, 580 –591. Snibbe, A. C., & Markus, H. R. (2005). You can’t always get what you want. Journal of Personality and Social Psychology, 88, 703–720. Sonnemans, J., & Frijda, N. H. (1994). The structure of subjective emotional intensity. Cognition & Emotion, 8, 329 –350. Sperber, D. (1996). Explaining culture: A naturalistic approach. Cambridge, MA: Blackwell Publishers. Uchida, Y., Kitayama, S., Mesquita, B., Reyes, J. A., & Morling, B. (2004). What puts the well in well-being? Self-esteem and emotional support. Koshien University, Momijigaoka Takarazuka Hyogo, Japan. Wierzbicka, A. (1994). Emotion, language, and cultural scripts. In S. Kitayama & H. R. Markus (Eds.), Emotion and culture: Empirical studies of mutual influence (pp. 133–194). Washington, DC: American Psychological Association.
Received May 7, 2003 Revision received March 19, 2006 Accepted March 31, 2006 䡲
Journal of Personality and Social Psychology 2006, Vol. 91, No. 5, 904 –917
Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.5.904
Will You Be There for Me When Things Go Right? Supportive Responses to Positive Event Disclosures Shelly L. Gable
Gian C. Gonzaga
University of California, Los Angeles
University of California, Los Angeles, and Eharmony.com
Amy Strachman University of California, Los Angeles Close relationship partners often share successes and triumphs with one another, but this experience is rarely the focus of empirical study. In this study, 79 dating couples completed measures of relationship well-being and then participated in videotaped interactions in which they took turns discussing recent positive and negative events. Disclosers rated how understood, validated, and cared for they felt in each discussion, and outside observers coded responders’ behavior. Both self-report data and observational codes showed that 2 months later, responses to positive event discussions were more closely related to relationship well-being and break-up than were responses to negative event discussions. The results are discussed in terms of the recurrent, but often overlooked, role that positive emotional exchanges play in building relationship resources. Keywords: social support, close relationships, positive emotions, intimacy, marital satisfaction
conflict, criticisms; Gottman, 1994; Karney & Bradbury, 1997; Notarius & Markman, 1989), studies such as those of capitalization processes offer emerging evidence that important dyadic relationship processes take place in the context of positive emotional experiences and deserve continued empirical investigation.1 In the present article, we examined the role that positive emotional exchanges play in relationship functioning through an observational study of couples’ interactions when sharing positive events. In addition, couples’ responses to positive event disclosures were compared with their responses to negative event disclosures— what is traditionally known as social support—to determine whether the association between positive event responses and relationship well-being are independent of the associations between social support and relationship well-being. That is, previous research has shown convincingly that a characteristic of satisfying relationships is believing that the partner will be there when things go wrong (e.g., Collins & Feeney, 2000; Pasch, Bradbury, & Davila, 1997), but it has not yet been shown that having a partner who will be there when things go right has independent effects on relationship functioning.
Good things happen, and when they do, people often share the positive event with someone else—a process that has been called capitalization (Langston, 1994). Capitalizing on positive events has been linked to increases in positive affect and well-being independent of the positive events themselves; however these effects rest, in large part, on the reactions of persons with whom the events are shared (Gable, Reis, Impett, & Asher, 2004). Moreover, the targets of capitalization are almost always close relationship partners, such as spouses, parents, best friends, or roommates. Research has shown that when close relationship partners, specifically romantic partners, regularly respond to positive event disclosures in a supportive manner, disclosers report feeling closer, more intimate, and generally more satisfied with their relationships than those whose partners typically respond in a nonsupportive manner (Gable et al., 2004). These effects have also been shown to be independent of the well-established association between partners’ responses to each other’s negative behavior and the health of the relationship (Rusbult, Verdette, Whitney, Slovic, and Lipkus, 1991). Whereas previous research has focused primarily on couples’ management of negative emotional experiences (e.g., jealousy,
Capitalization Responses and Traditional Social Support When people experience a negative or stressful event, they often turn to others for aid and comfort. The provision of emotional,
Shelly L. Gable and Amy Strachman, Department of Psychology, University of California, Los Angeles; Gian C. Gonzaga, Department of Psychology, University of California, Los Angeles, and Eharmony.com, Pasadena, California. This research was supported by a Young Scholars Grant from the Templeton Foundation and Positive Psychology Network awarded to Shelly L. Gable. We are grateful to Natalya Maisel and Harry Reis for comments on earlier versions of this article. Correspondence concerning this article should be addressed to Shelly L. Gable, Department of Psychology, University of California, Los Angeles, 4560 Franz Hall, Box 951563, Los Angeles, CA 90095-1563. E-mail:
[email protected]
1
We do not intend to suggest that all work on close relationships has focused on negative processes. There are certainly numerous examples of work on positive emotional processes in close relationships (e.g., A. Aron, Norman, E. N. Aron, McKenna, & Heyman, 2000; Drigotas, Rusbult, Wieselquist, & Whitton, 1999; Hatfield & Rapson, 1993; Sternberg, 1986). However, it remains the case that most research on close relationships targets the management of negative emotions (for a review and discussion, see Gable & Reis, 2001 and Reis & Gable, 2003). 904
SUPPORTIVE RESPONSES TO POSITIVE EVENT DISCLOSURES
tangible, and informational assistance from the social network has come to be known as social support. An abundance of research shows that the perception that one has supportive others to turn to in times of stress (i.e., perceived support) buffers against the harmful effects of stress (e.g., Cohen, 1992; Collins & Feeney, 2000; Sarason, Sarason, & Gurung, 1997). Of the different types of support, emotional support may play a particularly important role in the stress–adjustment link (Uchino, Cacioppo, & KiecoltGlaser, 1996). Moreover, in terms of the quality of close relationships, perceptions that a partner provides good support in times of distress are correlated with better functioning relationships (e.g., Barbee & Cunningham, 1995; Collins & Feeney, 2000, 2004; Cutrona, 1986; Pasch et al., 1997; Reis & Franks, 1994). However, although it is apparent that believing others will be available in bad times is beneficial for the person and the relationship, the associations among enacted support, perceived support, and well-being are mixed (e.g., Lakey, McCabe, Fisicaro, & Drew, 1996). In fact, many studies show that actual support transactions are not associated with better adjustment, or worse, they are negatively correlated with well-being (e.g., Barbee, Rowlett, & Cunningham, 1998; Bolger, Zuckerman, & Kessler, 2000; Coyne, Wortman, & Lehman, 1988). One possible reason that receiving social support may have neutral or detrimental effects is that it may be a signal to the recipient that he or she is unable to cope with the stressor, which can be a blow to self-worth and self-esteem (e.g., Fisher, Nadeler, & Whitcher-Alagna, 1982). Moreover, because a romantic partner is often a primary support provider, the perception of a diminished sense of self-worth in the eyes of the partner (real or imagined) may be especially problematic. For example, Murray and colleagues (e.g., Murray, Holmes, & Griffin, 2000; Murray et al., 2005) have shown that feeling inferior to one’s partner is associated with less commitment, less relationship satisfaction, and less love for the partner. These costs may offset the tangible or emotional benefits of a partner’s assistance. Bolger, Zuckerman, and Kessler (2000) reasoned that one way around the catch-22 inherent in support receipt may be to provide help to a distressed partner without his or her awareness. Bolger and colleagues have labeled this invisible support and have offered evidence that the most effective support is that which goes unnoticed by the distressed recipient. However, it may be difficult to provide support to a distressed individual without his or her knowledge. For example, in a daily experience study, when partners reported providing support to distressed New York Bar examinees, the distressed examinees reported receiving that support 65% of the time, and they even reported receiving support on 44% of the days that their partners denied providing it (Bolger et al., 2000, Table 1). We propose that another way around this apparent catch-22 is for the provision of support to occur in a situation free of threats to self-worth. That is, individuals who receive supportive responses from their partners in response to positive event disclosures can reap the relational benefits associated with perceived support without the blow to self-esteem. In contrast to negative event disclosure discussions, supportive responses to positive events actually highlight and play up the capitalizer’s strengths. Note that there are still risks involved in sharing a positive event; the partner could respond in an unsupportive manner or not respond at all. However, these risks are equivalent to the risks of a partner responding in an unsupportive manner when a negative
905
event is shared. Thus, there are unique threats to the self associated with seeking social support in times of stress that are not inherent in capitalization situations. Finally, because supportive responses to positive events can and should be out in the open (see below), enacted support may be more strongly linked to perceptions of support; which is in contrast to findings in traditional social support research that show that the link between enacted and perceived support is neither consistent nor clear (e.g., Kaul & Lakey, 2003).
Reactions to Capitalization Attempts and Perceived Responsiveness What constitutes a supportive response to capitalization attempts? In previous work, we used a modified framework that was originally used to describe responses to another person’s negative behavior (e.g., Hirschman, 1970; Rusbult, Zembrodt, & Gunn, 1982) to categorize responses along two dimensions: constructive– destructive and active–passive. Therefore, responses to capitalization attempts can be differentiated into four types: active– constructive (e.g., enthusiastic support), passive– constructive (e.g., quiet, understated support), active– destructive (e.g., demeaning the event), and passive– destructive responses (e.g., ignoring the event). These four different responses are illustrated in the following example. Maria comes home from her job as an associate in a law firm and excitedly tells her husband, Robert, that the senior partners called her into a meeting today and assigned her to be the lead lawyer for an important case filed on behalf of their most prestigious client. An active– constructive response from Robert might be, “Wow, this is great news! Your skills and hard work are definitely paying off; I am certain that your goal to make partner will happen in no time. What is the case about?” A passive– constructive response could be a warm smile followed by a simple, “That’s nice, dear.” An active– destructive response might be, “Wow, I bet the case will be complicated; are you sure you can handle it? It sounds like it might be a lot of work; maybe no one else wanted the case. You will probably have to work even longer hours this month.” A passive– destructive response might be, “You won’t believe what happened to me today,” or “What do you want to do for dinner?” Previous studies have found that only responses that were perceived to be active and constructive were associated with personal well-being and higher relationship quality, whereas the other three types of responses were negatively associated with these outcomes (Gable et al., 2004). In this study, we examined possible explanations for why responses perceived as active and constructive were beneficial to close relationships, whereas passive or destructive ones were detrimental. We reasoned that active and constructive responses convey two types of information to the discloser. First, active– constructive responses communicate positive information about the event itself through confirmation of the event’s importance and elaboration on potential implications of the event. Second, active-constructive responses convey positive information about the responder’s relationship with the capitalizer through displayed knowledge of the personal significance of the event to the capitalizer and a demonstration of the responder’s own feelings toward the capitalizer. On the other hand, passive or destructive responses fail to convey this information or, worse, convey the reverse. A passive or destructive response may signify (explicitly or implicitly) that (a) the event itself is not significant, either in the
906
GABLE, GONZAGA, AND STRACHMAN
present or in its future value; (b) the responder does not have intimate knowledge of what is important to the capitalizer; or (c) the capitalizer’s emotions, thoughts, and life are not of concern to the responder. In short, sharing personal positive events provides prime opportunities to obtain understanding, validation, and caring—a construct termed perceived partner responsiveness to the self in Reis and Shaver’s (1988) transactional model of intimacy. Perceived responsiveness to the self (responsiveness, for short) includes three overlapping elements: beliefs about others’ understanding of oneself, including one’s qualities, opinions, goals, emotions, and needs; thoughts about the degree to which others value, respect, and validate the self; and the perception that others care about and support the self. As Reis, Clark, and Holmes (2004) pointed out in a comprehensive literature review, the perception that close others appreciate and care for us lies at the heart of many processes in close relationships, including expectancies in social interaction, self-verification theory, individual differences in attachment security, and communal relationships (Holmes, 2002; Swann, 1990; Collins & Read, 1990; and Clark & Mills, 1979, respectively). Given that responsiveness seems to be central to relationship functioning, we hypothesized that capitalization exchanges play a significant role in the development and maintenance of healthy relationships. Therefore, in the present study, we investigated whether perceptions of partner responsiveness during discussions of personal positive events and negative events predicted relationship health. We also hypothesized that when partners reacted in an active– constructive manner to disclosers’ positive events, as coded by outside observers, the disclosers would report more perceived responsiveness than when their partners reacted in a passive or destructive manner.
ship itself becomes stronger. These relationship resources, such as commitment, satisfaction, intimacy, and love, can be drawn on in the future. Thus, the context of positive events seems central to relationship health.
The Current Investigation We tested our hypotheses with an observational study. Specifically, dating couples participated in four videotaped interactions. They each took turns sharing a recent positive event and a recent negative event. After each interaction, the discloser rated how understood, validated, and cared for (i.e., perceived responsiveness) he or she had felt during the interaction. Before their interactions, we assessed the well-being of the participants’ relationship using standard measures and the participants’ perceptions of how their partners typically respond to the participants’ capitalization attempts. We had three primary hypotheses: Hypothesis 1. The measure of how a partner typically reacts to positive event disclosures would predict perceived partner responsiveness during the positive event discussion but not during the negative event discussion (i.e., discriminant validity). Hypothesis 2. Ratings of responsiveness in the positive event discussion would be a better predictor of relationship well-being than ratings of responsiveness in the negative event disclosure. Hypothesis 3. Our new behavioral coding system designed to assess active versus passive and constructive versus destructive behavior of the partner would predict the discloser’s reports of perceived responsiveness in the positive event discussion.
Method
The Context of Positive Events and Positive Emotions When positive events occur, individuals are likely to experience positive emotions. For example, previous research has shown that when rewarding events occur, people experience an increase in positive affect, but negative affect remains unchanged (e.g., Gable, Reis, & Elliot, 2000). Fredrickson’s (1998, 2001) broaden-andbuild functional model of positive emotions posits that positive emotions broaden an individual’s scope of cognition, attention, and action and build the individual’s physical, intellectual, and social resources. Isen and colleagues’ (Isen & Daubman, 1984; Isen, Daubman, & Nowicki, 1987) research provided early evidence for the broadening aspects of positive emotions such that induced positive emotions led to more flexible and creative processing. More recently, Fredrickson and Joiner (2002) found that broader and more flexible coping was associated with increased positive emotional experiences. Although most empirical investigations have focused on broadening functions, a recent study of post–September 11th resilience found that experiencing some positive emotions (such as interest, hope, contentment) following the terrorist attacks led to increases in psychological resources (such as optimism, life satisfaction; Fredrickson, Tugade, Waugh, & Larkin, 2003). We suggest that capitalization presents opportunities to build social resources. That is, when an individual discloses a positive event to his or her partner, and the partner responds in an active– constructive manner, both partners experience positive emotions, and the relation-
Participants Seventy-nine couples were recruited via advertisements in the campus newspaper and flyers posted throughout the campus of a large public university. The recruitment materials specified that participation required couples to have been dating exclusively for a minimum of 6 months. Although advertisements did not specify sexual orientation, only heterosexual couples responded to the ads and participated in the study. On average, participants had been dating 25.1 months (SD ⫽ 22.3 months, Mdn ⫽ 18, range ⫽ 6 –98). The mean age of the women was 21.3 years (SD ⫽ 2.69 years) and the mean age of the men was 22.2 years (SD ⫽ 2.80 years). Participants were of diverse ethnicity (41.1% White, 36.1% Asian/ Pacific Islander, 6.3% Hispanic, 5.1% African American, and 10.1% other or declined to answer) that reflected, roughly, the ethnic composition of the university community. Approximately one third of participants (38.0%) described themselves as full-time students, 13.9% were employed fulltime, 3.2% were unemployed, and the remaining participants split their time between school and employment. Forty-three percent of the couples were cohabitating; 3 couples were engaged. Couples received $50 for participation in the study.
General Laboratory Session Procedure Couples attended a single laboratory session that lasted approximately 1.5 hr. After a brief introduction to the study and completion of consent procedures, couples were led into separate rooms to complete the packet of demographic, individual difference, and relationship measures. After completing the measures, couples were reunited and seated in two chairs angled
SUPPORTIVE RESPONSES TO POSITIVE EVENT DISCLOSURES to face each other. Two small cameras were mounted on the wall approximately 4 feet above the ground, with one camera pointed at each participant at an angle to allow for full frontal recording. The cameras were visible to the couple and captured an image of the participants from the top of their heads to their feet. The cameras were controlled by experimenters in an adjacent control room who could see and hear the activities in the experiment room, adjust the cameras to follow participants if they shifted positions in their chairs, and communicate with couples via an intercom. Couples then participated in seven separate interactions, each lasting a maximum of 5 min. After each interaction, they completed brief questionnaires independently; we used appropriate measures to ensure particpants’ confidentiality.
Measures in Initial Packet Perceived Responses to Capitalization Attempts Scale (PRCA; Gable et al., 2004). Participants completed the PRCA scale, a recently developed and validated 12-item scale measuring perceptions of a partner’s typical response to the sharing of positive events. Participants rated each item using the stem, “When I tell my partner about something good that has happened to me . . . ,” and a 7-point scale on which 1 is labeled as not at all true and 7 is labeled as very true. The scale includes three active– constructive responses (e.g., “I sometimes get the sense that my partner is even more happy and excited than I am”), three passive– constructive responses, (e.g. “My partner tries not to make a big deal out of it but is happy for me”), three active– destructive responses (e.g., “My partner reminds me that most good things have their bad aspects as well”), and three passive– destructive responses (e.g., “My partner often seems disinterested”). Previous research has shown that active and constructive responses are positively correlated with relationship well-being, whereas the remaining three types are negatively correlated with relationship wellbeing (Gable et al., 2004). Thus, a single composite capitalization score was created by subtracting the mean of the passive– constructive, active– destructive, and passive– destructive scales from the active– constructive scales. Higher numbers indicated more active– constructive and less passive– destructive responses. The composite scores ranged from ⫺2.22 to 6.00 for men and from ⫺3.22 to 5.67 for women, and the scale showed good reliability for both men (␣ ⫽ .84) and women (␣ ⫽ .81). Relationship quality measures. Participants completed three measures of the quality of their relationship with their partners. They completed the seven-item (e.g., “How good is your relationship compared with most?”) Relationship Satisfaction Scale (Hendrick, 1988). Statements were rated on a 7-point scale ranging from 1 (low satisfaction/never/not at all/none) to 7 (very high satisfaction/very often/a great deal/very many), and reliabilities were good for both men (␣⫽ .90) and women (␣⫽ .92). Participants also completed a seven-item commitment measure (e.g., “I want our relationship to last for a very long time.”) from the Investment Model Scale (Rusbult, Martz, & Agnew, 1998) on a scale ranging from 1 (not at all true/never true) to 7 (very true/true all of the time), and reliabilities were good for both men (␣⫽ .91) and women (␣⫽ .92). Passionate love was also measured using seven items (e.g., “I have an endless appetite for affection from my partner” from the Passionate Love Scale (Hatfield & Sprecher, 1986) on a scale ranging from 1 (not al all true of our relationship/never true) to 7 (very true/true all of the time), and reliabilities were good for both men (␣⫽ .85) and women (␣⫽ .83). Principal component analyses were computed separately for men and women, and all three measures loaded on a single factor for both sexes (loadings ⫽ .93, .89, and .86 for men and .94, .91, and .86 for women, respectively). The single factor accounted for 79.8% of the variance in male responses and 81.3% of the variance in female responses. Thus, a single composite score was calculated by averaging the three measures into one score, Time 1 relationship well-being (RWB). Individual difference measures. For discriminant validity purposes, we included two sets of individual difference measures variables that would theoretically be predicted to influence partners’ active– constructive behav-
907
ior after the disclosure of a positive event: attachment dimensions and the Big Five personality variables. Attachment. Participants completed the Experiences in Close Relationships Scale (Brennan, Clark, & Shaver, 1998). This standard 36-item attachment measure assesses the two primary dimensions of attachment: avoidance (␣ ⫽ .86 for men and ␣ ⫽ .91 for women) and anxiety (␣⫽ .89 for men and ␣ ⫽.91 for women). Participants responded to each statement on a scale ranging from 1 (strongly disagree) to 7 (strongly agree). Personality. Personality was measured with the Big Five Inventory (BFI; John, Donahue, & Kentle, 1991). This 44-item measure assesses five major dimensions of personality. Participants are asked to rate the degree to which they agree or disagree with each of the statements on a scale ranging from 1 (disagree strongly) to 5 (agree strongly). These dimensions are extraversion (8 items measuring qualities such as sociability), agreeableness (9 items measuring qualities such as helpfulness and unselfishness), conscientiousness (9 items measuring qualities related to reliability), neuroticism (8 items measuring predisposition to anxiety), and openness to experience (10 items measuring qualities such as curiosity about new things). The scales showed good reliability for both men and women in the present sample: alphas for men were .82, .80, .79, .81, and .79 and alphas for women were .87, .78, .80, .81, and .79, respectively. The BFI scales have shown convergent validity with other measures of personality and predict meaningful life outcomes (John & Srivastava, 1999).
Measures and Procedure for Videotaped Interactions Participants completed seven videotaped interactions. In the first interaction, couples were asked to describe their first date in an unstructured discussion for up to 5 min. This interaction was designed to allow couples to become acquainted with the cameras and videotape task and is not discussed further. In the last two of the seven interactions, each member of the couple took a turn describing their favorite characteristic of the partner. This interaction was designed so that all couples ended the laboratory session on a positive note and is not discussed further. The four interactions in the middle of the laboratory session are the focus of the current study. In Interactions 2–5, each member of the couple took turns discussing a recent personal negative event and a recent personal positive event. The order of these discussions was randomly assigned and counterbalanced such that in 21 couples, the woman discussed her positive event first; in 21 couples, the man discussed his positive event first; in 20 couples, the woman discussed her negative event first; and in 17 couples, the man discussed his negative event first. The next discussion was the other partner discussing his or her event in the same category. Then, the participants took turns discussing the other event, with the same individual who went first in the first round going first in the second round. We examined mean differences of all the interaction variables (see below) and found no significant mean differences that were based on order (all ps ⬎ .05). Pre-event instruction and measures. Before completing the personal positive event discussion, participants were given the following instructions: In these next set of interactions, we are interested in how couples discuss positive things that happen to them. We are not interested in how couples discuss positive things that happen to the both of you, such as going on vacation, or something that the other has done for you. Rather, we are interested in how couples talk about the positive events that one member has in his or her life. We would like you to choose some recent positive event from your life. Your positive event may be something that happened to you recently or in the past that continues to make you happy, something going on now, or something you anticipate will happen in the future. Examples of positive events would be receiving a good grade in a class, a work promotion, or a financial windfall; being offered a job, internship, or scholarship; being accepted into graduate school; or even being given a compliment from someone other than your partner. Please pick something
908
GABLE, GONZAGA, AND STRACHMAN that has been on your mind recently, no matter how big or small you may think it is.
discussion is about your partner’s concern, you can respond to, add to, or talk about as much or as little as you would under normal circumstances.
Each participant was then given a form to complete on which he or she was asked to briefly describe the positive event and to rate how important the event was on a 7-point scale ranging from 1 (not very important ) to 7 (extremely important). The average positive event importance rating was 5.54 (SD ⫽ 1.34) for women and 5.75 (SD ⫽ 1.49) for men. Personal positive events were content-coded by trained raters, whose rate of agreement was 100%. The positive events discussed were academic accomplishments (37.1%), work or financial success (29.8%), family and friends (8.6%), personal travel (4.0%), receiving compliments (3.3%), athletic accomplishments (3.3%), and other accomplishments and miscellaneous positive events (e.g., health, housing, receiving gifts; 13.9%). We also asked participants the degree to which they had previously discussed this event with their partner on a 7-point scale ranging from 1 (never) to 4 (a fair amount) to 7 (a great deal). All but 4 participants (1 man, 3 women) had previously discussed their positive event with their partner, and 85% of participants reported discussing it a fair amount or more. The average rating on the previous discussion item was 5.09 (SD ⫽ 1.67) for women and 5.75 (SD ⫽ 1.49) for men. Each participant then took a turn discussing his or her positive event (see description below). Before completing the personal negative event discussion, participants were given the following instructions:
The couple then discussed the man’s negative event for up to 5 min. Both members of the couple completed a postinteraction form (see below). Then, the couple discussed the woman’s negative event for up to 5 min, and both members of the couple completed a postinteraction form (see below). The procedure was then repeated for the positive events, with identical instructions given. Postevent discussion responsiveness measure. After each discussion, the disclosing participants independently completed a measure of how responsive their partner had been during the interaction.2 Specifically, they rated 10 items from Reis’s (2003) 18-item Responsiveness Scale on a scale ranging from 1 (not at all) to 5 (very much). The measure was designed to assess how understood, validated, and cared for individuals feel when interacting with their intimate partners, and it is theoretically modeled on Reis and Shaver’s (1988) Intimacy Model. The postevent form was completed independently by each partner. To ensure confidentiality, we instructed the couples to turn and angle away from each other and use clipboards. Experimenters observed the couples during this time via the video–audio equipment to verify that forms were completed independently. Each time a form was completed, participants placed it inside a covered box next to their chairs to ensure continued confidentiality. The items were as follows: “My partner . . . saw the ‘real’ me; ‘got the facts right’ about me; focused on the ‘best side’ of me; was aware of what I was thinking and feeling; understood me; really listened to me; expressed liking and encouragement for me; valued my abilities and opinions; respected me; was responsive to my needs.” The mean responsiveness score following the negative event was 4.33 (SD ⫽ 0.58) for men (␣ ⫽ .94) and 4.33 (SD ⫽ 0.73) for women (␣⫽ .89). The mean responsiveness score following the positive event was 4.20 (SD ⫽ 0.82) for men (␣ ⫽ .94) and 4.37 (SD ⫽ 0.67) for women (␣⫽ .95).3 Coding of the responding partner’s behavior in positive event interactions. The personal positive event discussions were coded for how active and constructive the responding partner was during the interaction in which the disclosing partner discussed his or her positive event (i.e., the partner listening to the positive event). Judges were given the following instructions on coding how passive or active the respondent was on a scale ranging from 1 (extremely passive) to 7 (extremely active):
In these next set of interactions, we are interested in how couples discuss their personal concerns. We are not interested in the concerns you may have about your relationship or your partner, but rather we are interested in concerns that affect one of you. We would like you to choose some current problem, concern, or stressor you are facing in your life. This may be something that happened before but continues to bother you, something going on now, or something you anticipate will happen in the future. Some examples could be a recent argument with a friend or family member, a grade in class, work or financial problems, or personal illness. Please pick something that has been on your mind recently, no matter how big or small you may think it is. Each participant was then given a form on which he or she was asked to briefly describe the negative event and to rate how important the event was on a 7-point scale ranging from 1 (not very important ) to 7 (extremely important). The average negative event importance rating was 5.86 (SD ⫽ 1.44) for women and 5.82 (SD ⫽ 1.29) for men. Negative events focused on work or financial issues (41.4%), academic difficulties (27.6%), family or friends (21.7%), personal illness (3.3%), or other or general concerns (5.9%). We also asked participants the degree to which they had previously discussed this event with their partner on a 7-point scale ranging from 1 (never) to 4 (a fair amount) to 7 (a great deal). All but 4 participants (2 men, 2 women) had previously discussed their concern with their partner, and 84% of participants reported discussing it a fair amount or more. The average rating on the previous discussion item was 5.38 (SD ⫽ 1.60) for women and 5.09 (SD ⫽ 1.67) for men. Each participant then took a turn discussing his or her negative event (see the description below). Event discussions. As noted, the order of discussion of positive and negative events was counterbalanced. For example, in the negative event– man first discussion, participants were given the following instructions after completing the pre–negative event discussion form: In the first interaction, the discussion will be about (man’s name)’s concern. When you have finished with that discussion, you will complete another short form, and then you will repeat the process, except the discussion will center on the (woman’s name)’s concern. Again, while you are interacting, please feel free to talk about anything related to the personal concern. Some suggestions for the person who has the concern would be to discuss the circumstances surrounding the concern, how you feel and what you think about the concern, and any other details or issues that you think are important. When the
Rate the activity, both verbal and nonverbal, on the scale provided. This rating should be devoid of content, made irrespective of positivity or negativity of the interactions. Look for head nodding/shaking, emotional displays, animation, hand gestures, laughing or scoffing, questions, or statements. Judges were given the following instructions when coding how destructive or constructive the respondents were on a scale ranging from 1 (extremely destructive) to 7 (extremely constructive): Rate the valence of the verbal and nonverbal displays using the scale provided. Destructive units include negative suggestions and questions, turning the discussion away from the target, and displays of negative emotion. Constructive units include elaboration of positives, linking to other positive events, smiling, laughing. The eight judges completed a set of 10 cases and then discussed their ratings in a group (which also included authors Shelly L. Gable and Gian C. Gonzaga) and came to a consensus on the rating to be given to each
2
The nondisclosing participants also completed a form at this time, the contents of which are not the focus of this article. 3 Six couples mistakenly received the incorrect version of the post– positive event discussion form. Therefore, the sample size for the post– positive event discussion responsiveness ratings was 73 men and 73 women.
SUPPORTIVE RESPONSES TO POSITIVE EVENT DISCLOSURES case. To increase the independence of ratings and decrease direct comparisons between men and women, raters coded either the male responders or the female responders, and they either coded the passive–active dimension or the destructive– constructive dimension. Thus, two raters coded the passive–active dimension of the men responding to the women; two raters coded the destructive– constructive dimension for men responding to the women; two raters coded the passive–active dimension of the women responding to the men, and two raters coded the destructive– constructive dimension for women responding to the men. Interrater reliability was good. On the passive–active dimension, the intraclass correlations were .83 for ratings of the female responder and .87 for ratings of the male responder; on the destructive– constructive dimension, the intraclass correlations were .68 for ratings of the female responder and .70 for ratings of the male responder. The scores of two independent judges of each target were averaged to create one passive–active dimension and one destructive– constructive score. These two dimensions were uncorrelated for both men and women, r(78) ⫽ .06 and r(77) ⫽ .02, respectively; ps ⬎ .60. As stated earlier, only responses that were active and constructive have been positively correlated to relationship quality in previous research, whereas passive and destructive responses have been negatively correlated with relationship quality. Therefore, a single “observed partner reactions” score was created by adding the two codes (new range ⫽ 2–14), higher scores indicated more active or constructive and less passive or destructive responding. The average observed partner reaction score was 9.57 (SD ⫽ 1.67) for women’s behavior during men’s positive event disclosure and was 9.41 (SD ⫽ 1.98) for men’s behavior during women’s positive event disclosure.4 It should be noted that we did not code partners’ behavior during the negative event disclosure. We had no reason to predict that active and constructive responses to negative event disclosures would be positively related to relationship outcomes. In fact, an enthusiastic response to a discussion of recent problems is likely to have negative consequences for the person and the relationship. There are existing systems for coding social support provisions (e.g., Barbee & Cunningham, 1995). Understanding capitalization behaviors was the focus of the current research, and comparisons of an existing social support behavioral coding schemes to our own would have been difficult, thus reactions to negative event disclosures were not examined. We refer the reader instead to existing literature (e.g., Collins & Feeney, 2000; Simpson, Rholes, & Nelligan, 1992) that does examine social support provision with observational methods.
Follow-Up Assessment Eight weeks after their participation in the study, both members of the couple were independently sent follow-up relationship questionnaires. Of the 158 people who participated in the laboratory portion of the study, 88 individuals (38 men and 50 women) completed the follow-up measures. In exchange for returning their follow-up assessment, participants were mailed a $5 gift certificate to the campus store. At least 1 member of 4 additional couples (8 individuals) indicated that they had broken up by the time of the follow-up and therefore could not complete the follow-up measures (they were still mailed their compensation), and the remaining 62 participants (37 men and 25 women) did not respond at all to the follow-up survey. To determine whether participants who responded to the follow-up survey differed from those who did not (excluding the 8 individuals who had broken up), we conducted a series of t tests for independent groups on the Time 1 measures. As seen in Table 1, the two groups did not differ significantly on the Time 1 relationship quality variables, the postinteraction ratings, observer ratings of partners’ behavior, attachment, or length of time dating. The follow-up questionnaires included the commitment, satisfaction, and passionate love measures described above. A second set of principal component analyses was computed, and all three measures again loaded on one factor (loadings ⫽ .94, .83, and .85 for men and .94, .93, and .93 for women, respectively). The single factor accounted for 76.6% of the vari-
909
ance in male responses and 87.2% of the variance in the female responses. Thus, a single composite score was calculated by averaging the three measures into one score: Time 2 RWB.
Results Data Analysis Strategy Our data violated assumptions of independence because both members of the romantic couple participated in all interactions. More important, men’s postinteraction reports were taken from two separate videotaped interaction sessions: the man sharing his positive event and the man sharing his negative event. Women’s postinteraction reports were taken from two additional videotaped interaction sessions: the woman sharing her positive event and the woman sharing her negative event. Because the data originated from different interactions in which participants were playing different roles (i.e., discloser or responder) and discussing different events, the most appropriate data analytic strategy was to analyze the man-as-discloser interactions and the woman-as-discloser interactions separately. This strategy had the advantage of being the most conservative in terms of avoiding problems associated with nonindependent data and allowing the examination of male and female patterns of associations separately.
PRCA and Relationship Quality Measures Before testing our major hypotheses, we thought it was important to replicate the findings of Gable et al. (2004) by testing whether PRCA ratings predicted relationship RWB at Time 1 and Time 2. As we described above, before participants engaged in the videotaped interactions, they completed the 12-item PRCA measure of how their partner typically responded to news of the participants’ positive events as well as measures of relationship well-being. Indeed, the PRCA scores were positively correlated with the composite relationship well-being measure at Time 1, r(79) ⫽ .41 for men and r(79) ⫽ .41 for women, ps ⬍ .001, such that the more active and constructive (and less passive or destructive) participants rated their partners’ typical response to positive event sharing, the more commitment, satisfaction, and passionate love they also reported feeling. A similar finding emerged when we predicted Time 2 RWB, r(37) ⫽ .53 for men, p ⬍ .01, and r(50) ⫽ .27 for women, p ⬍ .06. Finally, we used a multiple regression analysis to predict change in Time 2 RWB, relative to Time 1 RWB, by entering Time 1 RWB in the first step and PRCA in the second step. For men, the addition of PRCA was significant, ⌬R2 ⫽ .06, F(1, 34) ⫽ 4.84, p ⬍ .05, PRCA  ⫽ .27, p ⬍ .05. For women, the addition of the PRCA in Step 2 was not significant ⌬R2 ⫽ .01, F(1, 47) ⫽ 1.92, p ⫽ ns, PRCA  ⫽ ⫺.13, p ⫽ ns. Examination of change scores shows that on average Time 2 RWB 4 One couple, although fluent in English, reported that they typically spoke to each other in Korean when at home. They requested that their videotaped interactions also be in Korean. Therefore, these two interactions were not coded by our raters, who were not fluent in Korean. We experienced a technical difficulty (loss of sound) during one woman’s positive event disclosure and thus could not code the male partner’s behavior in this interaction. Therefore, the final sample size was 78 ratings of women’s behavior during men’s disclosures and 77 ratings of men’s behavior during women’s disclosures.
GABLE, GONZAGA, AND STRACHMAN
910
Table 1 Comparison of Means, Standard Deviations, and t Tests of Scores for Couples Who Completed the Follow-Up Measures Versus Couples Who Did Not Completed Time 1 measure
M
Not Completed
SD
M
SD
t(73)
p
5.73 2.28 4.21 4.31 3.50 2.40 22.1
1.08 2.21 0.78 0.62 1.08 0.72 19.6
⫺0.27 0.61 ⫺0.47 ⫺0.33 0.51 0.01 ⫺1.31
.79 .54 .64 .75 .61 .97 .19
5.74 2.40 4.37 4.14 3.85 2.63 25.3
0.95 2.14 0.57 0.75 1.40 0.79 24.3
⫺1.07 ⫺0.12 ⫺0.01 ⫺1.63 0.08 1.05 ⫺0.02
.29 .91 .99 .11 .94 .30 .98
Men RWB PRCA Positive event responsiveness Negative event responsiveness Attachment-anxiety Attachment-avoidance Months dating
5.79 1.98 4.30 4.36 3.38 2.40 28.9
0.82 1.94 0.84 0.52 1.05 0.80 24.8 Women
RWB PRCA Positive event responsiveness Negative event responsiveness Attachment-anxiety Attachment-avoidance Months dating
5.98 2.45 4.37 4.44 3.82 2.39 25.5
0.88 1.20 0.73 0.72 0.97 1.00 22.5
Note. Among the men, 38 completed and 37 did not complete the follow-up measures; among the women, 50 completed and 25 did not complete the follow-up measures. Comparison excludes 4 couples (8 participants) who indicated they had broken up at Time 2. Degrees of freedom for positive event responsiveness ⫽ 67. RWB ⫽ relationship well-being score; PRCA ⫽ Perceived Responses to Capitalization Attempts Scale.
decreased 0.23 points for men and 0.22 points for women. Thus, the appropriate interpretation of the significant effect of PRCA on men’s Time 2 RWB scores, controlling for Time 1, is that men with higher PRCA scores decreased in RWB less than those with lower PRCA scores.
PRCA and Postdiscussion Perceived Partner Responsiveness To test Hypothesis 1, we examined the discriminant validity of the PRCA measure. Specifically, participants who reported that their partners typically responded actively and constructively (and not passively or destructively) should have felt more understood, validated, and cared for following the interaction in which they discussed their own positive event but not necessarily following the discussion of their own negative event. Indeed, participants’ feelings of responsiveness following discussion of their positive event were positively correlated with their ratings of their partners’ typical reactions using the PRCA, r(73) ⫽ .41 for men and r(73) ⫽ .31, for women, ps ⬍ .01. Responsiveness ratings following the positive event discussion and responsiveness ratings following the negative event discussion were positively correlated, r(73) ⫽ .68 for men and r(73) ⫽ .72, for women, ps ⬍ .001. That is, the more understood, validated, and cared for participants felt after disclosing their positive event, the more understood, validated, and cared for they felt after disclosing their negative event.5 Not surprisingly (given the high correlation between the two postdiscussion responsiveness measures), responsiveness ratings after the negative event were positively correlated with PRCA ratings of their partners’ typical reactions to positive event disclosure, r(79) ⫽ .37 for men, p ⬍ .01, and r(79) ⫽ .22, for women, p ⬍ .06.
However, the critical test of discriminant validity for the PRCA measure, and thus of Hypothesis 1, was the independent variability that is accounted for when both ratings of responsiveness were entered simultaneously into a multiple regression equation. As seen in Table 2, responsiveness ratings after the negative event discussion were no longer significant predictors of the PRCA for either men or women,  ⫽ .11 for men, p ⫽ .47, and  ⫽ ⫺.02 for women, p ⫽ .92. However, the association between post–positive event discussion responsiveness ratings and the PRCA remained significant for men,  ⫽ .34, p ⬍ .05, and marginal for women,  ⫽ .32, p ⫽ .057. Thus, when controlling for the relationship between the two postevent discussion responsiveness ratings, only the positive event responsiveness ratings predicted the PRCA measures, showing that the PRCA is assessing variance uniquely associated with a partner’s ability to effectively respond to capitalization attempts. A question related to Hypothesis 1 was whether the PRCA was associated with the actual behavior of partners during disclosures of positive events. To test this, we correlated outsider observer codes of the partner’s behavior during participants’ disclosure of positive events (higher numbers indicate more active and constructive behavior) with the PRCA measures. Additional analyses were also done in which we controlled for the participants’ rating of the 5
An additional regression was performed in which postevent responsiveness and the rating of how much they had discussed the event before the laboratory session were entered as predictors of the PRCA score. Prior discussion of an event was not a significant predictor of the PRCA score for men or women, ps ⬎ .20, and postevent responsiveness remained a significant predictor of PRCA for men and women, ps ⬍ .05.
SUPPORTIVE RESPONSES TO POSITIVE EVENT DISCLOSURES
importance of the event because we reasoned that the partner’s response may have been related to the importance of the event. Men’s ratings of their female partner on the PRCA were significantly correlated with observer codes of her behavior during the discussion of the man’s event, r(78) ⫽ .29, p ⬍ .05, and when controlling for male ratings of the importance of the event discussed, the partial correlation remained significant, pr(75) ⫽ .29, p ⬍ .05. Women’s ratings of their male partner on the PRCA were also positively correlated with observer ratings of his behavior during the discussion of the woman’s event, r(77) ⫽ .18; however, this was not significant ( p ⫽ .12). Although when women’s ratings of the importance of the event were controlled for, the partial correlation was marginally significant, pr(75) ⫽ .20, p ⬍ .09 (interactions with event importance are explored more fully below). Thus, there is evidence for both men and women (marginally) that the more active and constructive (and less passive or destructive) participants described their partners’ typical reaction to their good fortune on the 12-item PRCA, the more active and constructive their partners actually behaved in the laboratory interaction.
Postinteraction Responsiveness and RWB The next set of analyses was designed to test Hypothesis 2 via the relationships among postinteraction responsiveness ratings and the RWB composite measure. First, post–positive event responsiveness ratings and post–negative event responsiveness ratings were entered simultaneously as predictors of the Time 1 RWB measure in a multiple regression equation. The results are presented in Table 3. For men, only the positive event responsiveness ratings were a significant predictor of Time 1 RWB, but for women, both positive event and negative event responsiveness ratings were significant predictors of Time 1 RWB. This analysis was repeated using Time 2 RWB as the outcome measure (see Table 3). For men, neither postinteraction responsiveness rating was a unique significant predictor of Time 2 RWB; however, both predictors jointly accounted for a significant portion of the variability in the overall model. For women, only postpositive event responsiveness ratings were significant predictors of Time 2 RWB.6 We then examined whether responsiveness ratings predicted change in Time 2 RWB, controlling for Time 1. However, when Time 1 RWB was entered into the equation, postinteraction responsiveness ratings (positive or negative) were no longer signif-
Table 2 Postinteraction Responsiveness Ratings Predicting Perceived Responses to Capitalization Attempts Scale (PRCA) PRCA score Predictor
Men
Women
Positive event responsiveness Negative event responsiveness Total R2 for model
.34** .11 .18**
.32* ⫺.02 .09**
Note. n ⫽ 73 for each group. Numbers are standardized regression weights (s). Predictor variables were entered simultaneously. * p ⬍ .06. ** p ⬍ .05.
911
Table 3 Positive and Negative Event Disclosure Postinteraction Responsiveness Ratings Predicting Relationship Well-Being at Time 1 and Time 2 Relationship well-being Men Predictor Positive event responsiveness Negative event responsiveness R2 for responsiveness predictors
Women
Time 1 Time 2 Time 1 Time 2 .54*** .08 .36***
.19 .34 .25**
.35** .34** .40***
.62*** .05 .43***
Note. Relationship well-being score ⫽ a composite of scores on scales of commitment, satisfaction, and passionate love. Time 1 n ⫽ 73 for each group, and Time 2 n ⫽ 34 men and 45 women. Numbers are standardized regression weights (s). Predictor variables were entered simultaneously. ** p ⬍ .05 *** p ⬍ .01.
icant predictors of Time 2 RWB for either men or women. These results indicate that feeling understood, validated, and cared for following a positive event disclosure is more strongly and consistently associated with RWB than ratings of responsiveness following a negative event disclosure.
Observer Ratings of Partner’s Active–Constructive Behavior in Response to Participants’ Disclosure of Positive Event Hypothesis 3 predicted that the behavior of participants’ partners during the disclosure of the positive event, as coded by the outside observers, would be associated with participants’ feelings of responsiveness directly following the interaction. To examine this question, we conducted hierarchical multiple regression equations separately for men and women in which the observed ratings of partners’ reactions predicted responsiveness ratings of the participant. As seen in the first two rows of Table 4, in couples in which the woman’s behavior was rated as more active and constructive by outside observers, her male partner’s feelings of responsiveness were significantly higher,  ⫽ .35, R2 ⫽ .12, p ⬍ .01. In couples in which the man’s behavior was rated as more active and constructive by outside observers, his female partner’s feelings of responsiveness were significantly higher,  ⫽ .35, R2 ⫽ .12, p ⬍ .01.7 6
All the analyses reported in Table 3 were rerun, entering ratings of how much participants had discussed their events before the laboratory sessions into the regression. Neither prior positive event nor personal concern discussion ratings predicted Time 1 RWB for men or women; prior personal concern discussion did not predict Time 2 RWB for men or women; and prior positive event discussion ratings was a significant predictor of women’s (but not of men’s) Time 2 RWB ( ⫽ .29, p ⬍ .05). Most important, all significant effects in Table 3 remained significant when controlling for prior event discussion. 7 Prior positive event discussion was not a significant predictor of men’s or women’s behavior, as coded by observers, ps ⬎ .45. Moreover, observer codes of behavior remained significant predictors of responsiveness ratings when we controlled for prior event discussion.
GABLE, GONZAGA, AND STRACHMAN
Table 4 Observer Ratings of Partners’ Active and Constructive Reactions and Importance of Event Predicting Participants’ Perceived Responsiveness Perceived responsiveness Predictor Step 1 Observed partner reactions R2 for model Step 2 Importance of event discussed ⌬R2 for model Step 3 Interaction of reactions and importance ⌬R2 for model Total R2 for model
Men
Women
.35*** .12***
.35*** .12***
⫺.05 .002 ⫺.12 .02 .14**
⫺.03 .001 .27** .06** .19***
Note. n ⫽ 72 men and 71 women for each group. Numbers are standardized regression weights (s). ** p ⬍ .05. *** p ⬍ .01.
Men 0.8 Responsiveness
To further explore Hypothesis 3, we examined the relationship between the importance of the event disclosed because this may have been an important factor moderating both how the responder behaved and how the responder’s behavior was interpreted. Thus, in Step 2 of the regression equation, we entered the importance rating that the disclosing participants gave their own event before discussing it in the interaction, but it was not a significant predictor of responsiveness ratings for either men or women (see Table 4). For each person, an interaction term was created by multiplying the z scores of event importance and observer codes, and this score was entered in Step 3 of the equation. Also as seen in Table 4, the interaction term was not significant for men, nor did R2 change significantly. However, for women, the interaction term was significant,  ⫽ .27, p ⬍ .05, and this was a significant change, ⌬R2 ⫽ .06, p ⬍ .05. To interpret the interaction, we calculated predicted scores for men and women 1 standard deviation above and below the mean on importance ratings of events and on the observed active– constructive behavior of their partners using the  weights from the final step of the regression equation. These scores are shown in Figure 1. For men, there was no significant interaction; only the active– constructive behavior of their partner predicted feelings of responsiveness. However, for women, there was a significant interaction, reflecting that women felt most responded to when their partners were active and constructive in discussing their important events and least responded to when their partners were not active and constructive in discussing their important events. Thus, the man’s behavior was particularly influential when the woman’s event was important, as per her own ratings of the event. Finally, to further investigate discriminant validity, we examined the association between our observer ratings of partners’ behavior during the positive event disclosure and self-ratings of responsiveness after the negative event disclosure. Neither men’s nor women’s ratings of responsiveness after their negative event discussions were significantly correlated with their partners’ behavior during the positive event discussions, r(78) ⫽ .08, p ⬎ .50, and r(76) ⫽ .118, p ⬎ .35, respectively. This result indicates that active– constructive responses to positive event disclosures are
0.6 0.4 High Event Importance Low Event Importance
0.2 -0 -0.2 -0.4 -0.6 -0.8
High Low Partner’s Active/Constructive Behavior
Women 0.8 Responsiveness
912
0.6 0.4 0.2 -0 -0.2 -0.4
High Event Importance Low Event Importance
-0.6 -0.8 High Low Partner’s Active/Constructive Behavior
Figure 1. Predicted ratings of partners’ responsiveness during positive event discussion by observer ratings of active– constructive behavior and importance of event.
uniquely related to feeling understood, validated, and cared for in the capitalization context, not in the social support context and that partners who respond actively and constructively to capitalization attempts are not necessarily the same partners who provide effective social support.
Additional Analyses Predictors of partners’ behavior during participants’ disclosures of positive events. We examined two sets of individual difference variables that were theoretically predicted to influence active– constructive behavior following the disclosure of a positive event: attachment dimensions and the Big Five personality variables. For the first set of analyses, observer codes of participants’ behavior during their partners’ positive event disclosure were correlated with their own attachment anxiety and attachment avoidance scores. For men, anxiety was not a significant predictor of behavior, r(76) ⫽ ⫺.05, p ⫽ .65, and avoidance was marginally associated with behavior,  ⫽ ⫺.22, p ⬍ .06, such that men who scored high on avoidance were rated as less active and constructive by our observers. For women, neither the avoidance nor the anxiety dimensions significantly correlated with their behavior, rs (78) ⫽ ⫺.06 and .01, respectively, ps ⬎ .60. Thus, it appears that the attachment dimensions are not consistent or strong predictors of behavior during the positive event discussion.
SUPPORTIVE RESPONSES TO POSITIVE EVENT DISCLOSURES
Men 5 Responsiveness
Correlations among the Big Five and participants’ behavior during their partners’ positive event disclosure were also calculated. For men, agreeableness was correlated with their behavior, r(78) ⫽ .23, p ⬍ .05, such that more agreeable men were rated as more active and constructive responders by our coders. The remaining four personality dimensions were not correlated with behavior, all ps ⬎ .25. For women, none of the correlations with personality factors and behavior reached traditional significance levels. However, agreeableness, r(76) ⫽ .19, p ⬍ .10, and neuroticism, r(76) ⫽ ⫺.20, p ⬍ .08, were marginally associated with behavior. Overall, personality was not strongly correlated with behavior during partners’ positive event disclosures. Break-ups at Time 2. Four couples broke up (as reported by at least 1 member of the couple); both members of 34 couples completed measures and 1 member of an additional 20 couples completed the follow-up (4 men, 16 women), for a total of 88 individuals who completed the follow-up. Neither member of the remaining 21 couples returned the follow-up survey. Therefore, we had at least one follow-up measure completed by 54 of the 75 couples that were not verifiably broken up at Time 2. We used t tests to compare the Time 1 RWB score, the PRCA score, and the positive and negative event postinteraction responsiveness ratings of the 54 couples who we were certain remained together at the follow-up with those of the 4 couples who broke up. Because of the large difference in samples sizes, unequal variances were assumed, and separate analyses were performed for men and women. We note here that the results reported should be considered preliminary, given the small portion of the sample that broke up. The women from the dissolved couples did not differ from those in the intact couples on Time 1 RWB (Ms ⫽ 5.61 and 6.01, respectively), t(56) ⫽ 1.25, p ⬎ .25; post–positive event responsiveness ratings (Ms ⫽ 4.28 and 4.38, respectively), t(52) ⫽ 0.35, p ⬎ .75; or post–negative event responsiveness (Ms ⫽ 4.25 and 4.45, respectively), t(52) ⫽ 0.69, p ⬎ .50. However, the two groups were marginally different on the PRCA ratings (Ms ⫽ 1.56 vs. 2.55), t(56) ⫽ 2.23, p ⫽ .068, such that women in couples whose relationships dissolved before the follow-up rated their partners’ typical response to their sharing of positive events as less active and constructive and more passive and destructive than the women in couples who remained intact at the follow-up. The men from the dissolved couples did not differ from those in intact couples on Time 1 RWB (Ms ⫽ 4.70 and 5.79, respectively), t(56) ⫽ 1.73, p ⫽ .176, or post–negative event responsiveness (Ms ⫽ 4.13 and 4.34, respectively), t(56) ⫽ 0.51, p ⫽ .65. However, like the women, the two groups of men scored significantly differently on the PRCA scale (Ms ⫽ ⫺0.25 vs. 2.09), t(56) ⫽ 2.97, p ⬍ .05. Also interesting was the finding that the two groups significantly differed on the postpositive event discussion responsiveness ratings of their partners (Ms ⫽ 3.32 vs. 4.26), t(52) ⫽ 3.04, p ⬍ .05. That is, as shown in Figure 2, men in the couples who broke up before the follow-up felt less understood, validated, and cared for following their positive event discussion (but not following the negative event discussion) than those men in the couples who remained intact at Time 2. Moreover, women in the dissolved couples actually behaved less actively and constructively as rated by the outside observers during the male partner’s positive event discussion than those in the couples that remained intact at Time 2 (Ms ⫽ 8.75 and 9.62, respectively), t(56) ⫽ 2.44, p ⫽ .05.8
913
4 Broken Up Together
3 2 1 Post-positive Event Discussion
Post-negative Event Discussion
Figure 2. Difference between ratings of responsiveness during postevent discussions for men who were in couples who had broken up (n ⫽ 4) and who had remained together (n ⫽ 54) at Time 2.
The analyses of relationship stability over time should be interpreted with caution because the vast majority of couples remained together, and the couples who dissolved their relationships were a small group (4 couples). Nonetheless, the results suggest that if the two groups differed on any of the variables, they differed on the variables concerning how partners respond to the disclosure of positive events. These results, although preliminary, are supportive of our hypotheses concerning the context of positive events in relationship processes.
Discussion Data from the present study replicate and extend findings from previous research on capitalization. First, we found that a measure of partners’ typical responses to capitalization attempts (the PRCA) was correlated with both concurrent and future relationship commitment, satisfaction, and love and, for men, with change in these outcomes over time. These findings are consistent with previous research showing that when individuals rate their partners as active and constructive responders (and not as passive or destructive), they feel more intimacy and trust, are more satisfied with their relationships on a daily basis, report fewer daily conflicts, and engage in more fun and relaxing activities on a daily basis (Gable et al., 2004). Perhaps more important, we found both discriminant and convergent validity for the PRCA. First, the PRCA predicted the 8
Because there was a small number of couples who broke up, we were concerned our significant effects may have been driven by an outlier in the group. Thus, we carefully inspected the data from the four broken-up couples on the variables on which they differed from the intact couples: women’s and men’s PRCA scores, men’s post–positive event responsiveness ratings, and observer’s codes of women’s behavior during men’s positive event disclosure. For each variable, scores from all four broken-up couples were below the mean and median scores for the intact couples. Thus, it is unlikely that significant differences were driven by an outlier; all broken up participants scored similarly low on these variables.
914
GABLE, GONZAGA, AND STRACHMAN
participants’ feeling understood, validated, and cared for (i.e., responsiveness) during the positive event disclosure but not during the negative event disclosure. Thus, it seems that supporting the partner in times of stress is not necessarily the same thing as supporting the partner in times of good fortune. This argues against the idea that reacting supportively to a partner’s positive events is due solely to some overall ability to respond effectively and argues for the notion that the context of the interaction is important. Some partners may be particularly comfortable responding in a supportive manner when their partners talk about their positive events but less effective at traditional support (or vice versa). Future research might focus on determining which individual difference factors and relationship variables are associated with effective support in multiple contexts. Evidence for convergent validity of the PRCA came from our finding that outside observers’ codes of active and constructive responding during taped positive event discussions was predicted by the PRCA. This is important for two reasons. First, people seem capable of reporting on the typical behaviors of their partner during positive event disclosures. Their accuracy may indeed stem from repeated experiences. That is, people regularly seek out their partners when good things happen to them, and they likely have a large pool of experience on which to draw when describing their partners’ behavior (for example, on the PRCA). Second, these results suggest that during the videotaped conversations in the laboratory, individuals behaved in a manner consistent with their behavior in more natural settings. Thus, despite the artificial setting in which these data were collected, it appears that the procedures at the very least approximated in situ behavior. Of course, it is very difficult to obtain data in the context of real, everyday life; however, daily experience studies come closer than laboratory studies, and future research might assess partner responses (with both self- and partner ratings of behavior) in an event-contingent study to determine if the PRCA also predicts online reporting of behavior in more ordinary circumstances. Our data also showed that feeling understood, validated, and cared for during the positive event discussion was strongly and consistently associated with relationship well-being (satisfaction, commitment, and love). In fact, for men, only perceived responsiveness in the positive event discussion (and not in the negative event discussion) uniquely predicted relationship well-being. For women, perceived responsiveness in both the positive and negative event discussions predicted concurrent relationship well-being, but only positive event responsiveness predicted future relationship health. Thus, it is fair to say that positive event responsiveness was more strongly and consistently associated with relationship health than with perceived responsiveness in the social support interaction. This provides good evidence for our hypotheses regarding the special opportunities offered in the context of positive event disclosure. That is, compared with sharing a problem, fewer risks are involved in sharing a recent good event. The benefits of a partner’s active and constructive response could be garnered without the costs to self-worth inherent in seeking out help for a recent problem or stressor (like the invisible support reported by Bolger and colleagues, 2000). Moreover, providing social support to a distressed partner without his or her knowledge may be difficult (i.e., the majority of supportive responses are noticed; Bolger et al., 2000). Unlike traditional social support situations, capitalization responses are
actually more effective when they are transparent and obvious (i.e., active– constructive but not passive– constructive responses are perceived as supportive), and we would argue that the context of sharing positive events occurs more regularly than social support situations. For example, Gable and Haidt (2005) reported that daily reports of positive event occurrences outnumber negative event occurrences 5 to 1, a ratio that is similar to those found in other daily experience studies (e.g., Gable & Nezlek, 1998; Nezlek & Gable, 2001). In short, positive event disclosures offer all of the benefits that traditional social support exchanges confer without the same costs to self-esteem; they do not need to be concealed in order to be effective; and they are likely to take place far more often than negative event exchanges. To put it colloquially, they seem to offer a lot more bang for the buck. When individuals share positive events with their partners, they are sharing their strengths. Perceiving that the partner validates a strength could be particularly beneficial for one’s sense of selfworth. Murray and colleagues’ (e.g., Murray et al., 2000; Murray & Holmes, 1993) work has clearly shown that a positive sense of self is integral to feeling secure in a relationship. This also suggests that the capitalization context may be more important for some people than others—those with low self-esteem. In a series of recent studies, Murray and colleagues (2005) showed that when individuals’ own strengths were pointed out, they reported feeling more commitment to and more secure in their relationships, but this was only the case for those with chronically low self-esteem. Future research may wish to examine whether perceived partner responses to capitalization attempts are more closely tied to relationship well-being for individuals with low self-esteem. On a related note, some disclosures of positive events may be more difficult for the responder to provide supportive responses to than others. That is, it may be threatening to the responder’s self-worth if he or she did not have success in the particular domain of the discloser’s event (e.g., it may be hard to respond enthusiastically when a partner gets a promotion on the same day that you find out you did not get your promotion). Tesser et al.’s ( Tesser, Millar, & Moore, 1988; Beach & Tesser, 1995) self-evaluation maintenance model makes predictions of differential processes of reflection (“basking in reflected glory”) and comparison (i.e., envy) depending on the self-relevance of the event. Future research might focus on these more competitive situations. Most research on responding to one’s partner has focused on traditional social support or support in times of stress. However, our results suggest that feeling responded to when good things happen plays a vital role in relationship well-being. Thus, sharing positive events with one another provides prime opportunities for partners to offer support and convey understanding, validation, and caring. Other contexts are likely to also provide this opportunity, such as Feeney’s (2004) recent work showing that when partners are responsive to each other’s expressions of personal goals (e.g., career promotion plans, losing 5 pounds), they experience greater self-efficacy and self-worth. Thus, responding effectively to personal goal disclosures may be another mechanism for building social resources through the dyadic regulation of positive emotions. Our results are also highly consistent with Fredrickson’s (1998, 2001) broaden-and-build functional theory of positive emotions. Specifically, we believe that capitalization attempts and the responses to them build relationship resources. The resources take the form of increased intimacy, satisfaction, love, and commit-
SUPPORTIVE RESPONSES TO POSITIVE EVENT DISCLOSURES
ment, which can then be called on in times of stress and uncertainty. Indeed, it is quite possible that capitalization exchanges serve as a primary mechanism through which traditional social support networks are built. Moreover, these resources may lead to an overall sense of perceived support from the partner. Not only can capitalization exchanges provide building opportunities, they can also provide “safe” opportunities to test the social support system. That is, analogous to the emergency broadcast system with which Americans have become so familiar, a safety alarm should be tested when there is no emergency. Future research might measure social support networks at two different time points to determine if capitalization exchanges mediate changes in the size and or quality of the networks. The question of whether perceived responsiveness is real or imagined is an important one. On one hand, perceptions of whether the partner understands, validates, and cares for one may be all that matters in terms of satisfaction with the relationship. On the other hand, forming and maintaining perceptions of responsiveness with little or no basis in the reality of the partner’s actual behavior may be difficult. Our data suggest that perceived responsiveness in the positive event discussion was based, in part, on the partner’s behavior. That is, we found that perceived partner responsiveness was correlated with our judges’ ratings of active– constructive behavior (s ⫽ .35 for both men and women). The more active and constructive and the less passive and destructive individuals reacted when their partners disclosed a positive event, the more responsiveness the partners reported. These results are consistent with social support data that have shown that a support provider’s actual behavior in support exchanges (as coded by outside observers) does predict the recipient’s perceptions of being supported (e.g., Collins & Feeney, 2000; Simpson et al., 1992). We are quick to point out that our outside observers’ ratings of partner behavior accounted for only 12% of the variance in perceived responsiveness. Although the effect of active– constructive behavior is strong and consistent, it is certainly not the whole story in perceived responsiveness. There are two explanations for this finding. First, it is possible that our coding scheme missed important behaviors that contribute to responsiveness. We did attempt to code both verbal and nonverbal behavior; however, it is difficult (more accurately, impossible) to capture all relevant behavior in a coding scheme. Moreover, couples may have idiosyncratic ways of communicating that no coding scheme designed for nomothetic use could pick up. Second, and more interesting, is the possibility that perceived responsiveness is a function of both the partner’s response and factors within the discloser, such as schemas, expectations, mood, and individual differences. Thus, consistent with Reis and Shaver’s (1988) transactional intimacy model, factors internal to the discloser act as a filter through which the partner’s behavior is interpreted. Future research could focus on variables that may influence perceptions of a partner’s behavior during capitalization attempts. However, to the degree that actual behavior does matter, what is it about active and constructive reactions that convey responsiveness to the discloser? We believe active– constructive responses convey important information about the event, the discloser, and the responder’s relationship with the discloser. First, enthusiastically supportive reactions indicate that the responder believes the event is significant. By asking questions about the event and expressing a sense of pleasure about the event, the responder
915
conveys to the discloser that the event itself is significant, either presently or in its future value. Second, through recognition of the importance of the event to the discloser in particular, the responder shows that he or she has intimate knowledge of what is important to the discloser. Finally, when the responder displays positive emotions about the event and the discloser, he or she conveys that both the discloser and the responder’s relationship to the discloser are important. In short, an active– constructive response is unique in its capacity to convey all the components of responsiveness— understanding, validation, and caring. Although most of our data did not reveal significant or reliable sex differences, two interesting distinctions between men and women did emerge. For men, the importance of the event did not matter in terms of the impact of their partner’s behavior. Specifically, regardless of whether men talked about a big or small positive event, active– constructive responses from their partners led to perceived responsiveness. Women reported similar and average levels of responsiveness when they discussed a recent event that was not particularly important. However, if they discussed an event that was important, the response of their male partners was crucial: Active– constructive responses led to high feelings of responsiveness, but if the partner responded passively or destructively, women felt particularly low responsiveness. One possible explanation that is consistent with the filters in the intimacy model is that men expect their partners to respond actively and constructively regardless of event importance, but women only expect active and constructive responding when they themselves view the event as important. We had no assessments of expectations of responses, but future studies should include such measures. The other gender difference that emerged was that for men only responsiveness to positive event discussion was associated with current relationship well-being, but for women responsiveness in both the positive and negative event discussion was associated with well-being. One possible explanation for this is that for men, disclosing a negative event in social support situations may present particularly salient threats to self-esteem. Thus, the costs of discussing a negative event may impede the relationship-enhancing benefits of social support. Of course, both of our findings of gender differences should be interpreted with some caution because they may be confounded with heterosexual relationship variables; one limitation of this study was that we examined only heterosexual dating couples. The final set of results that deserves some attention is the set concerning break-ups at Time 2. The intact couples did not differ from the broken-up couples on any of the measures except the PRCA, male-perceived responsiveness during discussion of the man’s positive event, and woman’s behavior during the man’s positive event discussion. That is, the only discriminating variables in terms of who would remain together were those having to do with capitalization responses. It is possible that effectively managing positive emotional experiences is of vital importance to the health of a relationship, and future research might examine other ways in which couples cope during good times, such as anniversaries, birthdays, and other happy occasions. Again, because there were only 4 couples who broke up out of the 58 couples whom we were able to contact after the study, we view these results as preliminary but encouraging.
GABLE, GONZAGA, AND STRACHMAN
916 Limitations
Several limitations of the current study need to be addressed. One possible confound is that in the preinteraction questionnaire session, participants completed a measure of their partners’ typical responses to their positive event disclosures. This measure was imbedded among many measures of both individual differences and relationship variables; however, this measure could have made participants more aware of these processes during the interaction. Second, the nature of the observational portion of the study compelled participants to discuss a recent positive event, which may be a type of discussion that is less likely to occur spontaneously, at least for some couples. In our sample, however, only a very small portion of the participants had not already discussed their event with their partner (⬍3%), and previous daily experience studies indicate that people share their most positive event of each day on that day with someone else 80% of the time (Gable et al, 2004, Study 4). Nonetheless, a daily experience study examining capitalization attempts in situ, specifically with romantic couples, would further illuminate the issue. Finally, our sample was a dating sample, albeit a stable dating sample (mean length of relationship was more than 2 years). Married participants and friendship dyads may show different behaviors in positive event discussions. However, we do not suspect the associations among the variables to be extremely different between married and dating couples because previous research has found, for example, similar associations among the PRCA score and relationship outcomes (e.g., satisfaction) with both dating and married couples (Gable et al., 2004, Studies 2 and 3).
Concluding Comments How couples deal with positive emotional experiences has received considerably less attention than how couples deal with negative emotional experiences. The disproportionate focus on processes such as conflict, social support, and jealousy, although clearly important, may have unintentionally led to our failure to empirically notice the importance of positive experiences and the dyadic regulation of positive emotions in the lives of couples. The results of the present study indicate that feeling that your partner is there for you when things go right and that your partner actually being there for you when things go right play important roles in the health of relationships. Moreover, because our previous research has shown that individuals share news of positive events with close others at a very high rate, capitalization processes likely play a central role in relationship formation and maintenance. Indeed, positive emotional exchanges may serve as a foundation on which stable and satisfying relationships rest.
References Aron, A., Norman, C. C., Aron, E. N., McKenna, C., & Heyman, R. E. (2000). Couples’ shared participation in novel and arousing activities and experienced relationship quality. Journal of Personality and Social Psychology, 78, 273–284. Barbee, A. P., & Cunningham, M. R. (1995). An experimental approach to social support communications: Interactive coping in close relationships. Communication Yearbook, 18, 381– 413. Barbee, A. P., Rowatt, T. L., & Cunningham, M. R. (1998). When a friend is in need: Feelings about seeking, giving, and receiving social support.
In P. A. Andersen, & L. K. Guerrero (Eds.), Handbook of communication and emotion: Research, theory, applications, and contexts (pp. 281–301). San Diego, CA: Academic. Beach, S. R. H., & Tesser, A. (1995). Self-esteem and the extended self-evaluation maintenance model: The self in social context. In M. H. Kernis (Ed.), Efficacy, agency, and self-esteem (pp. 145–170). New York: Plenum. Bolger, N., Zuckerman, A., & Kessler, R. C. (2000). Invisible support and adjustment to stress. Journal of Personality and Social Psychology, 79, 953–961. Brennan, K. A., Clark, C. L., & Shaver, P. R. (1998). Self-report measurement of adult attachment: An integrative overview. In J. A. Simpson & W. S. Rholes (Eds.), Attachment theory and close relationships (pp. 46 –76). New York: Guilford. Clark, M. S., & Mills, J. (1979). Interpersonal attraction in exchange and communal relationships. Journal of Personality and Social Psychology, 37, 12–24. Cohen, S. (1992). The content and context of effective espousal support. In K. W. Schaie & D. G. Blazer (Eds.), Aging, health behaviors, and health outcomes. Social structure and aging (pp. 127–134). Hillsdale, NJ: Erlbaum. Collins, N. L., & Feeney, B. C. (2000). A safe haven: An attachment theory perspective on support seeking and caregiving in intimate relationships. Journal of Personality and Social Psychology, 78, 1053–1073. Collins, N. L., & Feeney, B. C. (2004). Working models of attachment shape perceptions of social support: Evidence from experimental and observational studies. Journal of Personality and Social Psychology, 87, 363–383. Collins, N. L., & Read, S. J. (1990). Adult attachment, working models, and relationship quality in dating couples. Journal of Personality and Social Psychology, 58, 644 – 663. Coyne, J. C., Wortman, C. B., & Lehman, C. R. (1988). The other side of support: Emotional overinvolvement and miscarried helping. In B. H. Gottlieb (Ed.), Marshaling social support: Formats, processes, and effects (pp. 305–330). Thousand Oaks, CA: Sage. Cutrona, C. E. (1986). Behavioral manifestations of social support: A microanalytic investigation. Journal of Personality and Social Psychology, 51, 201–208. Drigotas, S. M., Rusbult, C. E., Wieselquist, J., & Whitton, S. W. (1999). Close partner as sculptor of the ideal self: Behavioral affirmation and the Michelangelo phenomenon. Journal of Personality and Social Psychology, 77, 293–323. Feeney, B. C. (2004). A secure base: Responsive support of goal strivings and exploration in adult attachment. Journal of Personality and Social Psychology, 87, 631– 648. Fisher, J. D., Nadler, A., & Whitcher-Alagna, S. (1982). Recipient reactions to aid. Psychological Bulletin, 91, 27–54. Fredrickson, B. L. (1998). What good are positive emotions? Review of General Psychology, 2, 300 –319. Fredrickson, B. L. (2001). The role of positive emotions in positive psychology: The broaden-and-build theory of positive emotions. American Psychologist, 56, 218 –226. Fredrickson, B. L., & Joiner, T. (2002). Positive emotions trigger upward spirals toward emotional well-being. Psychological Science, 13, 172– 175. Fredrickson, B. L., Tugade, M., Waugh, C. E., & Larkin, G. R. (2003). What good are positive emotions in crisis? A prospective study of resilience and emotions following the terrorist attacks on the United States on September 11th, 2001. Journal of Personality and Social Psychology, 84, 365–376. Gable, S. L., & Haidt, J. (2005). What (and why) is positive psychology? Review of General Psychology, 9, 103–110. Gable, S. L., & Nezlek, J. B. (1998). Level and instability of day to day
SUPPORTIVE RESPONSES TO POSITIVE EVENT DISCLOSURES psychological well-being and risk for depression. Journal of Personality and Social Psychology, 74, 129 –138. Gable, S. L., & Reis, H. T. (2001). Appetitive and aversive social interaction. In J. Harvey & A. Wenzel (Eds.), Close romantic relationships: Maintenance and enhancement (pp. 169 –194). Mahwah, NJ: Erlbaum. Gable, S. L., Reis, H. T., & Elliot, A. J. (2000). Behavioral activation and inhibition in everyday life. Journal of Personality and Social Psychology, 78, 1135–1149. Gable, S. L., Reis, H. T., Impett, E., & Asher, E. R. (2004). What do you do when things go right? The intrapersonal and interpersonal benefits of sharing positive events. Journal of Personality and Social Psychology, 87, 228 –245. Gottman, J. M. (1994). What predicts divorce? The relationship between marital processes and marital outcomes. Hillsdale, NJ: Erlbaum. Hatfield, E., & Rapson, R. L. (1993). Love, sex, and intimacy: Their psychology, biology, and history. New York: HarperCollins. Hatfield, E., & Sprecher, S. (1986). Measuring passionate love in intimate relationships. Journal of Adolescence, 9, 383– 410. Hendrick, S. S. (1988). A generic measure of relationship satisfaction. Journal of Marriage and the Family, 50, 93–98. Hirschman, A. O. (1970). Exit, voice, and loyalty: Responses to decline in firms, organizations, and states. Cambridge, MA: Harvard University Press. Holmes, J. G. (2002). Interpersonal expectations as the building blocks of social cognition: An interdependence theory perspective. Personal Relationships, 9, 1–26. Isen, A. M., & Daubman, K. A. (1984). The influence of affect on categorization. Journal of Personality and Social Psychology, 47, 1206 – 1217. Isen, A. M., Daubman, K. A., & Nowicki, G. P. (1987). Positive affect facilitates creative problem solving. Journal of Personality and Social Psychology, 52, 1122–1131. John, O. P., Donahue, E. M., & Kentle, R. P. (1991). The “Big Five” Inventory: Versions 4a and 54 (Tech. rep.). Berkeley: University of California, Institute of Personality and Social Research. John, O. P., & Srivastava, S. (1999). The Big Five taxonomy: History, measurement, and theoretical perspectives. In L. A. Pervin (Ed.), Handbook of personality, theory, and research (pp. 102–138). New York: Guilford. Karney, B. R., & Bradbury, T. N. (1997). Neuroticism, martial interaction, and the trajectory of marital satisfaction. Journal of Personality and Social Psychology, 72, 1075–1092. Kaul, M., & Lakey, B. (2003). Where is the support in perceived support? The role of generic relationship satisfaction and enacted support in perceived support’s relation to low distress. Journal of Social and Clinical Psychology, 22, 59 –78. Lakey, B., McCabe, K. M., Fisicaro, S. A., & Drew, J. B. (1996). Environmental and personal determinants of support perceptions: Three generalizability studies. Journal of Personality and Social Psychology, 70, 1270 –1280. Langston, C. A. (1994). Capitalizing on and coping with daily-life events: Expressive responses to positive events. Journal of Personality and Social Psychology, 67, 1112–1125. Murray, S. L., & Holmes, J. G. (1993). Seeing virtues in faults: Negativity and the transformation of interpersonal narratives in close relationships. Journal of Personality and Social Psychology, 65, 707–722. Murray, S. L., Holmes, J. G., & Griffin, D. W. (2000). Self-esteem and the quest for felt security: How perceived regard regulates attachment processes. Journal of Personality and Social Psychology, 78, 478 – 498. Murray, S. L., Rose, P., Holmes, J. G., Derrick, J., Podchaski, E. J., Bellavia, G., & Griffin, D. W. (2005). Putting the partner within reach: A dyadic perspective on felt security in close relationships. Journal of Personality and Social Psychology, 88, 327–347.
917
Nezlek, J. B., & Gable, S. L. (2001). Depression as a moderator of relationships between positive daily events and day-to-day psychological adjustment. Personality and Social Psychology Bulletin, 27, 1692– 1704. Notarius, C. I., & Markman, H. J. (1989). Coding marital interaction: A sampling and discussion of current issues. Behavioral Assessment, 11, 1–12. Pasch, L. A., Bradbury, T. N., & Davila, J. (1997). Gender, negative affectivity, and observed social support behavior in marital interaction. Personal Relationships, 4, 361–378. Reis, H. T. (2003). A self-report measure of perceived partner responsiveness. Unpublished data, University of Rochester. Reis, H. T., Clark, M. S., & Holmes, J. G. (2004). Perceived partner responsiveness as an organizing construct in the study of intimacy and closeness. In D. J. Mashek & A. P. Aron (Eds.), Handbook of closeness and intimacy (pp. 201–225). Mahwah, NJ: Erlbaum. Reis, H. T., & Franks, P. (1994). The role of intimacy and social support in health outcomes: Two processes or one? Personal Relationships, 1, 185–197. Reis, H. T., & Gable, S. L. (2003). Toward a positive psychology of relationships. In C. L. Keyes & J. Haidt (Eds.). Flourishing: The positive person and the good life (pp. 129 –159). Washington, DC: American Psychological Association. Reis, H. T., & Shaver, P. (1988). Intimacy as an interpersonal process. In S. Duck & D. F. Hay (Eds.), Handbook of personal relationships: Theory, research, and interventions (pp. 367–389). New York: Wiley. Rusbult, C. E., Martz, J. M., & Agnew, C. R. (1998). The investment model scale: Measuring commitment level, satisfaction level, quality of alternatives, and investment size. Personal Relationships, 5, 357–391. Rusbult, C. E., Verette, J., Whitney, G. A., Slovic, L. F., & Lipkus, I. (1991). Accommodation processes in close relationships: Theory and preliminary empirical evidence. Journal of Personality and Social Psychology, 60, 53–78. Rusbult, C. E., Zembrodt, I. M., & Gunn, L. K. (1982). Exit, voice, loyalty, and neglect: Responses to dissatisfaction in romantic involvements. Journal of Personality and Social Psychology, 43, 1230 –1242. Sarason, B. R., Sarason, I. G., & Gurung, R. A. R. (1997). Close personal relationships and health outcomes: A key to the role of social support. In S. Duck (Ed.), Handbook of personal relationships: Theory, research, and interventions (2nd ed., pp. 547–573). New York: Wiley. Simpson, J. A., Rholes, W. S., & Nelligan, J. S. (1992). Support seeking and support giving within couples in an anxiety-provoking situation: The role of attachment styles. Journal of Personality and Social Psychology, 62, 434 – 446. Sternberg, R. J. (1986). A triangular theory of love. Psychological Review, 93, 119 –135. Swann, W. B., Jr. (1990). To be adored or to be known? The interplay of self-enhancement and self-verification. In E. T. Higgins & R. M. Sorrentino (Eds.), Handbook of motivation and cognition: Foundations of social behavior (Vol. 2, pp. 408 – 448). New York: Guilford. Tesser, A., Millar, M., & Moore, J. (1988). Some affective consequences of social comparison and reflection processes: The pain and pleasure of being close. Journal of Personality and Social Psychology, 54, 49 – 61. Uchino, B. N., Cacioppo, J. T., & Kiecolt-Glaser, J. K. (1996). The relationship between social support and physiological processes: A review with emphasis on underlying mechanisms and implications for health. Psychological Bulletin, 119, 488 –531.
Received October 14, 2005 Revision received April 1, 2006 Accepted April 10, 2006 䡲
Journal of Personality and Social Psychology 2006, Vol. 91, No. 5, 918 –928
Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.5.918
When Inclusion Costs and Ostracism Pays, Ostracism Still Hurts Ilja van Beest
Kipling D. Williams
Leiden University
Purdue University
Recent research indicates that ostracism is painful even in the face of mitigating circumstances. However, in all previous experiments, there have been no costs to inclusion or benefits for ostracism. If being included meant losing money and being ostracized meant retaining money, would individuals still be distressed when ostracized? In 2 studies, the authors attempted to “load the dice” against inclusion in favor of ostracism. Participants played a variant of Cyberball called €yberball (pronounced Euroball), in which ostracism and inclusion were crossed with whether the participants earned or lost money for each ball toss they received. In 2 experiments, the authors found that even when being ostracized meant retaining more money than the other players, it was painful. In Study 2, the authors also introduced conditions in which participants were overincluded. In these conditions, participants were sensitive to financial incentives. However, even then participants felt worse when given no positive attention than when given punitive attention. Keywords: gain–loss framing, social exclusion, bullying
2004; in Internet chat rooms, Williams et al., 2002; or in an Internet game of ball toss—Cyberball—Williams, Cheung, & Choi, 2000; Williams & Jarvis, in press). In Cyberball, participants play a game of virtual toss with two other ostensible players whom they do not know and do not expect to meet. Despite these minimal conditions, participants show strong negative effects after less than 4 min of ostracism (i.e., a period in which they do not get the ball thrown to them while the other two participants continue to play). Eisenberger, Lieberman, and Williams (2003) used functional magnetic resonance imaging to examine the brain activity of ostracized Cyberball players. They found an increase in blood flow in the same region of the brain (the dorsal anterior cingulate cortex) that is activated when individuals experience physical pain. This increase occurred even when people knew that the others were not intentionally ostracizing them but were unable to include them because of a technical malfunction (Eisenberger & Liebman, 2005). Ostracism is distressing even when the individual is ostracized by outgroup members (Williams et al., 2000) or by those whom they despise (e.g., Ku Klux Klan members; Gonsalkorale & Williams, in press). Ostracism even hurts when a computer is the source (Zadro, Williams, & Richardson, 2004); in fact, it hurts just as much as when participants believe the ostracizers are humans.
Being ostracized—ignored and excluded—is painful and distressing. It severs our sense of belonging and feelings of connection with others; it makes us realize that others do not value us and consequently lowers our own self-esteem; it takes away any sense of control that we think we have in our social interaction with the others; and at perhaps a deeper level, it challenges our sense of existence. Several groups of researchers have amassed considerable insight into how we react to ostracism, social exclusion, and rejection. Our analysis relies on a fast-growing literature in social psychology on the impact of being ostracized, socially excluded, and rejected (for an overview, see Williams, Forgas, & von Hippel, 2005). Being ostracized causes pain and distress during the ostracism episode itself, regardless of individual predispositions of the target of ostracism, the social context in which it occurs, or who is the source of ostracism. Ostracism threatens four fundamental needs (belonging, self-esteem, control, and meaningful existence) and increases sadness and anger when it occurs face-to-face (Warburton, Williams, & Cairns, 2006; Williams & Sommer, 1997) or when it occurs remotely (e.g., over cell phones, Smith & Williams,
Ilja van Beest, Department of Social and Organizational Psychology, Leiden University, Leiden, The Netherlands; Kipling D. Williams, Department of Psychological Sciences, Purdue University. This research was supported by Veni Grant NWO-451-04-069 from the Netherlands Organization for Scientific Research awarded to Ilja van Beest and by National Science Foundation Grant 0519209-BCS awarded to Kipling D. Williams. We thank Jan-Willem van Prooijen for his comments on an earlier version of this article and Anouk Hofman for conducting Study 1. Correspondence concerning this article should be addressed to either Ilja van Beest, Department of Social and Organizational Psychology, Leiden University, P. O. Box 9555, Leiden 2300 RB, The Netherlands, or Kipling D. Williams, Department of Psychological Sciences, Purdue University, West Lafayette, IN 47907. E-mail:
[email protected] or
[email protected]
The Immediate Impact of Ostracism Williams (1997, 2002) proposed that ostracism is uniquely capable of simultaneously threatening four fundamental needs. Ostracism threatens the need for belonging because it divorces the individual from the group. It threatens self-esteem because individuals interpret their exclusion as a result of being unlikable. It threatens the need for control because unlike an argument or even a physical entanglement, ostracism is unilateral and unaffected by the individual’s response. Finally, ostracism not only is a metaphor for death (Case & Williams, 2004), threatening individuals’ sense of existence and recognition, but also in extreme cases, leads to death in humans and animals. 918
WHEN INCLUSION COSTS
In addition to threatening these fundamental needs, it is likely that all social animals, including humans, have evolved to detect ostracism because it threatens survival (Gruter & Masters, 1986). Being able to detect ostracism quickly is adaptive to the extent that it guides effective coping responses. Thus, it appears that the brain is well equipped to detect even the slightest hint of ostracism (Eisenberger & Lieberman, 2005) by registering it as painful. This alarm then directs the individual to focus attention on the circumstances of ostracism and appropriate coping responses. Research has uncovered a variety of coping responses that direct the ostracized individual toward fight (Gaertner & Iuzzini, 2005; Leary, Kowalski, Smith, & Phillips, 2003; Twenge & Baumeister, 2005; Twenge, Baumeister, Tice, & Stucke, 2001); flight (Predmore & Williams, 1983); freezing (Twenge, Catanese, & Baumeister, 2003), or, in many instances, becoming more socially attentive and pliable so that the chances for future inclusion are enhanced (Lakin & Chartrand, 2005; Pickett & Gardner, 2005; Pickett, Gardner, & Knowles, 2004; Ouwerkerk, Kerr, Gallucci, & Van Lange, 2005; Williams et al., 2000; Williams & Sommer, 1997). Our focus in the present set of studies was on the immediate impact of ostracism to determine whether the large body of research showing the uniformly distressful reactions to ostracism is due primarily to the fact that ostracism is perceived to be costly, whereas inclusion is perceived to be beneficial. That is, if we were to create a situation in which people were punished for being included and rewarded for being ostracized, would the negative reactions to ostracism be reduced, eliminated, or even reversed? But how could one make ostracism beneficial and inclusion costly? For the sake of argument, imagine a group of individuals playing Russian roulette. Would an individual in that group feel unhappy if he or she were not passed the gun? In a less dramatic fashion, we investigated a pairing of ostracism and inclusion with financial penalties in the following studies.
Making Inclusion Costly and Ostracism Rewarding In the present investigation, we attempted to turn the tables, so to speak, on the rewarding nature of inclusion and the costly nature of ostracism. In two studies, participants played a variant of Cyberball called €yberball (pronounced Euroball) in which catching a ball-toss had financial consequences. In Study 1, we actually deducted money from individuals every time they were included. Thus, ostracized individuals would have ended up being wealthier than the other ostracizing players. In Study 2, we threatened individuals with the loss of all their experimental payoffs if they were thrown a ball toss. Again, ostracized individuals would have been better off than included individuals because they were not under constant threat of losing all their experimental payoffs. Would these variations of Cyberball be enough to minimize or reverse the negative impact of ostracism? From a rational cost– benefit perspective, making inclusion costly ought to have relieved ostracized participants of all distress. People should not feel unhappy when ostracized if ostracism prevents them from losing money. From a rational perspective, it may therefore be argued that financial incentives should have mitigated ostracism. In fact, applying prospect theory (Kahneman & Tversky, 1979; Ku¨hberger, 1998; Tversky & Kahneman, 1991), arguing that losses loom larger than gains, one could even predict that the difference between inclusion and ostracism might have
919
been reversed. If people would rather minimize losses than maximize gains, then study participants should have felt worse when included in a loss game than when ostracized from a gain game. We did not believe, however, that either of these outcomes would happen. As we noted above, research on ostracism indicates that the immediate reflexive reaction to ostracism appears to be precognitive: Information that if incorporated would reduce or dismiss the impact of ostracism either is not incorporated or is disregarded by what appears to be an alarm reaction to the pain of ostracism. Given this reasoning, then all that matters is that participants perceive themselves to be ignored and excluded. Any mitigating factors related to this perception should be inconsequential to the immediate experience of ostracism. Taking this perspective, we argued that even if we created a situation in which ostracism is financially beneficial and inclusion is financially costly, ostracism would still be painful, thwart fulfillment of fundamental needs, and increase sadness and anger.
Study 1 The main goal of Study 1 was to test whether immediate responses to ostracism can be mitigated by the manipulation of financial payoff valences. People were informed that their experimental pay was contingent on the number of ball tosses they received. In the gain conditions, they were informed that they would earn 50 euro cents for each ball toss that they received. In the loss conditions, they were informed that they would lose 50 euro cents for each ball toss that they received. Consequently, people made money when they were included in the gain game but lost money when they were included in the loss game. If individuals focus only on the monetary consequences as rational-person theorists would have us believe, then study participants should prefer to be included in a gain game and ostracized in a loss game. Additionally, such a prediction would also be accompanied by commensurate assessments of fundamental needs and mood. However, if immediate reactions to ostracism are indeed precognitive, as we believe, and override reflections of whether a situation has financial benefits, we would expect to find only a main effect of ostracism. It should lower the sense of belonging, self-esteem, control, and meaningful existence and should increase negative mood, regardless of the financial implications of the situation.
Method Participants and design. Participants were 135 students (28 men, 107 women; mean age ⫽ 20.90 years, SD ⫽ 2.33) from Leiden University who were randomly assigned to a 2 (€yberball experience: ostracized, included) ⫻ 2 (payoff valence: gain, loss) between-S design. Procedure. The general outline of the procedure was based on previous research on Cyberball (e.g., Williams et al., 2000; Williams & Jarvis, in press; Zadro et al., 2004). Participants were seated behind a computer in separate cubicles. All instructions were presented on the computer screen. The participants were told that they were participating in a study about the relation between mental visualization and task performance and informed that this would be tested by means of a three-player Internet ball-tossing game called €yberball. In this game, players see an animated ball-toss game. Depicted on the screen are two other ostensible players (represented by Cyberboy icons), and the participant is represented as an animated hand at the bottom of the screen). Participants were asked to use this game as a means of engaging in mental visualization (i.e., they were encouraged to
920
VAN BEEST AND WILLIAMS
visualize whom the others were, what they looked like, where they were playing, what the temperature was like, and so on). In this game, there were a total of 30 throws. €yberball experience manipulation. The €yberball experience was manipulated by the number of ball tosses thrown to the participant. In the ostracism conditions, the €yberball game was programmed such that the participant received two tosses at the beginning of the game and then never received another toss. In the inclusion conditions, the participant received one third of the tosses. Payoff valence manipulation. The payoff valence manipulation was based on a procedure used by van Beest, Van Dijk, De Dreu, and Wilke (2005), who examined the impact of gain and losses on the willingness to ostracize people in a three-player coalition game. Participants were endowed with 0 euros in the gain game and 6 euros in the loss game. It was made clear that their experimental earnings would be based on the number of received ball tosses. In the gain game, participants were informed that they would earn 50 euro cents for each ball toss they received. In the loss game, participants were told that they had to pay 50 euro cents for each ball toss they received. Combined with our ostracism manipulation, this arrangement meant that in the gain game, participants gained 5 euros when included but gained only 1 euro when ostracized and that in the loss game, participants lost 5 euros when included but only 1 euro when ostracized. The implication of this manipulation was that participants entered the next phase of the experiment believing that they possessed 1 euro (i.e., when ostracized in the gain game or when included in the loss game) or 5 euros (i.e., when ostracized in the loss game or when included in the gain game). Dependent variables. After the game ended, we asked participants to provide self-reports concerning their current satisfaction levels with belonging, self-esteem, meaningful existence, and control on 7-point scales (see Appendix, ␣ ⫽ .92).1 Next, we asked participants to assess their emotional state (also on 7-point scales) during the game. This mood index contained three items assessing negative emotions (sad, angry, hurt) and three assessing positive emotions (happy, elated, cheerful; ␣ ⫽ .88). We reverse-scored the negative emotions. Finally, we checked whether the participants had understood the most important elements of the experiments. To check the €yberball experience manipulation, we asked participants to recall the percentage of ball throws that they received. To check the payoff valence manipulation, we asked participants how financially rewarding it was to catch a ball (1 ⫽ not rewarding, 7 ⫽ rewarding). At the end of the experiment, we fully debriefed the participants. We explained that they had played against preprogrammed computer players. In addition, we informed them that their experimental pay would not be based on the number of ball throws and that they would be paid 5 euros instead.
Results A multivariate analysis and subsequent univariate analyses on the separate needs and the separate emotions showed that each need and each emotion yielded identical results. This is common in research on ostracism; therefore, following the procedure in previous research, we only report an analysis of the combined needs and the combined emotions (Williams et al, 2000; Zadro, Boland, & Richardson, in press). We included gender in a first run of our analyses. This factor did not yield any significant results, apart from an occasional main effect of gender (i.e., men were more positive or less negative during the game, and their need satisfaction levels were higher regardless of condition) and was therefore dropped from the reported analyses. Manipulation checks. Both manipulations were successful. A 2 (€yberball experience) ⫻ 2 (payoff valence) analysis of variance (ANOVA) on the percentage of balls caught yielded only a main effect of €yberball experience, F(1, 133) ⫽ 90.45, p ⬍ .000, 2 ⫽
.41. Included participants (M ⫽ 25.52, SD ⫽ 1.39) reported being thrown a higher percentage of ball tosses than did ostracized participants (M ⫽ 7.09, SD ⫽ 1.38). A 2 ⫻ 2 ANOVA on the financial consequences of catching a ball yielded only a main effect of payoff valence, F(1, 133) ⫽ 338.70, p ⬍ .000, 2 ⫽ .66. Participants in the gain conditions (M ⫽ 6.41, SD ⫽ 1.33) believed it was more rewarding to catch a ball than participants in the loss conditions did (M ⫽ 1.68, SD ⫽ 1.67). Fundamental needs. Means and standard deviations of our need satisfaction index are given in Table 1. A 2 ⫻ 2 ANOVA on needs yielded a small but (marginally) significant main effect of payoff valence, F(1, 133) ⫽ 3.87, p ⫽ .051, 2 ⫽ .03, such that need satisfaction levels were higher in the loss game (M ⫽ 3.74, SD ⫽ 1.08) than in the gain game (M ⫽ 3.43, SD ⫽ 1.07). Apparently, needs were more thwarted in the gain game than in the loss game. The more important finding was that the results yielded a main effect of €yberball experience, F(1, 133) ⫽ 51.64, p ⫽ .000, 2 ⫽ .28, and not an interaction effect of €yberball experience and payoff valence (F ⬍ 1). The €yberball experience effect showed that needs were more thwarted in the ostracism condition (M ⫽ 3.02, SD ⫽ 0.81) than in the inclusion condition (M ⫽ 4.14, SD ⫽ 1.02), supporting our hypothesis that ostracism is painful even when it pays. Mood. Means and standard deviations of our mood index are given in Table 2. A 2 ⫻ 2 ANOVA on the emotion index yielded a main effect of payoff valence, F(1, 133) ⫽ 8.99, p ⫽ .003, 2 ⫽ .06. Participants were more positive during the loss game (M ⫽ 4.90, SD ⫽ 1.15) than during the gain game (M ⫽ 4.30, SD ⫽ 1.15). As predicted, we observed a main effect of €yberball experience, F(1, 133) ⫽ 11.74, p ⫽ .001, 2 ⫽ .08, and no interaction between €yberball experience and payoff valence (F ⬍ 1). Participants were less positive when ostracized (M ⫽ 4.29, SD ⫽ 1.17) than when included (M ⫽ 4.95, SD ⫽ 1.12). This is again consistent with our prediction that individuals would respond negatively to ostracism even when it was financially beneficial to be ostracized. Mediation. The results showed that being ostracized lowered need satisfaction and mood. Our next step was to determine if the need satisfaction levels mediated mood, as was hypothesized and has been shown in other studies (e. g., Williams et al, 2000). To test for mediation, we used the procedure of Baron and Kenny (1986). Following the recommendations of Aiken and West (1991), we centered participants’ answers on the needs satisfaction and mood index and effect-coded our independent variables: €yberball experience manipulation and payoff valence. First, we regressed our mood index on €yberball experience. Second, we regressed our needs satisfaction index on €yberball experience. As could be deduced from the above ANOVAs these regression analyses showed that €yberball experience affected mood,  ⫽ .27, t ⫽ 3.32, p ⬍ .000, and needs,  ⫽ .52, t ⫽ 7.12, p ⬍ .000. Third, we regressed mood on €yberball experience while controlling for needs. This regression analysis showed that needs 1 The need questions were based on previous research using the Cyberball paradigm. Note that the questions measure how much belonging, control, self-esteem, and meaningful existence people are experiencing during Cyberball and, as such, only serve as a proxy of whether these needs are threatened.
WHEN INCLUSION COSTS
affected mood,  ⫽ .68, t ⫽ 8.74, p ⬍ .000, and that €yberball experience did not affect mood when we controlled for needs,  ⫽ .08, t ⫽ 1.03, ns. A subsequent Sobel test showed that this reduction was statistically significant, Z ⫽ 5.50, p ⬍ .000. Finally, we also regressed needs on €yberball experience while controlling for mood. This analysis revealed that the effect of €yberball experience on needs remained significant,  ⫽ .37, t ⫽ 6.15, p ⬍ .000. Taken together, these analyses show that needs mediate mood, not vice versa. It shows that mood is formed as an appraisal of needs.
921
Table 2 Means and Standard Deviations of Mood Index by €yberball Experience (Ostracism, Inclusion) and Payoff Valences (Gain, Loss) in Study 1 Ostracism
Inclusion
Variable
M
SD
M
SD
Gain Loss
3.93 4.66
1.03 1.22
4.66 5.25
1.17 1.01
Discussion Our results are consistent with those of previous research showing that ostracism appears to overwhelm factors that, rationally, ought to mitigate distressful reactions (Zadro, Williams, & Richardson, 2005). Again, ostracism seems to be a powerful experience, overriding even a pervasive incentive such as earning money. Regardless of whether people earned money by being ostracized or not, ostracism lowered need satisfaction levels and mood. Additionally, self-reported mood levels were mediated by the self-reported need satisfaction levels, and not the reverse, providing support for Williams’s need threat model of ostracism (Williams, 2001; Williams, in press; Zadro, Williams, & Richardson, 2004). In addition to the main effects of €yberball experience, the results also yielded main effects of payoff valence. Apparently, people felt better when playing a loss game than when playing a gain game. This may be a side effect of our payoff valence manipulation. People were endowed with 6 euros in the loss game and 0 euros in the gain game, and such an endowment may have caused participants to appreciate the experiment more in the loss conditions than in the gain conditions.
Study 2 Our first goal of Study 2 was to replicate our findings using a different payoff valence manipulation. In Study 1, participants gained or lost 50 euro cents each time they caught the ball. It could be argued that participants became used to losing or gaining a small amount of money and therefore 50 euro cents may not have been strong enough to moderate ostracism. Furthermore, we endowed participants with 6 euros in the loss conditions and 0 euros in the gain conditions in Study 1. A possible side effect of this may have been that people were more elated during the loss game than during the gain game. To address these issues, we altered our payoff valence manipulation in Study 2 such that its effect would
Table 1 Means and Standard Deviations of Need Satisfaction Index by €yberball Experience (Ostracism, Inclusion) and Payoff Valences (Gain, Loss) in Study 1 Ostracism
Inclusion
Variable
M
SD
M
SD
Gain Loss
2.85 3.20
0.74 0.86
4.02 4.28
1.03 1.02
be stronger but would not endow participants with different amounts of money prior to participating in €yberball. In Study 2, all participants were endowed with 5 euros at the beginning of €yberball. They were informed that the ball-tossing game could stop at any moment and that the person holding the ball would either lose all his or her money or double all his or her money, depending upon the condition. In contrast with Study 1 in which participants were confronted with small increments or decrements of money to which participants may have habituated, participants were now under constant threat of losing all their money (or were under the constant promise of doubling their money). Our second goal of Study 2 was to extend our interest in the inclusion– ostracism continuum by adding an overinclusion condition. Overinclusion means being included more than what one would expect, given equal participation among the members of the group. Similar to ostracism, overinclusion is likely to make participants feel conspicuous. The difference, however, is that overincluded participants are conspicuous by excessive attention, whereas ostracized participants are conspicuous by excessive inattention. Williams et al. (2000) added an overinclusion condition to their Internet Cyberball design (but for which there were no financial benefits or costs) and found that overinclusion tended to be a more positive experience than inclusion, although this effect was not significant. They included overinclusion to rule out that ostracism was distressing simply because it made participants feel conspicuous. Our reason for adding an overinclusion condition is that participants should be able to register whether they are overincluded in a good situation or overincluded in a bad situation when asked to evaluate their satisfaction with needs and mood. We thought that participants would have unmoderated reflexive responses only to ostracism, not to overinclusion. It also rules out the possibility that our need and mood measures are insensitive to detecting payoff valence differences. Our third goal of Study 2 was to test the specificity of our reasoning by assessing reflective behavioral responses. That is, after the initial pain and distress of ostracism that we think is unmoderated by situational constraints like gain or loss of payoff, we expect moderation for reflective behavioral responses. Several studies have shown that when ostracized individuals have little control over their future inclusion or have been sufficiently deprived of a sense of control over any outcome, their reactions are hostile. Not only are they more willing to lash out, but they are also less willing to donate to charity (Twenge, 2005; Twenge & Baumeister, 2005; Twenge et al., 2001; Warburton et al., 2006). Thus, we expected that differences in payoff valence (i.e., gain vs. loss) would be likely to moderate behavioral responses to ostracism. More specifically, we predicted that participants should be
VAN BEEST AND WILLIAMS
922
more willing to retaliate and less willing to donate to charity when they are ostracized from a gain game than from a loss game. This expected interaction would again underscore our argument that only the immediate responses to ostracism are reflexive. It would also help to rule out the possibility that our payoff valence manipulation cannot elicit moderation when people are ostracized. Our fourth and final goal of Study 2 was to directly compare the condition in which participants are overincluded in a loss game with the condition when they are ostracized from a gain game. We wondered which is worse: getting punitive attention or getting no positive attention? One could argue on the basis of the prospect theory that participants should react more negatively to punitive attention than to a lack of positive attention. However, as James (1890/1950) stated, it may also be the other way around. He wrote the following: If no one turned round when we entered, answered when we spoke, or minded what we did, but if every person we met ‘cut us dead,’ and acted as if we were non-existing things, a kind of rage and impotent despair would ere long well up in us, from which the cruelest bodily tortures would be a relief; for these would make us feel that, however bad might be our plight, we had not sunk to such a depth as to be unworthy of attention at all. (pp. 293–294)
In line with this reasoning, Williams and Zadro (2001) reported that targets of long-term ostracism voluntarily state that they would have preferred to be beaten than to have been ostracized. In Study 2, we put this notion to the test.
Method Participants and design. There were 167 students from Leiden University (62 men, 105 women; mean age ⫽ 20.93 years, SD ⫽ 2.95) who were randomly assigned to the conditions of a 3 (€yberball experience: ostracized, included, over-included) ⫻ 2 (payoff valence: gain, loss) design. Participants were given 5 euros before the start of the experiment. Procedure. The procedure was similar to Study 1 with some notable exceptions. The first difference was that we used a different payoff valence manipulation. Instead of losing or gaining money each time a participant got the ball, participants were now informed that the game could stop at any moment and that the person holding the ball would then lose or double his or her endowment of 5 euros. In addition, we also introduced an overincluded condition. In this condition, both computer players always threw the ball to the participant. In all conditions, the game would stop when the participant was not holding the ball. That is, the threat of losing all payoffs (or promise of doubling payoffs) was never realized. Note that we did not realize threats or promises because it is impossible to realize treats or promises in the ostracism conditions and because it would obscure the reason that people reacted to the inclusion and overinclusion conditions. Contrary to the procedure in Experiment 1 in which some participants possessed more money than others after the ostracism manipulation, this procedure ensured that all participants still possessed 5 euros the moment they started answering the dependent measures. Similar to our procedure in Study 1, we first asked participants to fill out questionnaires about need satisfaction levels (␣ ⫽ .92) and mood (␣ ⫽ .92). After answering these questions, the participants were asked whether they felt like taking revenge on the other players and punishing them. These questions were used as our measure of retaliation (␣ ⫽ .92). After the experiment ended and participants were paid 5 euros, participants were informed that they could make a donation to the people of Darfur (Sudan, Africa). At the time of the experiment, news about Darfur was televised daily in the Netherlands because of the war and famine, and
people were called on to make donations to alleviate the suffering. Participants were told that they could help out by making a donation to one of the major funds whose appeals were being televised by putting some of their experimental earnings in drop box that was positioned on the way out of the laboratory. Participants were asked to use an envelope that was present in their experimental cubicle. To ensure anonymity, we asked the participants to drop the envelope in the drop box regardless of whether they actually put money in the envelope. The amount given to Darfur was used as our measure of prosocial behavior, and after all participants completed the experiment, the money was indeed given to Darfur.
Results As in Study 1, we analyzed need satisfaction levels and mood levels with index scores of each.2 Furthermore, we also included gender in a first run of the analyses. However, similar to the results in Study 1, this factor yielded only some main effects (men were more positive and less thwarted in needs than women). Gender did not interact with €yberball experience or with our payoff valence manipulation, so this factor was dropped from further analyses. Manipulation checks. The manipulations were successful. A 3 (€yberball experience) ⫻ 2 (payoff valence) ANOVA on the percentage of balls caught yielded only a main effect of €yberball experience, F(2, 161) ⫽ 253.07, p ⬍ .000, 2 ⫽ .86. Tukey’s honestly significant difference (HSD) tests showed that the overincluded participants (M ⫽ 86.94, SD ⫽ 20.63) reported being thrown a higher percentage of balls than did the included participants (M ⫽ 35.70, SD ⫽ 10.11), who in turn reported being thrown a higher percentage of balls than did the ostracized participants (M ⫽ 6.64, SD ⫽ 4.96). A 3 ⫻ 2 ANOVA on the question as to whether it was financially profitable to catch a ball yielded only a main effect of payoff valence, F(1, 161) ⫽ 133.95, p ⬍ .000, 2 ⫽ .45. Participants in the gain conditions (M ⫽ 5.22, SD ⫽ 2.06) believed that it was more profitable to catch a ball than did participants in the loss conditions (M ⫽ 2.01, SD ⫽ 1.44). Fundamental needs. Means and standard deviations of our need satisfaction index are given in Table 3. A 3 ⫻ 2 ANOVA on the needs satisfaction index resulted in a main effect of €yberball experience, F(2, 161) ⫽ 66.06, p ⫽ .000, 2 ⫽ .45. Tukey’s HSD tests showed that the fundamental needs of ostracized participants (M ⫽ 3.00, SD ⫽ 0.81) were more thwarted than the fundamental needs of included (M ⫽ 4.76, SD ⫽ 0.80) and overincluded participants (M ⫽ 4.65, SD ⫽ 1.10). This replicates the findings of Study 1. The analysis yielded also an interaction effect of €yberball experience and payoff valence, F(2, 161) ⫽ 3.87, p ⫽ .05, 2 ⫽ .03. Simple main effects analysis to interpret this interaction showed that our payoff valence manipulation moderated the overinclusion conditions, F(1, 161) ⫽ 3.89, p ⫽ .05, 2 ⫽ .024, but not the inclusion and ostracism conditions. Participants reported lower need levels when overincluded in a loss game than when overin2 As in Study 1, we first conducted a 2 ⫻ 3 multivariate analysis of variance on the separate needs and moods. Again, the result of each separate need was identical when comparing the ostracism conditions with the inclusion conditions. There was a slight difference between the needs in the overinclusion conditions. Receiving punitive attention was especially detrimental for self-esteem and belonging. There were no differences whatsoever between the results on positive or negative mood.
WHEN INCLUSION COSTS
cluded in a gain game. As predicted, this result shows that individuals are capable of making rational cost– benefits analyses when overincluded. Finally, we compared participants who were ostracized from a gain game with participants who were overincluded in a loss game. This comparison showed that participants reported lower need satisfaction levels when ostracized from gains than when overincluded with financial penalties, indicating that at least in this experiment, getting no positive attention had greater negative impact on need satisfaction levels than getting punitive attention, t(53) ⫽ 6.00, p ⫽ .000. Mood. Means and standard deviations of our mood index are given in Table 4. As in Study 1, the results of mood mirrored the results of needs. A 3 ⫻ 2 ANOVA on our mood index yielded a main effect of €yberball experience, F(2, 161) ⫽ 17.78, p ⫽ .000, 2 ⫽ .18, and an interaction effect of €yberball experience and payoff valence, F(2, 161) ⫽ 6.01, p ⫽ .003, 2 ⫽ .07. Tukey’s HSD tests performed to interpret the main effect of €yberball experience showed that participants felt worse when ostracized (M ⫽ 4.21, SD ⫽ 1.15) than when included (M ⫽ 5.35, SD ⫽ 0.83) or overincluded (M ⫽ 5.08, SD ⫽ 1.23). Simple main effects analysis to interpret the interaction revealed that our manipulation of payoff valence only affected the overinclusion condition, F(1, 161) ⫽ 8.51, p ⫽ .004, 2 ⫽ .05. Participants felt worse when overincluded in a loss game than when overincluded in a gain game. Furthermore, participants felt worse when ostracized from gains than when overincluded in losses, indicating that at least in this experiment, getting no positive attention caused moods to plummet more than getting punitive attention, t(53) ⫽ 1.71, p ⬍ .05. Mediation. To test whether needs mediated mood, we centered the responses of the participants and effect-coded the independent variables and their interaction. To facilitate a direct comparison with Study 1, we first focused on the conditions of €yberball experience in which participants were either ostracized or included. Both the regression analysis on needs,  ⫽ .74, t ⫽ 11.49, p ⬍ .000, and the regression analysis on moods,  ⫽ .49, t ⫽ 6.00, p ⬍ .000, yielded significant €yberball experience effects. The regression on mood controlling for needs showed that needs affected mood,  ⫽ .78, t ⫽ 7.96, p ⬍ .000, and made the €yberball experience effect on needs disappear,  ⫽ ⫺.08, t ⫽ ⫺.80, ns. A subsequent Sobel test showed that this reduction was statistically significant, Z ⫽ 6.54, p ⬍ .001. Furthermore, a regression analysis on need in which mood was controlled failed to show a reduction in mood,  ⫽ .50, t ⫽ 8.54, p ⬍ .000. As in Study 1, these analyses show that needs mediate mood when participants are ostracized. Table 3 Means and Standard Deviations of Need Satisfaction Index by €yberball Experience (Ostracism, Inclusion, Overinclusion) and Payoff Valences (Gain, Loss) in Study 2 Ostracism
Inclusion
Overinclusion
Variable
M
SD
M
SD
M
SD
Gain Loss
2.86 3.15
0.86 0.77
4.65 4.87
0.81 0.80
4.90 4.41
1.12 1.05
923
Table 4 Means and Standard Deviations of Mood Index by €yberball Experience (Ostracism, Inclusion, Overinclusion) and Payoff Valences (Gain, Loss) in Study 2 Ostracism
Inclusion
Overinclusion
Variable
M
SD
M
SD
M
SD
Gain Loss
4.07 4.36
1.17 1.14
5.14 5.58
0.71 0.91
5.50 4.66
0.92 1.38
Next, we focused on the conditions of €yberball experience in which participants were either included or overincluded. Note that in these conditions, €yberball experience interacted with payoff valence. We therefore focused on this interaction term. In the first step, we regressed mood on €yberball experience, payoff valence, and their interaction,  ⫽ .30, t ⫽ 3.35, p ⬍ .000. In the second step, we regressed the mediator needs on our independent variables,  ⫽ .19, t ⫽ 1.96, p ⫽ .05. In the third step, we regressed mood and needs on the independent variables. This regression showed that needs affected mood,  ⫽ .60, t ⫽ 6.00, p ⬍ .000, and showed that the interaction between €yberball experience and payoff valence on mood disappeared,  ⫽ .13, t ⫽ 1.37, ns. A Sobel test showed that this reduction was (marginally) significant, Z ⫽ 1.90, p ⫽ .06. These regression analyses show that need satisfaction levels mediated mood when participants were overincluded. Retaliation. Means and standard deviations of willingness to retaliate are given in Table 5. The 2 ⫻ 3 ANOVA on self-reported desire to punish fellow game players yielded a main effect of €yberball experience, F(1, 161) ⫽ 7.59, p ⬍ .001, 2 ⫽ .09. Tukey’s HSD tests performed to interpret this main effect showed that ostracized participants were more likely retaliate (M ⫽ 2.23, SD ⫽ 1.68) than participants who were either included (M ⫽ 1.46, SD ⫽ 0.87) or overincluded (M ⫽ 1.55, SD ⫽ 0.78). The main effect was qualified by an interaction of €yberball experience and payoff valence, F(1, 161) ⫽ 6.08, p ⬍ .003, 2 ⫽ .07. Simple main effects analysis showed that our €yberball experience manipulation moderated willingness to retaliate only in the gain game, F(1, 161) ⫽ 12.76, p ⬍ .09, 2 ⫽ .13. Tukey’s HSD tests to interpret this simple main effect showed that ostracized participants were more likely to punish fellow game players than included or overincluded participants in a gain game. Simple main effects analysis also showed that our payoff valence manipulation only lowered willingness to retaliate in the ostracism conditions, F(1, 161) ⫽ 12.18, p ⬍ .001, 2 ⫽ .07. This interaction is consistent with our reasoning that in contrast to needs and emotions, subsequent coping responses to ostracism can be mitigated by situational concerns. Finally, a direct comparison between being ostracized from gains and being overincluded in losses revealed that participants were more likely to punish fellow participants when ostracized from gains than when overincluded in losses, t(53) ⫽ 2.35, p ⬍ .01. This shows that also on our retaliation measure, getting no positive attention had more negative consequences than getting punitive attention. Prosocial behavior— donating to a charitable cause. Means and standard deviations of prosocial behavior are given in Table 6.
VAN BEEST AND WILLIAMS
924
To analyze how much participants contributed to Darfur, we first performed a log transformation on the data because the data were heavily skewed. A 3 ⫻ 2 ANOVA yielded only an interaction trend of €yberball experience and payoff valence, F(1, 161) ⫽ 6.04, p ⬍ .09, 2 ⫽ .06. Simple main effects analysis to interpret this interaction showed that €yberball experience only moderated prosocial behavior in the gain game, F(1, 161) ⫽ 3.78, p ⬍ .025, 2 ⫽ .045. Further Tukey’s HSD tests comparing conditions within the gain game showed that ostracized participants donated less to charity than overincluded participants. Simple main effects analysis also revealed that payoff valence only moderated prosocial behavior in the overinclusion condition, F(1, 161) ⫽ 9.49, p ⬍ .002, 2 ⫽ .056. Participants donated more to charity when overincluded in a gain game than when overincluded in a loss game. This is again in agreement with our reasoning that coping responses to €yberball experience are moderated by financial implications. Finally, a direct comparison between being ostracized from losses and overincluded in gains was not statistically significant, t(53) ⫽ .32, ns. That is, different from our findings with the measure of antisocial intent, ostracized participants were not less prosocial than participants who received punitive attention.
Discussion We replicated the main results of Study 1 using a different method of payment. Participants found ostracism unpleasant even when it prevented them from losing all their money. Subsequent mediation analyses replicated the findings of Study 1 indicating that ostracism lowers need satisfaction levels and that this in turn negatively affects mood. We also found that whereas participants’ negative reactions were not mitigated by payoff valence when being ostracized, they were mitigated when being overincluded. Compared with people who were being included, ostracized people felt worse regardless of payoff valence, whereas overincluded people only felt worse in a loss game. This result provides further evidence that ostracism is indeed a powerful experience, overriding rational cost– benefit evaluations of a situation. It also speaks to the sensitivity of our dependent variables because of their ability to reveal differences between the overinclusion conditions. We also observed interactions between €yberball experience and payoff valence on antisocial intent and prosocial behavior. In agreement with our reasoning that only the immediate responses to ostracism are highly resistant to moderation, we observed that the financial implications of a situation influenced only subsequent coping responses. People were especially aggressive when ostra-
Table 5 Means and Standard Deviations Willingness to Retaliate by €yberball Experience (Ostracism, Inclusion, Overinclusion) and Payoff Valences (Gain, Loss) in Study 2 Ostracism
Inclusion
Overinclusion
Variable
M
SD
M
SD
M
SD
Gain Loss
2.77 1.70
2.00 1.09
1.53 1.38
1.08 0.60
1.33 1.78
0.59 0.89
Table 6 Means and Standard Deviations of Prosocial Behavior (in Euros) by €yberball Experience (Ostracism, Inclusion, Overinclusion) and Payoff Valences (Gain, Loss) in Study 2 Ostracism
Inclusion
Overinclusion
Variable
M
SD
M
SD
M
SD
Gain Loss
1.68 1.73
1.58 1.21
1.83 1.60
1.58 1.47
2.40 1.63
1.28 1.46
cized from a gain game and more prosocial when overincluded in a gain game. Another finding concerned our comparison between the ostracized from gain condition (i.e., ostracized from a chance of doubling payoffs) and the overincluded in loss condition (i.e., overincluded in a chance of losing all payoffs). Direct comparisons showed that participants felt worse in terms of need satisfaction and mood when they could not double their payoffs than when they were under the constant threat of losing all their payoffs. Moreover, participants who could not double their payoffs were more willing to retaliate than participants who were under constant threat of losing all their payoffs. Combined, these results add empirical support to the notion that people might prefer getting punitive attention to getting no attention at all (Williams & Zadro, 2001).
General Discussion In two studies, we loaded the dice in favor of ostracism by making ostracism more financially rewarding than inclusion. Results of both studies indicated that penalizing inclusion or making ostracism financially beneficial did not mitigate the phenomenological experience of being ostracized. Regardless of whether people lost small amounts of their experimental earnings (Study 1) or were threatened with losing all their money (Study 2), people felt worse when ostracized from a costly situation than when included or even overincluded in a costly situation. In addition, we assessed the relation between needs and mood. Both Study 1 and Study 2 showed that need satisfaction levels mediated mood and not vice versa when people were ostracized. This indicates that, at least in our studies, mood was lowered because of an appraisal of needs, not that people’s needs were thwarted because of lowered mood.
Immediate Responses to Ostracism The general thrust of our data is consistent with other research that attempted to find moderation of need satisfaction levels and mood during or immediately following ostracism. Again, ostracism was not moderated by a factor that should logically mitigate the stress of being ostracized (Eisenberger et al., 2003; Gonsalkorale & Williams, in press; Williams et al., 2000; Zadro et al., 2004). This provides further support for the notion that immediate reactions to ostracism may indeed be precognitive and linked to pain (McDonald & Leary, 2005). The current findings also address a possible criticism on research failing to find mitigating circumstances of ostracism. It may be argued that these studies are merely
WHEN INCLUSION COSTS
reporting null effects. In Study 2, we found moderation on selfreports of need satisfaction and mood in our overinclusion conditions, suggesting that our measures are sensitive to social pressures.
Subsequent Responses to Ostracism We also observed that payoff valences interacted with €yberball experience on retaliation and donations to Darfur. People reported desiring retaliation following ostracism but less so when ostracized from a loss game. Furthermore, overincluded participants, especially those who were overincluded in a gain game, donated more money to charity than ostracized participants donated. This is consistent with previous research on anti- and prosocial behavior that found that people seem to become more antisocial and less prosocial when excluded from a good situation (Catanese & Tice, 2005; Twenge et al., 2001; Warburton et al., 2006). An interesting contribution of our findings is that aggression was moderated by payoff valences when people were ostracized, whereas donations to charity were moderated by payoff valences when people were overincluded. Although speculative, this finding might suggest that positive behavior is more likely to occur after positive antecedents, whereas negative behavior is more likely to occur after negative antecedents. Whether this is indeed the case or whether this is perhaps due to the fact that our measure of negative behavior was directed at fellow participants whereas our measure of positive behavior was not directed at fellow participants remains to be examined in future research. What is important here is that ostracism increased aggression and decreased prosocial behavior but that these behavioral measures were mitigated by financial incentives.
Should Ostracism Be Viewed as One End of an Inclusionary Status Continuum? Introducing an overinclusion condition also afforded the possibility to compare two different situations in which other people single out individuals — ostracism and overinclusion. In the ostracism condition, people are the “objects of others’ inattention” (Williams et al., 2000, p. 169) and are singled out by receiving less attention than expected. In the overinclusion condition, people are singled out because they receive more attention than expected. This allowed us to assess whether the experience of ostracism can be viewed as a continuum that ranges from ostracism, through inclusion, to overinclusion (see also Leary, 1990). Consistent with the findings of Williams et al. (2000) who showed that being overincluded in a ball-tossing game without financial incentives is more rewarding than being included in a ball-tossing game, we show that well-being is indeed a linear function of inclusionary status in situations in which people earn money whenever they get a ball toss. What is more important, however, is that we also introduced conditions in which individuals lost money whenever they were thrown the ball. In this situation, well-being was not a linear function of inclusionary status. Instead, well-being went down in both the overinclusion and the ostracism conditions. Moreover, the fact that financial incentives moderated immediate responses to overinclusion but not to ostracism suggests that individuals process information differently in these situations. It seems that people do not process information about the pros and
925
cons of a situation when singled out in terms of ostracism but do process such information when singled-out in terms of overinclusion. In this respect, we agree with Leary (2005) who argued that conceptualizing rejection as an index of inclusionary status ranging from maximal exclusion to maximal inclusion may be useful for capturing the effort people use to exclude others, but it may not be useful in accounting for what victims of exclusion experience.
Bridging Two Forms of Peer Rejection Our paradigm also allowed for a comparison between being ostracized from the possibility of doubling one’s money and being overincluded when loss of all of one’s money loomed. Results showed that participants responded more negatively both in terms of immediate responses and aggressive intentions when ostracized from the gain situation than when overincluded in the loss situation. This finding suggests that being ostracized is indeed a very negative experience, perhaps worse that being ganged up on in such a way that increases our chances of losing valued resources. We can link this finding to research on peer rejection that stresses that both ostracism and bullying are possible antecedents of aggressive behavior (Juvonen & Gross, 2005). For example, studies on the infamous school shootings in the United States reveal that many of the adolescents who hurt and killed their fellow schoolmates were allegedly rejected and bullied by their peers (Leary et al., 2003). To the extent that our manipulation of punitive attention can be viewed as one form of bullying (Olweus, 1978; Zadro, Williams, & Richardson, 2005), we attempted to bridge these two forms of peer rejection. The current findings concur that both these forms of peer rejection are perceived as painful. Yet, we also show that, at least in our paradigm, ostracism is worse than our form of bullying. Combined with the fact that in our paradigm, ostracized individuals were more willing to lash out than bullied individuals, our finding underscores the particularly negative impact and consequences of ostracism.
Allocating Gains and Losses At this point, it may be relevant to compare our current findings with other research on gain–loss framing. A general finding in this literature is that losses loom larger than gains (Kahneman & Tversky, 1979; Ku¨hberger, 1998; Tversky & Kahneman, 1991). In individual decision tasks, individuals are more likely to minimize their losses than to maximize their gains, and they react more strongly to a loss than to an equivalent gain. In interdependent decision tasks, in which participants’ actions affect the payoff of others, people may lower their own payoff so as not to harm others as long they are led to focus on the payoffs of others (De Dreu & McCusker, 1997; van Beest et al., 2005; van Beest, Wilke, & Van Dijk, 2003). Whereas a recent study showed that participants were more likely to ostracize others when negotiating about gains than when negotiating about losses (van Beest et al., 2005), the present findings suggest that targets of ostracism do not appear to discriminate between being ostracized in a gain or a loss domain. An interesting implication then is that ostracizers may underestimate the damage they inflict when they ostracize people in a gain domain (or overestimate the damage in a loss domain). On a more general level, our results have implications for situations in which people negotiate or divide payoffs (De Dreu &
926
VAN BEEST AND WILLIAMS
Carnevale, 2003). In such situations, getting payoffs or not getting them is often related to being included or excluded from an agreement. The current results suggest that people may indeed respond fiercely to violations of equality or equity (Messick & Sentis, 1979, 1983; Walster & Walster, 1975; Walster, Walster, & Berscheid, 1978) in negotiation or distribution tasks but that those who do not obtain payoffs at all may feel worse. The current results illustrate that, as in the Olympic spirit, it often is not about winning or losing but indeed just about participating.
Limitations and Boundary Conditions One concern that might be raised is that had we just offered enough money in the ostracism– gain condition, we would have eliminated the pain and distress of ostracism. We agree that there may certainly be some excessively high monetary amount that might obliterate the initial pain of ostracism, but we are skeptical that our chosen amounts were so low as to give unfair advantage to the ostracism manipulation for a few reasons. First, we used payoff valence manipulations that were successful in causing differential behavior in other research areas such as negotiation. In fact, we adapted a payoff valence procedure that did impact coalition formation and thus the willingness to ostracize (van Beest, et al. 2005). Second, our ostracism manipulation was intentionally minimal. The participants neither saw nor knew the other players, nor did they expect to see or meet them in the future. One could easily imagine a much stronger ostracism manipulation in which participants endure the pain of being ignored and excluded in the presence of others for whom they care. In light of other research, it seems to us most plausible that ostracism is at least momentarily painful despite offsetting factors that rationally should minimize individuals’ appraisal of the importance and relevance of the ostracism experience. As reviewed in the introduction, individuals are distressed by ostracism when a computer (rather than humans) ostracizes them (Zadro et al., 2004), when a despised outgroup ostracizes them (Gonsalkorale & Williams, in press), and when they know the others are not able to include them because of technical limitations (Eisenberger et al, 2003). In the present studies, the pain appears to endure even when the meaning of ostracism is symbolically linked with relative wealth.
Extensions The results of our studies may also be of interest to domains beyond social psychology. Indeed, there is ample literature and scholarly attention to ostracism in biology (Gruter & Masters, 1986), anthropology (Boehm, 1986; Gruter, 1986; Mahdi, 1986), law and politics (Anawalt, 1986; Kort, 1986; Rehbinder, 1986; Weisberger, 1986), and developmental psychology (Barner-Barry, 1986; Crick, Ostrov, Appleyard, Jansen, & Casas, 2004). The interest in ostracism is commensurate with its commonality, power, and consequences across all levels of social units and apparently among all social species. Our studies suggest that the need to belong, to feel good about oneself, to have some control over social interactions, and to be recognized as existing are potent motives that overwhelm rational thought when they are first threatened by ostracism. The next question for researchers and theorists will be to determine how ostracized individuals cope and respond once they have
had time for reflection and appraisal. In his summary of the current empirical literature in ostracism and social exclusion, Williams (in press) describes four different categories of subsequent reactions to ostracism. Responses to ameliorate the threats posed by ostracism may be to (a) increase one’s inclusionary status by attending more closely to social information and behaving more socially acceptably, perhaps to the extent of becoming overly susceptible to social influence; (b) reclaim control or provoke recognition of oneself by becoming more aggressive; (c) shut down emotionally and cognitively, as if in a state of stunned numbness; or (d) withdraw and ostracize oneself from others, preventing the possibility of further social pain. Although it would seem plausible that, on reflection, ostracism from a loss game would be more easily dismissed as nonthreatening or perhaps even regarded as protective and nurturing, it might still diminish one’s sense of self-worth and overall contribution to the group. For instance, in many U.S. jurisdictions, the elderly are not obligated to fulfill their duty to serve as jurors (Fingerman & Hanley, 2006). Although this might give the elderly temporary relief from an arduous task, it might also make them feel societally invisible and unimportant (Levy, 2003).
Conclusion The results of these two studies again point to the power of ostracism: Even when ostracism results in financial gain (relative to the other players), it hurts. Going back to our rather extreme hypothetical situation posed earlier about Russian roulette, we would tentatively conclude that an individual would feel unhappy about being left out of such a potential costly activity.
References Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting interactions. Thousand Oaks, CA: Sage. Anawalt, H. C. (1986). Ostracism and the law of defamation. Ethology and Sociobiology, 7, 181–190. Barner-Barry, C. (1986). Rob: Children’s tacit use of peer ostracism to control aggressive behavior. Ethology and Sociobiology, 7, 281–293. Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173– 1182. Boehm, C. (1986). Capital punishment in tribal Montenegro: Implications for law, biology, and theory of social control. Ethology and Sociobiology, 7, 305–320. Case, T. I., & Williams, K. D. (2004). Ostracism: A metaphor for death. In J. Greenberg, S. L. Koole, & T. Pyszczynski (Eds.), Handbook of experimental existential psychology (pp. 336 –351). New York: Guilford. Catanese, K. R., & Tice, D. M. (2005). The effect of rejection on antisocial behaviors: Social exclusion produces aggressive behaviors. In K. D. Williams, J. P. Forgas, & W. von Hippel (Eds.), The social outcast: Ostracism, social exclusion, rejection, and bullying (pp. 297– 305). New York: Psychology Press. Crick, N. R., Ostrov, J. M., Appleyard, K., Jansen, E. A., & Casas, J. F. (2004). Relational aggression in early childhood: “You can’t come to my birthday party unless . . .” In M. Putallaz. & K. L. Bierman (Eds), Aggression, antisocial behavior, and violence among girls: A developmental perspective. (pp. 71– 89). New York: Guilford. De Dreu, C. K. W., & Carnevale, P. (2003). Motivational bases of information processing and strategy in conflict and negotiation. In M. P.
WHEN INCLUSION COSTS Zanna (Ed.), Advances in experimental social psychology (Vol. 35, pp. 235–291). San Diego, CA: Academic. De Dreu, C. K. W., & McCusker, C. (1997). Gain–loss frames and cooperation in two-person social dilemmas: A transformational analysis. Journal of Personality and Social Psychology, 72, 1093–1106. Eisenberger, N. I., & Lieberman, M. D. (2005). Why it hurts to be left out: The neurocognitive overlap between physical and social pain. In K. D. Williams, J. P. Forgas, & W. von Hippel (Eds.), The social outcast: Ostracism, social exclusion, rejection, and bullying (pp. 109 –130). New York: Psychology Press. Eisenberger, N. I., Lieberman, M. D., & Williams, K. D. (2003, October 10). Does rejection hurt? An fMRI study of social exclusion. Science, 302, 290 –292. Fingerman, K. L., & Hanley, B. (2006). Jury duty exemption based on old age. Unpublished manuscript, Purdue University, West Lafayette, IN. Gaertner, L., & Iuzzini, J. (2005). Rejection and entitativity: A synergistic model of mass violence. In K. D. Williams, J. P. Forgas, & W. von Hippel (Eds.), The social outcast: Ostracism, social exclusion, rejection, and bullying (pp. 307–320). New York: Psychology Press. Gonsalkorale, K., & Williams, K. D. (in press). The KKK won’t let me play: Ostracism even by despised outgroups hurts. European Journal of Social Psychology. Gruter, M. (1986). Ostracism on trial: The limits of individual rights. Ethology and Sociobiology, 7, 271–279. Gruter, M., & Masters, R. D. (1986). Ostracism as a social and biological phenomenon: An introduction. Ethology and Sociobiology, 7, 149 –158. James, W. (1950). Principles of psychology: Vol. 1. New York: Dover. (Original work published 1890) Juvonen, J., & Gross, E. F. (2005). The rejected and the bullied: Lessons about social misfits from developmental psychology. In K. D. Williams, J. P. Forgas, & W. von Hippel (Eds.), The social outcast: Ostracism, social exclusion, rejection, and bullying (pp. 155–170). New York: Psychology Press. Kahneman, D., & Tversky, A. (1979). Prospect theory: Analysis of decision under risk. Econometrica, 47, 263–291. Kort, F. (1986). The politics of ostracism. Ethology and Sociobiology, 7, 219 –230. Ku¨hberger, A. (1998). The influence of framing on risky decisions: A meta-analysis. Organizational Behavior and Human Decision Processes, 75, 23–55. Lakin, J. L., & Chartrand, T. L. (2005). Exclusion and nonconscious behavioral mimicry. In K. D. Williams, J. P. Forgas, & W. von Hippel (Eds.), The social outcast: Ostracism, social exclusion, rejection, and bullying (pp. 279 –296). New York: Psychology Press. Leary, M. R. (1990). Responses to social exclusion: Social anxiety, jealousy, loneliness, depression, and low self-esteem. Journal of Social and Clinical Psychology, 9, 221–229. Leary, M. R. (2005). Varieties of interpersonal rejection. In K. D. Williams, J. P. Forgas, & W. von Hippel (Eds.), The social outcast: Ostracism, social exclusion, rejection, and bullying (pp. 213–226). New York: Psychology Press. Leary, M. R., Kowalski, R. M., Smith, L., & Phillips, S. (2003). Teasing, rejection, and violence: Case studies of the school shootings. Aggressive Behavior, 29, 202–214. Levy, B. R. (2003). Mind matters: Cognitive and physical effects of aging self-stereotypes. Journals of Gerontology: Psychological Sciences, 58, 203–211. Mahdi, N. Q. (1986). Pukhtunwali: Ostracism and honor among the Pathan hill tribes. Ethology and Sociobiology, 7, 147–156. McDonald, G., & Leary, M. R. (2005). Why does social exclusion hurt? The relationship between social and physical pain. Psychological Bulletin, 131, 202–223. Messick, D. M., & Sentis, K. P. (1979). Fairness and preference. Journal of Experimental Social Psychology, 15, 418 – 434.
927
Messick, D. M., & Sentis, K. P. (1983). Fairness, preference, and fairness biases. In D. M. Messick & K. S. Cook (Eds.), Equity theory (pp. 61–94). New York: Praeger. Olweus, D. (1978). Aggression in the schools: Bullies and whipping boys. Oxford, England: Hemisphere. Ouwerkerk, J. W., Kerr, N. L., Gallucci, M., & Van Lange, P. A. M. (2005). Avoiding the social death penalty: Ostracism and cooperation in social dilemmas. In K. D. Williams, J. P. Forgas, & W. von Hippel (Eds.), The social outcast: Ostracism, social exclusion, rejection, and bullying (pp. 321–332). New York: Psychology Press. Pickett, C. L., & Gardner, W. L. (2005). The social monitoring system: Enhanced sensitivity to social cues and information as an adaptive response to social exclusion and belonging need. In K. D. Williams, J. P. Forgas, & W. von Hippel (Eds.), The social outcast: Ostracism, social exclusion, rejection, and bullying (pp. 213–226). New York: Psychology Press. Pickett, C. L., Gardner, W. L., & Knowles, M. (2004). Getting a cue: The need to belong and enhanced sensitivity to social cues. Personality and Social Psychology Bulletin, 30, 1095–1107. Predmore, S. C., & Williams, K. D. (1983, May). The effects of social ostracism on affiliation. Paper presented at the Midwestern Psychological Association, Chicago. Rehbinder, M. (1986). Refusal of social cooperation as a legal problem: On the legal institutions of ostracism and boycott. Ethology and Sociobiology, 7, 173–180. Smith, A., & Williams, K. D. (2004). R U there? Effects of ostracism by cell phone. Group Dynamics: Theory, Research, & Practice, 8, 291– 301. Tversky, A., & Kahneman, D. (1991). Loss aversion in riskless choice: A reference-dependent model. Quarterly Journal of Economics, 106, 1039 –1061. Twenge, J. M. (2005, October). Why does social rejection lead to aggression? Paper presented at the Society of Experimental Social Psychology, San Diego, CA. Twenge, J. M., & Baumeister, R. F. (2005). Social exclusion increases aggression and self-defeating behavior while reducing intelligent thought and prosocial behavior. In D. Abrams, M. A. Hogg, & J. Marques (Eds.), The social psychology of inclusion and exclusion (pp. 27– 46). New York: Psychology Press. Twenge, J. M., Baumeister, R. F., Tice, D. M., & Stucke, T. S. (2001). If you can’t join them, beat them: The effects of social exclusion on aggressive behavior. Journal of Personality and Social Psychology, 81, 1058 –1069. Twenge, J. M., Catanese, K. R., & Baumeister, R. F. (2003). Social exclusion and the deconstructed state: Time perception, meaninglessness, lethargy, lack of emotion, and self-awareness. Journal of Personality and Social Psychology, 85, 409 – 423. van Beest, I., Van Dijk, E., De Dreu, C. K. W., & Wilke, H. A. M. (2005). Do-no-harm in coalition formation: Why losses inhibit exclusion and promote fairness cognitions. Journal of Experimental Social Psychology, 41, 609 – 617. van Beest, I., Wilke, H. A. M., & Van Dijk, E. (2003). The excluded player in coalition formation. Personality and Social Psychology Bulletin, 29, 237–247. Walster, E., & Walster, G. W. (1975). Equity and social justice. Journal of Social Issues, 31, 21– 43. Walster, E., Walster, G. W., & Berscheid, E. (1978). Equity: Theory and research. Boston: Allyn and Bacon. Warburton, W. A., Williams, K. D., & Cairns, D. R. (2006). When ostracism leads to aggression: The moderating effects of control deprivation. Journal of Experimental Social Psychology, 42, 213–220. Weisberger, J. M. (1986). Marital property discrimination: Reform for legally excluded women. Ethology and Sociobiology, 7, 205–218. Williams, K. D. (1997). Social ostracism. In R. M. Kowalski (Ed.), Aversive interpersonal behaviors (pp. 133–170). New York: Plenum.
VAN BEEST AND WILLIAMS
928
Williams, K. D. (2001). Ostracism: The power of silence. New York: Guilford. Williams, K. D. (in press). Ostracism. Annual Review of Psychology (Vol. 58). New York: Annual Reviews. Williams, K. D., Cheung, C. K. T., & Choi, W. (2000). CyberOstracism: Effects of being ignored over the Internet. Journal of Personality and Social Psychology, 79, 748 –762. Williams, K. D., Forgas, J. P., & von Hippel, W. (Eds). (2005). The social outcast: Ostracism, social exclusion, rejection, and bullying. New York: Psychology Press. Williams, K. D., Govan, C. L., Croker, V., Tynan, D., Cruickshank, M., & Lam, A. (2002). Investigations into differences between social and cyber ostracism. Group Dynamics: Theory, Research, & Practice, 6, 65–77. Williams, K. D., & Jarvis, B. (in press). Cyberball: A program for use in research on ostracism and interpersonal acceptance. Behavior Research Methods, Instruments, and Computers.
Williams, K. D., & Sommer, K. L. (1997). Social ostracism by one’s coworkers: Does rejection lead to loafing or compensation? Personality and Social Psychology Bulletin, 23, 693–706. Williams, K. D., & Zadro, L. (2001). Ostracism: On being ignored, excluded, and rejected. In M. R. Leary (Ed.), Interpersonal rejection (pp. 21–53). New York: Oxford University Press. Zadro, L., Boland, C., & Richardson, R. (in press). How long does it last? The persistence of the effects of ostracism in the socially anxious. Journal of Experimental Social Psychology. Zadro, L., Williams, K. D., & Richardson, R. (2004). How low can you go? Ostracism by a computer lowers belonging, control, self-esteem, and meaningful existence. Journal of Experimental Social Psychology, 40, 560 –567. Zadro, L., Williams, K. D., & Richardson, R. (2005). Riding the “O” train: Comparing the effects of ostracism and verbal dispute on targets and sources. Group Processes and Intergroup Relations, 8, 125–143.
Appendix Need Threat Scale Needs were assessed on 7-point scales ranging from 1 (do not agree) to 7 (agree). Questions ending with an “R” were recoded. Belongingness
Self-Esteem
1.
I felt as one with the other players.
1.
Playing the game made me feel insecure. (R)
2.
I had the feeling that I belonged to the group during the game.
2.
I had the feeling that I failed during the game. (R)
3.
I did not feel accepted by the other players. (R)
3.
I had the idea that I had the same value as the other players.
4.
During the game I felt connected with one of more other players.
4.
I was concerned about what the other players thought about me during the game. (R)
5.
I felt like an outsider during the game. (R) 5.
I had the feeling that the other players did not like me. (R)
Control
Meaningful Existence
1.
I had the feeling that I could throw as often as I wanted to the other players.
2.
I felt in control over the game.
3.
I had the idea that I affected the course of the game.
4. 5.
1.
During the game it felt as if my presence was not meaningful. (R)
2.
I think it was useless that I participated in the game. (R)
3.
I had the feeling that my presence during the game was important.
I had the feeling that I could influence the direction of the game.
4.
I think that my participation in the game was useful.
I had the feeling that the other players decided everything. (R)
5.
I believed that my contribution to the game did not matter. (R)
Received January 27, 2006 Revision received May 2, 2006 Accepted May 11, 2006 䡲
Journal of Personality and Social Psychology 2006, Vol. 91, No. 5, 929 –943
Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.5.929
When Sex Is More Than Just Sex: Attachment Orientations, Sexual Experience, and Relationship Quality Gurit E. Birnbaum
Harry T. Reis
Bar-Ilan University
University of Rochester
Mario Mikulincer
Omri Gillath
Bar-Ilan University
University of Kansas
Ayala Orpaz Bar-Ilan University The authors explored the contribution of individual differences in attachment orientations to the experience of sexual intercourse and its association with relationship quality. In Study 1, 500 participants completed self-report scales of attachment orientations and sexual experience. The findings indicated that whereas attachment anxiety was associated with an ambivalent construal of sexual experience, attachment avoidance was associated with more aversive sexual feelings and cognitions. In Study 2, 41 couples reported on their attachment orientations and provided daily diary measures of sexual experiences and relationship interactions for a period of 42 days. Results showed that attachment anxiety amplified the effects of positive and negative sexual experiences on relationship interactions. In contrast, attachment avoidance inhibited the positive relational effect of having sex and the detrimental relational effects of negative sexual interactions. The authors discuss the possibility that attachment orientations are associated with different sex-related strategies and goals within romantic relationships. Keywords: attachment, gender differences, romantic relationships, sexuality
tively little attention has been given to the subjective experience of sexual intercourse as well as to the association between sexual experience and relationship quality. The current research was intended to add to our understanding of the attachment–sexuality link within romantic relationships by examining the association between attachment orientations and the multifaceted emotional and cognitive components of subjective sexual experience and by exploring the possible role of attachment in moderating the complex linkage between sexuality and relationship quality.
Within attachment theory, adult romantic love involves the integration of three distinct behavioral systems: attachment, caregiving, and sexual mating (Bowlby, 1969/1982; Shaver, Hazan, & Bradshaw, 1988). Because the attachment system is the earliest developing social– behavioral system (Cassidy, 1999), it plays a crucial role in molding the functioning of the caregiving and sexual systems and shaping cognitive models for social life. Nevertheless, the sexual behavior system may also influence attachment by fostering the development of emotional bonds between sexual partners (Hazan & Zeifman, 1994). Indeed, empirical evidence points to a reciprocal relationship between the attachment system and the sexual system: Sexual satisfaction contributes to a relationship’s quality and stability (see review by Sprecher & Cate, 2004), and attachment orientations influence the way in which adolescents and adults construe their romantic relationships (see Feeney, 1999, for a review) and sexual interactions (see Feeney & Noller, 2004, for a review). However, although past research has provided substantial evidence about the role of attachment orientations in shaping sexual motives, attitudes, and behaviors, rela-
Contribution of Attachment Orientations to Relationship Quality and Sex According to attachment theory (Bowlby, 1969/1982, 1973), the quality of interactions with significant others in times of need shapes interaction goals, relational cognitions, and interpersonal behavior. When significant others are perceived as available and responsive to proximity-seeking attempts, a sense of attachment security is attained, intimacy and nurturance become primary interaction goals, and partners are thought to be trustworthy and reliable. However, when partners are felt to be emotionally unavailable, insecurities and doubts about close relationships predominate, leading to the adoption of either of two defensive strategies for dealing with these insecurities (Cassidy & Kobak, 1988; Mikulincer & Shaver, 2003). Under a hyperactivation strategy, the main goal is to get a relationship partner, perceived as insufficiently available and responsive, to provide support and
Gurit E. Birnbaum, Mario Mikulincer, and Ayala Orpaz, Department of Psychology, Bar-Ilan University, Ramat-Gan, Israel; Harry T. Reis, Department of Clinical and Social Sciences in Psychology, University of Rochester; Omri Gillath, Department of Psychology, University of Kansas. Correspondence concerning this article should be addressed to Gurit E. Birnbaum, Department of Psychology, Bar-Ilan University, Ramat-Gan 52900, Israel. E-mail:
[email protected] 929
930
BIRNBAUM, REIS, MIKULINCER, GILLATH, AND ORPAZ
protection. On the other hand, the main goal of deactivation strategies is to maintain emotional distance from relationship partners and to strive for self-reliance (Main, 1990; Mikulincer & Shaver, 2003). These behavioral strategies are thought to underlie many phenomena associated with attachment orientations: the systematic patterns of relational expectations, emotions, and behaviors that result from a particular attachment history (Fraley & Shaver, 2000; Shaver & Mikulincer, 2002). In turn, these different strategies are closely related to the two major dimensions thought to underlie attachment orientations. Research, beginning with Ainsworth, Blehar, Waters, and Wall (1978) and continuing through recent studies of adult attachment, indicates that individual differences in attachment orientations are appropriately conceptualized as regions in a two-dimensional space (Brennan, Clark, & Shaver, 1998; Fraley & Waller, 1998). The first dimension, typically called attachment avoidance, reflects the extent to which a person distrusts a relationship partners’ goodwill, strives to maintain behavioral independence, and relies on deactivating strategies for dealing with relational threats. The second dimension, typically called attachment anxiety, reflects the degree to which a person worries that partners will not be available in times of need and thereby hyperactivates cognitions and behaviors in an attempt to secure such availability (Mikulincer & Shaver, 2003). Attachment security is located in the region in which both anxiety and avoidance are low and is defined by comfort with closeness and faith in the reliability of caregivers. Previous research has shown that attachment orientations help explain variations in the construal and experience of romantic relationships (see Feeney, 1999, for a review). Securely attached individuals, compared with insecure individuals, are more likely to have long, stable, and satisfying relationships characterized by high involvement, trust, intimacy, warmth, support, and cohesion (e.g., Hazan & Shaver, 1987; Kirkpatrick & Davis, 1994; Mikulincer & Florian, 1999; Simpson, 1990). In contrast, in line with their goal of deactivating attachment concerns, highly avoidant individuals are less likely to fall in love and are less interested in being involved in long-term committed relationships (Hatfield, Brinton, & Cornelius, 1989; Shaver & Brennan, 1992). Accordingly, avoidant individuals tend to have relatively less stable relationships characterized by fear of intimacy and low levels of emotional involvement, trust, cohesion, and satisfaction (e.g., Collins & Read, 1990; Hazan & Shaver, 1987; Kirkpatrick & Davis, 1994; Mikulincer & Florian, 1999; Shaver & Brennan, 1992). Highly anxious individuals’ relationships, in contrast, tend to be organized around hyperactivation of the attachment system, manifested in obsessive and passionate romantic feelings; clinging, intrusive, and controlling patterns of relational behaviors; strong desire for merger with the partner; worries about rejection and abandonment; and bouts of jealousy and anger (e.g., Collins & Read, 1990; Hatfield et al. 1989; Hazan & Shaver, 1987; Mikulincer, Orbach, & Iavnieli, 1998). Paradoxically, highly anxious people’s demands for security, combined with frequent demonstrations of distrust and rage, may lead their partner to reject their proximity-seeking attempts, which in turn may intensify their own insecurities and exacerbate relationship conflicts (Downey, Freitas, Michaelis, & Khouri, 1998; Mikulincer & Shaver, 2003). Research has shown that attachment orientations are associated with the functioning of the sexuality system (e.g., D. Davis,
Shaver, & Vernon, 2004). In line with their pursuit of establishing intimate, faithful, and satisfying long-term relationships (Mikulincer & Shaver, 2003), securely attached people report preferring sexual activity in committed romantic relationships (e.g., Brennan & Shaver, 1995; Stephan & Bachman, 1999). As adolescents, securely attached individuals reported fewer one-night stands than their insecure counterparts (Cooper, Shaver, & Collins, 1998) and describe engaging in sexual intercourse mainly to express love for their partner (Tracy, Shaver, Albino, & Cooper, 2003). Tracy et al. (2003) also found that, compared with insecure adolescents, secure adolescents were less erotophobic (i.e., experienced fewer negative affective– evaluative responses to sexual cues) and experienced fewer negative emotions and more positive and passionate emotions during sexual activity. Similarly, in adulthood, secure individuals have more positive sexual self-schemas (Cyranowski & Andersen, 1998), report greater pleasure from the use of touch to express affection and sexuality (Brennan, Wu, & Loev, 1998; Hazan, Zeifman, & Middleton, 1994), and enjoy exploratory sexual activities with long-term partners (Hazan et al., 1994). Together, these findings suggest that securely attached individuals’ comfort with sexual intimacy and enjoyment of sexual interactions may contribute to their stable and satisfying romantic relationships. Sexual activity, with its inherent demand for physical and psychological intimacy, may create discomfort for avoidant people, who, as noted above, habitually seek physical and emotional distance from their partners. Consequently, they may attempt to deactivate the attachment system, manifested in two ways: abstaining from sexual activity (Kalichman et al., 1993; Tracy et al., 2003) or engaging in relatively emotion-free sex in the context of casual, short-term relationships (e.g., Brennan & Shaver, 1995; Gentzler & Kerns, 2004; Schachner & Shaver, 2002). There is empirical evidence that avoidant adolescents are relatively erotophobic, low in perceived sex drive, and less likely to participate in sexual interactions (Tracy et al., 2003). Moreover, when sexual intercourse does occur, avoidant adolescents report less enjoyment and greater focus on self-enhancing motives, such as losing their virginity, than relationship-focused motives, such as expressing love for their partner (Tracy et al., 2003). In adulthood, relatively avoidant participants tend to dismiss motives related to the promotion of emotional closeness, whereas they emphasize motives related to partner manipulation and control, protection of the self from partners’ negative affect, stress reduction, and prestige among peers (Cooper et al., 2006; D. Davis et al., 2004; Schachner & Shaver, 2004). Hence, avoidant persons are less likely to enjoy affectionate presexual activities (e.g., cuddling, kissing) and intimate copulatory positions (Brennan, Wu, & Loev, 1998; Hazan et al., 1994) and are more likely to make and respond favorably to short-term mate-poaching attempts (Schachner & Shaver, 2002). Overall, people high in avoidance may use sex to maximize control and distance even in the most intimate interactions. Highly anxious people’s construal of sexual activities reflects their attempts to fulfill unmet attachment-related needs for security and love. As adolescents, highly anxious persons are more likely to engage in sex to avoid abandonment (Tracy et al., 2003), which, in turn, leads to more common unwanted sexual behaviors (Feeney, Peterson, Gallois, & Terry, 2000) and interferes with the experience of passionate emotions during sex (Tracy et al., 2003). As adults, highly anxious individuals score relatively high in eroto-
ATTACHMENT, SEXUALITY, AND RELATIONSHIP QUALITY
philia (Bogaert & Sadava, 2002) and report using sex as a means to achieve emotional intimacy, approval, and reassurance; to elicit a partner’s caregiving behaviors; and to defuse a partner’s anger (D. Davis et al., 2004; Schachner & Shaver, 2004). Ironically, however, unfulfilled relational expectations or their inappropriate channeling into the sexual realm, when combined with worry about the partner’s reactions, make anxiously attached persons more prone to disappointing and dissatisfying sexual interactions (Birnbaum & Gillath, in press; Brennan, Wu, & Loev, 1998).
The Current Research Previous studies of the attachment–sexuality link have focused on attitudinal, motivational, and behavioral aspects of sexuality, but they have not fully captured the rich and complex picture of sexual experience described in theoretical and clinical literature. This experience includes a wide variety of positive and negative emotions and thoughts related to the self, the partner, the dyadic relationship, the sexual encounter, and the sexual response cycle (Birnbaum & Laser-Brandt, 2002). In addition, most prior studies have tended to be retrospective surveys that do not examine the dynamic interplay of sexuality and relationship quality in their natural context. The present research was designed to examine the association between attachment orientations and the cognitive and emotional components of sexual experience. This research also aimed to explore the possible role of attachment orientations in moderating the link between sexual experiences and daily relationship quality. Furthermore, to provide a more contextually informed view of how these processes unfold over time, we examined both retrospective and daily diary data. Our first goal was to examine associations between attachment orientations and relational, aversive, and pleasurable components of the experience of heterosexual intercourse (Birnbaum & LaserBrandt, 2002). On the basis of attachment theory and research, we made two main predictions. First, attachment avoidance would be related to a more aversive construal of sexual intercourse. Second, attachment anxiety would be related to a more complex and ambivalent construal of sexual experience. Whereas anxiously attached persons’ erotophilic tendencies may lead them to channel their relational expectations into the sexual realm and intensify the pleasurable aspects of sex, their attachment-related worries may lead simultaneously to aversive feelings during sexual intercourse. Our second goal was to explore the role of attachment orientations in explaining the link between sexual experiences and relationship quality. Growing empirical evidence has indicated that sexual satisfaction contributes to relationship’s quality and stability (for a review, see Sprecher & Cate, 2004), that sexual dysfunction may lead to disharmonious relationships (e.g., Hartman, 1983; Hassebrauck & Fehr, 2002), and that successful sex therapy may increase relationship satisfaction (Wright, Perrault, & Mathieu, 1977). Nevertheless, clinical evidence has suggested that some harmonious couples have relatively distressed sexual interactions, whereas other couples have turbulent relationships but good sex lives (e.g., Edwards & Booth, 1994; Kaplan, 1974). Indeed, the substantial body of research that has explored the association between sex and relationship quality has yielded conflicting results (e.g., Henderson-King & Veroff, 1994; LoPiccolo, Heiman, Hogan, & Roberts, 1985). These inconsistencies are somewhat difficult to resolve because of methodological discrepancies (e.g.,
931
use of specific vs. general relational measures) and limitations (e.g., failure to consider potential moderators). Furthermore, the majority of past research was correlational and retrospective and has examined the sex–relationship linkage aggregated across time and instances (for a review, see Sprecher & Cate, 2004). Such global assessments may be biased by cognitive and motivational processes (e.g., motivated construal, sentiment override), resulting in random but also systematic errors in the reconstructed assessment of sexual experiences and relational interactions (Reis & Gable, 2000). Beyond these and other methodological problems, there is still a lack of a compelling theoretical framework for understanding the functional meaning of sex within romantic relationships, particularly the interplay between sexual activity and relational problems (e.g., the possibility that sex might compensate for, or contribute to, relational difficulties). As described below, we believe that attachment theory can provide such a framework. The current studies adopted an attachment-theoretic perspective to clarify these inconsistencies. Because highly avoidant persons tend to engage in sexual intercourse for relatively self-enhancing, relationship-irrelevant reasons (D. Davis et al., 2004; Schachner & Shaver, 2004), we predicted that avoidant persons would experience a sense of disconnection between sexual activity and relationship quality. In contrast, as reviewed above, anxiously attached people use sex to meet attachment needs so that their sexual attitudes and behaviors are closely linked with strivings to induce partners to provide proximity, support, and protection (D. Davis et al., 2004; Schachner & Shaver, 2002, 2004; Tracy et al., 2003). In other words, by subordinating sexual activity to the attachment system, anxious persons’ hyperactivating strategies may strengthen the link between sexuality and relationship quality. This propensity may be fueled by highly anxious individuals’ tendency to report more pronounced vacillations in the daily quality of their romantic relationships following perceived positive or negative relational events (e.g., relationship conflict or support; Campbell, Simpson, Boldry, & Kashy, 2005). We therefore predicted that attachment anxiety would amplify the effects of positive and negative sexual experiences on relationship quality. Positive sexual interactions may temporarily satiate unmet attachment needs, mitigating relational worries, whereas frustrating and disappointing sexual experiences may be seen as additional signs of rejection and partner disapproval, thereby exacerbating attachment insecurities and relational worries. To date, the only evidence examining this hypothesis comes from one-time surveys. In examining the associations among attachment orientations, sexual experiences, and relationship quality, we also take into account evolutionary (e.g., Buss, 1998; Buss & Schmitt, 1993; Trivers, 1972) and socialization perspectives (e.g., DeLamater, 1987; Gagnon & Simon, 1973; Reiss, 1981) positing that men and women tend to experience sexual activity differently. Although these approaches explain the distal determinants of gender differences in sexuality differently, they generally agree that women tend to adopt a more emotional–interpersonal orientation to sexuality, emphasizing interpersonal factors related to sexual intercourse, whereas men tend to adopt a more recreational orientation toward sexuality, emphasizing the expression and fulfillment of sexual needs. Empirical studies have shown that, compared with men, women are more concerned with their romantic relationships during sexual intercourse and tend to experience intercourse as a reflection of relationship quality (e.g., Birnbaum & Laser-Brandt,
BIRNBAUM, REIS, MIKULINCER, GILLATH, AND ORPAZ
932
2002). Consequently, during sexual interactions, women are likely to be relatively more attuned to affect-related cues implying their partner’s intentions and willingness to invest resources (e.g., expressions of love; Birnbaum & Laser-Brandt, 2002; Birnbaum & Reis, in press). Men, on the other hand, have been shown to be relatively more motivated by physical release and to emphasize satisfaction derived from the sexual act itself (e.g., Carroll, Volk, & Hyde, 1985). On this basis, we expected that different qualities of sexual experience would predict relational interactions for highly anxious men and women. For anxious women, because they may be especially likely to use sex as an indicator of relationship quality (Birnbaum, in press; D. Davis et al., 2004), affects experienced during sex were expected to predict relational behaviors. On the other hand, for highly anxious men, relationship interactions were expected to be more strongly predicted by the fact of sexual intercourse having occurred (relative to feelings and cognitions experienced therein). Sexual experiences and relationship interactions were not expected to be associated for both highly avoidant men and women because both tend to dismiss the relational aspects of sexual interactions (Cooper et al., 2006; D. Davis et al., 2004; Schachner & Shaver, 2002). It might be noted that although the attachment literature generally shows few sex differences, we predicted that attachment anxiety would have sexdifferentiated effects when it comes to sexuality, simply because, as described above, men’s and women’s experience of sexuality differs substantially. In two studies, we examined the association between attachment orientations and the emotional and cognitive components of subjective sexual experience within close relationships. In Study 1, a sample of university and community adult participants completed self-report scales tapping attachment orientations and the multifaceted experiential aspects of subjective sexual intercourse. To control for retrospective, one-time recording bias characterizing survey studies, we used a daily experience methodology in our second study. This daily diary study can more accurately assess the processes underlying the sexuality–relationship linkage because aggregate correlations between sexual and relational quality found in past studies (as reviewed by Sprecher & Cate, 2004) might reflect unmeasured individual differences in attachment orientations. These results might also reflect the systematic effect of cognitive and motivational biases and predispositions. Study 2 provided the first empirical test of how individual differences in attachment orientations contributed to the dynamic and temporal interplay of sexuality and relationship quality, assessed in their natural context. Taken together, our research simultaneously examined between-persons and within-person processes and systematically examined the subjective construal and relational consequences of sexual experiences in everyday close relationships.
Study 1 The main goal of Study 1 was to examine the hypothesized associations between attachment orientations and the subjective experience of sexual intercourse. A large sample of adult participants completed self-report scales of attachment orientations (anxiety, avoidance) and three aspects of sexual experience (relational issues, sex-related worries, and pleasure-related feelings and cognitions). Consistent with the above theorizing, our main predictions were that (a) attachment anxiety would be associated with
stronger emphasis on desire for partner’s emotional involvement and aversive affect and cognitions during sexual intercourse, and (b) attachment avoidance would be associated with stronger emphasis on the aversive aspects of sex.
Method Participants Five hundred Israeli participants (224 women, 276 men) ranging from 17 to 48 years of age (M ⫽ 27.3, SD ⫽ 8.43) volunteered for the study without compensation. Participants were recruited from universities and community centers in the central area of Israel. All participants had had heterosexual intercourse either in a current or past relationship. Of the participants, 68.8% (n ⫽ 344) were currently involved in a romantic relationship and 31.4% were married. The length of the current relationship ranged from 1 to 96 months (M ⫽ 18.91, SD ⫽ 21.72). Education ranged from 9 to 19 years of schooling (M ⫽ 13.28, SD ⫽ 1.83).
Measures and Procedure Participants completed a randomly ordered battery of scales on an individual basis. Three scales assessed experience of heterosexual intercourse (Birnbaum & Laser-Brandt, 2002). Participants were instructed to recall a situation or a number of situations in which they had experienced sexual intercourse and to attempt to recollect, in as much detail as possible, what happened during the entire experience. Participants rated the extent to which each item matched their own experience on a 9-point scale, ranging from 1 (does not match at all) to 9 (closely matches). The first scale, the relationship-centered sexual experience scale, concerns relational components of sex, including 29 items organized around four factors: (a) feelings of being loved by the partner (11 items; e.g., “I feel I am important to my partner”), (b) focus on the partner’s state (8 items; e.g., “I’m focused on satisfying my partner”), (c) feelings of love toward partner (5 items; e.g., “I feel warmth toward my partner”), and (d) desire for partner’s involvement” (5 items; e.g., “I want to receive attention from my partner”). In the current study, Cronbach alphas indicated adequate internal consistency for the four factors (␣s ⫽ .76 –.88). Scores were computed by averaging the items of each factor. Correlations between the four factors were moderate and ranged from .24 to .41 ( ps ⬍ .001) The second scale, the worry-centered sexual experience scale, focuses on aversive components of sex and includes 26 items organized around four factors: (a) sense of estrangement and vulnerability (7 items; e.g., “I feel alienated and detached”), (b) negative feelings (8 items; e.g., “I feel self-hatred”), (c) disappointment from partner’s sexual behavior (5 items; e.g., “I feel my partner doesn’t know how to excite me”), and (d) worries and interfering thoughts” (6 items; e.g., “Bothersome thoughts disturb my concentration”). In the current study, Cronbach alphas indicated adequate internal consistency for the four factors (␣s ⫽ .74 –.84). Scores were computed by averaging the items of each factor. Correlations between the four factors were high and ranged from .55 to .68 ( ps ⬍ .001), reflecting the underlying presence of a global negative affectivity cluster. The third scale, the pleasure-centered sexual experience scale, focuses on sex-related pleasure and ecstasy and includes 24 items that are organized around three main factors: (a) pleasure-related feelings (11 items; e.g., “I feel satisfied”), (b) “letting go” state (7 items; e.g., “I’m in a state of ecstasy”), and (c) sense of strength and focus on one’s sexual needs (6 items; e.g., “I feel a sense of conquest”). In the current study, Cronbach alphas indicated adequate internal consistency for the three factors (␣s ⫽ .77–.88). Scores were computed by averaging the items of each factor. Correlations between the three factors were moderate, ranging from .15 to .41 ( ps ⬍ .001). Participants also completed Mikulincer, Florian, and Tolmacz’s (1990) 10-item adult attachment style scale. This scale includes 5 items tapping
ATTACHMENT, SEXUALITY, AND RELATIONSHIP QUALITY
933
Table 1 Predicting the Experience of Heterosexual Intercourse From Attachment Anxiety, Attachment Avoidance, and Gender Measure Relationship-centered scale Being loved Focus on partner’s state Love toward partner Desire for partner Involvement Worry-centered scale Sense of estrangement Negative feelings Disappointment Interfering thoughts Pleasure-centered scale Pleasure-related feelings Letting go Focus on one’s needs
Anxiety
Avoid
Gender
⫺.13** .11 ⫺.07 .24***
⫺.15*** ⫺.16*** ⫺.27*** ⫺.02
.30*** ⫺.29*** .18*** .13**
.26*** .32*** .30*** .26***
.32*** .24*** .24*** .28***
⫺.09 ⫺.16*** .03 ⫺.06
⫺.06 .20*** .26***
⫺.16*** .04 .21***
.10 .25*** ⫺.02
Anxiety ⫻ Avoid .01 ⫺.11 ⫺.08 ⫺.05
Anxiety ⫻ Gender
Avoid ⫻ Gender
Three-way interaction
.00 .08 .07 .08
.03 .01 .08 .09
.07 ⫺.08 .06 .02
.17*** .22*** .10 .07
.00 ⫺.08 ⫺.00 ⫺.10
⫺.07 ⫺.06 ⫺.11 .02
.02 ⫺.04 .02 .05
.04 .07 .13**
.05 ⫺.07 ⫺.08
.00 ⫺.01 ⫺.06
.07 .07 .00
Note. Entries are the standardized regression coefficients for each effect at the step in which it was entered into the regression. ** p ⬍ .01. *** p ⬍ .001.
the avoidance dimension (e.g., “I am somewhat uncomfortable being close to others”) and 5 items tapping the anxiety dimension (e.g., “I often worry that my partner doesn’t love me”). Items were constructed on the basis of Hazan and Shaver’s (1987) prototypical descriptions of attachment styles. Participants were asked to think about their close relationships and to rate the extent to which each item was self-descriptive. Ratings used 7-point scales, ranging from 1 (not at all) to 7 (very much). Previous studies have found this scale to be reliable, valid, and correlated with other adult attachment scales (e.g., Mikulincer & Florian, 2000). In our sample, Cronbach alphas were adequate for brief measures of anxiety (.73) and avoidance (.61). We computed two total scores by averaging the 5 items in each dimension. The two scores were not significantly correlated (r ⫽ .09, ns).
Results and Discussion The data were analyzed by three-step hierarchical regressions examining the unique and interactive effects of attachment anxiety, attachment avoidance, and gender. In the first step, we examined main effects for gender—a contrast code variable comparing women (⫺1) to men (1)—and the attachment scores of anxiety and avoidance (entered as standard scores). The two-way interactions were examined in the second step, and the three-way interaction was entered in the third step. Because of the large number of statistical tests conducted in this study, we set the significance level for all coefficients at ␣ ⬍ .01. Table 1 presents the standardized regression coefficients () for each effect at the step at which it was entered into the regression. With regard to the relationship-centered sexual experience scale, attachment anxiety significantly predicted feelings of being loved and desire for partner involvement: The higher the anxiety, the lower the feelings of being loved and the higher the desire for partner involvement during sexual intercourse (see Table 1). Attachment avoidance made a significant unique contribution to three of four factors, such that the higher the avoidance, the lower the feelings of being loved and love toward partner and the less the focus on partner’s state. Gender made a significant unique contribution to all four factors, such that women reported higher feelings
of being loved and love toward partner, higher desire for partner involvement, and less focus on partner’s state than men. No interaction effects were significant. With regard to the worry-centered sexual experience scale, both attachment anxiety and avoidance made significant unique contributions to all four factors: Higher anxiety and avoidance were associated with more aversive feelings and thoughts about sex. Also, women reported significantly less negative feelings about sexual intercourse than men did.1 The interaction between attachment anxiety and avoidance was significant for the sense of estrangement and negative feelings. Using Aiken and West’s (1991) procedure for examining simple slopes, we found that anxiety was significantly associated with sense of estrangement and negative feelings when attachment avoidance was high (one standard deviation above the mean; s of .43 and .54, ps ⬍ .01) but not when attachment avoidance was low (one standard deviation below the mean; s ⬍ .10). That is, the strongest aversive feelings were reported by participants scoring high on both attachment anxiety and avoidance dimensions (what Bartholomew & Horowitz, 1991, called “fearful avoidance,” p. 227). With regard to the pleasure-centered sexual experience scale, the higher the attachment anxiety, the stronger the letting go state of mind and the stronger the focus on one’s own needs during 1
Past research has shown that adolescent women have generally viewed their first sexual intercourse experience with more ambivalence, describing it as less enjoyable than did adolescent men (e.g., Guggino & Ponzetti, 1997; Sprecher, Barbee, & Schwartz, 1995). However, later sexual experience may minimize these differences (Darling, Davidson, & Passarello, 1992). In their meta-analysis of differences between men and women in sexuality, Oliver and Hyde (1993) found small to moderate differences for sexual anxiety and guilt, with women being generally more guilty and anxious, even though they did not find gender differences in sexual satisfaction. Baumeister, Catanese, and Vohs (2001) concluded that results regarding sexual enjoyment are mixed because women exhibit significantly higher within-person variance in correlates of sexual enjoyment than men.
BIRNBAUM, REIS, MIKULINCER, GILLATH, AND ORPAZ
934
sexual intercourse. Regressions also indicated that the higher the attachment avoidance, the weaker the pleasure-related feelings participants reported and the stronger the focus on their own needs (see Table 1). Gender made a significant unique contribution to letting go, with women reporting a stronger letting go state of mind during sexual intercourse than men did (see Table 1). Beyond these main effects, a significant interaction between anxiety and avoidance was found for focus on one’s own needs. Using Aiken and West’s (1991) procedure, we found that anxiety was significantly associated with focus on one’s needs when attachment avoidance was one standard deviation above the mean ( ⫽ .39, p ⬍ .01) but not when avoidance was one standard deviation below the mean ( ⫽ .13). That is, fearful avoidance was associated with the strongest focus on one’s own needs during sexual intercourse. Finally, to determine whether associations between the attachment and sexual systems are manifested only in the context of ongoing romantic relationships or can be evident even among people not currently involved in committed relationships, we conducted three-step hierarchical regressions examining unique and interactive effects of attachment anxiety, attachment avoidance, and relationship status—a contrast code variable comparing participants who were currently involved in a romantic relationship (code ⫽ 1; n ⫽ 344; 154 women and 190 men) to those who were not currently involved in such a relationship (code ⫽ –1; n ⫽ 156; 70 women and 86 men). Alpha was again set at .01. These analyses revealed that the significant main effects for attachment orientations reported in Table 1 did not notably change and were still significant after the statistical control for relationship status. In addition, beyond two significant main effects for relationship status (involved people were higher on being loved,  ⫽ .13, p ⬍ .01; and lower on negative feelings,  ⫽ ⫺.11, p ⬍ .01), the regression analyses revealed no other significant main effects for relationship status and no significant interaction between relationship status and attachment orientations. Overall, the associations between attachment orientations and sexual experience did not depend on participants’ current involvement in a romantic relationship. Overall, the findings of Study 1 were in line with predictions. Highly anxious persons reported relatively high levels of a letting go state of mind, attentional focus on their own needs, and desire for partner’s emotional involvement, but at the same time, they also reported relatively strong aversive feelings during sexual intercourse as well as doubts about being loved. Highly avoidant persons reported relatively strong aversive feelings during sexual intercourse, low levels of pleasure-related feelings, and serious doubts about loving and being loved. Furthermore, they reported relatively high levels of attentional focus on their own needs during sexual intercourse, in line with their egocentric stance toward relationship partners (e.g., Mikulincer & Shaver, 2003).
Study 2 In Study 2, we examined the hypothesis that attachment orientations would moderate the association between sexual experience and relationship quality, whereas we attempted to overcome the limitations of single-time, retrospective, cross-sectional studies (Sprecher & Cate, 2004) by relying on a daily diary methodology. Although past research has shown that sexual satisfaction is correlated with relationship’s quality and stability at the aggregate
level (i.e., between-persons; Sprecher & Cate, 2004), these correlations may be a result of either unmeasured individual differences in attachment orientations or cognitive and motivational biases. The current study combined both between-persons and withinperson levels of analysis, thereby enabling us to more accurately examine the processes underlying the sexuality–relationship linkage. Furthermore, this study addressed the limitations of crosssectional studies for considering causal hypotheses by conducting lagged day analyses that examined temporal effects and whether these temporal effects interacted with attachment orientations. To address these questions, we asked both members of heterosexual couples to report on their attachment orientations (anxiety, avoidance) and then to provide daily diary measures of relationship quality and sexual activity for a period of 42 consecutive days. In addition, each time they had sex during the 42-day study period, participants were asked to immediately report on their feelings and cognitions during that sexual intercourse. Study 2 examined whether attachment orientations moderated the contribution of sexual activity, as well as affects experienced during that activity, on a given day to next-day reports of relationship quality (after controlling for relationship quality on that day). We also explored gender differences in the effects of attachment orientations on the associations between sex and relationship quality. In particular, we tested the following hypotheses: (a) Attachment anxiety would moderate the effects of having sex on relationship behaviors and quality, such that more attachment-anxious men should show a stronger association between having sex and next-day reports of relationship quality. (b) Attachment anxiety should amplify the possible effects of positive and negative sexual experiences on daily relationship quality. That is, the next-day relational effects of sex-related feelings would be particularly strong among highly anxious women. (c) Sexual experiences and daily relationship behaviors and quality would not be correlated in highly avoidant persons.
Method Participants Fifty heterosexual cohabiting Israeli couples participated in Study 2 in exchange for 400 NIS (about $90). All participants were recruited via flyers or by word of mouth from universities, colleges, community centers, and sport clubs in the central area of Israel. Potential study participants were included in the sample if they (a) were in a steady monogamous relationship of longer than 6 months, (b) agreed to report on their daily relationship behaviors and quality each evening for a period of 42 days, (c) agreed to report on the feelings they experienced during sexual intercourse on each occurrence during the 42-day study period, and (d) were currently sexually active (defined as having had vaginal sex at least once a week in the 2 months preceding the study). Six couples were excluded from analyses because at least one partner failed to complete the diary protocol on at least 1 day. Data from noncompliant participants are routinely excluded from analysis in diary research (e.g., Gable, Reis, & Elliot, 2000; Reis & Gable, 2000). Three other couples were excluded because they reported having sex fewer than six times during the 6-week period, which left too few sexual events to analyze the sex–relationship linkage. The final sample consisted of 41 couples with no missing data across the 42-day study period and who had sex at least once a week during the study period. Although the statistical power of these analyses is low, these 41 couples did not differ significantly from the 9 dropped couples in their attachment scores or reports of relationship quality across the study period.
ATTACHMENT, SEXUALITY, AND RELATIONSHIP QUALITY Women ranged in age from 21 to 34 years (M ⫽ 25.97, SD ⫽ 3.15) and in education from 12 to 19 years of schooling (M ⫽ 14.31, SD ⫽ 2.11). Men ranged in age from 20 to 30 years (M ⫽ 26.58, SD ⫽ 2.63) and in education from 11 to 19 years of schooling (M ⫽ 13.88, SD ⫽ 1.95). Sixty-one percent of the couples were cohabiting, and 39% were married. None had children. Relationship length ranged from 6 to 138 months (M ⫽ 44.73, SD ⫽ 31.22). Number of reported sexual intercourses (vaginal, oral, or anal sex) during the 42-day study period ranged from 6 to 23 (M ⫽ 13.14, SD ⫽ 4.34). Overall, the two members of the 41 couples agreed about having had sex 520 times. Beyond these 520 episodes, there were 13 instances (2.5%) in which one partner reported having had sex and the other did not (no more than one instance in a couple). We coded these cases as days without sex and did not enter reports of sex-related feelings to the statistical analyses.
Measures and Procedure Couples who fulfilled the inclusion criteria were invited to the laboratory, were asked to fill out a background questionnaire, and were trained to complete the diary questionnaires. Participants were instructed to fill out forms independently and to refrain from discussing responses with their partner until completion of the study. They took the entire package of diaries to their home and every evening for 42 days reported on the behaviors and quality level that characterized their relationship on that day. In addition, immediately after every occasion in which they had sexual intercourse, participants completed a brief questionnaire assessing feelings experienced during that intercourse. We contacted couples by telephone every 2 days to improve compliance with the diary protocol. Participants reported full compliance with the protocol while explicitly telling us that they reported all the instances of sexual intercourse that they had had. At the end of each week, we collected completed forms from each participant. At the end of the study, participants were debriefed and thanked for their participation. They did not comment during the debriefing as to whether they had increased or decreased sexual activity as a result of participating in the study.2 Person-level measure. Attachment orientations were assessed with the 36-item Experience in Close Relationships Scale (Brennan, Clark, & Shaver, 1998), tapping variations in attachment anxiety and attachment avoidance (18 items per dimension). Participants rated the extent to which each item described their feelings in close relationships on a 7-point scale ranging from 1 (strongly disagree) to 7 (strongly agree). In the current sample, Cronbach’s alphas were high for both anxiety (.90 for women, .86 for men) and avoidance items (.91 for women, .85 for men). On this basis, two global attachment scores were computed for each participant by averaging the relevant items. Pearson correlations between anxiety and avoidance scores were not significant among women, r(39) ⫽ .22, nor among men, r(39) ⫽ .16. Correlations between couple members in attachment anxiety and avoidance approached statistical significance, r(39) ⫽ .28 for anxiety, r(39) ⫽ .30 for avoidance, ps ⬍ .10, indicating moderate correspondence in partner’s attachment orientations. Attachment avoidance was significantly higher among men (M ⫽ 3.25, SD ⫽ 0.85) than among women (M ⫽ 2.72, SD ⫽ 1.05), t(40) ⫽ 3.01, p ⬍ .01. No significant difference between partners was found in anxiety. Daily relationship measures. The diary questionnaire dealing with relational behaviors and quality included two parts. In the first part, participants rated the quality of the relationship with their partner on that day. Ratings were made on a 9-point scale, ranging from 1 ( poor) to 9 (excellent; M ⫽ 7.62, SD ⫽ 1.87 for women; M ⫽ 7.71, SD ⫽ 1.94 for men). The sample means of within-person variance for daily reports of relationship quality were 1.73 for women and 1.47 for men. The difference between variances was not significant. In the second part, participants reported whether or not they had enacted each of 19 specific behaviors toward their partner on a given day, and with
935
a parallel item, whether their partner had enacted each of the same 19 behaviors toward them. The list included 10 relationship-enhancing behaviors (e.g., “I told my partner I loved him/her—My partner told me he/she loved me”; “I was concerned about some problem, and my partner provided me support and reassurance—My partner was concerned about some problem and I provided support and reassurance to him/her”) and 9 relationship-damaging behaviors (e.g., “I was inattentive and unresponsive to my partner—My partner was inattentive and unresponsive to me”; “My partner criticized me—I criticized my partner”). These behaviors were sampled from previous diary studies examining couple interactions (Gable, Reis, & Downey, 2003; Tidwell, Reis, & Shaver, 1996). Participants indicated a behavior’s occurrence by checking a box next to the item. We computed two scores for each participant on each day. First, we counted the number of participants’ own and perceived partners’ relationship-enhancing behaviors (scores ranged from 0 to 16; M ⫽ 8.50, SD ⫽ 3.15 for women; M ⫽ 8.17, SD ⫽ 3.08 for men). The sample means of within-person variance for daily reports of relationship-enhancing behaviors were 9.01 for women and 9.72 for men. Second, we counted the number of participants’ own and perceived partners’ relationshipdamaging behaviors (scores ranged from 0 to 8; M ⫽ 2.03, SD ⫽ 1.21 for women; M ⫽ 2.04, SD ⫽ 0.96 for men). The sample means of withinperson variance for daily reports of relationship-damaging behaviors were 4.38 for women and 4.52 for men. For each type of behavior (enhancing, damaging), we decided to collapse a participant’s reports of what he or she did in the relationship and what he or she reported having received from his or her partner into a single score because these two scores were highly correlated, r(39) ⫽ .78 for relationship-enhancing behaviors, r(39) ⫽ .83 for relationship-damaging behaviors.3 In both sexes, daily reports of relationship quality showed significant positive associations with daily reports of relationship-enhancing behaviors (rs ⫽ .53 and .54, ps ⬍ .01) and significant inverse associations with daily reports of relationship-damaging behaviors (rs ⫽ ⫺.48 and ⫺.38, ps ⬍ .01). The two types of behaviors showed small but significant correlation (rs ⫽ ⫺.24 and ⫺.14, ps ⬍ .01). These associations strengthened our confidence in the construct validity of the computed scores and supported our decision to treat them separately in the main analyses. The dyadic correlations between men’s relational scores and women’s relational scores across the study period were strong (rs ⫽ .56 and .65, ps ⬍ .01). No significant difference between partners was found for the three relational variables. Sex diary measures. The sex diary questionnaire included 12 items tapping sex-related feelings and cognitions during the reported intercourse. The items included 6 positive sexual feelings and cognitions (e.g., “During the sexual intercourse, I felt passionately attracted to my partner”; “During
2 Notably, this diary format has been criticized because participants’ compliance rates cannot be confirmed, particularly with regard to the timing of diary reports (e.g., Stone, Shiffman, Schwartz, Broderick, & Hufford, 2002). Nevertheless, three studies conducted recently by Green, Rafaeli, Bolger, Shrout, and Reis (in press) have shown that both paper and electronic methods of collection yielded data that were comparable in compliance rates, psychometric properties, and pattern of results. Their findings suggested that compliance is more dependent on participant motivation than on methods of collection. Our participants’ motivation was fostered, and related compliance was facilitated through promoting their rapport with the researchers, generating a sense of personal involvement in the research, as well as constant deadline and partners’ reminders. Thus, although we did not have data verifying compliance, our research was likely to produce valid data that were less likely to be affected by the choice of data collection mode (Green et al., in press). 3 All of the correlations reported for daily measures were computed within person or within a dyad (in case of dyadic correlations) and then, by means of Fisher’s r to z transformation, were averaged across the sample.
936
BIRNBAUM, REIS, MIKULINCER, GILLATH, AND ORPAZ
the sexual intercourse, I reached a satisfying orgasm”) and 6 negative sexual feelings and cognitions (e.g., “During or after the sexual intercourse, I felt some frustration and disappointment”; “During sexual intercourse, I felt bored and apathetic”). Two total scores were computed for each participant on each day he or she reported having sex. First, we computed the total number of positive sexual feelings and cognitions reported during that intercourse (scores ranged from 0 to 6; M ⫽ 3.28, SD ⫽ 0.87 for women; M ⫽ 3.41, SD ⫽ 1.05 for men). The sample means of within-person variance for daily reports of positive sexual feelings behaviors were 2.14 for women and 2.02 for men. Second, we computed the total number of reported negative sexual feelings and cognitions (scores ranged from 0 to 4; M ⫽ 0.54, SD ⫽ 0.61 for women; M ⫽ 0.29, SD ⫽ 0.34 for men). The sample means of withinperson variance for daily reports of negative sexual feelings were 1.04 for women and 0.87 for men. In both sexes, the association between these two scores was small but significant (rs ⫽ ⫺.27 and ⫺.28, ps ⬍ .01). Again, these associations supported our decision to treat them separately in the main analyses. The dyadic correlations between men’s and women’s sexrelated feelings were significant but moderate (rs ⫽ .30 and .32, ps ⬍ .01). Although partners did not differ significantly in the overall amount of sex-related positive feelings, women tended to report significantly more negative feelings than men did, t(522) ⫽ ⫺5.09, p ⬍ .01.
the effects of having sex on relational behaviors and quality for men and women, but the estimates are determined simultaneously and take into account nonindependence of the couple members. In addition, this analysis simultaneously estimates (a) the unique contribution of each participant’s (man or woman’s) attachment anxiety and avoidance scores to the association between having sex and relational behaviors and quality and (b) the unique contribution of his or her partner’s attachment anxiety and avoidance scores to such an association.4 This analysis required two new dummy variables, one representing the male partner (where 1 ⫽ male and 0 ⫽ female) and the other representing the female partner (where 1 ⫽ female and 0 ⫽ male). To facilitate interpretation, we centered each partner’s attachment anxiety and avoidance and variables at the lower, or day, level around the sample grand mean. The equations for the Level 1 model were as follows: Y ⫽ B1*共male partner兲 ⫹ B2*共having sex*male partner兲 ⫹ B3*共having sex*female partner兲 ⫹ R. The Level 2 model was the following: B1 ⫽ G10 ⫹ U1
Results and Discussion The Contribution of Sexual Intercourse to Relational Behaviors and Quality In this section, we examine (a) whether relational behaviors (enhancing, damaging) and relationship quality on a given day were affected by having had sexual intercourse on the previous day and (b) whether this association was moderated by participants’ and partners’ attachment orientations (anxiety, avoidance). These questions were analyzed as follows. First, to examine how sexual intercourse on a given day relates to changes in relational behaviors and quality from that day to the next day, we controlled for prior-day reports of relational behaviors and quality (i.e., prior day’s quality and relational behaviors were entered as an additional predictor to control for its contribution to next-day changes in relational behaviors and quality). Therefore, we regressed for each participant his or her daily reports of each relational variable on his or her previous-day reports of the same variable, which effectively makes the unexplained residual variance the dependent variable for the main statistical analyses. Second, we coded a having sex variable by assigning ⫺1 to days in which at least one partner reported having no intercourse and 1 to days in which the two partners reported having sex. Third, we conducted hierarchical linear modeling (HLM6; Bryk & Raudenbush, 2002) and examined the main and interactive effects of having sex on a given day and dispositional attachment orientations (anxiety and avoidance) on the next day’s relational behaviors and quality (unexplained by the prior day’s relational behaviors or quality). These analyses included both between-participants (attachment anxiety and avoidance) and within-participant (daily reports of sexual intercourse and relational variables) variables. Because male and female partners’ behaviors were necessarily dependent on each other, we used a multilevel analysis (see Bolger, Davis, & Rafaeli, 2003, for a fuller description of the application of two-level nested models for couples’ research). Thus, we included both the male partner and female partner in the same HLM analysis. This procedure provides separate estimates of
B2 ⫽ G20 ⫹ G21*共male’s anxiety兲 ⫹ G22*共male’s avoidance兲 ⫹ G23*共 female’s anxiety兲 ⫹ G24*共 female’s avoidance兲 ⫹ U2 B3 ⫽ G30 ⫹ G31*共male’s anxiety兲 ⫹ G32*共male’s avoidance兲 ⫹ G33*共 female’s anxiety兲 ⫹ G34*共 female’s avoidance兲 ⫹ U2. The coefficient G10 represents gender differences in relational behaviors and quality across the study’s period. The coefficient G20 represents the main effect of having sex on men’s next-day reports of relational behaviors and quality and the coefficient G30 represents the main effect of having sex on women’s next-day reports of relational behaviors and quality. The remaining coefficients represent interactive effects between having sex and participants’ dispositional attachment orientations on their own and partners’ next-day reports of relational behaviors and quality. Within-participant coefficients are represented by G21 and G22 (the contributions of a male participant’s attachment anxiety and avoidance to the effects of sexual intercourse on his own next-day reports of relational behaviors and quality) and G33 and G34 (the contributions of a female participant’s attachment anxiety or avoidance to the effects of sexual intercourse on her own next-day reports of relational behaviors and quality). Dyadic coefficients are represented by G23 and G24 (the contributions of a female participant’s attachment anxiety and avoidance to the effects of sexual intercourse on her male partner’s next-day reports of relational 4
We also conducted three-level HLM analysis including within-couple effects for gender and interactions between gender and the other study variables. However, because of the small sample, we had relatively low levels of power to adequately examine high-level interactions among gender, attachment orientations, and day-level variables. On this basis, we decided to focus exclusively on the two-level effects reported in the text.
ATTACHMENT, SEXUALITY, AND RELATIONSHIP QUALITY
behaviors and quality) and G31 and G32 (the contributions of a male participant’s attachment anxiety or avoidance to the effects of sexual intercourse on his female partner’s next-day reports of relational behaviors and quality). Table 2 presents relevant HLM coefficients. Because the coefficient G10 (gender differences in relational behaviors and quality across the study period) was not significant in any of the HLM analyses, we present in Table 2 the main effects for having sex and its interactive effects with each partner’s attachment orientations. As Table 2 shows, sexual intercourse had significant main effects on men’s next-day reports of relationship-enhancing behaviors (unexplained by previous-day reports), and its effects on men’s next-day reports of relationship-damaging behaviors and relationship quality approximated statistical significance ( p ⬍ .10): Men reported heightened relationship-enhancing behaviors and relationship quality following days in which they had sex with their partner. Men also tended to report a decrease in relationshipdamaging behaviors following days in which they had sex with their partner. The main effect of having sex also approximated statistical significance for women’s next-day reports of relationship-damaging behaviors ( p ⬍ .10): Women tended to report a decrease in relationship-damaging behaviors following days in which they had sex with their partner. Having sex had no significant main effect on women’s next-day reports of relationship-enhancing behaviors and relationship quality. Examination of within-participant interactive effects revealed that men’s attachment anxiety significantly moderated the effects of having sex on men’s next-day reports of relationship-enhancing behaviors, relationship-damaging behaviors, and relationship quality (see Table 2). The simple slope relating sex to men’s next-day reports of relationship behaviors and quality was higher for men one standard deviation above the attachment anxiety mean (Bs ⫽ 0.80, ⫺0.64, and 0.27 for relationship-enhancing behaviors, relationship-damaging behaviors, and relationship quality, respectively) than for men one standard deviation below the attachment anxiety mean (Bs ⫽ ⫺0.12, 0.06, and ⫺0.05, respectively). In other words, more attachment-anxious men showed greater gains on their own daily reports of relationship-enhancing behaviors and relationship quality and a greater reduction in their own daily reports of relationship-damaging behaviors following days in which they had sex. In other words, attachment anxiety intensified the positive effects of sex on men’s daily reports of relationship behaviors and quality.
937
Table 2 also shows that men’s attachment avoidance did not significantly moderate the effects of having sex on their own next-day reports of relationship behaviors and quality. In addition, attachment anxiety or avoidance did not significantly moderate the effects of sexual intercourse on women’s next-day reports of relationship behaviors and quality. Examination of dyadic interactive effects revealed the following significant findings. First, women’s attachment anxiety significantly moderated the effects of having sex on men’s next-day reports of relationship-damaging behaviors and relationship quality (see Table 2). The simple slope relating sex to men’s next-day reports of relationship behaviors and quality was higher for women one standard deviation above the attachment anxiety mean (Bs ⫽ ⫺0.67 and 0.31 for relationship-damaging behaviors and relationship quality, respectively) than for women one standard deviation below the attachment anxiety mean (Bs ⫽ 0.09 and ⫺0.09). In other words, men with more anxiously attached female partners showed greater positive gains on their own daily reports of relationship quality and a greater reduction in their own daily reports of relationship-damaging behaviors following days in which they had sex. Women’s attachment anxiety intensified the positive effects sex had on men’s relational behaviors and appraisals. Second, men’s attachment avoidance significantly moderated the effects of having sex on women’s next-day reports of relationship-damaging behaviors (see Table 2). The simple slope relating sex to women’s next-day reports of relationship-damaging behaviors was higher for men one standard deviation below the attachment avoidance mean (B ⫽ ⫺0.65) than for men one standard deviation above the attachment avoidance mean (B ⫽ 0.01). In other words, women whose male partner was less avoidant showed a greater reduction in their daily reports of relationshipdamaging behaviors following days in which they had sex. Men’s attachment avoidance inhibited the positive effects sex had on women’s relational behaviors. It is important that men’s attachment anxiety and women’s attachment avoidance did not significantly moderate the effects of having sex on their partners’ next-day reports of relationship behaviors and quality. Overall, HLM analyses revealed that having sex had significant positive effects on daily reports of relationship behaviors and quality among highly anxious men or among men whose partner was highly anxious. In addition, having sex led to a decrease in the daily report of relationship-damaging behavior among women
Table 2 Hierarchical Linear Modeling Coefficients Predicting Fluctuations in Daily Relational Behaviors and Quality From Previous-Day Sexual Intercourse and Dispositional Attachment Orientations Effects and interactions of previous-day sexual intercourse Main effects Interactions with Women’s anxiety Women’s avoidance Men’s anxiety Men’s avoidance † p ⬍ .10.
* p ⬍ .05.
Relationship-enhancing behaviors Women
Relationshipdamaging behaviors
Relationship quality
Men
Women
Men
Women
Men
0.14
0.34**
⫺0.32†
⫺0.29†
⫺0.01
0.11†
0.15 ⫺0.14 ⫺0.14 0.11
0.02 ⫺0.07 0.46** 0.03
0.07 ⫺0.03 0.19 0.33*
⫺0.38* 0.09 ⫺0.35* 0.19
0.07 0.01 0.01 0.02
** p ⬍ .01.
0.20** 0.05 0.16* ⫺0.01
BIRNBAUM, REIS, MIKULINCER, GILLATH, AND ORPAZ
938
and avoidance. In another series of HLM analyses, we examined the main effects of negative sex-related feelings (main effects of men’s reports of negative sex-related feelings on their own nextday reports of relationship behaviors and quality; main effects of women’s reports of negative sex-related feelings on their own next-day reports of relationship behaviors and quality) and the extent to which these effects were moderated by men’s and women’s attachment anxiety and avoidance. Table 3 presents the relevant HLM coefficients. With regard to positive sex-related feelings, HLMs revealed significant associations between women’s feelings during sex and next-day relationship behaviors and appraisals (see Table 3): The higher the positive feelings reported during sex, the greater the gains on their own next-day reports of relationship-enhancing behaviors and relationship quality and the greater the reduction in their own next-day report of relationship-damaging behavior. Men’s positive sex-related feelings had no significant main effect on their own next-day reports of relationship behaviors and quality (see Table 3). Women’s attachment anxiety significantly moderated the effects of their positive sex-related feelings on next-day reports of relationship quality (see Table 3). The simple slope was greater for women one standard deviation above the attachment anxiety mean (B ⫽ 0.23) than for women one standard deviation below the attachment anxiety mean (B ⫽ 0.05). In other words, more attachment-anxious women showed a stronger association between positive feelings during sexual intercourse and next-day reports of relationship quality. Women’s attachment anxiety seemed to intensify the relational benefits of experiencing positive feelings during sex. Other within-participant interactive effects were not significant (see Table 3).
whose partner scored relatively low in attachment avoidance. It bears noting that because all four attachment scores were included in the same analyses, we can be sure that both partner’s scores had independent effects; that is, there is no confound from the correlation between partners’ attachment scores.
The Contribution of Sex-Related Feelings to Daily Relational Behaviors and Quality In these analyses, we focused on days in which couples had sex and examined (a) whether relational behaviors (enhancing, damaging) and relationship quality on the next day were affected by the feelings and cognitions (positive, negative) participants reported having experienced during intercourse and (b) whether this association was moderated by participants’ and partners’ attachment orientations. For these purposes, we conducted a series of HLM analyses similar to those described in the previous section: The dependent variables were residuals of next-day reports of relationship behaviors and quality unexplained by previous-day reports of these variables, and the between-persons level variables were male partner’s and female partner’s attachment anxiety and avoidance scores. Instead of a dichotomous code for whether or not the participant reported having sex, the within-person level included the extent to which participants reported experiencing either positive or negative feelings during sex. In one series of HLM analyses, we examined the main effects of positive sex-related feelings (main effects of men’s reports of positive sex-related feelings on their own next-day reports of relationship behaviors and quality; main effects of women’s reports of positive sex-related feelings on their own next-day reports of relationship behaviors and quality) and the extent to which these effects were moderated by men’s and women’s attachment anxiety
Table 3 Hierarchical Linear Modeling Coefficients Predicting Fluctuations in Relational Behaviors and Quality Following Days of Sexual Intercourse From Feelings During the Intercourse and Dispositional Attachment Orientations Relationshipenhancing behaviors Effects and interactions Positive feelings during previous-day intercourse Main effects Interactions with Women’s anxiety Women’s avoidance Men’s anxiety Men’s avoidance Negative feelings during previous-day intercourse Main effects Interactions with Women’s anxiety Women’s avoidance Men’s anxiety Men’s avoidance * p ⬍ .05.
** p ⬍ .01.
Women
Relationship quality
Women
Men
Women
Men
0.20
⫺0.23**
⫺0.01
0.14**
0.08
0.01 ⫺0.02 ⫺0.08 0.08
0.07 0.01 0.01 0.05
0.04 0.01 0.02 0.05
⫺0.05 0.02 0.07 0.02
0.09* ⫺0.06 0.03 ⫺0.02
0.12** ⫺0.04 0.06 0.03
⫺0.15
0.19
0.13
0.18
⫺0.11*
⫺0.09
0.43** ⫺0.23** ⫺0.07 0.15
0.06 0.13 0.17 0.02
⫺0.10* 0.10* ⫺0.03 0.03
⫺0.26** 0.06 0.04 0.02
0.36**
⫺0.49** 0.13 0.19 ⫺0.05
Men
Relationshipdamaging behaviors
⫺0.16 ⫺0.21 0.15 0.01
ATTACHMENT, SEXUALITY, AND RELATIONSHIP QUALITY
Examination of dyadic interactive effects revealed that women’s attachment anxiety significantly moderated the effects of men’s positive sex-related feelings on their own next-day reports of relationship quality (see Table 3). The simple slope was greater for men whose female partner was one standard deviation above the attachment anxiety mean (B ⫽ 0.20) than for men whose female partner was one standard deviation below the attachment anxiety mean (B ⫽ ⫺0.04). In other words, male partners of more anxiously attached women showed a stronger association between positive feelings during sexual intercourse and next-day reports of relationship quality. Women’s attachment anxiety intensified their male partner’s relational gains from experiencing positive feelings during sex. Other dyadic interactive effects were not significant (see Table 3). With regard to negative sex-related feelings, HLMs revealed only one significant main effect, that between women’s feelings during sex and their next-day report of relationship quality (see Table 3): The more negative feelings reported during sex, the lower their next-day reports of relationship quality. The examination of within-participant interactive effects revealed that women’s attachment anxiety significantly moderated the effects of negative sex-related feelings on their own next-day reports of relationship behaviors and quality (see Table 3). The simple slope was greater for women one standard deviation above the attachment anxiety mean (Bs ⫽ ⫺0.64, 0.56, and ⫺0.21 for relationship-enhancing behaviors, relationship-damaging behaviors, and relationship quality, respectively) than for women one standard deviation below the attachment anxiety mean (Bs ⫽ 0.34, ⫺0.30, and ⫺0.01, respectively). In other words, women scoring high in attachment anxiety (as compared with less anxious women) showed lower levels of next-day relationship-enhancing behaviors and relationship quality and higher levels of next-day relationship-damaging behaviors after experiencing more negative feelings during sex. Among women, attachment anxiety seemed to intensify the detrimental relational effects of negative feelings experienced during sex. The HLMs revealed one additional within-participant interactive effect: Women’s attachment avoidance significantly moderated the effects of women’s negative sex-related feelings on their own next-day reports of relationship behaviors and quality (see Table 3). The simple slope was larger for women one standard deviation below the attachment avoidance mean (Bs ⫽ 0.36 and ⫺0.21 for relationship-damaging behaviors and relationship quality, respectively) than for women one standard deviation above the attachment avoidance mean (Bs ⫽ ⫺0.10 and ⫺0.01, respectively). In other words, women scoring low in attachment avoidance (as compared with more avoidant women) showed lower levels of next-day relationship quality and higher levels of nextday relationship-damaging behaviors when they experienced more negative feelings during sex. Women’s attachment avoidance seemed to inhibit the detrimental relational effects of experiencing negative feelings during sex. Our analyses of dyadic interactive effects revealed that women’s attachment anxiety significantly moderated the effects of men’s negative sex-related feelings on their own next-day reports of relationship quality (see Table 3). The simple slope was greater for men whose female partner was one standard deviation above the attachment anxiety mean (B ⫽ ⫺0.35) than for men whose female partner was one standard deviation below the attachment anxiety mean (B ⫽ 0.17). In other words, men whose female partner was
939
more anxiously attached showed a stronger association between negative feelings during sexual intercourse and next-day reports of relationship quality. Women’s attachment anxiety seemed to intensify their male partner’s relational distress after having experienced negative feelings during sex. Other dyadic interactive effects were not significant (see Table 3). In sum, our multilevel analyses revealed that sex-related feelings had significant relational effects among highly anxious women and among women who score relatively low in attachment avoidance. In these cases, women’s experience of positive feelings during sexual intercourse or the lack of sex-related negative feelings had strong positive effects on their next day relational behaviors and appraisals. The next-day relational effects of sex-related feelings were also particularly strong among men whose female partner was highly attachment anxious. In couples with highly anxious women, men’s experience of positive feelings during sexual intercourse or the lack of sex-related negative feelings had strong positive effects on their next-day relational behaviors and appraisals.5
General Discussion The current research adds to our understanding of romantic relationships by examining both retrospectively and on a daily basis the associations between attachment orientations, sexual experiences, and relationship quality. Study 1 showed that, regardless of gender, attachment anxiety was associated with a relatively ambivalent construal of sexual experience, whereas attachment avoidance was associated with more aversive sexual feelings and cognitions. Study 2 indicated that attachment anxiety amplified the effects of positive and negative sexual experiences on relationship interactions. Specifically, the relationship quality of couples with a more anxiously attached partner was more affected by daily fluctuations in sexual experiences, with some findings differing for men and women. Attachment avoidance, on the other hand, inhibited the effects of sexual experiences on daily relationship interactions, such that daily relationship interactions of couples with a highly avoidant partner were less affected by sexual experiences. As expected, attachment avoidance was associated with aversive sexual experiences. Our findings expand the existing picture of avoidant people’s sexuality by portraying its aversive experiential nature and related foci of distress. Past research has indicated that highly avoidant people exhibit a more erotophobic disposition and are less likely to enjoy sex, in general (e.g., Tracy et al., 2003), and affectionate sexual activities, in particular (Brennan, Wu, & Loev, 1998; Hazan et al., 1994). This pattern may reflect discomfort with the intimacy imposed inherently by the sexual interaction, which of course includes relational aspects beyond the sexual activity itself. The current findings imply that this discomfort may also be manifested in aversive feelings (e.g., sense of estrangement and disappointment), intrusive thoughts, doubts about love and being 5 Relationship length was not significantly associated with either men’s or women’s attachment scores (rs ⬍ .07). Moreover, the introduction of relationship length as an additional Level 2 variable in HLM analyses revealed that this variable did not significantly moderate the effects of sex-related variables on relationship quality and behaviors and did not modify the moderating effects of attachment orientations reported in Tables 2 and 3.
940
BIRNBAUM, REIS, MIKULINCER, GILLATH, AND ORPAZ
loved, lack of pleasurable feelings, and difficulty focusing on a partner’s needs. Whereas highly avoidant people’s aversive sexual experience seems to reflect tension between the demand for closeness implied by sexual interactions and fears about intimacy and closeness, the ambivalent nature of highly anxious people’s sexuality seems to be more of a manifestation of their difficulty meeting attachment needs. Study 1 showed that in describing their experiences of sexual intercourse, highly anxious people reported strong aversive feelings and doubts about being loved. At the same time, they also reported relatively high levels of a letting go state of mind; attending to their own needs; and desires for emotional involvement, warmth, and attention from their partner. Highly anxious individuals’ erotophilic tendencies (Bogaert & Sadava, 2002) may reflect their use of sex to satisfy unmet attachment-related needs (D. Davis et al., 2004; Schachner & Shaver, 2004). However, sex, a prominent route for seeking proximity and attaining emotional and physical closeness, may fail to meet their unrealistic relational expectations and seemingly endless demands for greater closeness and result in frustrated attachment needs and sexual disappointment along with feelings of loneliness and alienation. It is interesting that although both attachment avoidance and attachment anxiety were positively associated with aversive sexual experiences, attachment anxiety and avoidance differed in the way they moderated the association between daily sexual experiences and relationship interactions. Having sex or the feelings experienced during sexual intercourse had stronger relational effects among highly anxious persons, which is consistent with other available evidence. Highly anxious persons rely heavily on the sexual route to fulfill attachment needs (D. Davis et al., 2004; Schachner & Shaver, 2004), and these powerful, sometimes obsessional motives may lead them to experience rejection and abuse (e.g., K. E. Davis, Ace, & Andra, 2002; Feeney et al., 2000). However, our findings also imply that sex may be beneficial for anxiously attached persons in established relationships. Having sex or experiencing positive feelings during sexual intercourse was shown to satisfy highly anxious persons’ needs for intimacy, closeness, reassurance, and caregiving, bringing at least next-day relief from relationship worries and related destructive behaviors. Whether this relief is temporary or more lasting is a question for future research. Although both highly anxious men and women showed the expected association between daily sexual experiences and relationship quality, the nature of this linkage differed somewhat between the two sexes. Whereas the quality of highly anxious women’s relationship interactions was mainly affected by feelings and cognitions experienced during sexual intercourse, highly anxious men’s relationship interactions were more strongly affected by the mere engagement in sexual intercourse than by the feeling and cognitions experienced therein. These findings support evolutionary models (e.g., Buss, 1998; Buss & Schmitt, 1993; Trivers, 1972), as well as more social approaches (Gagnon & Simon, 1973; Reiss, 1981), to gender differences in human sexuality, which posit that women are more likely than men to rely on sexual experiences as a means for evaluating (or reevaluating, in the case of ongoing relationships) the suitability of relationship partners and to expect partners to be responsive to their emotional needs during sexual intercourse. Women may thus react to positive and negative feelings during sexual activity with a congruent increase or decrease in
perceived relationship quality (Birnbaum & Reis, in press). Experiencing a frustrating and dissatisfactory sexual intercourse may raise doubts about the suitability of the sexual partner for longterm relationships, which in turn may impair relationship quality. In contrast, positive feelings during sexual intercourse, such as perceiving a sexual partner as caring and responsive, may signal an advantageous mating choice (Birnbaum & Reis, in press) and thereby contribute to relationship-enhancing interactions. Our findings show that these heightened relational reactions to sexual experiences are more pronounced among anxiously attached women who tend to view sexual interactions as a barometer of relationship quality (Birnbaum, in press; D. Davis et al., 2004). This pattern also accords with a recent retrospective study that has found that sexual satisfaction mediated the association between women’s attachment anxiety and relationship satisfaction (Birnbaum, in press). Study 2’s findings reveal that attachment anxiety may intensify the association between sexual and relational interactions. This pattern of results is consistent with Campbell et al.’s (2005) findings that more attachment-anxious people weigh daily relational events more heavily when judging the quality of their relationships. This tendency may reflect highly anxious individuals’ sensitivity to cues that imply changes in perceived rejection and support from attachment figures (Fraley & Shaver, 2000). Highly anxious persons may therefore use sex as one particularly potent barometer of a partner’s feeling toward them as well as a means for earning security. Alternatively, highly anxious people, particularly women, may have difficulty differentiating between sexual desire and other relational components, such as affection, intimacy, caregiving, and commitment. These associations between sexual experiences and one’s own relationship outcomes were paralleled by partner effects that revealed that the partners of highly anxious women, as compared with partners of less anxious women, also displayed stronger associations between sexual experiences and daily relationship interactions. Women’s attachment anxiety amplified their male partners’ relational distress after having experienced negative feelings during sex. As mentioned above, we found that highly anxious women reacted to negative sexual experiences with increases in daily relationship-damaging behaviors. Their partners might react negatively to this pattern of destructive behavior, thereby heightening relationship conflicts. In this way, a self-exacerbating dyadic cycle of sexual and relational dissatisfaction could be created, similar to the pattern shown more generally for rejection sensitivity (Downey et al., 1998). Women’s attachment anxiety also intensified their male partner’s relational gains from positive sexual interactions and merely having sex. These findings suggest that just as there may be a negative cycle, so there may also be a favorable self-amplifying dyadic cycle of positive sexual experiences and relationship-enhancing behaviors. Although these processes warrant deeper scrutiny, the current findings indicate that attachment anxiety makes sex more influential to relational wellbeing. Attachment avoidance, on the other hand, inhibited both the detrimental relational effects of negative sexual interactions and the positive relational effect of having sex. Highly avoidant women’s relationships were less adversely affected by negative sexual experiences as compared with those of less avoidant women. Furthermore, women whose male partner was more avoidant, as
ATTACHMENT, SEXUALITY, AND RELATIONSHIP QUALITY
compared with women whose partner was less avoidant, showed lesser reductions in daily reports of relationship-damaging behaviors following days in which they had had sex. This pattern of relative disconnection between sexual and relationship interactions may indicate that avoidant persons engage in sexual intercourse for relationship-irrelevant, extraneous reasons (i.e., reasons other than intimacy and attachment; Cooper et al., 2006; D. Davis et al., 2004; Schachner & Shaver, 2004). Having sex with a highly avoidant male partner may therefore contribute minimally to women’s intimacy goals in close relationships (Cooper et al., 2006). Highly avoidant people’s detached stance may have the somewhat ironic benefit of inhibiting the translation of negative sex-related experiences into relationship distress. However, this apparently defensive strategy impedes the experience of genuine intimate interactions. On the whole, our findings suggest that attachment avoidance and attachment anxiety may represent two extremes relevant to the linkage between sex and relationships. Highly anxious individuals, particularly women, tend to conflate sex and other relationship qualities, such that sex-related feelings and cognitions are more likely to be transferred onto the broader functioning of romantic relationships. In contrast, highly avoidant individuals tend to detach sex from other relationship qualities, even within the context of ongoing romantic relationships. More generally, the fact that the associations between sexual experiences and relationship interactions varied as a function of attachment dimensions also implies that although there may be reciprocal relationships between sex and other components of romantic love (e.g., intimacy, caregiving, lust, etc.), the nature and strength of these interconnections among components may themselves vary in ways that are theoretically and pragmatically important. This conclusion is consistent with Fisher’s (Fisher, 1998; Fisher, Aron, Mashek, Li, & Brown, 2002) contention that lust and attachment are separate emotion– motivation systems that became independent during human evolution. It is also consistent with Diamond’s (2003) theory that the processes underlying sexual desire and affectional bonding are functionally distinct. However, it suggests that rather than focusing attention on questions of separability, researchers might constructively consider how these (and other) behavioral systems coordinate and mutually influence each other in the context of ongoing close relationships. After all, in adulthood romantic partners typically function simultaneously as sexual partners and attachment figures (Hazan & Zeifman, 1994). Our findings suggest that optimal functioning of the attachment system involves neither high nor low levels of dependence between the various relationshiprelevant behavioral systems. Instead, among established couples, intermediate levels of interdependence between sexual and relationships interactions provided more adaptive relationship maintenance mechanisms. Of course, there may well be differences for newly established or later life couples (Kotler, 1985; Reedy, Birren, & Schaie, 1981; Sternberg, 1986).
Limitations Our results should be interpreted in the context of several limitations. The sample of Study 2 was relatively small. Although participants were instructed to complete a diary each day, we collected the completed forms only weekly and therefore were unable to verify levels of timeliness (see Green et al., in press, for
941
discussion of these issues.) Additionally, although participants were instructed to fill out their forms independently, their responses may still reflect worry about their partner’s reaction, should the records become known, as well as other types of reactivity (Catania, Gibson, Chitwood, & Coates, 1990). Future studies should use methods of administration that minimize these potential sources of error (see Reis & Gable, 2000, for alternatives). Finally, although our research used both university and community samples, it is unclear how well our findings would generalize to a broader range of couples. Further research with more diverse and representative samples is needed to establish and extend the robustness of these findings across different groups (e.g., long-term married couples, distressed couples, couples with sexual dysfunction).
Conclusions This research raises important questions about the interplay between attachment and sexual behavioral systems within romantic relationships. Although pair bonding and sexual mating systems represent distinctive behavioral systems with different primary functions (Bowlby, 1969/1982; Diamond, 2003; Fisher, 1998), their impact on relationship well-being may reflect interdependence more than independence. We found that the functional significance of the sexual system for relationship maintenance and deterioration was influenced by attachment-related concerns about acceptance and closeness. This association suggests other questions that might be profitable for future research. For example, do chronic anxieties and worries about one’s sexual attractiveness, the extent to which one is able to gratify one’s partner, and the partner’s responses to one’s sexual appeals contribute to relationship quality? Can attachment-related behaviors compensate for sexual difficulties and temper sexual anxieties and worries? Does a long-term pattern of sexual satisfaction contribute to earned security among initially anxious persons? Do aversive sexual experiences trigger highly avoidant persons’ mate-poaching tendencies and indirectly contribute to their relatively low relationship quality or are these merely unrelated manifestations of deactivating strategies? Alternatively, does poor relationship quality lead avoidant persons to become more interested in extramarital involvement? Does anxious persons’ dissatisfaction lead them to pursue more promising alternative partners? Although a growing body of research links attachment orientations with sexual motivation, emotions, cognitions, and behaviors (Shaver & Mikulincer, in press), much more research is needed to explore the intricate interplay between these two behavioral systems.
References Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting interactions. Newbury Park, CA: Sage. Ainsworth, M. D. S., Blehar, M. C., Waters, E., & Wall, S. (1978). Patterns of attachment: Assessed in the strange situation and at home. Hillsdale, NJ: Erlbaum. Bartholomew, K., & Horowitz, L. M. (1991). Attachment styles among young adults: A test of a four-category model. Journal of Personality and Social Psychology, 61, 226 –244. Baumeister, R. F., Catanese, K. R., & Vohs, K. D. (2001). Is there a gender difference in strength of sex drive? Theoretical views, conceptual dis-
942
BIRNBAUM, REIS, MIKULINCER, GILLATH, AND ORPAZ
tinctions, and a review of relevant evidence. Personality and Social Psychology Review, 5, 242–273. Birnbaum, G. E. (in press). Attachment orientations, sexual functioning, and relationship satisfaction in a community sample of women. Journal of Social and Personal Relationships. Birnbaum, G. E., & Gillath, O. (in press). Measuring subgoals of the sexual behavioral system: What is sex good for? Journal of Social and Personal Relationships. Birnbaum, G. E., & Laser-Brandt, D. (2002). Gender differences in the experience of heterosexual intercourse. The Canadian Journal of Human Sexuality, 11, 143–158. Birnbaum, G. E., & Reis, H. T. (in press). Women’s sexual working models: An evolutionary-attachment perspective. The Journal of Sex Research. Bogaert, A. F., & Sadava, S. (2002). Adult attachment and sexual behavior. Personal Relationships, 9, 191–204. Bolger, N., Davis, A., & Rafaeli, E. (2003). Diary methods: Capturing life as it is lived. Annual Review of Psychology, 54, 579 – 616. Bowlby, J. (1973). Attachment and loss: Vol. 2. Separation: Anxiety and anger. New York: Basic Books. Bowlby, J. (1982). Attachment and loss: Vol. 1. Attachment (2nd ed.). New York: Basic Books. (Original work published 1969) Brennan, K. A., Clark, C. L., & Shaver, P. R. (1998). Self-report measurement of adult attachment: An integrative overview. In J. A. Simpson & W. S. Rholes (Eds.), Attachment theory and close relationships (pp. 46 –76). New York: Guilford Press. Brennan, K. A., & Shaver, P. R. (1995). Dimensions of adult attachment, affect regulation, and romantic relationship functioning. Personality and Social Psychology Bulletin, 21, 267–283. Brennan, K. A., Wu, S., & Loev, J. (1998). Adult romantic attachment and individual differences in attitudes toward physical contact in the context of adult romantic relationships. In J. A. Simpson & W. S. Rholes (Eds.), Attachment theory and close relationships (pp. 248 –256). New York: Guilford. Bryk, A. S., & Raudenbush, S. W. (2002). Hierarchical linear models: Applications and data analysis methods (2nd ed.). Thousand Oaks, CA: Sage. Buss, D. M. (1998). Sexual strategies theory: Historical origins and current status. Journal of Sex Research, 35, 19 –31. Buss, D. M., & Schmitt, D. P. (1993). Sexual strategies theory: An evolutionary perspective on human mating. Psychological Review, 100, 204 –232. Campbell, L., Simpson, J. A., Boldry, J. G., & Kashy, D. (2005). Perceptions of conflict and support in romantic relationships: The role of attachment anxiety. Journal of Personality and Social Psychology, 88, 510 –531. Carroll, J. L., Volk, K. D., & Hyde, J. S. (1985). Differences between males and females in motives for engaging in sexual intercourse. Archives of Sexual Behavior, 14, 131–139. Cassidy, J. (1999). The nature of the child’s ties. In J. Cassidy & P. R. Shaver (Eds.), Handbook of attachment: Theory, research, and clinical applications (pp. 3–20). New York: Guilford Press. Cassidy, J., & Kobak, R. R. (1988). Avoidance and its relationship with other defensive processes. In J. Belsky & T. Nezworski (Eds.), Clinical implications of attachment (pp. 300 –323). Hillsdale, NJ: Erlbaum. Catania, J. A., Gibson, D. R., Chitwood, D. D., & Coates, T. J. (1990). Methodological problems in AIDS behavioral research: Influences on measurement error and participation bias in studies of sexual behavior. Psychological Bulletin, 108, 339 –362. Collins, N. L., & Read, S. J. (1990). Adult attachment, working models, and relationship quality in dating couples. Journal of Personality and Social Psychology, 58, 644 – 663. Cooper, M. L., Pioli, M., Levitt, A., Talley, A., Micheas, L., & Collins, N. L. (2006). Attachment styles, sex motives, and sexual behavior:
Evidence for gender specific expressions of attachment dynamics. In M. Mikulincer & G. S. Goodman (Eds.), Dynamics of love: Attachment, caregiving, and sex (pp. 243–274). New York: Guilford Press. Cooper, M. L., Shaver, P. R., & Collins, N. L. (1998). Attachment styles, emotion regulation, and adjustment in adolescence. Journal of Personality and Social Psychology, 74, 1380 –1397. Cyranowski, J. M., & Andersen, B. L. (1998). Schemas, sexuality, and romantic attachment. Journal of Personality and Social Psychology, 74, 1364 –1379. Darling, C., Davidson, J., & Passarello, L. (1992). The mystique of first intercourse among college youth: The role of partners, contraceptive practices, and psychological reactions. Journal of Youth and Adolescence, 21, 97–117. Davis, D., Shaver, P. R., & Vernon, M. L. (2004). Attachment style and subjective motivations for sex. Personality and Social Psychology Bulletin, 30, 1076 –1090. Davis, K. E., Ace, A., & Andra, M. (2002). Stalking perpetrators and psychological maltreatment of partners: Anger-jealousy, attachment insecurity, need for control, and break-up context. In K. E. Davis & I. H. Frieze (Eds.), Stalking: Perspectives on victims and perpetrators (pp. 237–264). New York: Springer. DeLamater, J. (1987). Gender differences in sexual scenarios. In K. Kelley (Ed.), Females, males, and sexuality (pp. 127–140). Albany, NY: State University of New York Press. Diamond, L. M. (2003). What does sexual orientation orient? A biobehavioral model distinguishing romantic love and sexual desire. Psychological Review, 110, 173–192. Downey, G., Freitas, A. L., Michaelis, B., & Khouri, H. (1998). The self-fulfilling prophecy in close relationships: Rejection sensitivity and rejection by romantic partners. Journal of Personality and Social Psychology, 75, 545–560. Edwards, J. N., & Booth, A. (1994). Sexuality, marriage, and well-being: The middle years. In A. S. Rossi (Ed.), Sexuality across the life course (pp. 233–259). Chicago: The University of Chicago Press. Feeney, J. A. (1999). Adult romantic attachment and couple relationships. In J. Cassidy & P. R. Shaver (Eds.), Handbook of attachment: Theory, research, and clinical applications (pp. 355–377). New York: Guilford Press. Feeney, J. A., & Noller, P. (2004). Attachment and sexuality in close relationships. In J. H. Harvey, A. Wenzel, & S. Sprecher (Eds.), Handbook of sexuality in close relationships (pp. 183–201). Mahwah, NJ: Erlbaum. Feeney, J. A., Peterson, C., Gallois, C., & Terry, D. J. (2000). Attachment style as a predictor of sexual attitudes and behavior in late adolescence. Psychology and Health, 14, 1105–1122. Fisher, H. E. (1998). Lust, attraction, and attachment in mammalian reproduction. Human Nature, 9, 23–52. Fisher, H. E., Aron, A., Mashek, D., Li, H., & Brown, L. L. (2002). Defining the brain systems of lust, romantic attraction, and attachment. Archives of Sexual Behavior, 31, 413– 419. Fraley, R. C., & Shaver, P. R. (2000). Adult romantic attachment: Theoretical developments, emerging controversies, and unanswered questions. Review of General Psychology, 4, 132–154. Fraley, R. C., & Waller, N. G. (1998). Adult attachment patterns: A test of the typological model. In J. A. Simpson & W. S. Rholes (Eds.), Attachment theory and close relationships (pp. 77–114). New York: Guilford Press. Gable, S. L., Reis, H. T., & Downey, G. (2003). He said, she said: A quasi-signal detection analysis of daily interactions between close relationship partners. Psychological Science, 14, 100 –105. Gable, S. L., Reis, H. T., & Elliot, A. J. (2000). Behavioral activation and inhibition in everyday life. Journal of Personality and Social Psychology, 78, 1135–1149.
ATTACHMENT, SEXUALITY, AND RELATIONSHIP QUALITY Gagnon, J. H., & Simon, W. (1973). Sexual conduct: The social sources of human sexuality. Chicago: Aldine. Gentzler, A. L., & Kerns, K. A. (2004). Associations between insecure attachment and sexual experiences. Personal Relationships, 11, 249 – 265. Green, A. S., Rafaeli, E., Bolger, N., Shrout, P. E., & Reis, H. T. (in press). Paper or plastic? Data equivalence in paper and electronic diaries. Psychological Methods, 11, 87–105. Guggino, J. M., & Ponzetti, J. J. (1997). Gender differences in affective reactions to first coitus. Journal of Adolescence, 20, 189 –200. Hartman, L. M. (1983). Effects of sex and marital therapy on sexual interaction and marital happiness. Journal of Sex and Marital Therapy, 9, 137–151. Hassebrauck, M., & Fehr, B. (2002). Dimensions of relationship quality. Personal Relationships, 9, 253–270. Hatfield, E., Brinton, C., & Cornelius, J. (1989). Passionate love and anxiety in young adolescents. Motivation and Emotion, 13, 271–289. Hazan, C., & Shaver, P. R. (1987). Romantic love conceptualized as an attachment process. Journal of Personality and Social Psychology, 52, 511–524. Hazan, C., & Zeifman, D. (1994). Sex and the psychological tether. In K. Bartholomew & D. Perlman (Eds.), Advances in personal relationships: Vol. 5. Attachment processes in adulthood (pp. 151–177). London: Jessica Kingsley. Hazan, C., Zeifman, D., & Middleton, K. (1994, July). Adult romantic attachment, affection, and sex. Paper presented at the 7th International Conference on Personal Relationships, Groningen, the Netherlands. Henderson-King, D. H., & Veroff, J. (1994). Sexual satisfaction and marital well being in the first years of marriage. Journal of Social and Personal Relationships, 11, 509 –534. Kalichman, S. C., Sarwer, D. B., Johnson, J. R., Ali, S. A., Early, J., & Tuten, J. T. (1993). Sexually coercive behavior and love styles: A replication and extension. Journal of Psychology and Human Sexuality, 6, 1993. Kaplan, H. S. (1974). The new sex therapy. New York: Brunner/Mazel. Kirkpatrick, L. A., & Davis, K. E. (1994). Attachment style, gender, and relationship stability: A longitudinal analysis. Journal of Personality and Social Psychology, 66, 502–512. Kotler, T. (1985). Security and autonomy within marriage. Human Relations, 38, 299 –321. LoPiccolo, J., Heiman, J. R., Hogan, D. R., & Roberts, C. W. (1985). Effectiveness of single therapists versus cotherapy teams in sex therapy. Journal of Consulting and Clinical Psychology, 53, 287–294. Main, M. (1990). Cross-cultural studies of attachment organization: Recent studies, changing methodologies, and the concept of conditional strategies. Human Development, 33, 48 – 61. Mikulincer, M., & Florian, V. (1999). The association between spouses’ self-reports of attachment styles and representations of family dynamics. Family Process, 38, 69 – 83. Mikulincer, M., & Florian, V. (2000). Exploring individual differences in reactions to mortality salience—Does attachment style regulate terror management mechanisms? Journal of Personality and Social Psychology, 79, 260 –273. Mikulincer, M., Florian, V., & Tolmacz, R. (1990). Attachment styles and fear of personal death: A case study of affect regulation. Journal of Personality and Social Psychology, 58, 273–280. Mikulincer, M., Orbach, I., & Iavnieli, D. (1998). Adult attachment style and affect regulation: Strategic variations in subjective self– other similarity. Journal of Personality and Social Psychology, 75, 436 – 448. Mikulincer, M., & Shaver, P. R. (2003). The attachment behavioral system in adulthood: Activation, psychodynamics, and interpersonal processes. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 35, pp. 53–152). New York: Academic Press.
943
Oliver, M. B., & Hyde, J. S. (1993). Gender differences in sexuality: A meta-analysis. Psychological Bulletin, 114, 29 –51. Reedy, M. N., Birren, J. E., & Schaie, K. W. (1981). Age and sex differences in satisfying love relationships across the adult lifespan. Human Development, 24, 52– 66. Reis, H. T., & Gable, S. L. (2000). Event-sampling and other methods for studying everyday experience. In H. T. Reis & C. M. Judd (Eds.), Handbook of research methods in social and personality psychology (pp. 190 –222). New York: Cambridge University Press. Reiss, I. L. (1981). Some observations on ideology and sexuality in America. Journal of Marriage and the Family, 43, 271–283. Schachner, D. A., & Shaver, P. R. (2002). Attachment style and human mate poaching. New Review of Social Psychology, 1, 122–129. Schachner, D. A., & Shaver, P. R. (2004). Attachment dimensions and motives for sex. Personal Relationships, 11, 179 –195. Shaver, P. R., & Brennan, K. A. (1992). Attachment styles and the “Big Five” personality traits: Their connections with each other and with romantic relationship outcomes. Personality and Social Psychology Bulletin, 18, 536 –545. Shaver, P. R., Hazan, C., & Bradshaw, D. (1988). Love as attachment: The integration of three behavioral systems. In R. J. Sternberg & M. Barnes (Eds.), The psychology of love (pp. 68 –99). New Haven, CT: Yale University Press. Shaver, P. R., & Mikulincer, M. (2002). Attachment-related psychodynamics. Attachment and Human Development, 4, 133–161. Shaver, P. R., & Mikulincer, M. (in press). A behavioral systems approach to romantic love relationships: Attachment, caregiving, and sex. In R. J. Sternberg & K. Weis (Eds.), The psychology of love (2nd ed.). New Haven, CT: Yale University Press. Simpson, J. A. (1990). Influence of attachment styles on romantic relationships. Journal of Personality and Social Psychology, 59, 971–980. Sprecher, S., Barbee, A., & Schwartz, P. (1995). “Was it good for you, too?” Gender differences in first sexual intercourse experiences. Journal of Sex Research, 32, 3–15. Sprecher, S., & Cate, R. M. (2004). Sexual satisfaction and sexual expression as predictors of relationship satisfaction and stability. In J. H. Harvey, A. Wenzel, & S. Sprecher (Eds.), Handbook of sexuality in close relationships (pp. 235–256). Mahwah, NJ: Erlbaum. Stephan, C. W., & Bachman, G. F. (1999). What’s sex got to do with it? Attachment, love schemas, and sexuality. Personal Relationships, 6, 111–123. Sternberg, R. J. (1986). A triangular theory of love. Psychological Review, 93, 119 –135. Stone, A. A., Shiffman, S., Schwartz, J. E., Broderick, J. E., & Hufford, M. R. (2002). Patient non-compliance with paper diaries. British Medical Journal, 324, 1193–1194. Tidwell, M. C. O., Reis, H. T., & Shaver, P. R. (1996). Attachment, attractiveness, and social interaction: A diary study. Journal of Personality and Social Psychology, 71, 729 –745. Tracy, J. L., Shaver, P. R., Albino, A. W., & Cooper, M. L. (2003). Attachment styles and adolescent sexuality. In P. Florsheim (Ed.), Adolescent romance and sexual behavior: Theory, research, and practical implications (pp. 137–159). Mahwah, NJ: Erlbaum. Trivers, R. L. (1972). Parental investment and sexual selection. In B. Campbell (Ed.), Sexual selection and the descent of man (pp. 136 –179). Chicago: Aldine. Wright, J., Perrault, R., & Mathieu, M. (1977). The treatment of sexual dysfunction. Archives of General Psychology, 34, 881– 890.
Received June 6, 2005 Revision received May 3, 2006 Accepted May 11, 2006 䡲
PERSONALITY PROCESSES AND INDIVIDUAL DIFFERENCES
Social Role and Birth Cohort Influences on Gender-Linked Personality Traits in Women: A 20-Year Longitudinal Analysis Stephanie Kasen, Henian Chen, Joel Sneed, Thomas Crawford, and Patricia Cohen Columbia University and New York State Psychiatric Institute
Growth curve modeling was used to examine the impact of social role experiences (e.g., marital support, occupational prestige) and birth cohort on mean-level differences and age-related changes in positive personality traits indicative of either femininity or masculinity in 758 mothers heterogeneous in age, assessed 4 times over 2 decades. Both femininity and masculinity increased significantly from mean ages 39 through 59; each was predictive of an age change in the other. Low masculinity was associated with a more rapid increase in femininity, whereas high occupational prestige decreased the magnitude of association between masculinity and femininity. Femininity increased with more marital support but decreased with unmarried status, more children at home, and working full or part time; among full-time workers, that effect was modified by marital support. Masculinity increased with full-time work and high occupational prestige. A trend for differing levels of femininity, and contrasting associations of masculinity with femininity and marital conflict in women born after 1944 compared with those born earlier, suggests shifting social norms and gender relations in the marital role. Keywords: middle age, women, femininity, masculinity, marital role
norms (Baltes, 1987): For example, compared with their counterparts in past cohorts, modern wives hold more sway in the marital relationship, whereas modern husbands are more invested in family roles (Amato & Booth, 1995; Rogers & Amato, 2000). Consequently, the notion that work roles and family roles each contribute exclusively to masculine- and feminine-linked traits, respectively, in recent cohorts of women (and men) may be outdated. Finally, findings that support role associations with genderlinked traits often are based on nonrepresentative samples, many of which are college samples, thus limiting their generalizability to women of varying ages, socioeconomic positions, or both. Here we attempt to address those issues by examining normative age changes in gender-linked traits and estimating the potential influence of social role experiences and birth cohort membership on mean trait levels and age-related changes in a community sample of 758 mothers, assessed four times between 1983 and 2003. We focus on socially desirable characteristics that denote positive communal traits thought to be primarily feminine in nature and thus more typical of women, and on positive agentic traits thought to be primarily masculine in nature and thus more typical of men. Use of longitudinal data to investigate age-related change in personality minimizes the risk of confounding generational differences and maturational change, as may happen when conclusions regarding change are based on cross-sectional comparisons of different age groups. This is especially important in the immediate aftermath of a historical period of rapid and related social change such as occurred for the current study with regard to the contemporaneous decline in gender-role restrictions.
Nearly four decades ago, Neugarten (1968) argued that the manner in which the female personality unfolds is, to a substantial extent, contingent on existing societal demands for labor. That perspective is compatible with recent assertions (and empirical evidence; e.g., see Twenge, 1997, 2001b) that socially determined roles continue to influence the rise and fall of self-reported genderlinked personality traits in women (Barnett & Hyde, 2001; Eagly, Wood, & Diekman, 2000). An explicit or implicit assumption underlying this position is that work roles promote traits thought to be more characteristic of men than women, whereas family roles foster traits thought to be more characteristic of women than men. Nonetheless, support for that hypothesis is based primarily on research related to work role experiences, with surprisingly limited evidence of family role influences. Further, how roles are defined and appraised may vary as a function of historical shifts in social
Stephanie Kasen, Henian Chen, Joel Sneed, Thomas Crawford, and Patricia Cohen, Department of Psychiatry, Columbia University College of Physicians and Surgeons, and Department of Epidemiology, New York State Psychiatric Institute. This study was supported by National Institute of Child Health and Human Development Grant HD-40685. We are indebted to the women in this sample for their continued participation in the Children in the Community Study and to our dedicated field staff for their tireless efforts in the collection of data. Correspondence concerning this article should be addressed to Stephanie Kasen, Unit 47, New York State Psychiatric Institute, 1051 Riverside Drive, New York, NY 10032. E-mail:
[email protected]
Journal of Personality and Social Psychology, 2006, Vol. 91, No. 5, 944 –958 Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.5.944
944
GENDER-LINKED TRAITS IN TWO BIRTH COHORTS
Evidence of Longitudinal Change in Gender-Linked Traits The most noteworthy evidence of longitudinal change in gender-linked personality traits in women has come from the Mills College Longitudinal Study sample, born in 1936 and 1939 and followed since their undergraduate years by Helson and her colleagues: Significant shifts in femininity and masculinity were reported between the ages of 27 and 43 (Helson & Moane, 1987), 43 and 52 (Helson & Wink, 1992), 21 and 52 (Roberts, Helson, & Klohnen, 2002), and 21 and 61 (Helson, Jones, & Kwan, 2002). Findings, based on the Femininity/Masculinity scale of the California Personality Inventory (CPI; Gough, 1987), suggest that masculinity tends to increase in the middle adult years whereas femininity tends to decline. On the CPI, however, femininity and masculinity traits are at opposite poles of a unidimensional continuum; thus, high scores on one pole necessarily constrain high scores on the other. Further, extreme scores at either end reflect relatively negative traits, such as overly sensitive and dependent on the feminine pole versus skeptical and extrapunitive on the masculine pole (Gough & Bradley, 1996); consequently, it often is considered to be a measure of emotional vulnerability.
Femininity and Masculinity: Independent but Related Constructs Alternatively, femininity and masculinity also have been conceptualized as related but distinct constructs, both of which reflect characteristics basic to the healthy development of human personality (Bakan, 1966). One of two widely used self-report measures of gender-linked traits, the Personal Attributes Questionnaire (PAQ; Spence, Helmreich, & Stapp, 1974, 1975), provides an index of relatively positive traits thought to be more typical of women (e.g., understanding of others) and an index of relatively positive traits thought to be more typical of men (e.g., independent). Similarly, the Bem Sex-Role Inventory (BSRI; Bem, 1974), the other widely used self-report measure, has separate indexes of feminine- and masculine-linked traits, many of which overlap with the PAQ and thus are positive; however, the BSRI also has been criticized for its inclusion of socially undesirable items (e.g., childlike and gullible on the Femininity Scale; Pedhazur & Tetenbaum, 1979). The considerable and distinct overlap of those indexes with the broader dimensions of the five-factor personality model proposed by McCrae and Costa (1987) attests to the basic dispositional nature of both feminine- and masculine-linked traits (Lippa, 1991; Roberts, Robins, Trzesniewski, & Caspi, 2003).
Family Roles and Gender-Linked Traits The marital role is ranked higher in importance than the work role by both men and women (Thoits, 1992). It is the primary role of our society in which a positive sense of identity, self-worth, and mastery is developed (Gove, Style, & Hughes, 1990). Moreover, for women, the quality of the marital relationship is a particularly potent predictor of mental health status (Greenberger & O’Neil, 1993). Indeed, there is substantial and consistent evidence that marital conflict is associated with increased risk of depression and other psychiatric disorders in women (Balog et al., 2003; Beach & Fincham, 1998; Fincham, Beach, Harold, & Osborne, 1997; Hammen, 2003; Kendler et al., 1995; O’Leary, Christian, & Mendell,
945
1994; Whisman & Bruce, 1999) and that marital support is associated with their psychological well-being (Barnett & Baruch, 1985; Baruch & Barnett, 1986; Gove et al., 1990; Martire, Stephens, & Townsend, 1998). Marriage also is the central arena in which gender roles are enacted (Eagly et al., 2000); thus, it may be the most salient interpersonal context for continued expression of feminine-linked traits by adult women. Further, wives in recent cohorts wield more power in the marital relationship than wives in past cohorts (Amato & Booth, 1995; Rogers & Amato, 2000); thus, the expression of masculine-linked traits also may be influenced in modern marriages. Abele (2003) used the PAQ to assess socially desirable genderlinked traits in female graduate students: High femininity predicted involvement in family roles 18 months later; however, that association was not reversible (i.e., involvement in family roles did not predict increased femininity). Mori, Nakashima, and Kurita (2002) reported that femininity, assessed by the BSRI, was highly correlated with family support, but the measure of support used did not clarify which family member offered support. On the other hand, increased femininity (as assessed by the CPI and thus potentially moving in the direction of emotional vulnerability) also has been linked to marital tension, but only in young adults (Roberts et al., 2002). Overall, more satisfactory marital adjustment has been linked to sex-congruent or androgynous (i.e., high in both feminine and masculine) role traits and behaviors (Auster & Ohm, 2000; Barrett & White, 2002; Baucom & Aiken, 1984; Kalin & Lloyd, 1985; Peterson, Baucom, Elliott, & Farr, 1989). With regard to the parent role, mothers typically assume the principal responsibility for child-care activities. Research that focuses on the impact of parenting on gender-linked traits and role behavior has been conducted primarily within the context of being a working mother. For the most part, working mothers are judged to be less agentic and less committed to their jobs than childless working women (Crittenden, 2002; Fuegen, Biernat, Haines, & Deaux, 2004) and also less communal and less effective parents than mothers who are full-time homemakers (Bridges & Etaugh, 1995; Bridges, Etaugh, & Barnes-Farrell, 2002). McHale and Huston (1984) reported that mothers who worked more hours or held less traditional attitudes regarding women’s roles were less involved with their children and engaged in fewer child-care activities; nevertheless, they also reported that self-ascribed feminine- and masculine-related traits, as assessed by the PAQ, were not related to parenting behaviors.
The Work Role and Gender-Linked Traits Recognition of work role influences on longitudinal personality change was facilitated by the seminal work of Kohn and Schooler (1978, 1982): They followed over 3,000 employed men and found that those engaged in self-directed (i.e., less closely supervised, independent) work increased in ideational flexibility and goal orientation, traits considered to be masculine in nature. Vandewater and Stewart (1998) compared women in differing occupations and found that those in career track occupations, which emphasize self-directed work, were rated as more instrumental and ambitious by their supervisors, whereas those in job track occupations, which emphasize closely supervised work, were rated as less self-reliant and more dependent. Roberts (1997) reported that women who increased participation in the workforce or increased in occupa-
KASEN, CHEN, SNEED, CRAWFORD, AND COHEN
946
tional prestige experienced a significant rise in their sense of agency. Other findings suggest that the degree to which women self-ascribe either feminine- or masculine-linked traits is more closely associated with employment status or education than with involvement in family roles (Abele, 2003; Cunningham & Antill, 1984).
The Current Study Here we draw on longitudinal data obtained from a representative sample of 758 women assessed on measures of femininity and masculinity, work status and occupational prestige, and marital status and quality (support and conflict) in four waves of data collection in 1983, 1986, 1993, and 2003, when they were mean ages 39, 42, 49, and 59, respectively. Because these women all are mothers, we were not able to examine differences in mean trait levels or degree of change between women with and women without children. However, number of children at home affects mothers’ emotional and physical resources (Sameroff, 1998), with higher numbers contributing to role overload and conflict (Barnett, Brennan, & Marshall, 1994), psychological distress (Gove & Geerken, 1977; Guelzow, Bird, & Koball, 1991; Russo & Zierk, 1992), and marital dissatisfaction (Glenn & McLanahan, 1982; Miller, 1975); consequently it is plausible that role-related traits also may be influenced. Thus, number of children at home was included in the analyses to control for those potential effects. We also examine the impact on gender-linked traits of historical shifts in role norms for women by considering birth cohort membership. Women’s role options increased dramatically in the latter half of the 20th century, owing in part to broad social forces that spurred events such as the women’s movement that emerged in 1966, undertaken to raise awareness of gender-based role restrictions and behaviors that prevailed at that time (Buechler, 1990; Chafe, 1991). The women’s movement has been implicated in personality change among women (Agronick & Duncan, 1998; Duncan & Agronick, 1995), with exposure in early adulthood, a life stage characterized by identity formation (Mannheim, 1972), exerting more influence than exposure in early middle adulthood, after life role commitments have been made. By 1969, the movement had gained momentum; therefore, the women in our sample were assigned to one of two birth cohorts on the basis of their age at that time: those who were born between 1945 and 1958 and were, on average, age 20 and thus more likely to be receptive to personality restructuring (referred to herein as baby boomers), or those who were born between 1931 and 1944 and were, on average, age 30 and thus more likely to have made role commitments (referred to herein as preboomers). That classification was shown to be salient with regard to cohort differences in depression (Kasen, Cohen, Chen, & Castille, 2003) and in associations between depression and marital and work role status (Kasen, Cohen, Berenson, Chen, & Dufur, 2005). As noted above, we focus here on positive gender-linked traits. The research questions addressed are as follows: 1.
Are there normative age changes in femininity or masculinity in this sample of women during the middle adult years?
2.
Are age trajectories of femininity and masculinity predictive of each other?
3.
Do social role experiences predict mean-level differences or age changes in femininity or masculinity?
4.
Does birth cohort membership predict mean-level differences or age changes in femininity or masculinity?
5.
Do associations of social role experiences with femininity or masculinity vary by cohort?
Hypotheses Underlying our hypotheses is the life span conception of personality change, which posits that change is not restricted to earlier developmental levels but may occur throughout the life course and that broad social influences may alter personality traits beyond genetic and more proximal environmental effects (Baltes, 1987; Baltes & Schaie, 1973; Helson & Stewart, 1994). We also adopt the position of proponents of contextual theories, who have established that the immediate social context, namely, involvement in work, marriage, and parenting roles, is implicated in shaping adult personality (e.g., Caspi, 1987; Helson, Kwan, John, & Jones, 2002; Kohn & Schooler, 1978, 1982; Neugarten, 1968, 1972; Roberts et al., 2003). We focus here on both positive communal traits and positive agentic traits as each is basic to personality growth and to overall human development (Bakan, 1966); thus, an increased level of either is considered to be a desirable outcome. Women report increases in communal traits such as generativity and warmth (Haan, Millsap, & Hartka, 1986; Morfei, Hooker, Carpenter, Mix, & Blakeley, 2004; Stewart, Ostrove, & Helson, 2001) and agentic traits such as instrumental competence and independence (Helson, Kwan, et al., 2002; Helson & Wink, 1992; Parker & Aldwin, 1997; Roberts et al., 2002) in midlife; thus, we expected both femininity and masculinity to rise with age. We also expected that each would be predictive of the other, owing to evidence of increasing prevalence of androgyny in successive cohorts of women (Twenge, 1997) and positive associations in women between relatedness, a communal trait, and autonomy, an agentic trait (Rankin-Esquer, Burnett, Baucom, & Epstein, 1997). On the basis of the literature linking marital status and quality to women’s well-being (as noted above), we expected that unmarried women would report fewer feminine-linked traits than married women; that marital support would be positively related to femininity; and that marital conflict would be inversely related to both femininity and masculinity. Similarly, because having more children is related to psychological distress (as noted above), we also expected an inverse association between number of children and femininity. Empirical support for the impact of work on the development of masculine-linked traits, especially of career-track occupations, is substantial (e.g., Clausen & Gilens, 1990; Kohn & Schooler, 1978, 1982; Roberts, 1997; Vandewater & Stewart, 1998); thus, we expected that working full or part time or being in a high-prestige occupation would be related to less feminine- but more masculine-linked traits. Spousal support alleviates stress related to role overload and conflict in working mothers, especially those working full time (Barnett, Marshall, & Singer, 1992; Greenberger & O’Neil, 1993); thus, we expected that full-time working women with more marital support would report higher levels of femininity than full-time working women with less marital support.
GENDER-LINKED TRAITS IN TWO BIRTH COHORTS
Recent cohorts are more likely to endorse both feminine- and masculine-linked traits in men and women than earlier cohorts (Auster & Ohm, 2000); thus, we hypothesized that baby boomers would be higher in both femininity and masculinity than preboomers. Rogers and Amato (2000) found that although wives in more recent cohorts have acquired more influence and power in the marital relationship than wives in earlier cohorts, they also report more conflict. Increased conflict may be due to a greater proportion of wives in more recent cohorts juggling dual family and work roles. However, that takes its greatest toll among younger women; by midlife, conflict may diminish and more masculine traits emerge owing in part to those changes in gender relations. Thus, we also hypothesized that the expected inverse association between marital conflict and masculinity would be stronger in baby boomers.
Method Sample This sample of 758 mothers first was assessed on several measures of personality traits and work and marital role experiences in 1983 when they were, on average, 39 years old; subsequent follow-ups occurred in 1986, 1993, and 2003, at mean ages 42, 49, and 59, respectively. These women,
947
originally sampled in 1975 on the basis of family residence in one of two upstate New York counties and having a child between the ages of 1 and 10, were randomly selected and interviewed about their offspring for a study of childhood behavior. Detailed information regarding study methodology is available from previous reports (P. Cohen & Cohen, 1996; Kasen et al., 2003). Table 1 shows the number of women, mean age (and standard deviations), full- and part-time work status, marital status, and number of children at home at each assessment point for the total sample and by individual cohort: preboomers (born between 1931 and 1944) and baby boomers (born between 1945 and 1958). Accumulated mean years of education (and standard deviations) is shown in the wave 2003 column (with 12 years indicating high school graduate). The women in the sample are 91% White; reside in urban, suburban, and rural areas; span the full socioeconomic status range; and are representative of the northeastern region of the United States, from which they were sampled in 1975. Retention rates were 95% and 90% in 1986 and 1993, respectively. In the recent 2003 follow-up, 609 women were reinterviewed and another 73 women had died, accounting for 90% of the 1983 sample; those remaining were not reinterviewed owing to refusal to participate (31), serious illness (17), study time constraints (16), and failure to locate (12).
Procedure Trained lay interviewers collected study data in the women’s homes. Informed consent was obtained in adherence to institutional guidelines.
Table 1 Mean Scores (and Standard Deviations) on Predictor Scales and Demographic Status at the Four Assessments Measure Number assessed Total sample Preboomers Baby boomers Mean (SD) age Total sample Preboomers Baby boomers Mean years of education by 2003 Total sample Preboomers Baby boomers Mean number children at home Total sample Preboomers Baby boomers Percentage married Total sample Preboomers Baby boomers Percentage working full time Total sample Preboomers Baby boomers Percentage working part time Total sample Preboomers Baby boomers Mean (SD) of predictor scales Femininity Masculinity Marital Support Marital Conflict Occupational Prestige
1983
1986
1993
2003
758 401 357
723 378 345
684 361 323
605 293 301
39 (6.0) 44 (4.1) 34 (3.1)
42 (6.0) 47 (4.0) 37 (3.0)
49 (6.0) 53 (4.1) 43 (3.1)
59 (5.7) 63 (3.9) 53 (2.9) 12.7 12.8 12.6
3.4 3.6 3.0
2.3 2.2 2.5
1.9 1.5 2.2
0.3 0.2 0.3
79.5 82.6 76.8
77.7 80.9 74.7
76.7 79.1 74.1
74.3 69.0 77.9
41.2 40.3 42.6
51.6 50.0 54.3
53.2 48.9 60.2
42.6 26.0 60.3
19.7 19.6 19.9
17.4 17.8 17.0
13.9 13.2 15.1
7.4 7.3 7.3
22.50 (4.2) 15.66 (5.1) 11.96 (3.2) 3.31 (2.1) 3.47 (1.5)
22.67 (4.0) 15.81 (5.1) 11.85 (3.5) 3.23 (2.0) 3.60 (1.6)
22.80 (4.1) 15.98 (5.2) 11.72 (3.8) 3.22 (2.0) 3.99 (1.5)
23.73 (4.0) 16.22 (5.1) 12.19 (2.6) 2.55 (1.3) 3.75 (1.9)
948
KASEN, CHEN, SNEED, CRAWFORD, AND COHEN
Measures Self-attributions of gender-linked traits were assessed with 20 items from the BSRI (Bem, 1974), a widely used self-report measure of gender role classification composed of 60 adjectives classified as more descriptive of a feminine role orientation (20), more descriptive of a masculine role orientation (20), or neutral (20). Although it has been three decades since it was developed, the BSRI still is considered to be a useful index of self-perceived feminine- and masculine-linked traits that can be used to assess either men or women of varying ages (Oswald, 2004). Ten positive items each from the Feminine (affectionate, compassionate, feminine, gentle, loves children, shy, sensitive to the needs of others, tender, warm, yielding) and Masculine (acts as a leader, aggressive, ambitious, analytical, assertive, competitive, dominant, independent, masculine, self-reliant) BSRI scales were rated on a 4-point Likert scale ranging from 0 (never or almost never true about me) to 3 (always or almost always true about me) and summed to create Femininity and Masculinity scales, respectively. Estimates of internal consistency were .80 or above on each scale at all assessments. Zero-order correlations between the Femininity and Masculinity scales declined gradually over mean ages 39 (r ⫽ .22, p ⬍ .01), 42 (r ⫽ .18, p ⬍ .01), 48 (r ⫽ .12, p ⬍ .01), and 59 (r ⫽ –.03, ns); the magnitude of those associations validates the orthogonal structure of these scales. Selection of Femininity and Masculinity scale items from the BSRI by study investigators was based in large part on their social desirability. The Marital Support and Marital Conflict scales used here were adapted from the Marital-Adjustment and Marital-Prediction tests of Locke and Wallace (1959). Marital support was assessed with four questionnaire items indicating how often husband and wife help each other when there is trouble, talk with each other about everything, are very affectionate with each other, and engage in outside interests together, all rated on a 5-point Likert scale ranging from 0 (never) to 4 (almost always). Estimates of internal consistency were .80, .82, .83, and .83 across the four assessments. Marital conflict was assessed with three items indicating how often differences of opinion occur and how often arguing and yelling or “rough stuff” occur, rated on a 5-point Likert scale from 0 (never) to 4 (almost always). Estimates of internal consistency were .60, .63, .61, and .60 across the four assessments. The validity of the Marital Support and Marital Conflict scales has been supported by significant longitudinal associations with offspring emotional and behavioral disturbances and substance use in the expected directions (Brook, Brook, Gordon, Whiteman, & Cohen, 1990; Brook, Zheng, Whiteman, & Brook, 2001; Crawford, Cohen, Midlarsky, & Brook, 2001). Occupational prestige was assessed on a 7-point scale based on modified versions of job categories as indicated in Hollingshead and Redlich (1958). The lowest scores reflect low-level occupations that typically are low in challenge and high in supervision (e.g., waitress, health care aid); the highest scores reflect high-level occupations that typically are high in challenge, responsibility, and autonomy (e.g., business executives, major professions in academia, medicine, or law). Occupations were classified as follows: 1 ⫽ unskilled work, 2 ⫽ semiskilled, 3 ⫽ skilled, 4 ⫽ clerical/ sales, 5 ⫽ administrative, 6 ⫽ managers/lesser professionals, 7 ⫽ executives, proprietors of major concerns, major professions); interrater reliability ranged from .84 to .87 across the four assessments. Similar constructs have been used by others to examine qualitative aspects of the work role (Duncan & Agronick, 1995; Roberts, 1997; Stewart & Vandewater, 1993). Mean scores (and standard deviations) on the Femininity, Masculinity, Marital Support, Marital Conflict, and Occupational Prestige scales as well as information on number of women assessed, age, years of education, number of children at home, and marital and work status at all four assessments are shown in Table 1. The women were classified as married or unmarried (separated/divorced, widowed, never married) and as fulltime workers (paid employment 40 or more hours per week), part-time workers (paid employment for 20 or more hours but less than 40 hours per week), or not working (all others) at each assessment.
Conditional Missing Data For a substantial number of women, marital role occupancy varied across the four assessment points because of separation or divorce, widowhood, and remarriage; moreover, work role occupancy also was time varying. Questions regarding support from and conflict with marital partners or occupational prestige were conditional on role occupancy; thus, a woman would have data missing on these variables in any wave of data collection at which she was not married or not working. We used the missing-data/dichotomy method recommended by J. Cohen, Cohen, West, and Aiken (2003) to address this issue: Women not occupying the role were scored 0 on the role (marital or work) status variable(s), and cohortspecific means on the Marital Support, Marital Conflict, and Occupational Prestige scales were substituted for each woman in the corresponding cohort in any wave at which she was not married or working. Such statistical treatment of missing conditional data uses all available information, thus avoiding the loss of statistical power and precision of estimates and also minimizing the risk of obtaining biased results (J. Cohen et al., 2003); this method has been successfully implemented in other studies (e.g., Golding, 1989; Noor, 1995, 1997; Stumpf, 1978).
Data Analytic Models The PROC MIXED procedure from the SAS statistical package (Littell, Miliken, Stroup, & Wolfinger, 1996; Singer, 1998) was used to estimate the effects of social role predictors and birth cohort membership on mean levels and age change trajectories of femininity and masculinity over the two-decade period covered by the four follow-ups. In addition, we were able to estimate the impact of the alternate gender-linked trait by examining each trait as a dependent variable with the other incorporated into the model as a time-varying independent variable. This growth curve method has a number of advantages that allow fuller exploitation of longitudinal data as compared with traditional regression: It estimates both linear and nonlinear change, accommodates time-varying repeated measures, permits inclusion of individuals not assessed at all time points, allows data from individuals assessed at different ages to be combined, and is tolerant of unequal intervals between data points. Growth curve modeling provides estimates of both random effects (i.e., variation in individuals’ means and slopes and deviation from own slope of femininity or masculinity) and fixed effects (i.e., average effects of predictors on femininity or masculinity across participants); however, as the current study focuses on normative change in gender-linked traits and whether mean-level differences or age-related changes are attributable to variations in social role experiences or birth cohort membership, only the fixed effects are reported here. Basic models examined the fixed effects of linear and nonlinear (quadratic) age changes in feminine- and masculine-linked traits (e.g., Femininity ⫽ ␣1 ⫹ 11 linear age ⫹ 12 quadratic age). Main models examined the additional fixed effects of cohort, the alternate gender-linked scale, number of children at home, marital status and marital support and conflict, and full- and part-time work status and occupational prestige, all of which, with the exception of cohort, were time varying; in addition, years of education completed was controlled1 (e.g., Femininity ⫽ ␣2 ⫹ 21 linear age ⫹  22 quadratic age ⫹  23 education ⫹  24 cohort ⫹  25 masculinity ⫹ 26 number of children ⫹ 27 marital status ⫹ 28 marital support ⫹ 29 marital conflict ⫹ 210 full-time work ⫹ 211 part-time work ⫹ 212 occupational prestige). Interaction models examined the effects of two-way cross-product interaction terms, formed by multiplicative combinations of age, cohort, alternate gender scale, and marital and work role variables. If the set of
1 The zero-order correlation between years of educational and overall occupational level was .34 ( p ⬍ .001). Controlling for age and quadratic age only, there was a positive association between the control variable years of education and masculinity ( ⫽ .13, SE ⫽ .06, p ⬍ .05); however, that effect was not independent of social role predictors (see Table 3).
GENDER-LINKED TRAITS IN TWO BIRTH COHORTS
949
Table 2 Predictors of Feminine-Linked Traits in Two Cohorts of Women Basic model
Main model
Interaction model
Effect

SE

SE

SE
Intercept Age Age2 Years of education Cohorta Masculinity Number of children Unmarried Marital support Marital conflict Full-time work Part-time work Occupational prestige Masculinity–age change Masculinity–cohort Masculinity–occupational prestige Marital support–full-time work
22.995**** .051**** ⫺.0002
0.132 .008 .001
23.040**** .028**** ⫺.002** ⫺.004 .454† .162**** ⫺.109* ⫺.566** .089**** ⫺.038 ⫺.510*** ⫺.581*** ⫺.044
0.205 .010 .001 .049 .258 .016 .049 .188 .020 .042 .152 .189 .052
23.052**** .029** ⫺.002** ⫺.009 .468† .188**** ⫺.110* ⫺.508** .143**** ⫺.038 ⫺.514*** ⫺.576*** ⫺.079 ⫺.005**** ⫺.078* ⫺.031*** ⫺.090**
0.204 .010 .001 .049 .257 .023 .049 .187 .028 .042 .151 .188 .052 .001 .035 .009 .032
Note. All parameter entries are maximum-likelihood estimates using SAS PROC MIXED. Age was centered at 47.  ⫽ estimate. Preboomers ⫽ 0, baby boomers ⫽ 1. † p ⬍ .10. * p ⬍ .05. ** p ⬍ .01. *** p ⬍ .001. **** p ⬍ .0001. a
interactions significantly improved the fit to the data as compared with the predictor model, the significant interactions within the set were graphed and interpreted; only significant interactions were tabled. To facilitate interpretation, age was centered at 47, the mean age of the total sample; thus, the regression intercept in the basic model represents estimated overall mean levels of Femininity or Masculinity2 at age 47; all other scaled variables were centered at the mean. Cohort was set at 0 for preboomers and 1 for baby boomers, and marital status was set at 0 for married and 1 for unmarried, producing main effects for preboomers and married women, respectively, when interactions with these variables were examined. To analyze the effects of work status, we used dummy variable coding to compare both full-time workers and part-time workers with nonworking women, thus producing main effects for nonworking women when interaction effects were examined for full- and part-time workers.
Results Preliminary Analyses Preliminary analyses (not tabled) compared cohorts with regard to marital and work status and number of children at home at mean age 43 (using 1983 data for preboomers [N ⫽ 401] and 1993 data for baby boomers [N ⫽ 323], as indicated in Table 1) and mean age 53 (using 1993 data for preboomers [N ⫽ 361] and 2003 data for baby boomers [N ⫽ 301], as indicated in Table 1). More preboomers than baby boomers were in intact marriages at mean age 43 (82.6% vs. 74.1%), 2(1) ⫽ 8.8, p ⬍ .01, whereas more baby boomers than preboomers were working full time at mean ages 43 (60.2% vs. 40.3%), 2(1) ⫽ 27.8, p ⬍ .0001, and 53 (60.3% vs. 48.9%), 2(1) ⫽ 8.3, p ⬍ .01. At mean age 53, however, more preboomers than baby boomers were working part time (13.2% vs. 7.3%), 2(1) ⫽ 5.8, p ⬍ .05. Nonetheless, across full- and part-time work statuses, more baby boomers than preboomers were working at mean ages 43 (75.7% vs. 60.1%), 2(1) ⫽ 14.24, p ⬍ .001, and 53 (68.7% vs. 62.2%), 2(1) ⫽ 10. 45, p ⬍ .001. Baby boomers also had fewer children at home at
mean ages 43 (2.2 vs. 3.4), two-tailed t(1) ⫽ 12.64, p ⬍ .001, and 53 (0.3 vs. 1.5), two-tailed t(1) ⫽ 13.01, p ⬍ .001, than preboomers. Those differences are compatible with the increased divorce rate, increased workforce participation, and reduced family size reported in more recent cohorts of women relative to women in earlier cohorts (Aube, Fleury, & Smetana, 2000; Bond, Galinsky, & Swanberg, 1998).
Femininity The basic model examined linear and nonlinear (quadratic) age changes in Femininity (Table 2). The average Femininity score at age 47 was 22.995, with a significant .051 average unit increase per year from mean ages 39 to 59 ( ⫽ .051, SE ⫽ .008, p ⬍ .0001) but no significant quadratic age effect; trajectories are shown in Figure 1 (left panel) by cohort. The addition of fixed effects of years of education, cohort, masculinity, number of children at home, marital status and marital support and conflict, and full- and part-time work statuses and occupational prestige in the main model improved the fit to the data, 2(10, N ⫽ 2,770) ⫽ 159.9, p ⬍ .001. Mean level of Femininity rose significantly with each unit increase in masculinity ( ⫽ .162, SE ⫽ .016, p ⬍ .0001) and marital support ( ⫽ .089, SE ⫽ .020, p ⬍ .0001) and was higher among baby boomers than among preboomers, but only at a trend level ( ⫽ .454, SE ⫽ .258, p ⬍ .10) (albeit prior to consideration of other fixed effects, baby boomers reported a significantly higher mean level of Femininity than preboomers independent of age:  ⫽ .647, SE ⫽ .251, p ⫽ .01, not tabled). On the other hand, mean level of Femininity declined significantly with each additional child at home ( ⫽ –.109, SE ⫽ .049, p ⫽ 2
For clarity we capitalize Femininity and Masculinity when these terms are used as time-varying dependent variables.
KASEN, CHEN, SNEED, CRAWFORD, AND COHEN
950 25
16.5
Femininity
Masculinity
24 16.0
23
Preboomers
15.5
Babyboomers
22
21 30
39
48
57
66
15.0
75
30
39
48
Age
57
66
75
Ag e
Figure 1.
Age changes in Femininity and Masculinity by birth cohort.
addition, the inverse association of full-time work status with Femininity was modified by marital support: Nonworking women with high marital support had the highest mean level of Femininity; however, women working full time with high marital support had a higher mean level of Femininity than full-time working or nonworking women with low marital support (Figure 4).
.05) and was significantly lower among unmarried women than among married women ( ⫽ –.566, SE ⫽ .188, p ⬍ .01) and, compared with nonworking women, among women working full time ( ⫽ –.510, SE ⫽ .152, p ⫽ .001) or part time ( ⫽ –.581, SE ⫽ .189, p ⫽ .001). Compared with the main model, the interaction model significantly improved the fit to the data, 2(4, N ⫽ 2,770) ⫽ 34.7, p ⬍ .001. To illustrate interaction effects with scaled variables as high and low values (e.g., high marital support vs. low marital support), cutoffs of ⫾ 1 SD were used. Masculinity modified the trajectory of Femininity so that the linear increase in Femininity with age was more rapid in women with low masculinity than in women with high masculinity (0.034 per year vs. 0.024 per year, respectively) (Figure 2, left panel vs. right panel). Moreover, although the positive association of masculinity with Femininity remained significant across cohorts, among women with high masculinity there was virtually no cohort disparity in mean level of Femininity, whereas among women with low masculinity, preboomers reported a significantly lower mean level of Femininity than baby boomers (Figure 2, right panel vs. left panel). Occupational prestige modified the positive association of masculinity with Femininity such that the association was of lesser magnitude among women in high-prestige occupations than among women in lowprestige occupations (Figure 3, right panel vs. left panel). In
Low masculinity
Masculinity The basic model examined the fixed effects of the linear and nonlinear (quadratic) age change in Masculinity (Table 3). The average Masculinity score at age 47 was 16.011, with a significant .019 average unit increase per year ( ⫽ .019, SE ⫽ .009, p ⬍ .05) but no quadratic effect of age; trajectories are shown in Figure 1 (right panel) by cohort. The addition of fixed effects of years of education, cohort, femininity, number of children at home, marital status and marital support and conflict, and full- and part-time work statuses and occupational prestige in the main model improved the fit to the data, 2(10, N ⫽ 2,770) ⫽ 169.5, p ⬍ .001; however, the linear effect was no longer significant. Mean level of Masculinity rose significantly with each unit increase in femininity ( ⫽ .228, SE ⫽ .021, p ⬍ .0001) and occupational prestige ( ⫽ .224, SE ⫽ .058, p ⬍ .001) and was significantly higher among
High masculinity
Femininity
26
Preboomers
24
Baby boomers
22
20 30
39
48
57 Age
Figure 2.
66
75
30
39
48
57
66
75
Age
Age changes in Femininity by masculinity level and cohort differences in Femininity.
GENDER-LINKED TRAITS IN TWO BIRTH COHORTS
Low occupational status
29
951
H igh occupational status
28
Femininity
27
High masculinity
26
Low masculinity
25 24 23
25
35
45
Figure 3.
55
Age
65
75
25
35
45
55
Age
65
75
Differences in Femininity level by masculinity and occupational prestige.
women working full time than among nonworking women ( ⫽ .927, SE ⫽ .171, p ⬍ .0001). Compared with the main model, the interaction model significantly improved the fit to the data, 2(1, N ⫽ 2,770) ⫽ 5.5, p ⬍ .05. As shown in Figure 5, among preboomers, high marital conflict was associated with a higher mean level of Masculinity (the reverse of what was hypothesized), whereas among baby boomers, low marital conflict was associated with a higher mean level of Masculinity (as hypothesized). Moreover, baby boomers in low-conflict marriages had the highest mean level of Masculinity, whereas baby boomers in high-conflict marriages had the lowest.
Additional Analyses We found the opposing directions of the marital conflict 3 Masculinity association across cohorts to be quite provocative; thus, we undertook additional analyses to explore those differences. Because of space limitations and the ex post facto nature of these analyses, however, results are presented only briefly in the text (but are shown in Table 4). Marital support significantly increased with age in the basic model ( ⫽ .032, SE ⫽ .008, p ⬍ .0001), following an earlier decline (quadratic age effect:  ⫽ .0013, SE ⫽ .0006, p ⬍ .05). However, the main-interaction model significantly improved the fit to the data, 2(3, N ⫽ 2,770) ⫽ 12.0, p ⬍ .0001. As shown in Figure 6 (left panel), compared with preboomers, the upswing in marital support began at a younger age in baby 25
High support
boomers; moreover, after consideration of this age– cohort interaction effect, the increase in marital support with age was no longer significant among preboomers. Viewed alternatively, although there was no significant difference between cohorts in overall mean level of marital support, preboomers reported higher levels than baby boomers prior to reaching age 50, whereas baby boomers reported higher levels than preboomers after reaching age 50. Marital conflict significantly decreased with age in the basic model ( ⫽ –.024, SE ⫽ .004, p ⬍ .0001); however, there was no quadratic age effect. The main-interaction model significantly improved the fit to the data, 2(3, N ⫽ 2,770) ⫽ 13.3, p ⬍ .0001. Mean level of marital conflict was comparable across cohorts; however, as shown in Figure 6 (right panel), baby boomers reported a decline in conflict from about their late 40s, whereas a more stable level of conflict across the entire age span was reported by preboomers; moreover, after consideration of that effect, decline in marital conflict with age was no longer significant among preboomers.
Discussion Normative Change in Femininity and Masculinity With Age Our hypothesized finding of an increase in positive femininelinked traits with age is consistent with longitudinal evidence of increased warmth, generativity, and relatedness in women at midlife (Franz, 1997; Haan et al., 1986; Stewart et al., 2001).
Low support Working
Femininity
24
Not working 23
22
30
39
48
57
Age Figure 4.
66
75
21
30
39
48
57
66
75
Age
Differences in Femininity level by work status and marital support.
KASEN, CHEN, SNEED, CRAWFORD, AND COHEN
952
Table 3 Predictors of Masculine-Linked Traits in Two Cohorts of Women Basic model
Main model
Interaction model
Effect

SE

SE

SE
Intercept Age Age2 Education Cohorta Femininity Number of children Unmarried Marital support Marital conflict Full-time work Part-time work Occupational prestige Marital conflict–cohort difference
16.011**** .019* ⫺.001
0.172 .009 .001
15.495*** ⫺.007 .001 .068 ⫺.077 .228**** ⫺.070 ⫺.041 .012 ⫺.015 .927**** .275 .224****
0.256 .012 .001 .065 .332 .021 .056 .217 .022 .047 .171 .210 .058
15.486**** ⫺.009 .001 .073 ⫺.062 .227**** ⫺.073 .049 .012 .087 .911**** .268 .225**** ⫺.202*
0.256 .012 .001 .065 .331 .021 .056 .217 .022 .064 .171 .210 .058 .086
Note. All parameter entries are maximum-likelihood estimates using SAS PROC MIXED. Age was centered at 47.  ⫽ estimate. Preboomers ⫽ 0, baby boomers ⫽ 1. * p ⬍ .05. *** p ⬍ .001. **** p ⬍ .0001. a
social role or other predictors examined here, at least in this sample of mothers.
Moreover, our finding of a parallel increase in positive masculinelinked traits, also expected, is supported by considerable longitudinal evidence that women become more autonomous and competent at midlife relative to their earlier adult years (Helson, Jones, & Kwan, 2002; Helson & Moane, 1987; Helson & Wink, 1992; Parker & Aldwin, 1997; Roberts et al., 2002). Those findings have implications for how femininity and masculinity are conceptualized and measured. Women may not decline in positive femininelinked traits in midlife as is suggested when femininity and masculinity are defined as opposite ends of a single dimension, such as on the Femininity/Masculinity Scale of the CPI, where high scores denote negative feminine-linked traits and low scores denote negative masculine-linked traits. Gender role socialization pressures may suppress masculine-related traits during the younger adult years, especially in women with family role commitments; by midlife, however, such pressures tend to diminish, and women may feel relaxed enough to express their full personality, including both traditional and nontraditional facets. Nonetheless, we also found that the observed linear effect of increased masculinity was accounted for by other predictors in the main model, indicating that age-related changes in masculinity may be attributable to
Confluence of Femininity and Masculinity As hypothesized, femininity and masculinity each were predictive of an increase with age in the other. This finding corresponds to evidence of positive associations between traits indicative of femininity and masculinity (Rankin-Esquer et al., 1997) and supports the contention of others that feminine- and masculine-linked traits are both independent and related and are exhibited simultaneously (Barrett & White, 2002; Bem, 1974, 1978; Feldman & Aschenbrenner, 1983; Galambos, Almeida, & Petersen, 1990). We also found that women low in masculinity increased in femininity with age at a more rapid rate than women high in masculinity, thus exhibiting an increasingly extreme sex-congruent orientation. Women with low levels of both masculinity and femininity are reported to experience poorer mental health and lower self-esteem compared with women with high levels of either or both (Auster & Ohm, 2000; Barrett & White, 2002); thus, those with few
17.0
Preboomers
Baby boomers
Masculinity
16.5
30
39
48
57
Age Figure 5.
66
16.0
High conflict
15.5
Low conflict
15.0 75 30
39
48
57
66
75
Age
Differences in Masculinity level by birth cohort and marital conflict.
GENDER-LINKED TRAITS IN TWO BIRTH COHORTS
953
Table 4 Age Changes and Cohort Differences in Marital Support and Marital Conflict Among Women Basic model Effect Marital support Intercept Age Age2 Cohorta Age–cohort Marital conflict Intercept Age Age2 Cohort Age–cohort Age2–cohort difference
Main-interaction model

SE

10.954**** .032**** .0013*
0.123 .008 .0006
10.947**** ⫺.017 .004**** ⫺.010 .085***
0.159 .017 .001 .234 .024
3.104**** ⫺.024**** .000
0.055 .004 .000
3.107**** ⫺.015 ⫺.000 .034 ⫺.031** ⫺.002*
0.073 .009 .001 .112 .011 .0009
SE
Note. All parameter entries are maximum-likelihood estimates using SAS PROC MIXED. Age was centered at 47.  ⫽ estimate. a Preboomers ⫽ 0, baby boomers ⫽ 1. * p ⬍ .05. ** p ⬍ .01. *** p ⬍ .001. **** p ⬍ .0001.
masculine-linked traits may adhere even more so to traditional femininity in order to enhance positive feelings about themselves.
The Influence of Family Roles on Gender-Linked Traits As hypothesized, marital support was associated with more femininity, whereas unmarried status was associated with less. Spousal support is reported to enhance women’s feelings of selfefficacy in the wife role (Martire et al., 1998), which may spur the expression of traditional gender-role traits related to enactment of that role. Moreover, having a supportive spouse is associated with increased psychological well-being and reduced role stress in women (e.g., Gove et al., 1990), especially among those with offspring (Barnett et al., 1994). Thus, lack of a supportive partner also may be implicated in the lower level of femininity found here in unmarried mothers compared with married mothers. There was an inverse relation between number of children at home and femininity, also expected. In women, role-related stress is attributed more frequently to the parenting role than to the marital or worker role (Barnett & Baruch, 1985; Barnett et al., 1992); con16
sequently, in mothers with more versus fewer children at home, the increased potential for role overload or conflict may account in part for their lower levels of femininity. On the other hand, we did not find the expected inverse relation between marital conflict and femininity or masculinity for the sample as a whole; among baby boomers, however, less conflict was related to more masculinity, providing partial support for that hypothesis. The potential impact of changing gender relations in the marital role on perceived gender-linked traits is discussed more fully below (in Birth Cohort Effects).
The Influence of Work Roles on Gender-Linked Traits The finding that women working full time were higher in masculinity than nonworking women supports existing evidence suggesting that work experiences may play a role in shaping adult personality, especially in the direction of more agentic traits (e.g., Kohn & Schooler, 1978, 1982; Roberts, 1997). We also found that relative to nonworking women, full- and part-time working women were lower in femininity. Working mothers often bear the brunt of
4
Conflict
Support
3 14
Preboomers
2
Baby boomers
12 1
10 30
39
48
57
Age
Figure 6.
66
75
0 30
39
48
57
Age
66
75
Age changes in Marital Support and Marital Conflict by birth cohort.
954
KASEN, CHEN, SNEED, CRAWFORD, AND COHEN
child-care and household responsibilities, making them more vulnerable to role strain (Aneshensel, 1986); consequently, their expression of communal traits may be dampened. Marital support, however, modified the association in that full-time working mothers with high support endorsed more feminine-linked traits than full-time working or nonworking mothers with low support. Thus, in addition to increasing psychological well-being among women in multiple roles, emotional or instrumental support from husbands or other significant figures may counteract the negative effect of role overload or conflict on the expression of personality traits indicative of nurturance and warmth. As found by others (e.g., Roberts, 1997; Vandewater & Stewart, 1998), masculinity also increased in women working in highprestige occupations. The hypothesized association of occupational prestige and less femininity, however, was not observed; nonetheless, high occupational prestige weakened the positive relation of masculinity with femininity and thus, in effect, played an instrumental if indirect role in lowering femininity. For women in a high-powered professional or business position, the increase in agentic traits may come at the expense of a decrease in communal traits, perhaps because expression of the latter by working mothers, especially those in demanding career-track occupations, may heighten others’ perceptions of lessened commitment to the workplace (Crittenden, 2002).
aged owing to earlier gains in marital influence (at the expense of earlier conflict), which in turn may have fostered more agentic traits. In contrast, preboomers did not report a significant reduction in marital conflict (or a significant upswing in marital support) with age, yet there was a positive association between marital conflict and masculinity, suggesting that high levels of masculinity in wives from earlier cohorts may place strain on the marriage. Nonetheless, as shown here and in a previous study (Kasen et al., 2005), divorce rate was higher in baby boomers than in preboomers in this sample. Accordingly, an alternative explanation for this finding may be that baby boomers were more likely than preboomers to extricate themselves from high-conflict/lowsupport relationships; thus, baby boomer marriages that remained intact likely would be those characterized by relatively less conflict and more support than those that failed. Further, the lack of a cohort difference in divorce rate at age 53 after a significant difference at age 43 suggests that baby boomers may have entered into successful remarriages, also characterized by low conflict and high support. Nonetheless, the finding of a simultaneous decline in marital conflict and increase in marital support in baby boomers but not preboomers was based on exploratory analyses only; thus, it requires further replication.
Birth Cohort Effects
Compatibility With Theory and Evidence of Overall Personality Change
The expected higher level of femininity in baby boomers was only a trend; however, among women with very low masculinity, baby boomers reported significantly more feminine-linked traits than preboomers, thus supporting this hypothesis in part. Baby boomers may have responded in kind to the increased value placed on communal and nurturing traits by both men and women in more recent cohorts (Amato & Booth, 1995; Rogers & Amato, 2000), especially baby boomers who perceive themselves as less agentic at a time when such traits are increasingly endorsed in women. Albeit expected, no cohort disparity in masculinity was observed; associations of marital conflict with masculinity, however, did differ significantly by cohort: The inverse relation between marital conflict and masculinity hypothesized in the entire sample was observed in baby boomers only; moreover, among preboomers, marital conflict was positively related to masculinity. Further analyses of marital conflict and support across all assessment points revealed that although mean levels did not differ significantly between preboomers and baby boomers, age-related changes were significantly more favorable for baby boomers, namely, a decline in conflict with a parallel upswing in support. The literature on marital conflict suggests that wives are more likely to initiate and argue over controversial issues than husbands, who are more likely to avoid these issues (Christensen & Heavey, 1990; Heavey, Layne, & Christensen, 1993). Such patterns of behavior are thought to be a consequence of the power hierarchy that exists in marriage: Wives are more demanding of change owing to their less favorable positions, whereas husbands prefer to protect the status quo. However, recent cohorts of married couples have reported a more egalitarian distribution of influence, albeit a greater amount of conflict in their younger adult years (Greenberger & O’Neil, 1993; Rogers & Amato, 2000). Thus, baby boomers may have had fewer issues over which to argue as they
The findings here of normative increases in positive feminineand masculine-linked traits are compatible with the notion that both communal and agentic traits are basic to overall personality development in all individuals (Bakan, 1966; Spence & Helmreich, 1978, 1980; Spence et al., 1974, 1975) and with reports of normative change in broader personality dimensions in adulthood. In a recent meta-analysis (Roberts, Walton, & Viechtbauer, 2006) and review (Roberts et al., 2003) of longitudinal findings organized by the five-factor model of personality, increases in traits reflecting agreeableness and conscientiousness, both of which are related to femininity, and traits reflecting social dominance and emotional stability, both of which are related to masculinity, occurred through middle age. Similar trends based on crosssectional data were reported for those same dimensions (Srivastava, John, Gosling, & Potter, 2003) and for self-esteem (Robins, Trzesniewski, Tracy, Gosling, & Potter, 2002). Overall, both cross-sectional and longitudinal data support normative increases in personality traits confirming prevailing expectations that people become more socially competent and adapt more easily to changed circumstances with age (Helson & Wink, 1987; Roberts et al., 2003). In addition, the results here suggest that gender-linked traits may be influenced by the immediate social context of family and work and by broader social forces as assessed by birth cohort membership. Thus, these findings also support arguments and evidence that social role involvement affects adult personality development (e.g., Eagly et al., 2000; Havighurst, 1975; Helson & Picano, 1990; Helson & Roberts, 1992; Neugarten, 1968; Pals, 1999; Roberts, 1997; Robins, Caspi, & Moffit, 2002) and that exposure or timing of exposure to meaningful events or shifting social norms may account for generational differences (e.g., Baltes, 1987; Baltes & Schaie, 1973; Duncan & Agronick, 1995; Elder, 1979; Stewart & Healy, 1989; Twenge, 2001a, 2001b).
GENDER-LINKED TRAITS IN TWO BIRTH COHORTS
Summary and Conclusions We found normative increases with age in both feminine- and masculine-related traits that were, for the most part, positive traits; moreover, each predicted an increase in the other. Those findings support reports of personality development in women during midlife that reflect increases in generativity and warmth on the one hand and competence and independence on the other, and of an age-related rise in androgyny. Work role experiences influenced gender-linked traits in the expected directions, namely, more masculinity but less femininity, especially among those in full-time work and high-prestige occupations, thus corroborating the findings of others in a representative sample of mothers. The longitudinal impact of marital experiences on gender-linked traits found here suggests the need for more research in this area. Partner support had a direct effect on more femininity in all mothers and offset less femininity in those working full time. Lack of partner support also may have contributed to the lower level of femininity found here in unmarried mothers relative to married mothers, suggesting that the increased burdens inherent in single parenting may, paradoxically, reduce communal traits. We also found that birth cohort membership significantly influenced the impact of marital conflict on masculinity: An inverse association held for baby boomers, whereas a positive one held for preboomers. Further analyses revealed that although average levels of marital conflict and marital support did not differ across cohorts, agerelated changes did, namely, a simultaneous decline in conflict and upswing in support in baby boomers only, suggesting that changing role norms and expectations may be implicated in differential personality change among women. Overall, the findings here emphasize the longitudinal influence of social role experiences on personality trait change and also suggest that the repercussions of historical changes in gender role norms that have come about for both women and men may continue to affect personality trait change for some time to come.
Limitations The following limitations should be considered when interpreting these findings. First, we had no information regarding genderlinked personality traits of spouses of this study sample. The systems perspective of marriage indicates that complementary personality styles of husbands and wives may be more important to marital quality and adjustment than individual personality styles (Robins, Caspi, & Moffit, 2000). Second, these data were obtained from single, not multiple, informants; thus, it was actually perceptions of marital quality and self-attributions of gender-linked traits that were examined. Third, the women in the sample all were mothers; thus, findings may not generalize to women who are childless. Fourth, both number of offspring and offspring age may influence perceived gender-linked traits in mothers (Kessler & McRae, 1982); however, although we controlled for number of children at home, information about the youngest child at home was not available at all assessment points and so was not controlled. Nonetheless, this sample of women was followed longitudinally over two decades and is representative of the region from which sampling took place. In addition, the study emphasized normative changes in gender-linked traits with age, and findings highlight the influence of social role experiences and shifting social norms on personality change through the adult years.
955 References
Abele, A. E. (2003). The dynamics of masculine-agentic and femininecommunal traits: Findings from a prospective study. Journal of Personality and Social Psychology, 85, 768 –776. Agronick, G. S., & Duncan, L. E. (1998). Personality and social change: Individual differences, life path, and importance attributed to the women’s movement. Journal of Personality and Social Psychology, 74, 1545–1555. Amato, P. R., & Booth, A. (1995). Changes in gender role attitudes and perceived marital quality. American Sociological Review, 60, 58 – 66. Aneshensel, C. S. (1986). Marital and employment role-strain, social support, and depression among adult women. In S. E. Hobfoll (Ed.), Stress, social support, and women (pp. 99 –114). Washington, DC: Hemisphere. Aube, J., Fleury, J., & Smetana, J. (2000). Changes in women’s roles: Impact on and social policy implications for the mental health of women and children. Development and Psychopathology, 12, 633– 656. Auster, C. J., & Ohm, S. C. (2000). Masculinity and femininity in contemporary American society: A reevaluation using the Bem Sex-Role Inventory. Sex Roles, 43, 499 –528. Bakan, D. (1966). The duality of human existence. Chicago: Rand McNally. Balog, P., Janszky, I., Leineweber, C., Blom, M., Wamala, S. P., & Orth-Gome´r, K. (2003). Depressive symptoms in relation to marital and work stress in women with and without coronary heart disease. The Stockholm Female Coronary Risk Study. Journal of Psychosomatic Research, 54, 113–119. Baltes, P. B. (1987). Theoretical propositions of life-span developmental psychology: On the dynamics between life and decline. Developmental Psychology, 23, 611– 626. Baltes, P. B., & Schaie, K. W. (1973). Life-span developmental psychology: Personality and socialization. New York: Academic Press. Barnett, R. C., & Baruch, G. K. (1985). Women’s involvement in multiple roles and psychological distress. Journal of Personality and Social Psychology, 49, 135–145. Barnett, R. C., Brennan, R. T., & Marshall, N. L. (1994). Gender and the relationship between parent role quality and psychological distress: A study of men and women in dual-earner couples. Journal of Family Issues, 15, 229 –252. Barnett, R. C., & Hyde, J. S. (2001). Women, men, work, and family: An expansionist theory. American Psychologist, 56, 781–796. Barnett, R. C., Marshall, N. L., & Singer, J. D. (1992). Job experiences over time, multiple roles, and women’s mental health: A longitudinal study. Journal of Personality and Social Psychology, 62, 634 – 644. Barrett, A. E., & White, H. R. (2002). Trajectories of gender role orientations in adolescence and early adulthood: A prospective study of the mental health effects of masculinity and femininity. Journal of Health and Social Behavior, 43, 451– 468. Baruch, G. K., & Barnett, R. (1986). Role quality, multiple role involvement, and psychological well-being in midlife women. Journal of Personality and Social Psychology, 51, 578 –585. Baucom, D. H., & Aiken, P. A. (1984). Sex role identity, marital satisfaction, and response to behavioral marital therapy. Journal of Consulting and Clinical Psychology, 52, 438 – 444. Beach, S. R. H., & Fincham, F. D. (1998). Marital processes and depression. In L. L’Abate (Ed.), Family psychopathology: The relational roots of dysfunctional behavior (pp. 256 –279). New York: Guilford Press. Bem, S. L. (1974). The measurement of psychological androgyny. Journal of Consulting and Clinical Psychology, 42, 155–162. Bem, S. L. (1978). Bem Sex Role Inventory professional manual. Palo Alto, CA: Consulting Psychologists Press. Bond, J. T., Galinsky, E., & Swanberg, J. E. (1998). The 1997 National Study of the Changing Workforce. New York: Families and Work Institute.
956
KASEN, CHEN, SNEED, CRAWFORD, AND COHEN
Bridges, J. S., & Etaugh, C. (1995). College students’ perceptions of mothers: Effects of maternal employment– childrearing pattern and motive for employment. Sex Roles, 32, 735–751. Bridges, J. S., Etaugh, C., & Barnes-Farrell, J. (2002). Trait judgments of stay-at-home and employed parents: A function of social roles and/or shifting standards? Psychology of Women Quarterly, 26, 140 –150. Brook, J. S., Brook, S. W., Gordon, A. S., Whiteman, M., & Cohen, P. (1990). The psychosocial etiology of adolescent drug use: A family interactional approach. Genetic, Social, and General Psychology Monographs, 116, 111–267. Brook, J. S., Zheng, L., Whiteman, M., & Brook, D. W. (2001). Aggression in toddlers: Associations with parenting and marital relations. Journal of Genetic Psychology, 162, 228 –241. Buechler, S. M. (1990). Women’s movements in the United States: Woman suffrage, equal rights and beyond. New Brunswick, NJ: Rutgers University Press. Caspi, A. (1987). Personality in the life course. Journal of Personality and Social Psychology, 53, 1203–1213. Chafe, W. H. (1991). The paradox of change: American women in the twentieth century. New York: Oxford University Press. Christensen, A., & Heavey, C. L. (1990). Gender and social structure in the demand/withdraw pattern of marital conflict. Journal of Personality and Social Psychology, 59, 73– 81. Clausen, J. A., & Gilens, M. (1990). Personality and labor force participation across the life course: A longitudinal study of women’s careers. Sociological Forum, 5, 595– 618. Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences (3rd ed.). Mahwah, NJ: Erlbaum. Cohen, P., & Cohen, J. (1996). Life values and adolescent mental health. Mahwah, NJ: Erlbaum. Crawford, T. N., Cohen, P., Midlarsky, E., & Brook, J. S. (2001). Internalizing symptoms in adolescents: Gender differences in vulnerability to parental distress and discord. Journal of Research on Adolescence, 11, 95–118. Crittenden, A. (2002). The price of motherhood. New York: Holt. Cunningham, J. D., & Antill, J. K. (1984). Changes in femininity and masculinity across the family life cycle: A reexamination. Developmental Psychology, 20, 1135–1141. Duncan, L. E., & Agronick, G. S. (1995). The intersection of life stage and social events: Personality and life outcomes. Journal of Personality and Social Psychology, 69, 558 –568. Eagly, A. H., Wood, W., & Diekman, A. (2000). Social role theory of sex differences and similarities: A current appraisal. In T. Eckes & H. M. Trautner (Eds.), The developmental social psychology of gender (pp. 123–174). Mahwah, NJ: Erlbaum. Elder, G. H., Jr. (1979). Historical change in life patterns and personality. In P. Baltes & O. Brim (Eds.), Life-span development and behavior (Vol. 2, pp. 117–159). New York: Academic Press. Feldman, S., & Aschenbrenner, B. (1983). Impact of parenthood on various aspects of masculinity and femininity: A short-term longitudinal study. Developmental Psychology, 19, 278 –289. Fincham, F. D., Beach, S. R. H., Harold, G. T., & Osborne, L. N. (1997). Marital satisfaction and depression: Different causal relationships for men and women? Psychological Science, 8, 351–357. Franz, C. E. (1997). Stability and change in the transition to midlife: A longitudinal study of midlife adults. In M. E. Lachman & J. B. James (Eds.), Multiple paths of midlife development (pp. 45– 66). Chicago: University of Chicago Press. Fuegen, K., Biernat, M., Haines, E., & Deaux, K. (2004). Mothers and fathers in the workplace: How gender and parental status influence judgments of job-related competence. Journal of Social Issues, 60, 737–754. Galambos, N. L., Almeida, D. M., & Petersen, A. C. (1990). Masculinity,
femininity, and sex role attitudes in early adolescence: Exploring gender intensification. Child Development, 61, 1905–1914. Glenn, N. D., & McLanahan, S. (1982). Children and marital happiness: A further specification of the relationship. Journal of Marriage and the Family, 44, 63–72. Golding, J. M. (1989). Role occupancy and role-specific stress and social support as predictors of depression. Basic and Applied Social Psychology, 10, 173–195. Gough, H. G. (1987). Manual for the California Psychological Inventory. Palo Alto, CA: Consulting Psychologists Press. Gough, H. G., & Bradley, P. (1996). CPI manual (3rd ed.). Palo Alto, CA: Consulting Psychologists Press. Gove, W. R., & Geerken, M. R. (1977). The effect of children and employment on the mental health of married men and women. Social Forces, 56, 66 –76. Gove, W. R., Style, C. B., & Hughes, M. (1990). The effect of marriage on the well-being of adults: A theoretical analysis. Journal of Family Issues, 11, 4 –35. Greenberger, E., & O’Neil, R. (1993). Spouse, parent, worker: Role commitments and role-related experiences in the construction of adults’ well-being. Developmental Psychology, 29, 181–197. Guelzow, M. G., Bird, G. W., & Koball, E. H. (1991). An exploratory path analysis of the stress process for dual-career men and women. Journal of Marriage and the Family, 53, 151–164. Haan, N., Millsap, R., & Hartka, E. (1986). As time goes by: Change and stability in personality over fifty years. Psychology and Aging, 1, 220 – 232. Hammen, C. (2003). Social structure and women’s risk for recurrent depression. Archives of Women’s Mental Health, 6, 9 –13. Havighurst, R. J. (1975). The future aged: The use of time and money. Gerontologist, 15, 10 –15. Heavey, C. L., Layne, C., & Christensen, A. (1993). Gender and conflict structure in marital interaction: A replication and extension. Journal of Consulting and Clinical Psychology, 61, 16 –27. Helson, R., Jones, C., & Kwan, V. S. Y. (2002). Personality change over 40 years of adulthood: Hierarchical linear modeling analyses of two longitudinal samples. Journal of Personality and Social Psychology, 83, 752–766. Helson, R., Kwan, V. S. Y., John, O., & Jones, C. (2002). The growing evidence for personality change in adulthood: Findings from research with personality inventories. Journal of Research in Personality, 36, 287–306. Helson, R., & Moane, G. (1987). Personality change in women from college to midlife. Journal of Personality and Social Psychology, 53, 176 –186. Helson, R., & Picano, J. (1990). Is the traditional role bad for women? Journal of Personality and Social Psychology, 59, 311–320. Helson, R., & Roberts, B. (1992). The personality of young adult couples and wives’ work patterns. Journal of Personality, 60, 575–597. Helson, R., & Stewart, A. J. (1994). Personality change in adulthood. In T. Heatherton & J. Weinberger (Eds.), Can personality change? (pp. 201– 225). Washington, DC: American Psychological Association. Helson, R., & Wink, P. (1987). Two conceptions of maturity examined in the findings of a longitudinal study. Journal of Personality and Social Psychology, 53, 531–541. Helson, R., & Wink, P. (1992). Personality change in women from the early 40s to early 50s. Psychology and Aging, 7, 46 –55. Hollingshead, A. B., & Redlich, F. C. (1958). Social class and mental illness: A community study. New York: Wiley. Kalin, R., & Lloyd, C. A. (1985). Sex role identity, sex role ideology, and marital adjustment. International Journal of Women’s Studies, 8, 32–39. Kasen, S., Cohen, P., Berenson, K., Chen, H., & Dufur, R. (2005). Dual work and family roles and depressive symptoms in two birth cohorts of women. Social Psychiatry and Psychiatric Epidemiology, 40, 300 –307.
GENDER-LINKED TRAITS IN TWO BIRTH COHORTS Kasen, S., Cohen, P., Chen, H., & Castille, D. (2003). Depression in adult women: Age changes and cohort effects. American Journal of Public Health, 93, 2061–2066. Kendler, K. S., Kessler, R. C., Walters, E. E., MacLean, C., Neale, M. C., Health, A. C., & Eaves, L. J. (1995). Stressful life events, genetic liability, and onset of an episode of Major Depression in women. American Journal of Psychiatry, 152, 833– 842. Kessler, R. C., & McRae, J. A. (1982). The effect of wives’ employment on the mental health of married men and women. American Sociological Review, 47, 216 –227. Kohn, M. L., & Schooler, C. (1978). The reciprocal effects of the substantive complexity of work and intellectual flexibility: A longitudinal assessment. American Journal of Sociology, 84, 24 –52. Kohn, M. L., & Schooler, C. (1982). Job conditions and personality: A longitudinal assessment of their reciprocal effects. American Journal of Sociology, 87, 1257–1286. Lippa, R. (1991). Some psychometric characteristics of gender diagnosticity measures: Reliability, validity, consistency across domains, and relationship to the Big Five. Journal of Personality and Social Psychology, 61, 1000 –1011. Littell, R. C., Miliken, G. A., Stroup, W. W., & Wolfinger, R. D. (1996). SAS system for fixed models. Cary, NC: SAS Institute. Locke, H. J., & Wallace, K. M. (1959). Short marital adjustment and prediction tests: Their reliability and validity. Marriage and Family Living, 21, 251–255. Mannheim, K. (1972). The problem of generations. In P. G. Altbach & R. S. Laufer (Eds.), The new pilgrims: Youth protest in transition (pp. 101–138). New York: David McKay. Martire, L. M., Stephens, M. A. P., & Townsend, A. L. (1998). Emotional support and well-being in midlife women: Role-specific mastery as a mediational mechanism. Psychology and Aging, 13, 396 – 404. McCrae, R. R., & Costa, P. T. (1987). Validation of the five-factor model of personality across instruments and observers. Journal of Personality and Social Psychology, 52, 81–90. McHale, S. M., & Huston, T. L. (1984). Men and women as parents: Sex role orientations, employment, and parental roles with infants. Child Development, 55, 1349 –1361. Miller, B. C. (1975). Child density, marital satisfaction, and conventionalization: A research note. Journal of Marriage and the Family, 37, 245–247. Morfei, M. Z., Hooker, K., Carpenter, J., Mix, C., & Blakeley, E. (2004). Agentic and communal generative behavior in four areas of adult life: Implications for psychological well-being. Journal of Adult Development, 11, 55–58. Mori, M., Nakashima, Y., & Kurita, H. (2002). Sex-role orientation, marital status, and mental health in working women. Archives of Women’s Mental Health, 5, 161–176. Neugarten, B. L. (1968). The awareness of middle age. In B. L. Neugarten (Ed.), Middle age and aging (pp. 93–98). Chicago: University of Chicago Press. Neugarten, B. L. (1972). Personality and the aging process. Gerontologist, 12, 9 –15. Noor, N. M. (1995). Work and family roles in relation to women’s well-being: A longitudinal study. British Journal of Social Psychology, 34, 87–106. Noor, N. M. (1997). Work and family roles in relation to women’s well-being: The role of negative affectivity. Personality and Individual Differences, 23, 487– 499. O’Leary, K. D., Christian, J. L., & Mendell, M. R. (1994). A closer look at the link between marital discord and depressive symptomatology. Journal of Social and Clinical Psychology, 13, 33– 41. Oswald, P. (2004). An examination of the current usefulness of the Bem Sex-Role Inventory. Psychological Reports, 94, 1331–1336. Pals, J. L. (1999). Identity consolidation in early adulthood: Relations with
957
ego-resiliency, the context of marriage, and personality change. Journal of Personality, 67, 295–329. Parker, R. A., & Aldwin, C. M. (1997). Do aspects of gender identity change from early to middle adulthood? Disentangling age, cohort, and period effects. In M. E. Lachman & J. B. James (Eds.), Multiple paths of midlife development (pp. 67–107). Chicago: University of Chicago Press. Pedhazur, E. J., & Tetenbaum, T. J. (1979). Bem Sex-Role Inventory: A theoretical and methodological critique. Journal of Personality and Social Psychology, 37, 996 –1016. Peterson, C. D., Baucom, D. H., Elliott, M. J., & Farr, P. A. (1989). The relationship between sex role identity and marital adjustment. Sex Roles, 21, 775–787. Rankin-Esquer, L. A., Burnett, C. K., Baucom, D. H., & Epstein, N. (1997). Autonomy and relatedness in marital functioning. Journal of Marital and Family Therapy, 23, 175–190. Roberts, B. W. (1997). Plaster or plasticity: Are adult work experiences associated with personality change in women? Journal of Personality, 65, 205–232. Roberts, B. W., Helson, R., & Klohnen, E. C. (2002). Personality development and growth in women across 30 years: Three perspectives. Journal of Personality, 70, 79 –102. Roberts, B. W., Robins, R. W., Trzesniewski, K. H., & Caspi, A. (2003). Personality trait development in adulthood. In J. Mortimer & M. Shanahan (Eds.), Handbook of the life course (pp. 579 –598). New York: Kluwer Academic. Roberts, B. W., Walton, K. E., & Viechtbauer, W. (2006). Patterns of mean level change in personality traits across the life course: A meta-analysis of longitudinal studies. Psychological Bulletin, 132, 1–25. Robins, R. W., Caspi, A., & Moffit, T. (2000). Two personalities, one relationship: Both partners’ personality traits shape the quality of their relationship. Journal of Personality and Social Psychology, 79, 252– 259. Robins, R. W., Caspi, A., & Moffit, T. (2002). It’s not just who you’re with, but who you are: Personality and relationship experiences across multiple relationships. Journal of Personality, 70, 925–964. Robins, R. W., Trzesniewski, K. H., Tracy, J. L., Gosling, S. D., & Potter, J. (2002). Global self-esteem across the life span. Psychology and Aging, 17, 423– 434. Rogers, S. J., & Amato, P. R. (2000). Have changes in gender relations affected marital quality? Social Forces, 79, 731–753. Russo, N. F., & Zierk, K. L. (1992). Abortion, childbearing, and women’s well-being. Professional Psychology: Research and Practice, 23, 269 – 280. Sameroff, A. (1998). Environmental risk factors in infancy. Pediatrics, 102, 1287–1292. Singer, J. D. (1998). Using SAS PROC MIXED to fit multilevel models, hierarchical models, and individual growth models. Journal of Educational and Behavioral Statistics, 23, 323–355. Spence, J. T., & Helmreich, R. L. (1978). Masculinity and femininity: Their psychological dimensions, correlates, and antecedents. Austin: University of Texas Press. Spence, J. T., & Helmreich, R. L. (1980). Masculine instrumentality and feminine expressiveness: Their relationships with sex role attitudes and behaviors. Psychology of Women Quarterly, 5, 147–163. Spence, J. T., Helmreich, R., & Stapp, J. (1974). The Personal Attributes Questionnaire: A measure of sex-role stereotypes and masculinity– femininity. Catalog of Selected Documents in Psychology, 4, 43– 44. Spence, J. T., Helmreich, R., & Stapp, J. (1975). Ratings of self and peers on sex-role attributes and their relations to self-esteem and conceptions of masculinity and femininity. Journal of Personality and Social Psychology, 32, 29 –39. Srivastava, S., John, O. P., Gosling, S. D., & Potter, J. (2003). Development of personality in early and middle adulthood: Set like plaster or
958
KASEN, CHEN, SNEED, CRAWFORD, AND COHEN
persistent change? Journal of Personality and Social Psychology, 84, 1041–1053. Stewart, A. J., & Healy, J. M. (1989). Linking individual development and social changes. American Psychologist, 44, 30 – 42. Stewart, A. J., Ostrove, J. M., & Helson, R. (2001). Middle aging in women: Patterns of personality change from the 30s to the 50s. Journal of Adult Development, 8, 23–37. Stewart, A. J., & Vandewater, E. A. (1993). The Radcliffe class of 1964: Career and family social clock projects in a transitional project. In K. D. Hulbert & D. Tickton Schuster (Eds.), Women’s lives through time: Educated American women of the twentieth century (pp. 235–258). San Francisco: Jossey-Bass. Stumpf, S. A. (1978). A note on handling missing data. Journal of Management, 4, 65–73. Thoits, P. A. (1992). Identity structures and psychological well-being: Gender and marital status comparisons. Social Psychology Quarterly, 55, 236 –256. Twenge, J. (1997). Changes in masculine and feminine traits over time: A meta-analysis. Sex Roles, 36, 305–325.
Twenge, J. (2001a). Birth cohort changes in extraversion: A cross-temporal meta-analysis, 1966 –1993. Personality and Individual Differences, 30, 735–748. Twenge, J. (2001b). Changes in women’s assertiveness in response to status and roles: A cross-temporal meta-analysis, 1931–1993. Journal of Personality and Social Psychology, 81, 133–145. Vandewater, E. A., & Stewart, A. J. (1998). Making commitments, creating lives: Linking women’s roles and personality at midlife. Psychology of Women Quarterly, 22, 717–738. Whisman, A., & Bruce, M. L. (1999). Marital dissatisfaction and the incidence of a major depressive episode in a community sample. Journal of Abnormal Psychology, 108, 674 – 678.
Received June 7, 2005 Revision received May 15, 2006 Accepted June 1, 2006 䡲
958
KASEN, CHEN, SNEED, CRAWFORD, AND COHEN
persistent change? Journal of Personality and Social Psychology, 84, 1041–1053. Stewart, A. J., & Healy, J. M. (1989). Linking individual development and social changes. American Psychologist, 44, 30 – 42. Stewart, A. J., Ostrove, J. M., & Helson, R. (2001). Middle aging in women: Patterns of personality change from the 30s to the 50s. Journal of Adult Development, 8, 23–37. Stewart, A. J., & Vandewater, E. A. (1993). The Radcliffe class of 1964: Career and family social clock projects in a transitional project. In K. D. Hulbert & D. Tickton Schuster (Eds.), Women’s lives through time: Educated American women of the twentieth century (pp. 235–258). San Francisco: Jossey-Bass. Stumpf, S. A. (1978). A note on handling missing data. Journal of Management, 4, 65–73. Thoits, P. A. (1992). Identity structures and psychological well-being: Gender and marital status comparisons. Social Psychology Quarterly, 55, 236 –256. Twenge, J. (1997). Changes in masculine and feminine traits over time: A meta-analysis. Sex Roles, 36, 305–325.
Twenge, J. (2001a). Birth cohort changes in extraversion: A cross-temporal meta-analysis, 1966 –1993. Personality and Individual Differences, 30, 735–748. Twenge, J. (2001b). Changes in women’s assertiveness in response to status and roles: A cross-temporal meta-analysis, 1931–1993. Journal of Personality and Social Psychology, 81, 133–145. Vandewater, E. A., & Stewart, A. J. (1998). Making commitments, creating lives: Linking women’s roles and personality at midlife. Psychology of Women Quarterly, 22, 717–738. Whisman, A., & Bruce, M. L. (1999). Marital dissatisfaction and the incidence of a major depressive episode in a community sample. Journal of Abnormal Psychology, 108, 674 – 678.
Received June 7, 2005 Revision received May 15, 2006 Accepted June 1, 2006 䡲
Journal of Personality and Social Psychology 2006, Vol. 91, No. 5, 959 –974
Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.5.959
Personality Development in Emerging Adulthood: Integrating Evidence From Self-Ratings and Spouse Ratings David Watson and John Humrichouse University of Iowa The authors examined self-ratings and spouse ratings in a young adult newlywed sample across a 2-year interval. Rank-order stability correlations were consistently high and did not differ across the 2 types of ratings. As expected, self-ratings showed significant increases in conscientiousness and agreeableness— and declines in neuroticism/negative affectivity— over time. Spouse ratings yielded a very different pattern, however, showing significant decreases in conscientiousness, agreeableness, extraversion, and openness across the study interval. Spouse ratings also showed evidence of a “honeymoon effect,” such that they tended to be more positive than self-ratings at Time 1. This effect had dissipated by the 2nd assessment; in fact, the spouse ratings tended to be more negative at Time 2. Analyses of individual-level change revealed little convergence between self- and spouse-rated change, using both raw change scores and reliable change index scores. Finally, correlational and regression analyses indicated that changes in spouse ratings were significantly associated with changes in marital satisfaction; in contrast, changes in self-ratings essentially were unrelated to marital satisfaction. These results highlight the value of collecting multimethod data in studies of adult personality development. Keywords: trait stability, mean-level change, personality development, marital satisfaction
Classic models of trait psychology promoted a relatively static view of personality. These traditional models acknowledged that change was prevalent—indeed, even the norm— during childhood and adolescence. Once individuals reached adulthood, however, traits were viewed as essentially being set like plaster and highly resistant to change (see Costa & McCrae, 1994; McCrae & Costa, 1990; Srivastava, John, Gosling, & Potter, 2003). As evidence has accumulated, however, it has become clear that a simple “plaster” model fails to capture the complexities of personality development across the life span. In fact, recent findings have established that personality traits are not static constructs but rather show meaningful change well into middle age (Caspi, Roberts, & Shiner, 2005; Clark & Watson, 1999; Fraley & Roberts, 2004; Roberts, Robins, Trzesniewski, & Caspi, 2003; Roberts, Walton, & Viechtbauer, 2006).
rank-order stability, that is, the extent to which individuals maintain their relative position on the trait continuum over time. In other words, rank-order stability data establish the extent to which individual differences persist over time (e.g., whether a highly conscientious job applicant is likely to remain a highly conscientious employee several years after being hired). The data on this issue are highly consistent and yield several clear conclusions. First, stability correlations for personality traits are moderate to strong in magnitude, even when assessed in childhood and adolescence (Caspi et al., 2005; Roberts & DelVecchio, 2000). Second, stability correlations decline in magnitude as the elapsed time interval increases (Caspi et al., 2005; Roberts & DelVecchio, 2000; Watson, 2004). This consistent pattern has helped to establish the existence of true change in personality, given that change is more and more likely to occur with increasing retest intervals (see Watson, 2004). It must be emphasized, however, that stability correlations never approach .00 and remain at least moderate in magnitude, even across intervals of several decades (Fraley & Roberts, 2004; see also Clark & Watson, 1999; Costa & McCrae, 1992). Third, stability correlations for personality increase systematically with age (Caspi et al., 2005; Clark & Watson, 1999; Roberts & DelVecchio, 2000). Older models of trait development assumed that most personality change occurred prior to the age of 30, after which stability correlations should be uniformly high (for discussions, see McCrae & Costa, 1990; Roberts & DelVecchio, 2000; Srivastava et al., 2003). However, the meta-analytic findings of Roberts and DelVecchio (2000) revealed that stability coefficients for personality continue to increase well into middle age. Fourth, stability estimates do not vary significantly by gender (Costa & McCrae, 1988; Roberts & DelVecchio, 2000; Schuerger, Zarrella, & Hotz, 1989). Indeed, in their meta-analytic review, Roberts and DelVecchio (2000) obtained identical population es-
Stability and Change in Adult Personality Rank-Order Stability In reviewing the prior literature on this topic, we consider three basic types of evidence. The first type of evidence concerns David Watson and John Humrichouse, Department of Psychology, University of Iowa. This research was supported by National Institute of Mental Health (NIMH) Grant 1-R01-MH61804-01 to Diane Berry, by NIMH Grant 1-R01-MH068472-1 to David Watson, and by NIMH Grant 1-R03MH068395-01 to Eva C. Klohnen. We thank Alex Casillas, Elizabeth Gray, Malik Haig, Daniel Heller, Eva C. Klohnen, Shanhong Luo, Ericka Nus Simms, and all of the Iowa Marital Assessment Project staff for their help in the preparation of this article. Correspondence concerning this article should be addressed to David Watson, Department of Psychology, E11 Seashore Hall, University of Iowa, Iowa City, IA 52242-1407. E-mail:
[email protected] 959
960
WATSON AND HUMRICHOUSE
timates of overall trait stability in men and women (see their Table 4). Finally, some trait measures consistently show greater stability than others (Watson, 2004). In this regard, it is important to emphasize that the traits comprising the influential five-factor model of personality—neuroticism, extraversion, openness, agreeableness, and conscientiousness—all show very similar levels of stability (see Roberts & DelVecchio, 2000, Table 5). In contrast, however, measures of trait affectivity consistently yield lower stability correlations (Watson, 2004). For instance, Vaidya, Gray, Haig, and Watson (2002) followed a large sample of young adults across an average retest interval of slightly more than 2.5 years. They obtained stability correlations ranging from .59 to .72 (mean r ⫽ .64) on the Big Five traits, which were assessed using the Big Five Inventory (BFI; John & Srivastava, 1999); in contrast, trait affectivity scales from the Positive and Negative Affect Schedule (PANAS; Watson, Clark, & Tellegen, 1988) yielded significantly lower stability correlations of .51 (Positive Affect scale) and .49 (Negative Affect scale). This pattern of differential stability was replicated in another sample that completed these same measures twice across a 2-month interval (Watson, 2004). It is particularly striking that BFI Neuroticism showed significantly greater stability than the PANAS Negative Affect scale in both samples, despite the fact that the two scales assess very similar content and were strongly interrelated. We return to this issue of differential stability later.
Mean-Level Change The second line of evidence concerns mean-level change, that is, whether or not the average levels of a trait change systematically with age. These data clarify how personality evolves over time as a function of changing biological and/or environmental conditions. Once again, recent reviews of this evidence have yielded several clear conclusions (see Caspi et al., 2005; Clark & Watson, 1999; Roberts, Robins, et al., 2003; Roberts et al., 2006). The evidence is especially consistent for the three traits—neuroticism, agreeableness, and conscientiousness—that define the “Alpha” superfactor identified by Digman (1997; see also Markon, Krueger, & Watson, 2005). First, trait levels of neuroticism and negative affectivity show a significant decline with age. The bulk of this decline appears to occur in adolescence and early adulthood (Caspi et al., 2005; Roberts, Robins, et al., 2003); nevertheless, decreases continue to be seen later in life (Clark & Watson, 1999; Roberts et al., 2006; Srivastava et al., 2003). Second, levels of both agreeableness and conscientiousness increase across young adulthood and middle age (Roberts, Robins et al., 2003; Roberts et al., 2006). In contrast, the two remaining Big Five traits show more complex patterns. The data for extraversion and positive emotionality have been particularly inconsistent (Clark & Watson, 1999; Vaidya et al., 2002). However, Roberts, Robins, et al. (2003) found that the evidence was more consistent at the specific trait level: Measures of dominance tended to increase from adolescence through early middle age, whereas levels of sociability increased during adolescence but then declined starting in young adulthood (see also Roberts et al., 2006). Finally, openness to experience tends to show a curvilinear pattern, exhibiting increases in adolescence and young adulthood (particularly for those individuals who
remain in school), but then declining later in life (see Caspi et al., 2005; Roberts et al., 2006). Taken together, these data demonstrate that mean-level changes are both meaningful and highly systematic across the life span (Caspi et al., 2005; Roberts, Caspi, & Moffitt, 2001). Indeed, Caspi et al. (2005) concluded that they reflect the influence of a maturity principle, arguing that these mean-level shifts “point to increasing psychological maturity over development, from adolescence to middle age” (p. 468). Finally, it is clear that the pace of change is not constant across the life span. Interestingly, the data indicate that much of this change occurs in young adulthood, a period roughly corresponding to the 20s (Arnett, 2000; Roberts, Caspi, & Moffitt, 2003). This period has been characterized as demographically dense, given that it typically involves more changes in identity and major life roles (e.g., the onset of marriage and parenthood; the initiation of one’s career) than any other period in life (Arnett, 2000; Caspi et al., 2005). In fact, the level of change during this period even exceeds that seen during adolescence (Caspi et al., 2005).
Individual-Level Change The third line of evidence is individual-level change. Several authors recently have emphasized the importance of examining change at the individual level (see Roberts et al., 2001; Roberts, Robins, et al., 2003; Robins, Fraley, Roberts, & Trzesniewski, 2001; Vaidya et al., 2002); these analyses help to clarify the specific processes that are associated with personality development across the life span. Roberts, Caspi, and Moffitt (2003), for example, demonstrated that changes in personality at the individual level are meaningful and can be systematically linked to ongoing life experiences. More specifically, they showed that several aspects of work experience—including occupational attainment, work satisfaction, work involvement, and financial security—were significantly associated with individual-level changes in personality between the ages of 18 and 26 years.
Beyond Self-Report: Evidence From Other Methods Although the evidence we have considered thus far is impressive, it is limited in a number of ways (see Caspi et al., 2005). A particularly important limitation is that most of the data are based on self-ratings. In this regard, several recent authors have bemoaned the overreliance on self-report in this area and have emphasized the importance of examining stability versus change across multiple methods (e.g., McCrae et al., 2004; McCrae, Terraciano, & Members of the Personality Profiles of Cultures Project, 2005; Roberts, Caspi, & Moffitt, 2003; Roberts, O’Donnell, & Robins, 2004). Before reaching any fundamental conclusions about the nature of personality development, it obviously is important to examine other types of evidence. What do non-self-report data show? First, the available evidence indicates that rank-order stability coefficients essentially are invariant across methods (Caspi et al., 2005; Costa & McCrae, 1988; Roberts & DelVecchio, 2000). Most notably, in their meta-analytic review of the literature, Roberts and DelVecchio (2000) obtained virtually identical population estimates of overall trait stability across self-report ( ⫽ .52) and observer-rated ( ⫽ .48) data (see their Table 4).
SELF-RATINGS VERSUS SPOUSE RATINGS OF STABILITY
The data regarding mean-level change are both less plentiful and less consistent. Generally speaking, analyses based on observer ratings tend to show patterns that parallel those seen with selfreport—that is, age-related declines in neuroticism, extraversion, and openness, and increases in agreeableness and conscientiousness— but that are weaker in magnitude (McCrae et al., 2004, 2005). McCrae et al. (2004), for instance, examined age-related changes in peer ratings collected in Russia and the Czech Republic. They concluded that the observer-based effects were weaker and less consistent but that “whenever significant age differences in observer-rated personality traits appear, they follow the same direction as self-reports” (p. 155). The bulk of the available observer-rating data comes from cross-sectional designs in which personality scores are correlated with age (e.g., McCrae et al., 2004, 2005). Accordingly, we currently know very little about individual-level change using methods other than self-report. To date, the most comprehensive analysis of this issue was reported by McCrae (1993, Study 3), who examined longitudinal changes in the Big Five across a 6- to 7-year retest interval. McCrae began by establishing that personality changes showed some evidence of within-method consistency: For instance, individuals who indicated that they decreased on one facet of neuroticism (e.g., anxiety) also reported parallel changes on other facets within this domain (e.g., depression, hostility). Thus, these initial results suggest that these self-rated changes are systematic and do not simply reflect random measurement error. Subsequent analyses, however, revealed that these self-rated changes failed to show substantial convergence with spouse and peer ratings obtained across the same time interval. For instance, McCrae (1993) computed raw change scores for the self-ratings and spouse ratings on each trait. These analyses revealed a significant association between self- versus spouse-rated change on neuroticism, but the correlation was only .16; furthermore, the corresponding coefficients for extraversion and openness were even weaker and nonsignificant. It should be noted, however, that these results are based on relatively small sample sizes (the largest sample size in these analyses is only 135). Consequently, these results need to be replicated in a larger sample before any firm conclusions can be drawn.
The Current Study The current study adds to this growing literature by examining personality development in a large sample (N ⫽ 460) of young adults across a time span of approximately 2 years. We included measures of the Big Five and trait affectivity at both assessments, which allowed us to examine the three critical issues of (a) rankorder stability, (b) mean-level change, and (c) individual-level change. We highlight two unusual features of our study that represent significant extensions of the prior literature. First, most previous longitudinal studies of “emerging adulthood” have focused on the college and immediate postcollege years—that is, the period between the ages of 18 and 26 years (e.g., Roberts et al., 2001; Roberts, Caspi, & Moffitt, 2003; Robins et al., 2001; Vaidya et al., 2002). In contrast, our sample consisted of newlywed couples who were somewhat older than this. Specifically, the mean age of our sample at the Time 1 assessment was approximately 28 years (see
961
Watson et al., 2004, Table 1). Thus, we are able to examine issues related to personality stability versus change in a sample that is beginning to undergo the transition from emerging adulthood to full adulthood. Second, as discussed earlier, many recent authors have emphasized the importance of moving beyond self-report measures in studying issues related to personality development across the life span (McCrae et al., 2004; Roberts, Caspi, & Moffitt, 2003; Roberts et al., 2004). Because our sample consisted of newlywed couples, we were able to obtain spouse ratings of personality and trait affectivity for most of our participants (n ⫽ 301) at both assessments. Consequently, our data allow us to examine issues such as mean-level change and individual-level change in a large sample across two different methods. On the basis of previous evidence, we had two basic expectations concerning rank-order stability. First, based on the metaanalytic results of Roberts and DelVecchio (2000), we predicted that self-ratings and spouse ratings would produce very similar stability correlations across this 2-year interval. Second, because stability correlations tend to increase with age (Caspi et al., 2005; Clark & Watson, 1999; Roberts & DelVecchio, 2000), we expected these coefficients to be somewhat higher than those reported in previous longitudinal studies of young adulthood (e.g., Roberts et al., 2001; Robins et al., 2001; Vaidya et al., 2002). We also made two basic predictions related to our analyses of mean-level change. First, we expected our young adult sample to display significant (a) increases in conscientiousness and agreeableness and (b) decreases in neuroticism and negative affectivity across this 2-year interval (Caspi et al., 2005; Clark & Watson, 1999; Roberts, Robins, et al., 2003). In contrast, the findings related to extraversion and openness are more complex and inconsistent (see Roberts et al., 2006). Accordingly, we made no specific predictions regarding these traits. Second, on the basis of earlier results (McCrae et al., 2004, 2005), we expected the selfratings and spouse ratings to show the same basic patterns over time; however, we expected the magnitude of these age-related changes to be greater in the former than in the latter (i.e., we predicted weaker increases in agreeableness and conscientiousness and smaller decreases in neuroticism and negative affectivity in the spouse ratings). Finally, as discussed earlier, McCrae (1993) found weak convergence between self-rated versus spouse-rated change at the individual level. On the basis of these results, we expected to obtain significant (because of our larger sample size) but low correlations between self- and spouse-rated change scores on our measures of personality and trait affectivity.
Method Participants and Procedure Time 1 assessment. The original sample consisted of 291 married couples who participated in the Iowa Marital Assessment Project (IMAP; for more details regarding IMAP, see Watson et al., 2004). IMAP staff members identified recently married couples from the records of Johnson County and Linn County in eastern Iowa. Couples who met the inclusion criteria for the study (which required that they had been married less than a year at the time of initial contact and that both members of the couple were age 50 or younger) then were sent a letter inviting them to participate. At the time of assessment, the couples had been married an average of
962
WATSON AND HUMRICHOUSE
153.9 days (range ⫽ 25 to 452 days), that is, approximately 5 months. They indicated that they had known each other an average of 4.69 years (range ⫽ less than a year to 30 years) and had begun dating approximately 3.5 years earlier (M ⫽ 3.54 years; range ⫽ less than a year to 15 years). All participants were assessed in small-group sessions involving from one to three couples. These sessions were conducted from June 2001 through December 2001. The sessions typically lasted from 2 to 2.5 hr and included a battery of self-report measures, spouse ratings, and intelligence testing (see Watson et al., 2004). The couples were compensated $120 for their participation. To ensure honest and independent responding, each participant sat quietly at a separate desk when completing the self-ratings and spouse ratings. Because of missing data, complete responses were available from a total of 574 participants (98.6%) at this initial assessment. Time 2 assessment. The Time 2 assessment was conducted in two separate phases. In the first phase, we attempted to contact all of the original IMAP participants, who were invited to return to our laboratory for another small-group session; these Time 2 sessions involved a maximum of six couples. These sessions typically lasted approximately 2 hr and again included a battery of self-report measures and spouse ratings. Each individual was compensated $50 for his or her participation. These small-group laboratory sessions were conducted between July 2003 and November 2003, that is, approximately 2 years after the original Time 1 assessment. In most cases, the participants again were assessed as couples; in our invitation letter, however, we emphasized that they were welcome to participate as single individuals (a) if they were separated or divorced from their spouse or (b) if their spouse was unwilling to return for this assessment for any reason. A total of 314 individuals participated in these Time 2 laboratory sessions; this included 154 intact couples and 6 single women. Because of missing responses (including missing Time 1 responses in some instances), we had complete data on 301 of these participants (152 men, 149 women). Many of the original IMAP participants expressed an interest in being involved in the Time 2 assessment but indicated that they were unable to return to our laboratory for various reasons (most often because they had moved out of eastern Iowa). We therefore initiated the second phase of the Time 2 assessment, in which participants were sent a battery of questionnaires by mail. Because we could not ensure that spouses would complete their ratings independently of one another, these questionnaires were restricted to self-report and did not ask the respondents to provide any spouse ratings. Each participant was compensated $25 for his or her responses, which were returned to us by mail. We received mailed responses from an additional 159 participants (75 men, 84 women). The bulk of these mailed questionnaires were completed between December 2003 and January 2004 (n ⫽ 146); however, we received 13 additional questionnaires between February and April 2004. Thus, these participants completed the Time 2 assessment roughly 2 to 2.5 years after Time 1. Overall, between these two phases, we obtained Time 2 responses from a total of 460 participants (227 men, 233 women). This represents 80.1% of the 574 participants with complete Time 1 data. At Time 1, the men were a mean age of 29.1 years (SD ⫽ 6.5 years, range ⫽ 19 – 49) and the women were a mean age of 27.3 years (SD ⫽ 6.2 years, range ⫽ 18 –50).
Measures BFI. We used the BFI (Benet-Martinez & John, 1998; John & Srivastava, 1999) to assess the traits comprising the five-factor model. The BFI contains 8-item scales assessing Neuroticism and Extraversion, a 10-item Openness scale, and 9-item measures of Agreeableness and Conscientiousness. In the self-rating version, participants were asked to indicate “the extent to which you agree or disagree” with each item on a 5-point scale ranging from disagree strongly to strongly agree. The format for the spouse ratings was identical, except that the respondents were asked to “consider the feelings, behaviors, and preferences of your spouse” and then to evaluate the extent to which each item characterized the personality of their partner (again using the same 5-point disagree–agree scale). In the
Time 1 assessment, the BFI scales had coefficient alphas ranging from .78 (Agreeableness) to .88 (Neuroticism) in the self-ratings and from .83 (Conscientiousness) to .88 (Neuroticism) in the spouse ratings (see Watson et al., 2004). PANAS. We assessed affectivity using the trait form of the PANAS. The PANAS includes 10-item scales assessing the general dimensions of Negative Affect (e.g., nervous, upset, irritable, ashamed, scared) and Positive Affect (e.g., enthusiastic, active, interested, proud, determined). Self-raters were asked to indicate on a 5-point scale (ranging from very slightly or not at all to extremely) “to what extent you generally feel this way, that is, how you feel on average.” The format and instructions for the spouse ratings were identical, except that respondents were told to rate “to what extent your spouse generally feels or acts this way, that is, how your spouse feels or acts on the average.” In the Time 1 assessment, the Negative Affect scale had coefficient alphas of .89 and .88 in the selfratings and spouse ratings, respectively; parallel values for the Positive Affect scale were .85 and .87, respectively (see Watson et al., 2004).
Results Preliminary Analyses Attrition analyses. Before turning to our main results, we report several other analyses to explicate basic aspects of our data. First, we conducted two series of analyses to evaluate the representativeness of our Time 2 subsamples. In the first series, we compared the Time 1 BFI and PANAS scores of our Time 2 retest participants (RP; n ⫽ 460) with those of the nonparticipants (NP; n ⫽ 114). Across the 14 individual analyses (i.e., seven traits assessed using both self-ratings and spouse ratings), we obtained four significant differences. Specifically, the RP respondents rated themselves as more conscientious, RP M ⫽ 34.45, SD ⫽ 5.72; NP M ⫽ 32.93, SD ⫽ 6.51; t(572) ⫽ 2.46, p ⬍ .05, more agreeable, RP M ⫽ 35.58, SD ⫽ 5.39; NP M ⫽ 34.47, SD ⫽ 4.90; t(572) ⫽ 2.02, p ⬍ .05, and less open, RP M ⫽ 38.58, SD ⫽ 6.11; NP M ⫽ 39.90, SD ⫽ 5.55; t(572) ⫽ ⫺2.11, p ⬍ .05, than the NP group. Interestingly, only one of these differences was replicated in the spouse ratings; specifically, the spouses rated the RP group as more conscientious than the NP group, RP M ⫽ 35.17, SD ⫽ 6.17; NP M ⫽ 33.52, SD ⫽ 6.82; t(572) ⫽ 2.51, p ⬍ .05. Thus, replicating a pattern observed in previous longitudinal studies (e.g., Vaidya et al, 2002), we obtained consistent evidence that the Time 2 retest participants were significantly more conscientious than the nonparticipants. Second, we compared the Time 1 BFI and PANAS scores of the Time 2 laboratory (LAB; n ⫽ 301) and mail-out (MAIL; n ⫽ 159) subsamples. Across the 14 individual analyses, we obtained only one significant difference: Spouses in the laboratory subsample rated their partners as higher in trait Negative Affect than did those in the mail-out group, LAB M ⫽ 19.48, SD ⫽ 6.26, MAIL M ⫽ 18.14, SD ⫽ 6.31; t(458) ⫽ 2.18, p ⬍ .05. It is noteworthy, however, that this difference did not approach significance in the self-ratings, LAB M ⫽ 19.04, SD ⫽ 7.11; MAIL M ⫽ 18.35, SD ⫽ 6.44; t(458) ⫽ 1.06, ns. On the basis of this evidence, it appears that our two Time 2 subsamples are very similar in terms of their personality characteristics; this result is not surprising, given that these groups emerged primarily for pragmatic reasons (i.e., most of the participants in the MAIL group had moved out of the area). Spousal similarity. Spousal similarity is a potentially important consideration in studies of married couples because it produces statistical nonindependence in data analyzed at the individ-
SELF-RATINGS VERSUS SPOUSE RATINGS OF STABILITY
ual level (Kashy & Snyder, 1995; Kenny, 1995). That is, if scores between members of a dyad are systematically interrelated, then the assumption of independent observations is violated and significance tests may be biased and misleading. However, studies consistently have found very little evidence of spousal similarity on a wide range of personality traits (see Watson, Hubbard, & Wiese, 2000b; Watson et al., 2004). Consistent with this broader trend, analyses of the Time 1 IMAP data yielded spousal similarity correlations ranging between ⫺.17 and .18 (self-ratings) and between ⫺.14 and .13 (spouse ratings) on the BFI and PANAS scales. Because nonindependence is not a substantial problem in these data, we report our main analyses of personality at the individual level. However, to examine possible gender differences in our data, we also report many of our key results separately for men and women. Subsequently, we examine the relations between personality ratings and marital satisfaction. Not surprisingly, the husband’s and wife’s marital satisfaction were substantially related at both Time 1 (r ⫽ .27, n ⫽ 289, p ⬍ .01) and Time 2 (r ⫽ .47, n ⫽ 144, p ⬍ .01). Accordingly, for all analyses involving marital satisfaction, we do not present any findings on our overall sample but report separate results only for the wives and husbands. Self–spouse agreement. Finally, to examine the convergent validity of our trait measures, we computed self–spouse agreement correlations at both Time 1 and Time 2; these analyses were based on the 301 participants with complete data at both assessments. Consistent with previous analyses of married couples (Watson et al., 2000b), the Big Five traits showed strong self–spouse convergence at both assessments: Specifically, these agreement correlations ranged from .43 (Agreeableness) to .62 (Neuroticism; mean r ⫽ .53) at Time 1, and from .40 (Openness) to .64 (Extraversion; mean r ⫽ .52) at Time 2. Replicating previous research in this area (see Watson et al., 2000b), the PANAS scales showed more moderate— but still significant—agreement at both assessments. As would be expected with relatively low visibility traits (which are particularly susceptible to acquaintanceship effects; see Watson et al., 2000b), these affective scales tended to show better agreement at Time 2 (rs ⫽
963
.39 and .38 for Negative Affect and Positive Affect, respectively) than at Time 1 (rs ⫽ .32 and .27, respectively). These agreement correlations are reported in greater detail in Humrichouse and Watson (2006).
Rank-Order Stability Basic findings. Table 1 presents rank-order stability correlations for both the self-ratings and the spouse ratings across the 2-year study interval. The table displays stability coefficients computed in the overall sample, as well as separately for women and men; these latter results are organized by the target of assessment (i.e., spouse ratings for women represent the husbands’ ratings of their wives, whereas the spouse ratings for men reflect the wives’ ratings of their husbands). Several aspects of these data are noteworthy. First, these correlations are consistently high, ranging from .67 (PANAS Positive Affect in the spouse ratings) to .83 (BFI Extraversion in the spouse ratings) in the overall sample. Furthermore, consistent with previous investigations of this issue (Costa & McCrae, 1988; Roberts & DelVecchio, 2000; Schuerger et al., 1989), the stability coefficients did not differ substantially by gender. In fact, only 1 of the 14 individual comparisons yielded a statistically significant sex difference: Self-rated Openness was more stable in women (r ⫽ .83) than in men (r ⫽ .68; z ⫽ 3.84, p ⬍ .01). We therefore restrict further discussion to results based on the overall sample. Second, as predicted, the self-ratings and spouse ratings yielded very similar stability correlations; indeed, in every case the stability correlations for a given scale differed by no more than 兩.04兩 from each other. Overall, the mean stability correlations for the two sets of ratings were virtually identical for both the BFI (mean rs ⫽ .78 and .77 in the self-ratings and spouse ratings, respectively) and the PANAS (mean rs ⫽ .69 in both the self-ratings and spouse ratings). Third, we predicted that the 2-year stability correlations in this study would be significantly higher than those reported in younger adults. We tested this prediction by comparing these stability correlations to those reported by Vaidya et al. (2002) in a some-
Table 1 Rank-Order Stability Correlations Across the 2-Year Interval Self-rating Scale BFI Neuroticism Extraversion Openness Agreeableness Conscientiousness M PANAS Negative Affect Positive Affect M
Spouse rating
Overall
Women
Men
Overall
Women
Men
.79 .82 .76 .75 .76 .78
.78 .83 .83 .72 .77 .79
.75 .81 .68 .77 .74 .75
.77 .83 .78 .71 .76 .77
.76 .82 .75 .65 .80 .76
.69 .85 .80 .76 .71 .77
.68 .69 .69
.67 .71 .69
.70 .66 .68
.72 .67 .69
.68 .68 .68
.66 .66 .66
Note. Data are organized by target (e.g., spouse-rating correlations for women represent the husbands’ ratings of their wives). For self-ratings, N ⫽ 460, n ⫽ 233, women; n ⫽ 227, men. For spouse ratings, N ⫽ 301, n ⫽ 149, women; n ⫽ 152, men. All correlations are significant at p ⬍ .01. BFI ⫽ Big Five Inventory; PANAS ⫽ Positive and Negative Affect Schedule.
WATSON AND HUMRICHOUSE
964
what younger sample (mean age ⫽ 21 years at Time 2). These comparisons strongly confirmed our prediction. Vaidya et al. (see their Table 6) reported 2.5-year stability correlations ranging from .59 to .72 on the BFI (mean r ⫽ .64), and correlations of .49 (Negative Affect) and .51 (Positive Affect) on the PANAS. Follow-up tests indicated that all seven scales produced significantly higher stability correlations in the current study; this was true for both the self-ratings (zs ranged from 3.21 to 5.26; all ps ⬍ .01) and the spouse ratings (zs ranged from 2.73 to 4.84; all ps ⬍ .01). Thus, our data again show that stability correlations for personality increase systematically with age (see also Caspi et al., 2005; Roberts & DelVecchio, 2000). Fourth, our results again indicate that the BFI scales tend to show higher stability correlations than the PANAS (see Vaidya et al., 2002; Watson, 2004). Replicating the results of earlier studies, BFI Extraversion was significantly more stable than PANAS Positive Affect in both the self-ratings (.82 vs. .69; z ⫽ 4.70, p ⬍ .01) and the spouse ratings (.83 vs. .67; z ⫽ 4.81, p ⬍ .01), despite the fact that these scales consistently were strongly correlated (for the self-ratings, r ⫽ .52 and .57 at Time 1 and Time 2, respectively; for the spouse ratings, r ⫽ .55 and .53, respectively). Similarly, BFI Neuroticism had higher stability correlations than the PANAS Negative Affect scale in both sets of ratings; this difference was significant in the self-ratings (.79 vs. .68; z ⫽ 3.70, p ⬍ .01 ), but not in the spouse ratings (.77 vs. .72; z ⫽ 1.35, ns). This difference is particularly striking given the very high correlations between these scales in both the self-ratings (rs ⫽ .64 and .67 at Time 1 and Time 2, respectively) and the spouse ratings (rs ⫽ .72 and .79 at Time 1 and Time 2, respectively). It is important to note, moreover, that the BFI Neuroticism and PANAS Negative Affect scales both are overwhelmingly affective in nature and contain very similar item content (see Watson, 2004, Table 5); taken together with previous findings, these results illustrate the importance of wording, format, and instructional effects on temporal stability (see Watson, 2004). Moderator analyses of age. As noted earlier, our data are consistent with previous evidence indicating that stability correlations increase with age (Roberts & DelVecchio, 2000). Given that our participants varied widely in age (range ⫽ 18 –50 years at
Time 1), this raises the further possibility that the level of rankorder stability was substantially higher in our older respondents. We examined this issue in a series of moderated multiple regression analyses, using the Time 2 scores as criteria. We entered the two main effects (i.e., age and the corresponding Time 1 trait score) as predictors in Step 1, followed by the centered interaction term in Step 2. Across the 14 individual analyses, we found four significant moderators. Only Agreeableness yielded a replicable interaction effect across both the self-ratings ( ⫽ .070, ⌬R2 ⫽ .005, p ⬍ .05) and the spouse ratings ( ⫽ .109, ⌬R2 ⫽ .012, p ⬍ .01). In addition, Negative Affect ( ⫽ .070, ⌬R2 ⫽ .006, p ⬍ .05) and Openness ( ⫽ .093, ⌬R2 ⫽ .008, p ⬍ .05) displayed significant moderator effects in the self-ratings and spouse ratings, respectively. Thus, we obtained some scattered evidence indicating that stability correlations were somewhat higher in our older participants.
Mean-Level Change Analyses of the self-ratings. We turn now to the issue of mean-level change. Table 2 reports the mean Time 1 and Time 2 scores for each scale in the self-ratings, together with an index (Cohen’s d) that quantifies the magnitude of the difference between them. We again report results based on the overall sample, as well as separately for men and women. These results strongly support our predictions. As expected, paired t tests revealed significant increases in both Conscientiousness (d ⫽ .33) and Agreeableness (d ⫽ .12), as well as significant declines in Neuroticism (d ⫽ ⫺.19) and Negative Affect (d ⫽ ⫺.11) in the overall sample. Conscientiousness showed the largest amount of change; in fact, it displayed significant increases in both women (d ⫽ .31) and men (d ⫽ .34). In contrast, our data did not reveal any significant mean-level changes in Extraversion, Openness, and Positive Affect across the 2-year study interval. Thus, consistent with previous work in this area, our findings demonstrate systematic temporal changes in personality that essentially are confined to the three trait domains comprising the Alpha superfactor (Markon et al., 2005). Moreover, these results again
Table 2 Mean-Level Changes in Self-Rated Personality Over the 2-Year Interval Overall Time 1 Scale BFI Neuroticism Extraversion Openness Agreeableness Conscientiousness PANAS Negative Affect Positive Affect
Women
Time 2
Time 1
Men
Time 2
Time 1
Time 2
M
SD
M
SD
d
M
SD
M
SD
d
M
SD
M
SD
d
22.2 27.9 38.6 35.6 34.4
7.1 6.6 6.1 5.4 5.7
21.3 27.7 38.2 36.0 35.7
6.8 6.7 6.4 5.4 5.3
⫺.19** ⫺.06 ⫺.09 .12* .33**
24.8 28.1 38.3 36.2 35.4
6.9 6.7 6.3 5.1 5.5
23.4 27.9 37.9 36.7 36.6
6.7 6.7 6.5 5.2 5.4
⫺.31** ⫺.05 ⫺.11 .15* .31**
19.5 27.7 38.9 35.0 33.5
6.2 6.5 5.9 5.6 5.8
19.2 27.4 38.5 35.3 34.8
6.2 6.6 6.2 5.4 5.1
⫺.06 ⫺.07 ⫺.08 .09 .34**
18.6 38.1
6.7 5.7
18.0 37.8
6.3 6.2
⫺.11* ⫺.07
18.7 37.7
7.0 5.8
18.1 37.3
6.0 6.6
⫺.11 ⫺.10
18.5 38.6
6.3 5.5
17.9 38.4
6.6 5.8
⫺.12 ⫺.05
Note. N ⫽ 460, n ⫽ 233, women; n ⫽ 227, men. BFI ⫽ Big Five Inventory; PANAS ⫽ Positive and Negative Affect Schedule. * p ⬍ .05. ** p ⬍ .01.
SELF-RATINGS VERSUS SPOUSE RATINGS OF STABILITY
965
assessments. We therefore computed paired t tests to examine whether the average self-ratings and spouse ratings differed significantly from each other at Times 1 and 2; these analyses were based on the 301 participants with complete data at both assessments. Analyses of the Time 1 data revealed evidence of a significant honeymoon effect in these newlywed couples; that is, the mean spouse ratings were significantly greater than the average self-ratings for Agreeableness (d ⫽ .33, p ⬍ .01), Extraversion (d ⫽ .26, p ⬍ .01), and Conscientiousness (d ⫽ .13, p ⬍ .05). Thus, at Time 1, spouses were rating their partners more positively than the partners were rating themselves. It is noteworthy that only one of these effects persisted at Time 2: The average spouse rating on Extraversion still was significantly higher than the mean self-rating (d ⫽ .17, p ⬍ .01). In contrast, all of the other significant findings suggested that the spouse judgments now were more negative than the self-ratings at Time 2. Specifically, the spouse ratings now were significantly higher than the self-ratings on both Neuroticism (d ⫽ .15, p ⬍ .01) and Negative Affect (d ⫽ .13, p ⬍ .05) and significantly lower on Conscientiousness (d ⫽ ⫺.25, p ⬍ .01) and Openness (d ⫽ ⫺.23, p ⬍ .01). These results clearly suggest that the honeymoon was over at Time 2, so that the spouses now were rating their partners more harshly at this second assessment. We revisit this issue later.
show that these changes are positive in nature, such that adult personality development is characterized by increasing psychological maturity with age (see Caspi et al., 2005). Analyses of the spouse ratings. Table 3 presents parallel results for the spouse ratings, again computed in the overall sample, as well as separately for women and men; as in Table 1, the latter results are organized by target (i.e., the ratings for women represent the husbands’ ratings of their wives, whereas the ratings for men reflect the wives’ ratings of their husbands). We predicted that these data would show the same basic pattern as the self-ratings, albeit in attenuated form; that is, we expected smaller increases in agreeableness and conscientiousness, and weaker declines in neuroticism and negative affectivity. These predictions obviously were not supported. Indeed, the spouse ratings show an entirely different pattern from the self-reports. Looking at the results from the overall sample, the most striking aspect of these data is that they showed a significant decline in both Agreeableness (d ⫽ ⫺.28) and Conscientiousness (d ⫽ ⫺.22), which is completely opposite to the trend exhibited in the self-ratings. Agreeableness displayed the largest overall level of change and exhibited significant decreases in both women (d ⫽ ⫺.24) and men (d ⫽ ⫺.32). Moreover, BFI Neuroticism (d ⫽ .07) and PANAS Negative Affect (d ⫽ .09) both showed small, nonsignificant increases in these data. Furthermore, Extraversion (d ⫽ ⫺.20) and Openness (d ⫽ ⫺.18)—which did not exhibit significant changes in the self-reports— both showed significant declines in the overall sample. In fact, across the seven trait scales, the only consistent finding was that PANAS Positive Affect displayed small, nonsignificant decreases in both the self-ratings (d ⫽ ⫺.07) and the spouse ratings (d ⫽ ⫺.07). Comparisons of self-ratings and spouse ratings. Clearly, the spouse ratings paint a very different picture of adult personality development, and— unlike the self-reports—they certainly do not suggest that our participants were exhibiting greater psychological maturity over time. These discrepant findings are troubling, and they raise a basic conceptual/interpretative issue: How can these results be reconciled into a coherent model of adult personality development? One way to begin to address this question is to compare the mean self-rating and spouse rating scores at each of the two
Individual-Level Change Change score analyses. The self-ratings and spouse ratings obviously displayed very different patterns of mean-level change. Nevertheless, it still is possible that our participants agreed about which specific individuals showed the largest relative increases and decreases on each trait. For example, if a wife reported a very large increase in her level of conscientiousness, it is plausible to suggest that her husband noticed this marked change and also judged her level of the trait to be higher at Time 2. Consequently, it also is important to examine change at the individual level. We began by computing separate change scores (subtracting the Time 1 score from the Time 2 score) for each scale in the self-ratings and spouse ratings. Before assessing self–spouse convergence, it first is important to establish that these measures assess systematic and meaningful variance, given understandable
Table 3 Mean-Level Changes in Spouse-Rated Personality Over the 2-Year Interval Overall Time 1 Scale BFI Neuroticism Extraversion Openness Agreeableness Conscientiousness PANAS Negative Affect Positive Affect
Women
Time 2
Time 1
Men
Time 2
Time 1
Time 2
M
SD
M
SD
d
M
SD
M
SD
d
M
SD
M
SD
d
21.5 29.1 37.8 37.7 35.1
7.0 6.3 6.8 5.6 6.3
21.8 28.4 37.0 36.4 34.2
7.6 6.6 7.0 6.4 6.6
.07 ⫺.20** ⫺.18** ⫺.28** ⫺.22**
24.6 29.5 37.7 37.6 36.0
6.3 5.9 6.2 5.2 6.5
24.7 28.6 36.4 36.5 35.5
7.4 6.2 6.7 5.9 6.5
.03 ⫺.25** ⫺.28** ⫺.24** ⫺.11
18.5 28.8 37.9 37.7 34.3
6.4 6.7 7.3 5.9 6.0
19.0 28.2 37.5 36.3 32.8
6.8 6.9 7.3 6.8 6.4
.11 ⫺.16 ⫺.09 ⫺.32** ⫺.31**
18.1 38.4
6.3 5.7
18.6 38.1
7.0 6.1
.09 ⫺.07
20.8 38.5
6.7 5.4
21.2 38.0
7.5 5.9
.07 ⫺.11
15.5 38.4
4.6 5.9
16.1 38.2
5.6 6.4
.13 ⫺.03
Note. N ⫽ 301, n ⫽ 149, women; n ⫽ 152, men. BFI ⫽ Big Five Inventory; PANAS ⫽ Positive and Negative Affect Schedule. ** p ⬍ .01.
WATSON AND HUMRICHOUSE
966
concerns about the unreliability of change scores (see Asendorpf, 1992; McCrae, 1993). As discussed by McCrae (1993), one interesting way to investigate this issue is to examine relations among change scores computed within the same method. If these variables are psychologically meaningful and tap systematic variance, then we should observe significant associations between change scores computed on closely related traits. For instance, we should see significant correlations between self-rated changes on the BFI Neuroticism and PANAS Negative Affect scales. Accordingly, Table 4 reports the within-method correlations between the change scores in both the self-ratings (below the diagonal) and the spouse ratings (above the diagonal). The most noteworthy aspect of these data is that change scores on related traits were, in fact, significantly correlated with one another. For example, the change scores for BFI Neuroticism and PANAS Negative Affect were moderately correlated in both the self-ratings (r ⫽ .42) and the spouse ratings (r ⫽ .44). Similarly, these analyses revealed significant correlations between changes on Extraversion and Positive Affect (r ⫽ .37 and .32 in the self-ratings and spouse ratings, respectively) and between changes on Agreeableness and Conscientiousness (r ⫽ .26 and .33, respectively). These results are reassuring, as they strongly suggest that these change scores tap meaningful psychological variance and do not simply reflect random measurement error. With that in mind, we now consider the convergence between self- and spouse-rated changes in personality. Based on the results of McCrae (1993), we predicted low, but significant, correlations between these two sets of scores. Table 5 reports these coefficients, both in the overall sample and separately by gender. The Table 5 data offer mixed support for our prediction. As expected, the correlations consistently were low. However, only two of them were significant in the overall sample—those for Conscientiousness (r ⫽ .20) and Neuroticism (r ⫽ .19); it is noteworthy, moreover, that neither of these effects replicated across both men and women. In contrast, the coefficients for the five remaining traits all were quite weak, ranging from only ⫺.01 (Openness) to .08 (Extraversion) in the overall sample. Thus, consistent with our analyses of mean-level change, these data reveal little convergence across our two types of ratings. Analyses of reliable change index (RCI) scores. Individual scores may randomly fluctuate from one testing to the next simply as a function of measurement error. Accordingly, analyses of raw change scores may be largely tapping random fluctuations over
Table 5 Convergent Correlations Between Self- Versus Spouse-Rated Change Scores Scale BFI Neuroticism Extraversion Openness Agreeableness Conscientiousness M PANAS Negative Affect Positive Affect M
Overall
Women
.19* .08 ⫺.01 .04 .20** .10
.07 .11 .07 .04 .35** .13
.06 .07 .06
⫺.01 .10 .04
1. 2. 3. 4. 5. 6. 7.
Neuroticism Extraversion Openness Agreeableness Conscientiousness Negative Affect Positive Affect
1 — ⫺.15** ⫺.08 ⫺.20** ⫺.15* .42** ⫺.23**
2 ⫺.21** — .26** .20** .14* ⫺.02 .37**
.29** .06 ⫺.07 .04 .08 .08 .16 .04 .10
Note. N ⫽ 301 (overall), 149 (women), 152 (men). BFI ⫽ Big Five Inventory; PANAS ⫽ Positive and Negative Affect Schedule. * p ⬍ .05. ** p ⬍ .01.
time rather than true, meaningful change (Roberts et al., 2001; Robins et al., 2001; Vaidya et al., 2002). Consequently, several recent studies have reported analyses based on RCI scores, which allow one to determine how many individuals showed a statistically significant amount of change versus how many essentially stayed the same. We computed RCI scores on our 301 participants with complete data at both assessments (see Roberts et al., 2001, for details regarding the calculation of RCI scores). To calculate the standard error of measurement for each scale, we used (a) its average standard deviation across the two assessments and (b) retest reliability coefficients computed across a 2-week interval in a sample of 446 University of Iowa students (Chmielewski & Watson, 2006). As in previous studies (e.g., Roberts et al., 2001; Robins et al., 2001; Vaidya et al., 2002), we classified individuals as having changed significantly from one assessment to the next if the probability associated with that person’s RCI score was less than 5% (i.e., an RCI score of 兩1.96兩 or greater). Table 6 summarizes our RCI results. The first four columns show the number of individuals who showed significant increases or decreases on each scale in the self-ratings and spouse ratings. These results again highlight the marked discrepancy in the patterns of mean-level change across the two sets of ratings. Consider, for instance, the results for Conscientiousness. In the self-ratings,
Table 4 Within-Method (Mono-Rater) Correlations Between Change Scores Across Traits Scale
Men
3 ⫺.06 .17** — .22** .08 ⫺.03 .19**
4 ⫺.36** .18** .26** — .26** ⫺.25** .28**
5 ⫺.29** .18** .28** .33** — ⫺.18** .30**
6
7
.44** ⫺.23** ⫺.09 ⫺.35** ⫺.21** — ⫺.06
⫺.31** .32** .22** .28** .31** ⫺.21** —
Note. N ⫽ 301. Self-rating correlations are shown below the diagonal; spouse-rating correlations are presented above the diagonal. * p ⬍ .05. ** p ⬍ .01.
SELF-RATINGS VERSUS SPOUSE RATINGS OF STABILITY
967
Table 6 Summary of Results Showing the Number of Participants With Significant Reliable Change Index Scores Cross-method comparison Self-rating Scale BFI Neuroticism Extraversion Openness Agreeableness Conscientiousness PANAS Negative Affect Positive Affect
Spouse rating
Decreased
Increased
Decreased
Increased
Consistent change
Paradoxical change
Inconsistent change
No change
19 10 17 11 6
6 7 16 10 22
12 12 16 25 22
16 4 9 5 6
3 3 2 0 3
0 0 3 2 1
47 27 48 47 48
251 271 248 252 249
13 8
5 10
7 9
13 11
0 0
0 0
38 38
263 263
Note. N ⫽ 301. BFI ⫽ Big Five Inventory; PANAS ⫽ Positive and Negative Affect Schedule.
22 of the 28 individuals (78.6%) who showed significant change reported increased levels of the trait. The spouse ratings showed exactly the opposite pattern: Here, the large majority of reliable change (22 of 28 individuals, or 78.6%) reflected significant declines in Conscientiousness. Similarly, whereas the bulk of the reliable change in self-rated Neuroticism (76%) and Negative Affect (72.2%) indicated decreasing negative emotionality, most of the significant change in the spouse ratings was in the opposite direction (i.e., 57.1% and 65% of the significant RCI scores for Neuroticism and Negative Affect, respectively, were associated with higher scores at Time 2). Still, these results do not directly address the question of whether the self-raters and spouse raters agreed about which specific individuals showed significant change. This issue is addressed in the last four columns of Table 6, which summarize the cross-method comparisons into four categories: consistent change (i.e., the self-ratings and spouse ratings both yielded significant RCI scores in the same direction), paradoxical change (i.e., the two sets of ratings both yielded significant RCI scores, but one showed an increase and the other a decrease), inconsistent change (i.e., one rating yielded a significant RCI score, but the other did not), and no change (i.e., both ratings produced nonsignificant RCI scores). These results yield very little evidence of consistency across the two methods. Across the seven traits, there were a total of only 11 instances of consistent change (all on the BFI scales). It is noteworthy, moreover, that there actually were 6 cases of paradoxical change in our data. Taken together with our earlier analyses of raw change scores, these results indicate that individual-level change shows little consistency across our two rating methods.
Exploring the Self–Spouse Discrepancy: The Influence of Marital Satisfaction Personality ratings and marital satisfaction. How can we explain this marked divergence between self-rated versus spouserated changes in personality? Earlier, we suggested that the spouse ratings reflected a honeymoon effect at Time 1 that had dissipated by Time 2. This, in turn, suggests that marital satisfaction may play a key role in explaining our findings. In this regard, previous research has established strong links between current relationship satisfaction and personality ratings of romantic partners. For ex-
ample, Watson, Hubbard, and Wiese (2000a) analyzed ratings of the Big Five and trait affectivity in dating and married couples. They found that the judge’s current level of relationship satisfaction was only weakly related to the self-rated personality characteristics of their romantic partners. For instance, relationship satisfaction in the dating women correlated only .18 and .17 with the self-reported agreeableness and conscientiousness of their male partners; conversely, satisfaction among the dating men correlated only .11 and .21 with the self-rated agreeableness and conscientiousness, respectively, of their female partners. These data therefore suggest that a person’s self-perceived standing on these traits had relatively little effect on the satisfaction of their romantic partner. In sharp contrast, however, satisfaction was much better predictor of the judges’ ratings of their partners’ traits. Thus, a dating woman’s relationship satisfaction correlated .41 and .50, respectively, with her ratings of her male partner’s agreeableness and conscientiousness; similarly, satisfaction among dating men correlated .30 and .55, respectively, with their ratings of their female partners’ agreeableness and conscientiousness. The IMAP participants rated their current level of marital satisfaction at both assessments. Satisfaction was assessed using a single global rating derived from the Locke–Wallace Marital Adjustment Test (Locke & Wallace, 1959). Participants chose “the number which best describes the degree of happiness, everything considered, that you feel in your present marriage”; these ratings were made on a 7-point scale ranging from very unhappy to perfectly happy. Not surprisingly, given that the participants were newlyweds at the initial assessment, their mean level of satisfaction declined significantly from Time 1 to Time 2. Specifically, the mean level of satisfaction among the wives dropped from 5.68, SD ⫽ 1.02, to 5.39, SD ⫽ 1.21; t(150) ⫽ 2.98, p ⬍ .01; Cohen’s d ⫽ .24, whereas the husbands’ satisfaction decreased from 5.77, SD ⫽ 0.95, to 5.54, SD ⫽ 1.12; t(145) ⫽ 2.76, p ⬍ .01; Cohen’s d ⫽ .23. These data raise the possibility that spouse-rated changes in personality—which, in marked contrast to the self-reports, tended to indicate a negative developmental trajectory (e.g., lower agreeableness and conscientiousness)—may reflect, in part, the significant decline in satisfaction that occurred over the course of our study.
WATSON AND HUMRICHOUSE
968
Correlational analyses. We conducted several series of analyses to examine this possibility. In the first set, we sought to replicate the results of Watson et al. (2000a), demonstrating an especially strong link between marital satisfaction and personality ratings of the romantic partner. Table 7 presents hetero-rater correlations between a participant’s current marital satisfaction and the self-rated personality characteristics of his or her spouse. For instance, the coefficient in the first row and column of Table 7 represents the correlation between a wife’s marital satisfaction and her husband’s self-rated Neuroticism at Time 1. These correlations are reported separately for the husbands and wives at each assessment. These results reveal several significant associations between marital satisfaction and the spouse’s self-rated characteristics. Most notably, marital satisfaction tended to be positively associated with Agreeableness and negatively related to Neuroticism/ Negative Affectivity. Thus, participants tended to report greater satisfaction if they were married to spouses who were agreeable and emotionally stable. Overall, 14 of the 28 coefficients (50%) were significant at p ⬍ .05. At the same time, however, it also should be noted that these correlations tend to be relatively low in magnitude. Specifically, only five coefficients (17.9%) are as high as 兩.20兩, and only two (7.1%) exceed 兩.30兩. These results provide an interesting context for interpreting the mono-rater correlations between a participant’s marital satisfaction and his or her ratings of the spouse’s trait characteristics; these coefficients are reported in Table 8. For example, the coefficient in the first row and column of Table 8 reflects the correlation between a wife’s satisfaction and her ratings of her husband’s Neuroticism at Time 1. These correlations clearly are systematically stronger than those reported in Table 7. Of the 28 correlations, 27 (96.4%) are significant at p ⬍ .05. Furthermore, 25 coefficients (89.3%) exceed 兩.20兩, and 19 (67.9%) exceed 兩.30兩; indeed, 10 correlations (35.7%) are 兩.40兩 and greater. Follow-up tests indicated that 27 of the 28 correlations in Table 8 differed significantly from the corresponding values reported in Table 7 (zs ranged from 兩2.00兩 to 兩5.47兩; all ps ⬍ .05); the single exception was that the Table 7 Hetero-Rater Correlations Between Marital Satisfaction and the Spouse’s Self-Rated Personality Wife’s satisfaction: Husband’s personality Scale
Time 1
Time 2
Table 8 Mono-Rater Correlations Between Marital Satisfaction and the Judge’s Ratings of the Spouse’s Personality
Scale BFI Neuroticism Extraversion Openness Agreeableness Conscientiousness PANAS Negative Affect Positive Affect
⫺.18* ⫺.07 .06 .23** .11
⫺.15* ⫺.01 .10 .10 .03
⫺.33** .19** .05 .23** .17*
⫺.16** ⫺.02
⫺.26** .06
⫺.09 .15*
⫺.31** .18*
Note. N ⫽ 283 (Time 1); n ⫽ 151, wife’s Time 2 satisfaction; n ⫽ 149, husband’s Time 2 satisfaction. BFI ⫽ Big Five Inventory; PANAS ⫽ Positive and Negative Affect Schedule. * p ⬍ .05. ** p ⬍ .01.
Time 2
Time 1
Time 2
⫺.28** .21** .24** .45** .19**
⫺.39** .11 .22** .53** .31**
⫺.36** .26** .33** .37** .19**
⫺.49** .41** .42** .48** .37**
⫺.35** .34**
⫺.40** .41**
⫺.29** .30**
⫺.55** .55**
Table 9 Hierarchical Multiple Regressions: Predicting Time 2 Spouse-Rated Traits From Time 2 Marital Satisfaction (Husbands’ Ratings)
Time 2
⫺.12* .02 .02 .16** ⫺.07
Time 1
correlations between a wife’s satisfaction and her husband’s negative affectivity at Time 2 did not differ from one another (⫺.40 vs. ⫺.26; z ⫽ ⫺1.66, p ⬍ .10). Thus, our findings replicate those reported by Watson et al. (2000a). These results again indicate that spouse ratings are closely linked to current levels of marital satisfaction. Predicting changes in spouse-rated personality. Next, we conducted hierarchical regression analyses to test whether spouserated changes in personality could be predicted from changes in marital satisfaction between Time 1 and Time 2. The results of these analyses are presented in Tables 9 (husbands’ ratings) and 10 (wives’ ratings). The participant’s Time 2 ratings of his or her spouse’s trait characteristics served as the criteria in these regressions. In each case, the rater’s Time 1 trait rating was entered as a predictor in Step 1, followed by Time 1 marital satisfaction in Step
Scale BFI Neuroticism Extraversion Openness Agreeableness Conscientiousness PANAS Negative Affect Positive Affect
Husband’s satisfaction: Wife’s personality traits
Note. N ⫽ 283 (Time 1); n ⫽ 151, wife’s Time 2 satisfaction, n ⫽ 149, husband’s Time 2 satisfaction. BFI ⫽ Big Five Inventory; PANAS ⫽ Positive and Negative Affect Schedule. ** p ⬍ .01.
Husband’s satisfaction: Wife’s personality Time 1
Wife’s satisfaction: Husband’s personality traits
BFI Neuroticism Extraversion Openness Agreeableness Conscientiousness PANAS Negative Affect Positive Affect
Step 1 T1 trait
Step 2 T1 satisfaction
Step 3 T2 marital satisfaction
⌬R2
⌬R2
⌬R2

.554** .648** .552** .421** .628**
.003 .000 .030** .015 .001
.035** .017** .029** .097** .041**
⫺.234** .161** .204** .376** .245**
.452** .389**
.001 .047**
.091** .087**
⫺.375** .352**
Note. N ⫽ 146. T1 ⫽ Time 1; T2 ⫽ Time 2; BFI ⫽ Big Five Inventory; PANAS ⫽ Positive and Negative Affect Schedule. ** p ⬍ .01.
SELF-RATINGS VERSUS SPOUSE RATINGS OF STABILITY
Table 10 Hierarchical Multiple Regressions: Predicting Time 2 SpouseRated Traits From Time 2 Marital Satisfaction (Wives’ Ratings)
Scale BFI Neuroticism Extraversion Openness Agreeableness Conscientiousness PANAS Negative Affect Positive Affect
Step 3 T2 marital satisfaction
Step 1 T1 trait
Step 2 T1 satisfaction
⌬R2
⌬R2
⌬R2

.479** .723** .629** .569** .509**
.057** .000 .002 .010 .011
.012 .000 .011* .036** .012
⫺.127 .008 .123* .225** .128
.461** .477**
.064** .014*
.018* .029**
⫺.153* .198**
Note. N ⫽ 146. T1 ⫽ Time 1; T2 ⫽ Time 2; BFI ⫽ Big Five Inventory; PANAS ⫽ Positive and Negative Affect Schedule. * p ⬍ .05. ** p ⬍ .01.
2; the rater’s Time 2 satisfaction then was added in Step 3 to determine its incremental predictive power. For example, the first row in Table 9 shows an analysis in which the husbands’ ratings of their wives’ Neuroticism at Time 2 served as the criterion. We entered the husband’s Time 1 rating of his wife’s Neuroticism in Step 1 and the husband’s Time 1 marital satisfaction in Step 2; we then added the husband’s Time 2 marital satisfaction in Step 3. Thus, by controlling for both Time 1 satisfaction and the corresponding Time 1 personality rating, these analyses allow us to determine whether changes in marital satisfaction predict changes in the spouse ratings. We consider first the analyses of the husbands’ ratings (see Table 9). These results clearly establish that changes in spouse ratings can be predicted from changes in marital satisfaction. In every case, the inclusion of Time 2 marital satisfaction in Step 3 was associated with a significant increase in predictive power; across the seven analyses, Time 2 marital satisfaction contributed from 1.7% to 9.7% incremental variance (M ⫽ 5.7%). The largest incremental variance effects were seen for ratings of Agreeableness (9.7%), Negative Affect (9.1%), and Positive Affect (8.7%). These results are particularly impressive given the very strong rank-order stabilities of these scales. Indeed, Table 9 indicates that the Time 1 trait scores already accounted for 42.1% to 64.8% (M ⫽ 52.1%) of the criterion variance in Step 1. Table 10 presents parallel results for the wives’ ratings. The effects here clearly are weaker overall. Across the seven analyses, Time 2 marital satisfaction contributed from 0% to 3.6% incremental variance, with a mean value of 1.7%. Still, Time 2 satisfaction contributed significantly in four of the seven analyses. It is noteworthy, moreover, that Agreeableness, Negative Affect, and Positive Affect again showed the strongest incremental effects, suggesting that these ratings are especially sensitive to changes in marital satisfaction. The data presented in Tables 9 and 10 generally indicate that spouse-rated changes in personality reflect, in part, changes in marital satisfaction between Time 1 and Time 2. Given that marital satisfaction declined significantly across the study interval, these analyses help to explain why the spouse ratings—in marked con-
969
trast to the self-ratings—tended to be somewhat more negative in character at Time 2. Predicting changes in self-rated personality. Is this link with marital satisfaction specific to spouse ratings, or does it also characterize self-reports? We examined this important issue in a parallel series of hierarchical regression analyses. These regressions were identical in form to those presented in Tables 9 and 10, except that they were based on self-rated—rather than spouse rated—trait scores. In one analysis, for instance, the criterion was the husband’s self-rated Neuroticism at Time 2. We added his self-rated Time 1 Neuroticism in Step 1 and his Time 1 marital satisfaction in Step 2; his Time 2 marital satisfaction was then entered in Step 3. These analyses therefore allow us to determine whether changes in marital satisfaction predict changes in selfratings between Time 1 and Time 2. These analyses established that changes in marital satisfaction essentially were unrelated to self-rated change in personality. Time 2 marital satisfaction contributed from 0% to 2.2% incremental variance in the husbands’ ratings (M ⫽ 0.5%), and from 0% to 0.6% incremental variance in the wives’ ratings (M ⫽ 0.2%). Across the 14 analyses, only 1 produced a significant effect: Time 2 satisfaction was negatively related to change on the PANAS Negative Affect scale in the husbands’ data (⌬R2 ⫽ .022,  ⫽ ⫺.182, p ⬍ .05); in other words, husbands who reported greater negative affectivity also experienced a decline in marital satisfaction. Overall, however, our results establish that marital satisfaction is a much more powerful predictor of spouse-rated change than of self-rated change. Predicting changes in marital satisfaction. Earlier, we established that changes in marital satisfaction predict changes in spouse-rated personality traits. We conducted a final series of hierarchical regression analyses to determine whether similar effects could be observed in the opposite direction, that is, whether changes in spouse-rated personality traits also could predict changes in marital satisfaction. The rater’s Time 2 marital satisfaction served as the criterion in these regressions. In each case, the rater’s Time 1 marital satisfaction was entered as a predictor in Step 1, followed by his or her Time 1 trait rating of the partner in Step 2; the rater’s Time 2 trait rating of the spouse then was added in Step 3 to examine its incremental predictive power. For example, the wife’s Time 1 marital satisfaction was entered in Step 1, followed by her Time 1 rating of her husband’s Agreeableness in Step 2; finally, her Time 2 rating of her husband’s Agreeableness was entered in Step 3. Tables 11 (husbands’ ratings) and 12 (wives’ ratings) summarize the results from Steps 2 and 3 of these regressions. The pattern of these results is very similar to that reported in Tables 9 and 10, but the overall magnitude of the effects is slightly stronger in these analyses. In the husbands’ data, the inclusion of the Time 2 spouse rating in Step 3 was associated with a significant increase in predictive power in every case; across the seven analyses, the Time 2 trait ratings contributed from 3.2% to 11.8% incremental variance (M ⫽ 7.7%) in predicting Time 2 marital satisfaction. The largest effects again were seen for ratings of Agreeableness (11.8% incremental variance), Positive Affect (10.8%), and Negative Affect (10.7%). As in the earlier analyses, the wives’ ratings yielded weaker effects overall. Across the seven traits, the Time 2 spouse rating contributed from 0% to 6.0% incremental variance, with a mean value of 2.8%. Still, the Time 2 trait score contributed
WATSON AND HUMRICHOUSE
970
Table 11 Hierarchical Multiple Regressions: Predicting Time 2 Marital Satisfaction From Time 2 Spouse-Rated Traits (Husbands’ Ratings) Step 2 T1 trait Scale BFI Neuroticism Extraversion Openness Agreeableness Conscientiousness PANAS Negative Affect Positive Affect
Step 3 T2 trait
⌬R2
⌬R2

Final R
.069** .053** .005 .022* .024*
.051** .032** .049** .118** .076**
⫺.338** .301** .344** .458** .452**
.64 .61 .59 .66 .63
.073** .016
.107** .108**
⫺.443** .437**
.68 .64
Note. N ⫽ 146. T1 marital satisfaction was entered in Step 1 of the regression. T1 ⫽ Time 1; T2 ⫽ Time 2; BFI ⫽ Big Five Inventory; PANAS ⫽ Positive and Negative Affect Schedule. * p ⬍ .05. ** p ⬍ .01.
significantly in four of the seven analyses, with Agreeableness, Negative Affect, and Positive Affect again showing significant incremental effects. Overall, therefore, these results further demonstrate the significant link between spouse ratings and changes in marital satisfaction.
Discussion Rank-Order Stability Our findings regarding rank-order stability supported our predictions and were broadly consistent with previous work in this area. As expected, the stability correlations for the self-ratings and spouse ratings were very similar; in the overall sample, in fact, the average stability correlations were virtually identical for the self-ratings (mean rs ⫽ .78 and .69 for the BFI and PANAS, respectively) and the spouse ratings (mean rs ⫽ .77 and .69, respectively). Moreover, the correlations in our study generally were quite high, ranging from .67 to .83. As we predicted, all of our stability correlations exceeded those previously reported by Vaidya et al. (2002) for these same scales in a somewhat younger sample. Thus, our results again demonstrate that the stability of personality increases systematically with age (Roberts & DelVecchio, 2000). Our data also replicated previous research indicating that the BFI scales show higher stability correlations than the PANAS (see Vaidya et al., 2002; Watson, 2004). The findings for Extraversion and Neuroticism are particularly striking. BFI Extraversion was significantly more stable than the PANAS Positive Affect scale in both the self-ratings and the spouse ratings; BFI Neuroticism also had higher stability correlations than PANAS Negative Affect in both sets of ratings, although this difference was not significant in the spouse ratings. These differences in stability are noteworthy in light of the consistently strong correlations (which ranged from .52 to .79 in our study) between these BFI and PANAS scales. The stability gap between Extraversion and Positive Affect most likely reflects systematic differences in content between the two scales. The BFI Extraversion scale asks respondents to indicate
whether they are talkative, outgoing, and reserved (reverse-keyed); only two items ask directly about affect-related content (“is full of energy,” “generates a lot of enthusiasm”). More generally, Pytlik Zillig, Hemenover, and Dienstbier (2002) found that only 22.7% of the content in the BFI Extraversion scale was affective in nature. Thus, this gap likely reflects the fact that individual differences in sociability are more stable over time than positive emotionality. However, simple content-based considerations cannot explain the fact that BFI Neuroticism tends to be more stable than the PANAS Negative Affect scale. The content of the BFI Neuroticism scale is predominantly affective in nature; indeed, Pytlik Zillig et al. (2002) classified 75.3% of its item content as affective (see also Watson, 2004, for an examination of the stability of individual BFI Neuroticism and PANAS Negative Affect items). It therefore seems likely that noncontent considerations—such as instructional or format effects—are at least partly responsible for this evidence of differential stability. Watson (2004) tested this possibility by constructing a new instrument, the Temperament and Emotion Questionnaire (TEQ). The 60-item TEQ was created by taking individual mood descriptors from the Expanded Form of the PANAS (PANAS-X; Watson & Clark, 1999) and embedding them in complete sentences (which are rated using a 5-point agree/disagree format). For instance, the PANAS/ PANAS-X term irritable became the TEQ item “I have days on which I can be rather irritable.” Watson (2004) then compared retest correlations for the PANAS-X and TEQ negative affectivity scales across a 2-month interval. In three of five comparisons, the stability of the TEQ scale was significantly higher than that of its PANAS-X counterpart. These data demonstrate that stability correlations can be significantly influenced by changes in wording and format, even while maintaining the same basic item content. More generally, they suggest that stability researchers should carefully attend to issues such as wording and format when choosing measures to be used in studies of adult personality development.
Mean-Level Change Summary of findings. Our analyses of mean-level change revealed a striking discrepancy between the self-ratings and spouse Table 12 Hierarchical Multiple Regressions: Predicting Time 2 Marital Satisfaction From Time 2 Spouse-Rated Traits (Wives’ Ratings) Step 2 T1 trait Scale BFI Neuroticism Extraversion Openness Agreeableness Conscientiousness PANAS Negative Affect Positive Affect
Step 3 T2 trait
⌬R2
⌬R2

Final R
.010 .000 .005 .060** .026*
.020 .000 .028* .060** .018
⫺.207 .022 .251* .376** .196
.52 .49 .52 .60 .53
.014 .016
.028* .043**
⫺.241* .290**
.53 .54
Note. N ⫽ 146. T1 marital satisfaction was entered in Step 1 of the regression. T1 ⫽ Time 1; T2 ⫽ Time 2; BFI ⫽ Big Five Inventory; PANAS ⫽ Positive and Negative Affect Schedule. * p ⬍ .05. ** p ⬍ .01.
SELF-RATINGS VERSUS SPOUSE RATINGS OF STABILITY
ratings. Our self-report data were consistent with prediction and largely replicated previous findings in this area. Specifically, our participants reported significant increases in both conscientiousness and agreeableness, and significant declines in neuroticism/ negative affectivity. We did not observe any systematic change in extraversion, openness, and positive affectivity across this 2-year interval. Overall, therefore, our findings are consistent with the broader literature indicating that adult personality development is associated with increasing psychological maturity with age (see Caspi et al., 2005). However, the spouse ratings yielded a markedly different pattern and suggested very different conclusions about the nature of adult development. Specifically, these scores showed significant declines in agreeableness, conscientiousness, extraversion, and openness, a pattern that certainly would not be characterized as reflecting greater psychological maturity. Moreover, across the seven assessed scales, we obtained only one consistent finding: PANAS Positive Affect showed a small, nonsignificant decline in both sets of ratings. For reasons discussed earlier, we suspected that changes in marital satisfaction might provide at least a partial explanation for these discrepant results. We conducted correlational and hierarchical regression analyses to examine this possibility. We found that changes in marital satisfaction were significantly associated with changes in spouse-rated personality, with each predicting the other in a separate series of analyses; the effects were particularly strong for agreeableness and trait affectivity. Subsequent analyses revealed that changes in marital satisfaction essentially were unrelated to self-rated change in personality. Thus, our findings suggest that marital satisfaction played a key role in producing the discrepant findings presented in Tables 2 and 3. Given that marital satisfaction declined significantly across the study period, it makes sense that our spouse ratings tended to be more negative overall at Time 2. Explaining the findings. This, then, leads to a more basic set of questions: How can these findings be reconciled and integrated into an overall model of adult personality development? Is the nature of this development largely positive (as indicated by the self-ratings) or rather negative (as suggested by the spouse ratings)? Put differently, which rating source provides a more accurate picture of mean-level change in emerging adulthood? One possible explanation of our data is that the spouse raters were able to remain more objective and, thus, ultimately had better insight into the targets’ true personalities than the self-raters. In other words, our IMAP participants actually showed negative changes in their trait characteristics—including significant declines in agreeableness and conscientiousness—across the 2-year study interval. Because these changes are negative and socially undesirable, however, the self-raters understandably may have been reluctant to acknowledge them. In contrast, spouse raters were not as motivated to gloss over these negative developmental trends and so were able to provide a more accurate account of the targets’ true personalities at Time 2. This explanation is consistent with our hierarchical regression analyses that established that changes in spouse ratings predicted changes in marital satisfaction (see Tables 11 and 12); it also is supported by evidence establishing the existence of a selfenhancement bias in self-report data (John & Robins, 1993; Kwan, John, Kenny, Bond, & Robins, 2004; Paulhus, Harms, Bruce, &
971
Lysy, 2003). However, there are two significant problems with a simple self-enhancement explanation of our data. First, as we reviewed earlier, a large body of evidence— based on both selfreports and, to a lesser extent, observer ratings—indicates that personality development is associated with greater maturity (i.e., higher levels of conscientiousness and agreeableness, lower levels of neuroticism/negative affectivity) over time. The mean-level changes in our self-rating data obviously conformed much more closely to this typical pattern, and so they appear to be more credible than the results that emerged in the spouse ratings. Second, a self-enhancement explanation clearly cannot account for the “honeymoon effect” we observed at Time 1, such that the mean spouse ratings significantly exceeded the average self-ratings for Agreeableness, Extraversion, and Conscientiousness. In other words, the spouse ratings actually were more positive and socially desirable than the self-ratings at Time 1. In light of these considerations, we believe it is unlikely that our participants actually showed true negative changes in their trait characteristics from Time 1 to Time 2. A more likely explanation is that the spouse ratings were unduly positive during the honeymoon period of Time 1, and then declined to more realistic levels at Time 2. However, this then leads to the further question of what caused these changes in the spouse ratings over the course of our study. One possibility is that the actual day-to-day behavior of our participants changed significantly between Time 1 and Time 2. That is, our IMAP participants may have been on their “best behavior” during the honeymoon period of Time 1 and acted as more agreeable and conscientious—and less neurotic—than they really were; this would explain why the spouse ratings were more positive than the self-ratings at this initial assessment. As the marriage wore on and the honeymoon ended, however, the participants gradually reverted to their true baseline levels of behavior. If so, then this would naturally lead to declines in marital satisfaction and to less positive spouse ratings at Time 2. We suspect that behavioral changes of this type offer at least a partial explanation of our findings. However, these changes alone cannot explain why the spouse ratings were significantly more negative than the self-ratings (i.e., higher levels of Neuroticism and Negative Affect, lower levels of Conscientiousness and Openness) at Time 2. We believe that the observed decline in marital satisfaction played a key role in producing this Time 2 negativity in the spouse ratings. Although they know each other well, husbands still are likely to have significant gaps in their knowledge about their wives, and vice versa. We now have extensive evidence indicating that judges use various rating strategies or heuristics to fill in these informational gaps. Of particular relevance here, Watson et al. (2000a) reviewed evidence suggesting that relationship satisfaction represents a significant heuristic that can be used when rating romantic partners: That is, raters compensate for informational gaps by using their current level of satisfaction as a basis for making strongly evaluative inferences about the personalities of their partners. Thus, at Time 1, the participants were extremely satisfied with their marriages and so rated their spouses very positively (even more positively than their partners rated themselves). Because the spouses became significantly more dissatisfied with their marriages over time, however, they judged their partners more harshly at Time 2. This heuristic-based model also offers a parsimonious explanation for the well-established finding that marital satisfaction (a) is
972
WATSON AND HUMRICHOUSE
only weakly associated with the partner’s self-rated traits but (b) is much more substantially linked to the judge’s ratings of the partner’s characteristics (see Watson et al., 2000a); we replicated this pattern at both Time 1 and Time 2 (see Tables 7 and 8). It is noteworthy that these mono-rater correlations tend to be strong even when self– other agreement is poor and judges are rating low visibility traits that are difficult to observe in others. For example, Watson et al. (2000b) obtained relatively low self– other agreement correlations for ratings of trait affectivity in a sample of dating couples (rs ⫽ .22 and .33 on the PANAS Negative Affect and Positive Affect scales, respectively). Nevertheless, relationship satisfaction was strongly correlated with partner ratings of trait affectivity in this same sample, with coefficients ranging from 兩.40兩 to 兩.56兩 (see Watson et al., 2000a, Table 7). Thus, consistent with a heuristic-based explanation, we tend to see strong correlations between partner-rated personality and satisfaction even under conditions in which trait-related information is limited and the ratings are unlikely to be highly accurate. Consequently, we believe that this heuristic-based account also provides at least a partial explanation for our findings. We emphasize, however, that our data do not allow us to draw clear, general conclusions regarding the relative merits of these various explanations. We emphasize, moreover, that these models are not mutually exclusive, and they each may offer partial explanations for our findings. This is a crucial issue that needs to be investigated more thoroughly in future research. In order to weigh the relative merits of these models, it would be helpful to include a third rating source (e.g., friends’ ratings of both spouses) in future studies. It also would be highly informative to collect converging data using other assessment approaches (e.g., time sampling of trait-related behaviors and feelings). Finally, future research in this area should use longer, more reliable measures of marital satisfaction.
Individual-Level Change We also investigated change at the individual level. We began by examining the within-method correlations among the raw change scores in both the self-ratings and the spouse ratings (see Table 4). These data demonstrated that change scores on related traits were significantly correlated with one another. These results help to establish that these scores tap some systematic variance and do not simply reflect random measurement error. Having said that, however, we must emphasize that this does not necessarily mean that this systematic variance is valid. Among other things, it may reflect systematic measurement errors, such as transient error (e.g., Becker, 2000; Schmidt, Le, & Ilies, 2003). Transient error reflects the influence of time-limited factors, such as the current mood of the respondent. For instance, if some participants were more distressed and upset at Time 2 than at Time 1, this could have influenced their responses on both BFI Neuroticism and PANAS Negative Affect, thereby producing a positive correlation between the change scores for these two scales. Indeed, our subsequent analyses raised some significant concerns about the overall validity of change scores. We next examined the convergence between self- and spouse-rated change using these raw change scores. These findings offered mixed support for our predictions. We obtained significant convergent correlations for only two of seven scales in the overall sample: Conscientious-
ness (r ⫽ .20) and Neuroticism (r ⫽ .19). Moreover, the coefficients were consistently low in magnitude, ranging from ⫺.01 to .20, with a median value of only .07. Thus, our data revealed little convergence between change assessments across two different rating methods. Our subsequent analyses of RCI scores yielded the same basic conclusion: Again, we found little consistency in individual-level change across our two rating methods. Our results replicate those of McCrae (1993), who reported very low correlations between self- and spouse-rated changes on neuroticism, extraversion, and openness. It is noteworthy, moreover, that McCrae also found little convergence with change scores that were based on the ratings of two peers. It therefore appears that this problem may not be restricted to spouse ratings, but instead reflects a more general pattern. This poor convergence across methods is troubling and raises significant concerns about the meaningfulness of change assessed at the individual level. The available evidence remains quite limited, however, so we strongly encourage more extensive investigation of this issue in subsequent research. Paralleling our earlier discussion of mean-level change, we believe it would be particularly informative to assess individual-level change in personality across multiple methods (e.g., self-reports, spouse ratings, peer ratings, time sampling) in a reasonably large sample.
Strengths and Limitations This study contributes to the literature on personality development in emerging adulthood in several ways. First, we examined personality stability and change in a relatively large sample of young adults across a time span of approximately 2 years. Second, our participants were somewhat older than those typically investigated in studies of young adulthood. We therefore were able to investigate key developmental issues during the transitional period from “emerging adulthood” to full adulthood. Our data revealed that this critical developmental period is characterized by both very strong rank-order stability and significant mean-level change. Third, because our sample consisted of newlywed couples, we were able to examine stability and change across two different rating methods. Although these methods yielded virtually identical results in our examination of rank-order stability, they diverged sharply in our analyses of both mean-level change and individuallevel change. These data highlight the importance of multimethod assessment in studies of adult personality development. We therefore join others (e.g., McCrae, Terraciano, et al., 2005; Roberts et al., 2004) who recently have called for increased reliance on multisource data in this area. At the same time, we also must acknowledge two significant limitations of our study. First, we examined personality stability and change in a single sample across only two assessments. Accordingly, certain aspects of our results may reflect study-specific factors that may not generalize across other samples and occasions. Second, because we collected personality data from only two sources, we were unable to resolve the striking discrepancies that emerged in our analyses of the self- and spouse-rating data. As noted previously, we believe that it will be extremely informative for future studies to obtain relevant longitudinal data from multiple sources. An expanded design of this type would be invaluable in clarifying the true nature and course of adult personality development.
SELF-RATINGS VERSUS SPOUSE RATINGS OF STABILITY
References Arnett, J. J. (2000). Emerging adulthood: A theory of development from the late teens through the twenties. American Psychologist, 55, 469 – 480. Asendorpf, J. (1992). Beyond stability: Predicting inter-individual differences in intra-individual change. European Journal of Personality, 6, 103–117. Becker, G. (2000). How important is transient error in estimating reliability? Going beyond simulation studies. Psychological Methods, 5, 370 – 379. Benet-Martinez, V., & John, O. P. (1998). Los Cinco Grandes across cultures and ethnic groups: Multitrait multimethod analyses of the Big Five in Spanish and English. Journal of Personality and Social Psychology, 75, 729 –750. Caspi, A., Roberts, B. W., & Shiner, R. L. (2005). Personality development: Stability and change. Annual Review of Psychology, 56, 453– 484. Chmielewski, M., & Watson, D. (2006). Investigations of dependability: Comparisons of two-week versus two-month retest correlations for measures of personality and psychopathology. Manuscript in preparation. Clark, L. A., & Watson, D. (1999). Temperament: A new paradigm for trait psychology. In L. Pervin & O. P. John (Eds.), Handbook of personality: Theory and research (2nd ed., pp. 399 – 423). New York: Guilford Press. Costa, P. T., Jr., & McCrae, R. R. (1988). Personality in adulthood: A six-year longitudinal study of self-reports and spouse ratings on the NEO Personality Inventory. Journal of Personality and Social Psychology, 54, 853– 863. Costa, P. T., Jr., & McCrae, R. R. (1992). Trait psychology comes of age. In T. B. Sonderegger (Ed.), Nebraska Symposium on Motivation: Psychology and aging (pp. 169 –204). Lincoln: University of Nebraska Press. Costa, P. T., Jr., & McCrae, R. R. (1994). Set like plaster: Evidence for the stability of adult personality. In T. F. Heatherton & J. L. Weinberger (Eds.), Can personality change? (pp. 21– 40). Washington, DC: American Psychological Association. Digman, J. M. (1997). Higher-order factors of the Big Five. Journal of Personality and Social Psychology, 73, 1246 –1256. Fraley, R. C., & Roberts, B. W. (2004). Patterns of continuity: A dynamic model for conceptualizing the stability of individual differences in psychological constructs across the life course. Psychological Review, 112, 60 –74. Humrichouse, J., & Watson, D. (2006). Self-spouse ratings, trait visibility, and acquaintanceship: A longitudinal study of newlywed couples. Manuscript in preparation. John, O. P., & Robins, R. W. (1993). Determinants of interjudge agreement on personality traits: The Big Five domains, observability, evaluativeness, and the unique perspective of the self. Journal of Personality, 61, 521–551. John, O. P., & Srivastava, S. (1999). The Big Five trait taxonomy: History, measurement, and theoretical perspectives. In L. A. Pervin & O. P. John (Eds.), Handbook of personality (2nd ed.; pp. 102–138). New York: Guilford Press. Kashy, D. A., & Snyder, D. K. (1995). Measurement and data analytic issues in couples research. Psychological Assessment, 7, 338 –348. Kenny, D. A. (1995). The effect of nonindependence on significance testing in dyadic research. Personal Relationships, 2, 67–75. Kwan, V., John, O. P., Kenny, D. A., Bond, M. H., & Robins, R. W. (2004). Reconceptualizing individual differences in self-enhancement bias: An interpersonal approach. Psychological Review, 111, 94 –110. Locke, H., & Wallace, K. (1959). Short marital-adjustment and prediction tests: Their reliability and validity. Marriage and Family Living, 21, 251–255. Markon, K. E., Krueger, R. F., & Watson, D. (2005). Delineating the structure of normal and abnormal personality: An integrative hierarchi-
973
cal approach. Journal of Personality and Social Psychology, 88, 139 – 157. McCrae, R. R. (1993). Moderated analyses of longitudinal personality stability. Journal of Personality and Social Psychology, 65, 577–585. McCrae, R. R., & Costa, P. T., Jr. (1990). Personality in adulthood. New York: Guilford Press. McCrae, R. R., Costa, P. T., Jr., Hrebı´ckova´, M., Urba´nek, T., Martin, T. A., Oryol, V. E., et al. (2004). Age differences in personality traits across cultures: Self-report and observer perspectives. European Journal of Personality, 18, 143–157. McCrae, R. R., Terraciano, A., & Members of the Personality Profiles of Cultures Project (2005). Universal features of personality traits from the observer’s perspective: Data from 50 cultures. Journal of Personality and Social Psychology, 88, 547–561. Paulhus, D. L., Harms, P. D., Bruce, M. N., & Lysy, D. C. (2003). The over-claiming technique: Measuring self-enhancement independent of ability. Journal of Personality and Social Psychology, 84, 890 –904. Pytlik Zillig, L. M., Hemenover, S. H., & Dienstbier, R. A. (2002). What do we assess when we assess a Big 5 trait? A content analysis of the affective, behavioral, and cognitive processes represented in Big 5 personality inventories. Personality and Social Psychology Bulletin, 28, 847– 858. Roberts, B. W., Caspi, A., & Moffitt, T. E. (2001). The kids are alright: Growth and stability in personality development from adolescence to adulthood. Journal of Personality and Social Psychology, 81, 670 – 683. Roberts, B. W., Caspi, A., & Moffitt, T. E. (2003). Work experiences and personality development in young adulthood. Journal of Personality and Social Psychology, 84, 582–593. Roberts, B. W., & DelVecchio, W. F. (2000). The rank-order consistency of personality traits from childhood to old age: A quantitative review of longitudinal studies. Psychological Bulletin, 126, 3–25. Roberts, B. W., O’Donnell, M., & Robins, R. W. (2004). Goal and personality trait development in emerging adulthood. Journal of Personality and Social Psychology, 87, 541–550. Roberts, B. W., Robins, R. W., Trzesniewski, K., & Caspi, A. (2003). Personality trait development in adulthood. In J. Mortimer & M. Shanahan (Eds.), Handbook of the life course (pp. 579 –598). New York: Kluwer. Roberts, B. W., Walton, K. E., & Viechtbauer, W. (2006). Patterns of mean-level change in personality traits across the life course: A metaanalysis of longitudinal studies. Psychological Bulletin, 132, 1–25. Robins, R. W., Fraley, R. C., Roberts, B. W., & Trzesniewski, K. H. (2001). A longitudinal study of personality change in young adulthood. Journal of Personality, 69, 617– 640. Schmidt, F. L., Le, H., & Ilies, R. (2003). Beyond alpha: An empirical examination of the effects of different sources of measurement error on reliability estimates for measures of individual differences constructs. Psychological Methods, 8, 206 –224. Schuerger, J. M., Zarrella, K. L., & Hotz, A. S. (1989). Factors that influence the temporal stability of personality by questionnaire. Journal of Personality and Social Psychology, 56, 777–783. Srivastava, S., John, O. P., Gosling, S. D., & Potter, J. (2003). Development of personality in early and middle adulthood: Set like plaster or persistent change? Journal of Personality and Social Psychology, 84, 1041–1053. Vaidya, J. G., Gray, E. K., Haig, J., & Watson, D. (2002). On the temporal stability of personality: Evidence for differential stability and the role of life experiences. Journal of Personality and Social Psychology, 83, 1469 –1484. Watson, D. (2004). Stability versus change, dependability versus error: Issues in the assessment of personality over time. Journal of Research in Personality, 38, 319 –350. Watson, D., & Clark, L. A. (1999). The PANAS-X: Manual for the Positive and Negative Affect Schedule—Expanded Form. Retrieved May 26,
974
WATSON AND HUMRICHOUSE
2006 from University of Iowa, Department of Psychology Web site: www.psychology.uiowa.edu/faculty/watson/watson.html Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief measures of Positive and Negative Affect: The PANAS Scales. Journal of Personality and Social Psychology, 54, 1063–1070. Watson, D., Hubbard, B., & Wiese, D. (2000a). General traits of personality and affectivity as predictors of satisfaction in intimate relationships: Evidence from self- and partner-ratings. Journal of Personality, 68, 413– 449. Watson, D., Hubbard, B., & Wiese, D. (2000b). Self– other agreement in personality and affectivity: The role of acquaintanceship, trait visibility,
and assumed similarity. Journal of Personality and Social Psychology, 78, 546 –558. Watson, D., Klohnen, E. C., Casillas, A., Simms, E. N., Haig, J., & Berry, D. S. (2004). Match makers and deal breakers: Analyses of assortative mating in newlywed couples. Journal of Personality, 72, 1029 –1068.
Received September 12, 2005 Revision received May 26, 2006 Accepted June 1, 2006 䡲
Journal of Personality and Social Psychology 2006, Vol. 91, No. 5, 975–993
Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.5.975
Nostalgia: Content, Triggers, Functions Tim Wildschut and Constantine Sedikides
Jamie Arndt
University of Southampton
University of Missouri
Clay Routledge University of Southampton Seven methodologically diverse studies addressed 3 fundamental questions about nostalgia. Studies 1 and 2 examined the content of nostalgic experiences. Descriptions of nostalgic experiences typically featured the self as a protagonist in interactions with close others (e.g., friends) or in momentous events (e.g., weddings). Also, the descriptions contained more expressions of positive than negative affect and often depicted the redemption of negative life scenes by subsequent triumphs. Studies 3 and 4 examined triggers of nostalgia and revealed that nostalgia occurs in response to negative mood and the discrete affective state of loneliness. Studies 5, 6, and 7 investigated the functional utility of nostalgia and established that nostalgia bolsters social bonds, increases positive self-regard, and generates positive affect. These findings demarcate key landmarks in the hitherto uncharted research domain of nostalgia. Keywords: nostalgia, emotions, affect, self, relationships
anxiety, irregular heartbeat, anorexia, insomnia, and even smothering sensations (McCann, 1941). Hofer regarded nostalgia as “a cerebral disease” (p. 387) caused by “the quite continuous vibration of animal spirits through those fibers of the middle brain in which impressed traces of ideas of the Fatherland still cling” (p. 384). The physician J. J. Scheuchzer (1732), a contemporary of Hofer’s, proposed instead that nostalgia was due to “a sharp differential in atmospheric pressure causing excessive body pressurization, which in turn drove blood from the heart to the brain, thereby producing the observed affliction of sentiment” (cited in Davis, 1979, p. 2). Scheuchzer applied this theory to account for the supposedly high incidence of nostalgia among Swiss mercenaries who left their Alpine homes to fight on the plains of Europe. Finally, not content with either explanation, some military physicians proposed that nostalgia was largely confined to the Swiss because of the unremitting clanging of cowbells in the Alps, which inflicted damage upon the eardrum and brain (Davis, 1979). This view of nostalgia as a neurological affliction persisted throughout the 17th and 18th centuries. By the early 19th century, definitions of nostalgia had shifted. Nostalgia was no longer regarded as a neurological disorder but, instead, came to be considered a form of melancholia or depression (McCann, 1941; Rosen, 1975). Nostalgia remained relegated to the realm of psychological disorders for much of the 20th century. Scholars in the psychodynamic tradition described nostalgia as an “immigrant psychosis” (Frost, 1938, p. 801), a “mentally repressive compulsive disorder” (Fodor, 1950, p. 25), and “a regressive manifestation closely related to the issue of loss, grief, incomplete mourning, and, finally, depression” (Castelnuovo-Tedesco, 1980, p. 110). In part, this gloomy perspective can be attributed to the fact that nostalgia has long been equated with homesickness. It was only in the latter part of the 20th century that nostalgia acquired a separate conceptual status. Davis (1979), for instance, showed that college
Approximately 3 millennia ago, Homer (1921) composed his epic poem The Odyssey and with it created one of the most gripping literary accounts of nostalgia. The poem revolves around the adventures of Odysseus who, after emerging victoriously from the Trojan War, embarks on a quest to return to his homeland, the island of Ithaca, and reunite with his faithful wife, Penelope. This quest was to last 10 years, 7 of which were spent in the possessive arms of the seductive sea nymph Calypso. In an attempt to persuade Calypso to set him free, Odysseus confides to her, “Full well I acknowledge Prudent Penelope cannot compare with your stature of beauty, for she is only a mortal, and you are immortal and ageless. Nevertheless it is she whom I daily desire and pine for. Therefore I long for my home and to see the day of returning” (Homer, 1921, Book V, pp. 78 –79). The Greek words for return and suffering are nostos and algos, respectively. The literal meaning of nostalgia, then, is the suffering caused by the yearning to return to one’s place of origin.
A Brief History of Nostalgia The term nostalgia was actually introduced by the Swiss physician Johannes Hofer (1688/1934) to refer to the adverse psychological and physiological symptoms displayed by Swiss mercenaries who plied their trade on foreign shores. Hofer conceptualized nostalgia as a medical or neurological disease. Symptoms were thought to include persistent thinking of home, bouts of weeping,
Tim Wildschut, Constantine Sedikides, and Clay Routledge, School of Psychology, University of Southampton; Jamie Arndt, Department of Psychological Sciences, University of Missouri. We thank Denise Baden, David Clarke, Friederike Hinrichs, Ana Rios, and Maren Wolfram for their assistance in data collection and coding. Correspondence concerning this article should be addressed to Tim Wildschut, School of Psychology, University of Southampton, Southampton SO17 1BJ, England. E-mail:
[email protected] 975
976
WILDSCHUT, SEDIKIDES, ARNDT, AND ROUTLEDGE
students associated words like warm, old times, childhood, and yearning more frequently with nostalgia than with homesickness, suggesting that students could discriminate between these two concepts. Current dictionary definitions of homesickness and nostalgia also reflect their distinctness. The New Oxford Dictionary of English (1998) defines homesick as “experiencing a longing for one’s home during a period of absence from it” and nostalgia as “a sentimental longing for the past.” There is now a sizeable literature on homesickness, which concentrates on the psychological problems associated with transition to boarding school or university (Van Tilburg, Vingerhoets, & van Heck, 1996). Empirical research on nostalgia, on the other hand, remains scarce and largely confined to the field of advertising and consumer psychology (Holak & Havlena, 1998; Schindler & Holbrook, 2003). Focused mainly on accounting for the market success of certain consumer goods, research in this tradition has demonstrated how product styles (e.g., of music, motion pictures, and automobiles) that were popular during an individual’s youth influence the individual’s lifelong preferences. Although we do not mean to suggest that such findings are uninteresting or unimportant, we do believe that a broader perspective is required if one is fully to investigate and, ultimately, understand nostalgia and its postulated significance to the self (Davis, 1979), interpersonal relationships (Batcho, 1998), memory (Cavanaugh, 1989), and affect (H. A. Kaplan, 1987). Given that we found ourselves in largely uncharted territory, we perceived a need to address three fundamental questions about nostalgia. First, what is the content of nostalgic experiences (content question)? Second, what are the triggers of nostalgia (trigger question)? Third, what, if any, are the psychological functions of nostalgia (function question)? We addressed these questions in seven methodologically diverse studies. Studies 1 and 2 examined the content question. Studies 3 and 4 examined the trigger question. Finally, Studies 5, 6, and 7 examined the function question.
Studies 1 and 2: Content of Nostalgic Experiences Studies 1 and 2 sought to answer four questions about the content of nostalgic experiences. These related to the salience of the self in nostalgic experiences, the objects of nostalgia, the manner in which positive and negative affective states are juxtaposed in nostalgic experiences, and the affective signature of nostalgia.
Salience of the Self in Nostalgic Experience We take as our point of departure the idea that nostalgia refers to a personally experienced past. From the outset, then, we distinguish the case of personal nostalgia from other proposed forms of nostalgia such as organizational (Gabriel, 1993) or historical (Stern, 1992) nostalgia. We suggest that nostalgia is a prima facie selfrelevant emotion in the sense that the self is a salient protagonist in the nostalgic experience (Sedikides, Wildschut, & Baden, 2004).
Objects of Nostalgia We propose that nostalgia possesses an important social element. Qualitative descriptions of nostalgic experiences indicate that these experiences often involve interactions between the self and close others, such as family members, friends, and romantic
partners (Holak & Havlena, 1992). Although we expect close others to figure prominently in nostalgic experiences, other types of personal experience may also provide a fertile soil for nostalgia. Likely candidates include momentous events, such as anniversaries and births, and experiences involving specific settings, such as one’s hometown.
Redemption Versus Contamination in Nostalgic Experience Nostalgia pertains to a personally experienced past, but not all past experience evokes nostalgia. How can we delineate more precisely the domain of nostalgic experiences? One possibility relates to the manner in which affective states are juxtaposed in these experiences. Davis (1977) noted that, in those cases where the nostalgic experience contains negative elements, these “hurts, annoyances, disappointments, and irritations . . . are filtered forgivingly through an ‘it was all for the best’ attitude” (p. 418). Relevant to this point, McAdams and colleagues (McAdams, 2001; McAdams, Reynolds, Lewis, Patten, & Bowman, 2001) identified two narrative patterns or strategies that people use to give their life stories meaning and coherence. In a redemption sequence, the narrative progresses from a negative life scene to a positive or triumphant one. By contrast, in a contamination sequence, the narrative moves from an affectively positive life scene to an affectively negative one. McAdams et al. (2001) found that psychological well-being was positively correlated with redemption sequences in life narratives and negatively associated with contamination sequences. We propose, in keeping with Davis, that nostalgic experiences are more typically characterized by redemption than by contamination sequences.
Affective Signature of Nostalgia What is the affective signature of nostalgia? We distinguish three perspectives bearing on this question. These perspectives emphasize the positive, negative, and bittersweet affective correlates of nostalgia, respectively.
Positive Affect Davis (1979) defined nostalgia as a “positively toned evocation of a lived past” (p. 18) and argued that “the nostalgic . . . experience is infused with imputations of past beauty, pleasure, joy, satisfaction, goodness, happiness, love . . . . Nostalgic feeling is almost never infused with those sentiments we commonly think of as negative—for example, unhappiness, frustration, despair, hate, shame, and abuse” (p. 14). The point of view that nostalgia is associated with positive affect was shared by Batcho (1995), Gabriel (1993), Holak and Havlena (1998), and H. A. Kaplan (1987).
Negative Affect Other theorists, however, have highlighted the negative side of nostalgia. Ortony, Clore, and Collins (1988), for instance, viewed nostalgia as part of the negative subset of well-being emotions. Specifically, they categorized nostalgia under the distress and loss emotions. The affective signature of nostalgia is considered to be sadness or mourning about the past. Best and Nelson (1985), Hertz
NOSTALGIA
(1990), and Peters (1985) also endorsed the view that nostalgia involves the wounding realization that some desirable aspect of one’s past is irredeemably lost.
Mixed Affect Still, a third category of theorists emphasize the affectively mixed or bittersweet nature of nostalgia. Johnson-Laird and Oatley (1989) defined nostalgia as positive emotion with tones of loss. They viewed nostalgia as a complex emotion, characterized by high-level cognitive appraisal and propositional content. In their opinion, nostalgia is a happiness-related emotion, yet, at the same time, it is thought to invoke sadness because of the realization that some desirable aspects of the past are out of reach. A similar view was endorsed by Werman (1977), who proposed that nostalgia involves “wistful pleasure, a joy tinged with sadness” (p. 393).
977
the authors were brief and read “True personal experience, reflective, insightful.” The narratives were between 1,000 and 1,500 words in length. No author submitted more than one story. Twenty-nine authors were female, and 13 were male. No information regarding author age was available, but in some cases age could be estimated on the basis of the content of narratives and photographs of the authors. One of the youngest authors was in his early 20s, whereas one of the oldest authors was in his late 80s and described an experience from 1931.
Coding Two trained judges with experience in qualitative data analysis independently coded the autobiographical narratives. Coding items were selected a priori and refined following inspection of a random sample of 10 narratives. These items are described in the Results section together with the findings.
Study 1
Results
Study 1 is a content analysis of autobiographical narratives published in the periodical Nostalgia and serves as a preliminary investigation into the content of nostalgic experience. Like any psychological methodology, the study of autobiographical narratives has both strengths and limitations. It is, for example, difficult to assess the extent to which autobiographical narratives are free of systematic bias (e.g., because of selective encoding and/or retrieval). On the other hand, the subjectivity of autobiographical narratives can be construed as an asset. These narratives offer a window into the individual’s personal view of their everyday experiences and feelings. As such, narratives complement experimental methods that often involve studying behavior in a controlled laboratory environment (Baumeister, Wotman, & Stillwell, 1993). Indeed, autobiographical narratives have proved to be a valuable source of information regarding a wide range emotional states, including inspiration (Thrash & Elliot, 2003, 2004), anger (Baumeister, Stillwell, & Wotman, 1990), unrequited love (Baumeister, Wotman, & Stillwell, 1993), shame and guilt (Tangney, 1991), and hurt feelings (Leary, Springer, Negel, Ansell, & Evans, 1998).
Interrater reliability, as assessed by Cohen’s kappa ranged from .70 to .88 for the coded measures. Judges resolved remaining disagreements through discussion.
Method Sample Forty-two autobiographical narratives were retrieved from issues 24 –27 of Nostalgia, which appeared throughout 1998 and 1999. Instructions to
Salience of the Self Table 1 lists the four categories used to code salience of the self, the proportion of narratives coded into each category, and a brief example for each category. Paired comparisons (see Table 1) revealed that the self figured more frequently in a major role than in any of the three other roles (i.e., sole actor, minor role, or outside observer). Frequencies for these other roles did not differ significantly. The finding that the self was a salient protagonist in almost all narratives is consistent with the idea that nostalgia is a self-relevant emotion. The finding that the self rarely figured as either an outside observer or as sole actor reinforces the idea that nostalgia has important social facets.
Object of Nostalgia Table 2 lists the seven categories used to code objects of nostalgia, the proportion of narratives coded into each category, and a brief example for each category. The most common objects of nostalgia were persons (33%). Paired comparisons (see Table 2) revealed that persons were more frequently the object of nostalgia than all other objects except momentous events and, somewhat
Table 1 Salience of Self: Coding Categories, Proportions of Narratives Coded Into Categories, and Category Examples—Studies 1 and 2 Proportion Category
Study 1
Study 2
Example
Sole actor
.05b
.05b
Major role
.88a
.86a
Minor role
.05b
.08b
Outside observer
.04b
.01c
“Prior to making the phone call I was really nervous and it took me a while to make the call. I even wrote down my name on a piece of paper because I was so muddled. The relief was an amazing feeling. It was a massive weight off my shoulders.” “I felt like I was really important to him and that no one else was as close. We had our own sort of ‘code’ and would talk to each other so no one else knew what we were saying.” “At the end of the ceremony, she and her new husband each lit a candle and then they lit one together. It was very symbolic, and during the final hymn I cried.” “One night as we came home from work, my husband caught a glimpse of the cat. Amazingly, as we realize now, he managed to coax her to him and he brought her inside.”
Note. Within columns, proportions with different subscripts differ significantly ( p ⬍ .05).
WILDSCHUT, SEDIKIDES, ARNDT, AND ROUTLEDGE
978
Table 2 Objects of Nostalgia: Coding Categories, Proportions of Narratives Coded Into Categories, and Category Examples—Studies 1 and 2 Proportion Category
Study 1
Study 2
Example
Persons Momentous events
.33a .21a,b
.28a,b .34a
Settings Periods in life
.10b,c,d .02d
.19b,c .16c
Animals
.17a,b,c
.01d
Tangibles
.12b,c,d
.01d
Past selves
.05c,d
.01d
“A smile crinkled the corners of my father’s eyes. I hadn’t seen him smile in a long time.” “I handed the baton over to the third leg and watched as our team managed to achieve first place. The excitement was amazing.” “It was like two opposites, this amazing force of water hitting this calm, serene lake.” “There was hardly any Uni. work and it’s this that makes me think: those were the good old days” “My whole family went down to the yard and I groomed her (my loan horse) one last time before the vet came.” “Since there was only one coat and many young girls, they held a drawing for it . . . The coat was soft, plush and beautiful.” “The dress had made me feel like a princess one night long ago.”
Note. Within columns, proportions with different subscripts differ significantly ( p ⬍ .05).
surprisingly, animals. Momentous events were more frequently the object of nostalgia than past selves and periods in life, and animals were more frequently the object of nostalgia than periods in life only.
Redemption Versus Contamination Redemption and contamination sequences were treated as mutually exclusive, and judges made a single judgment as to whether a given narrative was characterized by a redemption sequence (i.e., negative progresses to positive), a contamination sequence (i.e., positive progresses to negative), or neither. Table 3 presents the proportion of narratives coded into each category and a brief example for each category. Paired comparisons (see Table 3) revealed that redemption sequences (67%) were significantly more prevalent than contamination sequences (29%).
Affective Signature Feelings expressed in the narratives were coded as a proxy for the affective signature of nostalgia. Judges rated on a 5-point scale the extent to which each of 20 feelings was expressed in the
narratives (1 ⫽ not at all, 5 ⫽ extremely). The feelings were taken from the Positive and Negative Affect Scheme (PANAS; Watson, Clark, & Tellegen, 1988). Five feelings (attentive, interested, alert, nervous, jittery) were omitted from our final analysis, because they proved difficult to code reliably. The average Spearman–Brown corrected interrater correlation across the final 15 feelings was .68 (range ⫽ .35–.95). We created composite measures of positive (␣ ⫽ .81) and negative (␣ ⫽ .84) affect by first averaging across judges and then averaging across positive and negative feelings, respectively. A paired comparison indicated that the narratives were richer in expressions of positive (M ⫽ 3.12) than negative (M ⫽ 1.25) affect, F(1, 41) ⫽ 147.62, p ⬍ .001.
Discussion This initial study of nostalgia paints a picture of a positively toned and self-relevant emotion that is often associated with the recall of experiences involving interactions with important others or of momentous life events. Although most narratives contained negative as well as positive elements, these elements were often juxtaposed so as to form a redemption sequence—a narrative
Table 3 Redemption Versus Contamination: Coding Categories, Proportions of Narratives Coded Into Categories, and Category Examples— Studies 1 and 2 Proportion Category
Study 1
Study 2
Redemption sequence
.67a
.76a
Contamination sequence
.29b
.15b
Neither
.05c
.09b
Example “My Nan died that weekend and even though it was awful, it was a type of relief for my Nan and us. When I look back at this in my mind, I feel so proud of my Mum and the way she coped, it showed her immense love and devotion to her own mother.” “Playing with my granddad in the back garden on the grass . . . The flowers were all in bloom and there was a large jug of juice on the patio table. Shortly after this event my granddad died. I was never allowed to go to the funeral (too young) and have felt like I have never said goodbye.” “After opening the presents dad had to go to bed as he was feeling ill. It seemed such a shame as I knew he really wanted to enjoy what I know now was to be his last Christmas. We all had a lovely day and that Christmas will always be very special to us all.”
Note. Within columns, proportions with different subscripts differ significantly ( p ⬍ .05).
NOSTALGIA
pattern that progresses from a negative to a positive or triumphant life scene. This finding may explain why, despite the descriptions of disappointments and losses that they contained, the overall affective signature of the nostalgic narratives was predominantly— albeit not purely—positive. Shakespeare (1609/1996) sublimely captured this intricate pattern of relationships between nostalgia, redemption, and affect in his Sonnet 30: When to the sessions of sweet silent thought I summon up remembrance of things past, I sigh the lack of many a thing I sought, And with old woes new wail my dear time’s waste; ... But if the while I think on thee, dear friend, All losses are restor’d and sorrows end. (p. 47)
Study 2 The objectives of Study 2 were threefold. The first objective was to provide a conceptual replication of the key Study 1 findings. To this end, we extended our investigation in three ways. First, we used a sample that differed from the Study 1 sample in several respects. Whereas the Study 1 sample consisted of U.S. nationals who submitted their narratives to the periodical Nostalgia, the Study 2 sample consisted of British nationals whose narratives were solicited. Furthermore, whereas in Study 1 we used a community sample characterized by a wide age range, in Study 2 we used an undergraduate sample characterized by a much narrower age range. The use of such diverse samples allows us to assess the generality of our findings. Second, we gave participants more detailed instructions. Whereas authors in Study 1 were instructed to write about “true personal experience,” Study 2 participants were asked specifically to write about a nostalgic experience, thus sharpening the focus of our inquiry. Furthermore, in Study 2 we asked participants to write specifically about the feelings they experienced as a result of remembering the nostalgic event. This allowed us to examine more directly than in Study 1 the affective signature of nostalgia. Third, we used multiple converging methodologies. Whereas Study 1 relied exclusively on content analysis, in Study 2 we also administered a series of self-report measures that were intended to supplement the content analysis. The second objective of Study 2 was to assess the frequency with which nostalgia is typically experienced. Is nostalgia an esoteric experience or is it something that most persons experience on a regular basis? Although Boym (2002) argues, in her recent literary and cultural treatise of nostalgia, that it is an emotion experienced by almost all adults, there is little empirical evidence to speak to this claim. The third objective of Study 2 was to conduct a preliminary investigation of the triggers and functions of nostalgia. We therefore solicited participants’ descriptions of circumstances that evoke nostalgia and of nostalgia’s desirable and undesirable features.
Method Participants and Procedure Participants were 172 University of Southampton undergraduate students (148 women, 23 men, 1 of undeclared gender) who received course
979
credit. Materials were administered in sessions ranging in size from 1 to 8 persons. Participants were seated at small desks separated by partitions and completed the materials anonymously at their own pace. Debriefing concluded the testing session.
Materials Nostalgic narrative. Materials were presented in a single printed booklet. Instructions on the cover sheet read as follows: Please think of a past event in your life that has personal meaning for you. This should be an event that you think about in a nostalgic way. Specifically, please try to think of an important part of your past (e.g., event or episode) that makes you feel most nostalgic. Please bring this nostalgic experience to mind and think it through. Take a few minutes to think about your nostalgic experience. Participants were instructed to write about the nostalgic experience “in all its vivid detail” and were encouraged to “be as detailed, thorough, and descriptive” as possible. The narratives were coded by the same two judges and using the same coding scheme as in Study 1. Affective signature of nostalgia. After completing the narratives, written instructions prompted participants to “articulate as well as you can the emotions and feelings that you are experiencing right now, due to remembering this nostalgic experience.” Once participants completed the description of their emotions and feelings, they were asked to complete the PANAS. Participants were instructed to indicate how they felt as a result of having remembered the nostalgic experience by rating the PANAS items on a 5-point scale (1 ⫽ not at all, 5 ⫽ extremely). Triggers of nostalgia. On the next page of the booklet, participants received written instructions prompting them to give a detailed description of the circumstances that trigger nostalgia. Development of coding categories was aided by descriptions of triggers collected in a pilot sample (N ⫽ 18). These coding categories are described in the Results section together with the findings. Desirable and undesirable features of nostalgia. Participants were then asked to list as many general desirable and undesirable features of nostalgia as possible (counterbalanced for order). After completing this task, participants were instructed to review their lists and rate the desirability of each feature on a 7-point scale (⫺3 ⫽ not at all desirable, 3 ⫽ extremely desirable). Frequency of nostalgia. Finally, participants indicated how often they brought to mind nostalgic experiences by checking one of the following seven options: at least once a day, three to four times a week, approximately twice a week, approximately once a week, once or twice a month, once every couple of months, and once or twice a year.
Results Given that there were no significant gender differences on any of the dependent measures, we omitted this variable from the analyses reported below. Inter-rater reliability, as assessed by Cohen’s kappa, ranged from .75 to .82. Judges resolved remaining disagreements through discussion.
Content of Nostalgic Experience Salience of the self. To facilitate a comparison with Study 1 findings, we present results for coded salience of the self in the rightmost column of Table 1. As in Study 1, the self was a central character in almost all narratives and was rarely isolated (i.e., figured as sole actor or outside observer). Paired comparisons (see Table 1) revealed that the self figured more frequently in a major role than in any of the three other roles (i.e., sole actor, minor role,
980
WILDSCHUT, SEDIKIDES, ARNDT, AND ROUTLEDGE
or outside observer). The self figured less frequently as outside observer than in a minor role or as sole actor. Object of nostalgia. Results pertaining to the object of nostalgia are presented in the rightmost column of Table 2. As in Study 1, the two most common objects of nostalgia were persons (28%) and momentous events (34%). These two objects again played a role in the majority of narratives. Paired comparisons (see Table 2) revealed that momentous events were more frequently the object of nostalgia than all other objects except persons. Persons, in turn, were more frequently the object of nostalgia than all other objects except settings and periods in life. Note, finally, that momentous events often involved the presence of close others, such that judges sometimes had difficulty distinguishing these categories. Redemption versus contamination. Results pertaining to the prevalence of redemption and contamination sequences are presented in the rightmost column of Table 3. As in Study 1, redemption sequences (76%) were significantly more prevalent than contamination sequences (15%). Affective signature: Coded affect. After participants completed the narrative, they were instructed to describe how writing about the nostalgic experience made them feel. Judges rated the extent to which participants expressed the 20 PANAS feelings (1 ⫽ not at all, 5 ⫽ extremely). The average Spearman–Brown corrected interrater correlation was .87 (range ⫽ .47–.97). We created composite measures of coded positive (␣ ⫽ .86) and negative (␣ ⫽ .82) affect by first averaging across judges and then averaging across the positive and negative feelings, respectively. Participants expressed significantly more positive (M ⫽ 2.37) than negative (M ⫽ 1.37) affect, F(1, 171) ⫽ 196.56, p ⬍ .001. Affective signature: Self-reported affect. After participants described how writing about the nostalgic event made them feel, they completed the 20-item PANAS. Reliability alphas for self-report measures of positive and negative affect were .89 and .85, respectively. Participants reported more positive (M ⫽ 3.06) than negative (M ⫽ 1.53) affect, F(1, 171) ⫽ 294.61, p ⬍ .001. Scale means suggest that, although positive affect exceeds negative affect by a wide margin, nostalgia gives rise to gentle contentment rather than exuberant exaltation. This corroborates the results for coded affect described above and obtained in Study 1.1 Ambivalence. We drew on the attitude-ambivalence literature (see Eagly & Chaiken, 1993, for a review) to assess the degree to which participants reported ambivalent affect after completing the nostalgic narrative. We calculated ambivalence based on the selfreport measures because we believe these provide the most direct reading of participants’ affective responses. Following K. J. Kaplan (1972), ambivalence was defined as A ⫽ PA ⫹ NA ⫺ 兩PA ⫺ NA兩. Here, A refers to ambivalence, PA to positive affect, and NA to negative affect. We subtracted a constant of 1 from PA and NA, so that the possible range of A becomes 0⫺8, with 0 indicating the absence of ambivalence. The mean for A was 0.90 (SD ⫽ 0.94), indicating that nostalgia evoked only mild affective ambivalence. In light of these findings, Werman’s (1977) characterization of nostalgia as “a joy tinged with sadness” (p. 393) seems particularly fitting.2
Frequency of Nostalgia Sixteen percent of participants indicated that they experienced nostalgia “at least once a day,” 26% that they experienced nostalgia “three or four times a week” (mode), 19% that they experienced it “approximately twice a week” (median), and 18% that they experienced it “approximately once a week.” Thus, 79% of participants indicated that they experienced nostalgia once a week or more. A further 17% of participants indicated that they experienced nostalgia “once or twice a month,” and only 4% indicated that they experienced nostalgia less frequently than that (2% each for “once every couple of months” and “once or twice a year”). For the vast majority of participants, then, nostalgia is a common experience.
Triggers of Nostalgia: Preliminary Findings Participants provided detailed descriptions of the circumstances under which they wax nostalgic. The coding categories, a brief description of the categories, and the proportion of descriptions coded into each category are presented in Table 4. The most common trigger of nostalgia was negative affect (38%). Paired comparisons (see Table 4) revealed that negative affect was reported more frequently than any other trigger of nostalgia. Two other common triggers—sensory inputs and social interaction— did not differ significantly from each other but were both significantly more prevalent than all less common triggers. Given the relative prominence of negative affect as a trigger of nostalgia, we examined more closely descriptions coded into this category. We made a distinction between discrete negative affective states (e.g., lonely, scared) and generalized affective states often referred to as negative mood (e.g., sad, depressed). In contrast to more discrete affective states, which “arise from appraisals of specific actual or contemplated states of the world” (Bodenhausen, Sheppard, & Kramer, 1994, p. 46), mood often lacks a clearly delineated referent or antecedent (Schwarz & Clore, 1988). Some participants mentioned both discrete and generalized negative affective states (e.g., “If I ever feel lonely or sad, I tend to think of my friends or family who I haven’t seen for a long time”), therefore the two categories were not treated as being mutually exclusive. Of those who listed negative affect as a trigger of 1 Correlations between coded and self-reported affect were .36 for positive affect and .72 for negative affect ( ps ⬍ . 001). The relatively strong correlation for negative affect is due in particular to between-measure agreement at low levels of negative affect. This may reflect the relative ease of coding accurately for the absence (or near absence) of negative affect. 2 We selected the measure proposed by K. J. Kaplan (1972) because it is parsimonious, yields a score that is readily interpretable, and possesses construct validity (Lipkus, Green, Feaganes, & Sedikides, 2001). Various alternative operationalizations of ambivalence have been proposed. Abelson, Kinder, Peters, and Fiske (1982), for instance, proposed that ambivalence may be indicated by low correlations between positive and negative affect. In our data, the correlation between coded positive and negative affect was ⫺.17 and the correlation between self-reported positive and negative affect was ⫺.30. Although these small-to-moderate negative correlations might be seen as a sign of ambivalence, this interpretation is clouded by restriction of range in both coded and self-reported negative affect.
NOSTALGIA
981
Table 4 Triggers of Nostalgia: Categories, Proportions of Narratives Coded Into Categories, and Category Examples—Study 2 Category
Proportion
Example
Negative affect Social interactions Sensory inputs Tangibles
.38a .24b .19b .09c
Similar events Inertia Positive affect Anniversaries Settings
.03c .03c .03c .02c .01c
“Generally I think about nostalgic experiences when things are not going very well—lonely or depressed.” “Meeting up with people who were there and discussing what happened and laughing/crying about it.” “I find that some of the strongest triggers are smells and music.” “Anything that reminds me of my nostalgic experiences, i.e. my bridesmaid dress, will bring up emotions and memories.” “I usually think of nostalgic experiences when something similar happens and I say ‘remember the time when . . .’” “If I have a lot of time to sit and think, like on a long journey, I may start to think of nostalgic memories.” “They usually come to mind when I am feeling happy. They remind me of the good times.” “The days could be my grandfather’s birthday or my grandparents’ wedding anniversaries.” “Whenever I go back to my home town, memories come flooding back of that period of my life.”
Note. Within columns, proportions with different subscripts differ significantly ( p ⬍ .05).
nostalgia, 78% referred to negative mood, and 58% referred to discrete negative affective states. Within the latter category, 59% of participants referred to loneliness, making it by far the most frequently mentioned discrete affective state.
Functions of Nostalgia: Preliminary Findings Participants listed as many desirable and undesirable features of nostalgia as possible and then rated the desirability or undesirability of each feature. Participants listed significantly more desirable (M ⫽ 5.60) than undesirable (M ⫽ 4.16) features, F(1, 171) ⫽ 34.33, p ⬍ .001. To compare the rated desirability of the desirable features to the rated undesirability of the undesirable features, we folded the desirability scale by reversing the sign of negative ratings (e.g., ⫺3 became ⫹3). Desirable features were rated as being more desirable (M ⫽ 2.46) than undesirable features were rated as being undesirable (M ⫽ 2.02), F(1, 156) ⫽ 59.91, p ⬍ .001.3 Thus, not only did participants list more desirable than undesirable features, but the desirable features were also regarded as being of greater consequence. Next, we examined which desirable and undesirable features of nostalgia were mentioned most frequently. A research assistant transcribed all 1,675 features listed and then sorted synonymous features into groups.4 This resulted in 34 groups of synonymous desirable features and 22 groups of synonymous undesirable features. From these groups of synonyms, we then distilled five broad categories of desirable features and five broad categories of undesirable features by assembling groups of synonyms that we judged to be conceptually related (see Table 5). Two independent judges then used this coding scheme to categorize all 1,675 features listed. For each broad category, we counted the number of listed features it comprised and expressed this number as a proportion of the number of features listed. We did this separately for desirable features (see column 3 of Table 5), undesirable features (see column 4), and all features combined (see column 5). The three most prominent categories of desirable features were positive affect, social bonds, and self-regard. Sixty-seven percent of all desirable features were captured by these broad categories. Additional desirable features of nostalgia related to its perceived capacity to conserve positive memories and to promote personal growth. Although participants listed fewer undesirable than desirable features of nostalgia, almost all participants (95%) could think of
at least one undesirable feature. Most prominent by far was negative affect. Furthermore, the number of desirable and undesirable features listed were positively correlated, r(172) ⫽ .27, p ⬍ .01. These findings suggest that, rather than evaluating nostalgia in predominantly positive or negative terms, participants showed considerable nuance—acknowledging simultaneously desirable and undesirable features of nostalgia. They also underscore a vital point: Nostalgia, despite its predominantly positive affective signature, is not a purely hedonic experience.
Discussion Study 2 findings corroborated the preliminary description of nostalgia that emerged from Study 1. In both studies, nostalgia was associated with memories in which the self figured prominently and that typically related to interactions with important others or to momentous events. Nostalgic narratives often contained descriptions of disappointments and losses but, in the vast majority of cases, these negative life scenes were redeemed or mitigated by subsequent successes or triumphs over adversity. Also, although nostalgic narratives did not always tell happy stories (see Table 3), they evoked considerably more positive than negative affect, with little trace of ambivalence. This between-study consistency is particularly noteworthy bearing in mind that (a) the two studies used samples that differed in terms of nationality, age range, and recruitment method; (b) participants in Study 2 but not those in Study 1 were requested explicitly to write about nostalgic experiences; and (c) results for self-report measures of affect collected in Study 2 converged with results from the content analysis of narratives in both studies. Results further indicated that, for the vast majority of participants, nostalgia is a common experience. Just under 80% of participants indicated that they experience nostalgia at least once a week, and close to half of all participants (42%) indicated that they experience nostalgia at least three or four times a week. These 3 Degrees of freedom are reduced by 15 because some participants did not rate the desirability of the listed features. 4 This number is lower than that obtained by multiplying the number of participants by the average number of features listed by each participant (172 ⫻ 9.76 ⫽ 1,679). The discrepancy arises because one booklet was lost before listed features were coded.
WILDSCHUT, SEDIKIDES, ARNDT, AND ROUTLEDGE
982
Table 5 Proportions of Desirable and Undesirable Features of Nostalgia—Study 2 Proportion of listed features Category
Example
n
Desirable
315 192 139 98 45 171 960
.33 .20 .14 .10 .05 .18 1.00
Undesirable
All
Desirable features Positive affect Social bonds Self-regard Positive memories Growth Other Total desirable
“Being really happy” “Feeling loved” “High self-esteem” “Remember fun times” “Helps develop the person I am”
.19 .11 .08 .06 .03 .10 .57
Undesirable features Negative affect Loneliness Loss Rumination Regret Other Total undesirable
“Sadness” “Makes me feel alone” “Feel loss” “Think of the past too much” “Reminds me of things I regret”
Grand total
findings show that nostalgia is not an esoteric phenomenon but, rather, a strand in the fabric of everyday life. Study 2 also provided insight into the triggers of nostalgia. Chief among these was negative affect. The high frequency with which this trigger was listed is consistent with Davis’s (1979) notion that nostalgia “occurs in the context of present fears, discontents, anxieties, and uncertainties” (p. 34) and suggests that participants may retrieve nostalgic memories in an attempt to counteract negative affect. Consistent with the latter idea, Josephson, Singer, and Salovey (1996) found that participants in a sad mood condition who retrieved positive autobiographical memories frequently described this as an attempt at mood repair. Other common triggers were social interaction (e.g., conversations with friends) and sensory input (e.g., smells). The role of social interaction underscores the interpersonal aspects of nostalgia. Not only are important others often the object of nostalgia, but frequently they are also a trigger of nostalgia. In the words of one astute participant, nostalgia often arises from “being in the presence of the people concerned.” Such social sharing of nostalgic episodes may help to maintain their accessibility (Walker, Skowronski, & Thompson, 2003). In addition, the role of sensory input is consistent with findings that tastes and odors can trigger vivid, affect-laden memories (i.e., the Proust phenomenon; Chu & Downes, 2000). Finally, Study 2 offered initial insights into the functions of nostalgia. Chief among the perceived benefits of nostalgia were its capacity to generate positive affect, bolster social bonds, and increase positive self-regard.
Studies 3 and 4: Triggers of Nostalgia In Studies 3 and 4, we sought to examine in more detail the notion that nostalgia occurs in reaction to negative affect. By focusing on negative affect, we do not mean to suggest that other
284 88 82 44 34 183 715 1,675
.40 .12 .11 .06 .05 .26 1.00
.17 .05 .05 .03 .02 .11 .43 1.00
triggers of nostalgia—such as social interaction or sensory input— do not merit further investigation or to claim that negative affect is always the primary trigger of nostalgia. Study 2 revealed, however, that negative affect is widely considered to be an important trigger of nostalgia, and it was this finding that provided the initial impetus for the present studies. Although there exists, to the best of our knowledge, no research on the link between negative affect and nostalgia, a sizeable literature on the link between affect and self-relevant cognitions has emerged. This literature generally supports the idea that negative affect is associated with negative self-relevant cognitions, including the retrieval of negative autobiographical memories (i.e., mood congruency; for reviews, see: Sedikides, 1992; Sedikides & Green, 2001; Rusting, 1998). There is, however, some evidence to suggest that negative affect can also increase the retrieval and accessibility of positive autobiographical memories under certain circumstances (i.e., mood incongruency). Research documenting a link between negative affect and retrieval of positive autobiographical memories has typically involved asking participants to write brief accounts of events that happened when they were in high school. The dependent variable in these studies was the valence (positive vs. negative) of the autobiographical narratives as rated by participants, independent judges, or both. Results indicated that negative affect can increase retrieval of positive autobiographical memories but only when self-esteem is high (Smith & Petty, 1995), when persons acknowledge rather than attempt to repress the negative affect (McFarland & Buehler, 1997), when persons are unaware of the relevance of their moods to the experiment (Parrott & Sabini, 1990), or when persons both believe that they will be successful in their efforts to regulate negative moods and engage in positive reappraisal of the mood-inducing event (Rusting & DeHart, 2000). Research docu-
NOSTALGIA
menting a link between negative affect and accessibility of positive autobiographical narratives has involved cued recall tasks. Memory accessibility was operationalized as the time elapsed between cue (e.g., “a time in your life when you were particularly happy”) and recall of the event. Results revealed that negative affect can increase the accessibility of positive autobiographical memories but only for persons who are inclined to repress rather than acknowledge negative affect (Boden & Baumeister, 1997), or for nondysphoric persons (Joormann & Siemer, 2004). The present studies share certain similarities with those referenced above. For instance, they share the assumption that persons may draw on certain aspects of their past to counteract negative affect. Still, there are also some important distinctions. Past research has focused almost exclusively on the valence of autobiographical memories. Findings from Studies 1 and 2 indicate, however, that it would be an oversimplification to regard nostalgic memories as either entirely positive or negative. Nostalgic narratives often contained both negative and positive ingredients (usually in that order; see the redemption vs. contamination findings of Studies 1 and 2). Furthermore, although Study 2 participants listed more desirable than undesirable features of nostalgia, almost all participants could think of some undesirable features. Another important distinction is that previous research was concerned with autobiographical memory for a narrowly delineated period of life (e.g., one’s high school years) or domain of skills (e.g., social skills). We recognize that, when research is concerned with the valence of autobiographical memory, it is reasonable to define clearly a specific period of life to which the autobiographical memory should pertain to facilitate betweencondition comparisons and eliminate potential sources of error variance. When the focus is on nostalgia, however, this methodological practice seems unnecessarily constraining. In Studies 3 and 4, we therefore focused on the relation between negative affect and feelings of nostalgia for a broad range of aspects of one’s past.
Study 3 We found in Study 2 that a majority of participants who listed negative affect as a trigger of nostalgia described diffuse affective states often referred to as negative mood (e.g., down, sad, depressed). This provides one rationale for exploring first the effect of negative mood on nostalgia. There is also a strong theoretical rationale. Research has uncovered a wealth of evidence that people respond to negative mood with a wide array of mood-regulation strategies (Larsen & Prizmic, 2004). Confirmation of the postulated effect of negative mood on nostalgia would raise the interesting possibility that nostalgia can serve to counteract negative mood. The key objective of Study 3 was to examine whether participants in a negative mood state experience more nostalgia than those in a neutral mood state. An additional objective of Study 3 related to the role of positive mood. We examined the possibility that nostalgia is triggered by both negative and positive mood states (i.e., deviations from neutral in either direction). In this case, participants in either a negative or positive mood state should experience more nostalgia than neutral mood participants.
983 Method
Participants and Design Participants were 62 female undergraduate volunteers enrolled at the University of Southampton. They were randomly assigned to either the negative, neutral, or positive mood condition.
Materials and Procedure Participants received a booklet containing the mood manipulation and dependent measures. In the negative mood condition, they read a news story based on the tsunami that struck coastal regions in Asia and Africa in December 2004. In the neutral mood condition, participants read a news story based on the January 2005 landing of the Huygens probe on Titan, one of Saturn’s moons. In the positive mood condition, participants read a news story based on the November 2004 birth of a polar bear in the Detroit Zoo (we substituted “London Zoo” for “Detroit Zoo”). Participants in all conditions were then instructed to write down three to five keywords that captured their “emotional response to this event” and to think about how the event made them feel. After approximately 2 min, participants completed measures of positive and negative affect, and of nostalgia. Manipulation check. The manipulation check comprised 12 items, 6 each to assess positive (e.g., “happy,” “content”) and negative (e.g., “sad,” “depressed”) affect. The items were rated on a 5-point scale (1 ⫽ not at all, 5 ⫽ very much). Reliability alphas were .96 and .94 for measures of positive and negative affect, respectively. Nostalgia. We administered two measures of nostalgia. Participants first completed Batcho’s (1995) Nostalgia Inventory (NI). They rated on a 5-point scale (1 ⫽ not at all, 5 ⫽ very much) the extent to which they missed 18 aspects of their past (␣ ⫽ .88). The items were “my family,” “not having to worry,” “places,” “music,” “someone I loved,” “my friends,” “things I did,” “my childhood toys,” “the way people were,” “feelings I had,” “my school,” “having someone to depend on,” “holidays I went on,” “the way society was,” “my pets,” “not knowing sad or evil things,” “past TV shows, movies,” and “my family house.”5 Batcho (1995, 1998) provided preliminary evidence for the validity of the NI. Nonetheless, we were concerned that certain properties of the NI could bias our findings. For instance, by instructing participants to rate the extent to which they miss aspects of their past, the NI focuses attention on just a single facet of what we consider to be a multifaceted emotion. For this reason, we constructed an additional measure of nostalgia comprising three items that were rated on a 6-point scale (1⫽ strongly disagree, 6 ⫽ strongly agree). The items were “Right now, I am feeling quite nostalgic,” “Right now, I am having nostalgic thoughts,” and “I feel nostalgic at the moment” (␣ ⫽ .95). The use of convergent operations of nostalgia can provide valuable information regarding the construct validity of our measures.
Results Manipulation Check The mood manipulation was successful. There was a significant effect for the mood manipulation on negative affect, F(2, 59) ⫽ 76.95, p ⬍ .001. Relevant means are presented in Table 6. Posthoc tests (Tukey’s honestly significant difference) revealed that participants in the negative mood condition experienced significantly more negative affect than those in the neutral and positive mood conditions. The neutral and positive mood conditions did not 5 The items “church/religion” and “heroes/heroines” were deleted from the original 20-item scale after pretesting revealed restriction of range as a result of extremely low ratings.
WILDSCHUT, SEDIKIDES, ARNDT, AND ROUTLEDGE
984
Table 6 Means for Negative Affect, Positive Affect, the Batcho (1995) Nostalgia Inventory, and the Three-Item Nostalgia Measure as a Function of Manipulated Mood Dependent variable
Negative mood
Neutral mood
Positive mood
Negative affect Positive affect Batcho (1995) Nostalgia Inventory Three-item nostalgia measure
3.27a 1.12a 2.31a 3.83a
1.30b 1.99b 1.41b 2.59b
1.23b 3.30c 1.70b 2.31b
Note. Within rows, means with different subscripts differ significantly at p ⬍ .05 (Tukey’s honestly significant difference).
differ significantly.6 There was also a significant effect for the mood manipulation on positive affect, F(2, 59) ⫽ 35.40, p ⬍ .001. Post-hoc tests revealed that participants in the positive mood condition experienced significantly more positive affect than those in the neutral and negative mood conditions. Participants in the neutral mood condition experienced significantly more positive affect than those in the negative mood condition. As intended, then, participants in the negative mood condition reported high levels of negative affect and low levels of positive affect, those in the positive mood condition reported high levels of positive affect and low levels of negative affect, and those in the neutral mood condition reported low levels of both negative and positive affect.
Nostalgia There was a significant effect for the mood manipulation on the average NI score, F(2, 57) ⫽ 12.53, p ⬍ .001. Relevant means are presented in Table 6. Consistent with the postulated link between negative affect and nostalgia, post-hoc tests (Tukey’s honestly significant difference) revealed that participants in the negative mood condition scored higher on the NI than those in the neutral and positive mood conditions. The neutral and positive mood conditions did not differ significantly. Although the NI forms an internally consistent measure, given the uncharted domain of this research, we wanted to explore the mood effect in greater detail. We therefore tested for each NI item a contrast between the negative mood condition and the neutral and positive mood conditions pooled. Fourteen out of 18 comparisons revealed higher ratings in the negative mood condition. Of these 14 comparisons, 8 were statistically significant (alpha ⫽ .05/18 ⫽ .0028). Negative mood increased significantly the ratings of the following items: “my family,” “someone I loved,” “my friends,” “the way people were,” “having someone to depend on,” “my family house,” “the way society was,” and “things I did.” These item-level results should be interpreted with caution, but they do suggest that negative mood increased in particular feelings of nostalgia associated with social aspects of participants’ past. Analysis of the second measure of nostalgia also revealed a significant effect for the mood manipulation, F(2, 59) ⫽ 7.23, p ⬍ .01. In agreement with NI results, post-hoc tests revealed that participants in the negative mood condition felt significantly more nostalgic than those in the neutral and positive mood conditions (see Table 6). The neutral and positive mood conditions did not differ significantly. Evidence for the convergent validity of the two
nostalgia measures was provided by their average within-cell correlation, which was substantial, r(62) ⫽ .55, p ⬍ .001.
Discussion Consistent with the postulated causal link between negative mood and nostalgia, there was a significant difference between the neutral and negative mood condition on each nostalgia measure. Inconsistent with the possibility that nostalgia is triggered by positive as well as negative mood states, there was no significant difference between the neutral and positive mood condition on either nostalgia measure. These findings lend credence to the descriptions of triggers provided by participants in Study 2 and raise the interesting possibility that nostalgia serves to counteract negative mood. There are, however, at least two limitations to the present study. First, we think it is important to look beyond global negative mood and focus on the role of discrete affective states to achieve a more detailed understanding of the link between negative affect and nostalgia. On this point, we find ourselves in agreement with other investigators who have emphasized the unique influences of discrete affective states on outcomes such as persuasion (Tiedens & Linton, 2001), intergroup hostility (Mackie, Devos, & Smith, 2000), and stereotyping (Bodenhausen et al., 1994). Second, our mood manipulation may have produced variation not only in mood but also in participants’ thoughts and motivations (Forgas & Ciarrochi, 2002). Such threats to construct validity are particularly relevant when, as in the present study, a novel rather than wellestablished manipulation is used. To corroborate the preliminary evidence for a causal link between negative affect and nostalgia, it is therefore desirable to use an alternative manipulation in a related experiment. We seek to address these two issues in Study 4.
Study 4 Study 4 had two related objectives. First, we sought to examine the effect of discrete rather than global negative affective states on nostalgia. Study 2 showed that participants who listed a discrete negative affective state as a trigger of nostalgia frequently described feelings of loneliness. This finding provides one rationale for exploring in greater detail the effect of loneliness on nostalgia. There is also a compelling theoretical reason for targeting loneliness. Research has shown that deficiencies in belongingness elicit compensatory mechanisms. For instance, Williams and Sommers (1997) found that women responded to rejection from a group by working harder on a subsequent collective task. In a similar vein, Gardner, Pickett, and Brewer (2000) found that rejection experiences resulted in selective retention of social information. Confirmation of the postulated effect of loneliness on nostalgia would 6
The mean negative-affect score in the negative mood condition was 3.27. On the basis of this score, one might question whether participants in this condition were truly experiencing negative mood. Our response to this and a similar question that may arise in Study 4 would be twofold. First, ethical guidelines prevent the use of very strong negative-affect manipulations, and second, the manipulation was successful in as far as participants in the negative mood condition experienced more negative affect than those in the control and positive mood conditions.
NOSTALGIA
raise the interesting possibility that nostalgia can serve to redress deficiencies in belongingness. The second objective of Study 4 was to provide a conceptual replication of Study 3 using an alternative manipulation of negative affect. Thus, we manipulated loneliness by offering participants false feedback from a personality test that allegedly assessed their level of loneliness.
Method Participants and Design Participants were 43 University of Southampton undergraduate students (40 women, 3 men) who received course credit. They were randomly assigned to one of two conditions (high vs. low loneliness).
Materials and Procedure Experimental session ranged in size from 1 to 10 persons. Participants completed, first, a questionnaire labeled Southampton Loneliness Scale. Participants indicated whether they agreed or disagreed with each of 15 statements drawn from the UCLA Loneliness Scale (Russell, Peplau, & Cutrona, 1980). Statements administered to participants in the high loneliness condition were phrased so as to elicit agreement. This was achieved by prefacing each statement with the words “I sometimes,” as in “I sometimes feel isolated from others.” Statements administered to participants in the low loneliness condition were phrased so as to elicit disagreement. This was achieved by prefacing each statement with the words “I always,” as in “I always feel alone.” As intended, participants in the high loneliness condition (M ⫽ 8.70) agreed with a greater number of statements than did participants in the low loneliness condition (M ⫽ 0.80), F(1, 41) ⫽ 121.66, p ⬍ .001. This set the stage for the second part of the loneliness manipulation. After participants completed the loneliness questionnaire, they were told that the experimenter would score their questionnaires and provide them with feedback regarding their level of loneliness. This feedback was provided on a form containing the following information: The University of Southampton Loneliness scale has been administered to a large number of University students over the last five years. Based on the responses of over twelve hundred students, we have developed a way of scoring your answers. This allows us to provide you with valid and detailed feedback regarding your level of loneliness. Participants in the high loneliness condition were informed that they were in the 62nd percentile of the loneliness distribution and that, compared with other University of Southampton students, they were therefore “above average on loneliness.” Participants in the low loneliness condition were informed that they were in the 12th percentile of the loneliness distribution and that, compared with other University of Southampton students, they were “very low on loneliness.” To strengthen the loneliness manipulation, we then asked participants to explain their loneliness score on a separate sheet of paper. Next, participants completed the manipulation check and a measure of nostalgia. The manipulation check consisted of three items (␣ ⫽ .91) designed to assess state loneliness (e.g., “Right now, I feel a bit lonely”). The items were rated on a 5-point scale (1 ⫽ strongly disagree, 5 ⫽ strongly agree). Nostalgia was measured with the 18-item NI (␣ ⫽ .83).
Results and Discussion Manipulation Check As intended, participants in the high-loneliness condition (M ⫽ 2.90) reported higher levels of loneliness than those in the lowloneliness condition (M ⫽ 1.28), F(1, 41) ⫽ 54.91, p ⬍ .001.
985
Nostalgia Consistent with a causal effect of loneliness on nostalgia, the average rating across all NI items was higher in the high-loneliness (M ⫽ 3.01) than in the low-loneliness (M ⫽ 2.56) condition, F(1, 41) ⫽ 6.11, p ⬍ .05. As before, we also conducted for each NI item a t test comparing the high- and low-loneliness conditions. Fifteen out of 18 comparisons revealed higher ratings in the high-loneliness condition. Of these 15 comparisons, 4 were statistically significant. High loneliness increased significantly the ratings of the following items: “my family,” “the way people were,” “having someone to depend on,” and “not having to worry.” Replicating conceptually Study 3, the present findings provide further corroborating evidence for a causal link between negative affect and nostalgia. These findings extend Study 3 findings in two important ways. First, whereas before we focused on the link between global negative mood and nostalgia, we now provide evidence for a link between a discrete negative affective state— loneliness—and nostalgia. This constitutes a critical first step beyond a singular focus on global negative mood and toward a more differentiated understanding of the discrete affective antecedents of nostalgia. Second, whereas before we manipulated negative mood by presenting participants with one of two factually-based news stories, in the present study we manipulated loneliness by providing participants with false feedback from a personality test. The use of such different yet converging manipulations in related studies is crucial, because it contributes toward establishing the construct validity of said studies. Why might negative mood and loneliness elicit feelings of nostalgia? Studies 3 and 4 raised the interesting possibility that nostalgia may serve to counteract negative mood and loneliness. They did not, however, test these ideas directly. In Studies 5–7, we made the functional utility of nostalgia the central focus of our investigation.
Studies 5, 6, and 7: Functions of Nostalgia In the search for functions of nostalgia, Study 2 provides valuable leads. Recall that participants in this study listed as many desirable and undesirable features of nostalgia as they could. Most desirable features of nostalgia referred to its perceived capacity to generate positive affect, increase positive self-regard, and bolster social bonds. It is these functional aspects of nostalgia that constitute the focus of Studies 5, 6, and 7. We do not mean to suggest that this set of functions exhausts all possibilities or that other possible functions of nostalgia are uninteresting or unimportant. However, both extant conceptual treatises of nostalgia and the wider social-psychological literature provide a sound theoretical rationale for targeting these particular functions.
Social Bonds Individuals have a fundamental need to belong (Baumeister & Leary, 1995). This is illustrated, for instance, by findings that persons form social bonds with relative ease (Festinger, Schachter, & Back, 1950) and are reluctant to break social bonds (Vaughan, 1986). However, life transitions (e.g., graduating from college, finding new employment) inevitably lead to changes in social settings. The deterioration or even dissolution of valued social
WILDSCHUT, SEDIKIDES, ARNDT, AND ROUTLEDGE
986
bonds that often accompanies such transitions can make people feel adrift and isolated (Colson, 1971). In addition, social bonds can be threatened by more momentary interpersonal rejections (Williams, 1997). We propose that by reigniting meaningful relationships nostalgia bolsters social bonds and renders accessible positive relational knowledge structures (i.e., working models of self and others in the context of relationships; Baldwin, Keelan, Fehr, Enns, & Koh-Rangarajoo, 1996). In nostalgic reverie, “. . . the mind is ‘peopled’” (Hertz, 1990, p. 195). Important figures from one’s past are brought to life and become part of one’s present (Cavanaugh, 1989). As such, nostalgia may even play a role in coping with bereavement (Mills & Coleman, 1994).
Self-Regard People are motivated to establish and maintain a positive selfconcept (Sedikides, Gaertner, & Toguchi, 2003; Sedikides & Strube, 1997). We propose that nostalgia offers a way to protect and increase self-regard by affirming valued aspects of the self that “reinforce one’s overall self-adequacy” (Steele, Spencer, & Lynch, 1993, p. 885). Nostalgia can bestow “an endearing luster” on the self and cast “marginal, fugitive, and eccentric facets of earlier selves in a positive light” (Davis, 1979, pp. 41– 46). Furthermore, nostalgic reverie can serve to affirm one’s positive qualities as a friend, family member, or member of other important groups (Kleiner, 1977).
Method Participants and Design Participants were 52 University of Southampton undergraduate volunteers (45 women, 7 men). They were randomly assigned to one of two conditions (nostalgia vs. control).
Materials and Procedure Participants received a booklet containing instructions relevant to the experimental manipulation of nostalgia, a manipulation check, and a set of dependent measures. In the nostalgia condition, participants were instructed to “. . . bring to mind a nostalgic event in your life. Specifically, try to think of a past event that makes you feel most nostalgic.” In the control condition, they were instructed to “. . . bring to mind an ordinary event in your daily life—an event that took place in the last week.” Participants were then instructed to write down four keywords relevant to the event and to take a few moments to think about the event and how it made them feel. Subsequently, they completed the manipulation check and the remaining dependent measures. Manipulation check. Participants rated on a 6-point scale (1 ⫽ strongly disagree, 6 ⫽ strongly agree) two items designed as a check on the nostalgia manipulation. The items were “Right now, I am feeling quite nostalgic” and “Right now, I’m having nostalgic feelings” (␣ ⫽ .96). Functions. Participants rated on a 5-point scale (1 ⫽ not at all, 5 ⫽ extremely) the extent to which thinking about the nostalgic or ordinary event made them feel “loved” and “protected” (to measure social bonding), “significant” and “high self-esteem” (to measure positive self-regard), “happy” and “content” (to measure positive affect), and “sad” and “blue” (to measure negative affect). All reliability alphas exceeded .75.
Positive Affect Positive affect is associated with a host of desirable outcomes. To name but a few, it facilitates approach behavior (Watson, Wiese, Vaidya, & Tellegen, 1999), increases subjective well-being (Diener, Sandvik, & Pavot, 1991), fosters psychological resiliency (Aspinwall & Taylor, 1997), and gives rise to thought patterns that are flexible, creative, integrative, and efficient (Isen, 2004). We propose that nostalgia serves as a store of positive affect. H. A. Kaplan (1987) characterized nostalgia as a “joyous” experience that gives rise to “an expansive state of mind” and “a feeling of elation” (p. 465). Similarly positive characterizations have been offered by Batcho (1995, 1998, Chaplin (2000), Davis (1977, 1979), Gabriel (1993), and Holak and Havlena (1998). The findings of Studies 1 and 2 further attest to the predominantly positive affective tone of nostalgia.
Study 5 Study 5 is a preliminary investigation of three functions of nostalgia. Participants were instructed to think about either a nostalgic or ordinary event from their past and then completed brief two-item measures of social bonding, self-regard, and positive and negative affect. Items were drawn from representative desirable and undesirable features of nostalgia listed by Study 2 participants (see Table 5). Negative affect was assessed because Study 2 identified it as the most undesirable feature of nostalgia. This suggests the possibility that nostalgia increases both positive and negative affect.
Results Manipulation Check Analysis of the manipulation check revealed that, as intended, participants in the nostalgia condition (M ⫽ 4.52) felt more nostalgic than those in the control condition (M ⫽ 2.81), F(1, 50) ⫽ 21.69, p ⬍ .001.
Functions Relative to participants in the control condition, those in the nostalgia condition scored higher on measures of social bonding (M ⫽ 3.79 vs. 2.65), F(1, 50) ⫽ 12.88, p ⬍ .001; positive self-regard (M ⫽ 3.81 vs. 2.81), F(1, 50) ⫽ 15.63, p ⬍ .001; and positive affect (M ⫽ 4.21 vs. 3.27), F(1, 50) ⫽ 8.05, p ⬍ .01. These results provide evidence for the three postulated functions of nostalgia. There was no significant difference between the nostalgia (M ⫽ 1.58) and control (M ⫽ 1.37) conditions for negative affect, F(1, 50) ⫽ 0.79, p ⬍ .38. The latter result is important in light of the Study 2 finding that participants considered negative affect to be the most undesirable feature of nostalgia. The present findings indicate that, although thinking about a nostalgic event may elicit some level of negative affect, this level does not exceed significantly that elicited by thinking about an ordinary event.
Discussion These findings advance in two important ways our understanding of nostalgia and its functions. First, they demonstrate the feasibility of manipulating in-the-moment feelings of nostalgia.
NOSTALGIA
Second, they provide vital preliminary support for the idea that nostalgia bolsters social bonds, increases self-regard, and generates positive affect. There are, however, at least two limitations to the present study. The first limitation concerns the construct validity of our brief, two-item measures. Although confirmation of the predicted nostalgia effects on measures of social bonding, self-regard, and positive affect can be regarded as evidence for the construct validity of said measures (Cronbach & Meehl, 1955), it is desirable to replicate these results with well-established and validated measures of the focal outcome variables. This was the first objective of Study 6. The second limitation concerns the manipulation of nostalgia. It is likely that this single manipulation underrepresented or failed to capture entirely the experience of nostalgia. For instance, the instruction to think for “a few moments” about the nostalgic (or ordinary) experience may not have elicited the deeper reflection that can accompany feelings of nostalgia. Furthermore, we did not provide participants with a definition of nostalgia. The reason for not doing so was that we did not want to constrain or steer participants’ personal conceptualizations of nostalgia. It is, however, important to rule out the possibility that the present findings stem from an idiosyncratic conceptualization of nostalgia specific to our sample. The second objective of Study 6, then, was to use instructions that were designed to immerse participants more deeply in the nostalgic experience and included a dictionary definition of nostalgia. This use of alternative instructions in related experiments allows us to establish the generalizability of our findings across procedures. This is particularly important when, as in the present case, there is no established body of research to inform our experimental manipulations.
987 Imagine the event as though you were an historian recording factual details (e.g., I got on the number 37 bus). Then, please write about this everyday event in the space below. Write a purely factual and detailed account (e.g., like in a court of law, avoiding emotionally expressive words).
Participants were given 6 min to complete their narratives. They then responded to a manipulation check (identical to Study 5; ␣ ⫽ .86) and filled out validated measures of social bonding, positive self-regard, and positive and negative affect. Social bonding. We administered the Revised Experiences in Close Relationships Scale (ECR-R; Fraley, Waller, & Brennan, 2000) as a measure of social bonding. The ECR-R assesses the dimensions of attachment anxiety (e.g., “I worry that romantic partners won’t care about me as much as I care about them”) and attachment avoidance (e.g., “I am very uncomfortable with being close to romantic partners”). The anxiety (␣ ⫽ .90) and avoidance (␣ ⫽ .96) subscales each consisted of 18 items that were rated on a 7-point scale (1 ⫽ strongly disagree, 7 ⫽ strongly agree). Positive self-regard. We administered the Rosenberg Self-Esteem Scale (RSE; Rosenberg, 1965) to measure self-regard (␣ ⫽ .88). Items were rated on a 6-point scale (1 ⫽ strong disagreement, 6 ⫽ strong agreement). Positive and negative affect. We administered the 20-item version of the PANAS to measure positive (␣ ⫽ .88) and negative affect (␣ ⫽ .82). Items were rated on a 5-point scale (1 ⫽ not at all, 5 ⫽ extremely).
Results Manipulation Check The manipulation check revealed that, as intended, participants in the nostalgia condition (M ⫽ 4.80) felt more nostalgic than those in the control condition (M ⫽ 2.92), F(1, 52) ⫽ 65.21, p ⬍ .001.
Study 6
Functions
Method
Results revealed a successful conceptual replication of Study 5 findings. Consistent with the postulated capacity of nostalgia to bolster social bonds and render accessible positive relational knowledge structures, participants in the nostalgia condition evinced a more secure attachment style than those in the control condition. Relative to controls, participants in the nostalgia condition reported lower levels of attachment anxiety (M ⫽ 2.40 vs. 2.90), F(1, 52) ⫽ 4.34, p ⬍ .05, and lower levels of attachment avoidance (M ⫽ 2.35 vs. 2.87), F(1, 52) ⫽ 3.14, p ⬍ .08. Consistent with the postulated capacity of nostalgia to increase self-regard, participants in the nostalgia condition (M ⫽ 5.08) reported significantly higher self-esteem than those in the control condition (M ⫽ 4.62), F(1, 52) ⫽ 5.71, p ⬍ .05. Consistent with the postulated capacity of nostalgia to generate positive affect, participants in the nostalgia condition (M ⫽ 2.81) reported more positive affect than those in the control condition (M ⫽ 2.29), F(1, 52) ⫽ 7.03, p ⬍ .01. Participants in the nostalgia (M ⫽ 1.24) and control conditions (M ⫽ 1.37) did not differ significantly on negative affect, F(1, 52) ⫽ 1.32, p ⬍ .26.
Participants and Design Participants were 54 University of Southampton undergraduates (46 women, 8 men) who received course credit. They were randomly assigned to one of two conditions (nostalgia vs. control).
Materials and Procedure On arrival at the laboratory, participants were seated in separate cubicles. In the nostalgia condition, participants received the following instructions: According to the Oxford Dictionary, ‘nostalgia’ is defined as a ‘sentimental longing for the past.’ Please think of a nostalgic event in your life—a nostalgic event that has personal meaning for you. Specifically, try to think of a past event that makes you feel most nostalgic. Bring this experience to mind. Immerse yourself in the nostalgic experience. How does it make you feel? Then, write about this experience in the space below. Describe the experience and how it makes you feel nostalgic. In the control condition, instructions read as follows: Please think of an ordinary event in your life that took place in the last week. Try to bring this event to mind and think it through as though you were an observer of the event, rather than directly involved.
Discussion By demonstrating that Study 5 results generalize across different measures and manipulations, the present findings offer vital reinforcement for the idea that nostalgia bolsters social bonds, in-
WILDSCHUT, SEDIKIDES, ARNDT, AND ROUTLEDGE
988
creases self-regard, and generates positive affect. There is, however, at least one remaining issue that should be addressed. This relates to the evenness with which the ECR-R, the RSE, and the PANAS assess the postulated functions of social bonding, selfregard, and positive affect, respectively. In particular, we were concerned that the ECR-R, because it relates exclusively to interactions with romantic partners, may not have captured strength of social bonds to the same extent as the RSE captured self-regard or the PANAS captured positive (and negative) affect. To address this issue, Study 7 investigated the effect of nostalgia on three domains of interpersonal competence in everyday social interactions. If nostalgia bolsters social bonds and renders accessible positive relational knowledge structures, it should increase people’s perceived ability to form, maintain, and develop successfully not only romantic relationships but interpersonal relationships in general. A secondary objective of Study 7 was to provide a strong test of gender differences. With the exception of Study 1, the high female-to-male ratio in our participant pool (⬇ 8:1) was reflected in our samples, rendering tests of gender differences either meaningless or underpowered. In Study 7, we succeeded in recruiting approximately equal numbers of female and male participants through campus-wide advertisements.
Study 7 Method Participants and Design Participants were 121 University of Southampton undergraduate volunteers (67 women, 52 men, 2 of undeclared gender). They were randomly assigned to one of two conditions (nostalgia vs. control).
Manipulation Check As intended, participants in the nostalgia condition (M ⫽ 3.90) felt more nostalgic than those in the control condition (M ⫽ 2.79), F(1, 118) ⫽ 19.39, p ⬍ .001.
Interpersonal Competence Relative to participants in the control condition, those in the nostalgia condition evinced greater interpersonal competence in the domains of initiation (M ⫽ 3.12 vs. 2.61), F(1, 118) ⫽ 6.97, p ⬍ .01; self-disclosure (M ⫽ 3.17 vs. 2.59), F(1, 119) ⫽ 9.81, p ⬍ .001; and emotional support (M ⫽ 3.47 vs. 2.97), F(1, 119) ⫽ 6.71, p ⬍ .05. These findings provide further corroborating evidence for the idea that nostalgia bolsters social bonds and, importantly, show that this effect generalizes beyond the realm of romantic relationships and across gender.
General Discussion Although the term nostalgia was not coined until the late 17th century (Hofer, 1688/1934), references to its meaning can be traced back as far as the writings of Shakespeare, Caesar, Hippocrates, and Homer. Indeed, it is surprising that nostalgia has long been neglected in psychological scholarship. Granted, there has been speculation, mostly from a psychodynamic perspective, about the nature of nostalgia but rarely have these ideas been tested empirically. To find our bearings in this novel territory, we sought to address three basic questions pertaining, respectively, to the content, triggers, and functions of nostalgia.
Summary of Findings
Procedure and Materials
The Content Question
Participants received a booklet containing instructions relevant to the manipulation of nostalgia, a manipulation check, and an assessment of three domains of interpersonal competence. The manipulation of nostalgia was identical to the manipulation used in Study 5. The manipulation check was identical to the one used in the two preceding studies (␣ ⫽ .96). We administered the Initiation, Disclosure, and Emotional Support subscales from the Interpersonal Competence Questionnaire (Buhrmeister, Furman, Wittenberg, & Reis, 1988) to assess perceived competence in, respectively, initiating interactions and relationships (e.g., “Going to parties or gatherings where you don’t know people well in order to start up new relationships”), self-disclosing personal information (e.g., “Telling a close companion how much you appreciate and care for him or her”), and providing emotional support to others (e.g., “Helping a close companion get to the heart of a problem he or she is experiencing”). These three domains of interpersonal competence were assessed with 8 items each (alphas ⬎ .92). Ratings were made on a 5-point scale (1 ⫽ disagree, 5 ⫽ agree).
Studies 1 and 2 sought to examine the content of nostalgic experience using a phenomenon-based approach (Sternberg & Grigorenko, 2001) that both acknowledged the breadth of lay conceptualizations of nostalgia and was informed by existing theoretical treatises on the topic. Study 1 was a content analysis of autobiographical narratives published in the periodical Nostalgia. Study 2 used a vivid-recall methodology in which participants recalled a nostalgic experience, wrote a narrative account of this experience, and completed self-report measures regarding the experience. Despite the fact that these studies were methodologically diverse in areas of recruitment (e.g., U.S. vs. British nationality, heterogeneous vs. homogeneous age composition), procedure (e.g., absence vs. presence of explicit nostalgia instructions), and measurement (e.g., absence vs. presence of self-report measures), they yielded remarkably consistent findings. In both Studies 1 and 2, nostalgic narratives typically featured the self as central character and revolved around interactions with important others (e.g., friends, loved ones) or momentous events (e.g., graduation ceremonies, birth of a child). The narratives often contained descriptions of disappointments and losses but, in a large majority of cases, these negative life scenes were redeemed or mitigated by subsequent triumphs over adversity. Furthermore, nostalgic narratives were richer in expressions of positive than negative affect
Results and Discussion Initial analyses revealed no significant or marginal effects involving gender. Therefore gender was not included as an independent variable in the final analyses reported below. Denominator degrees of freedom vary due to missing values.
NOSTALGIA
(Study 1), and participants reported experiencing more positive than negative affect after recalling a nostalgic event (Study 2).
989 Broader Implications
Emotions The Trigger Question Our second basic question related to the triggers of nostalgia and was addressed directly in Studies 3 and 4. Study 3 provided strong evidence for a causal link between negative mood and nostalgia. Following a mood manipulation, participants completed two measures of nostalgia: the NI (Batcho, 1995) and a three-item measure designed to assess in-the-moment feelings of nostalgia. Participants in the negative mood condition scored significantly higher on both measures of nostalgia than participants in the neutral and positive mood conditions. The latter two conditions did not differ significantly on either measure. Study 4 made two additional contributions. First, it examined the role of a discrete affective state—loneliness—to achieve a more detailed understanding of the link between negative affect and nostalgia. The second contribution of Study 4 was that our manipulation of loneliness, which involved giving participants false feedback from a personality test, departed considerably from the mood manipulation used in Study 3, which involved presenting participants with one of three factually-based news stories. Relative to participants in the low loneliness condition, those in the high loneliness condition scored higher on the NI. This finding constitutes a critical first step beyond a singular focus on global negative mood and toward a more differentiated understanding of the discrete affective triggers of nostalgia. Finally, the particular significance attached by participants in both studies to social aspects of their past is consistent with, and reinforces, the earlier finding that friends, family, and loved ones are important objects of nostalgia.
The Function Question The third question that we sought to answer related to the functions of nostalgia. We took as our point of departure the idea that nostalgia bolsters social bonds, increases positive self-regard, and generates positive affect. In Study 5, we manipulated nostalgia by instructing participants to bring to mind either a nostalgic or ordinary event in their lives. Results revealed that, relative to participants in the control condition, those in the nostalgia condition scored higher on brief measures of social bonding, positive self-regard, and positive affect. There was no significant difference between the nostalgia and control condition for negative affect. Study 6 replicated these findings with a more immersive manipulation of nostalgia and validated measures of social bonding, positive self-regard, and affect. Relative to participants in the control condition, those in the nostalgia condition reported less attachment anxiety and avoidance, higher self-esteem, and more positive affect. As before, there was no significant difference for negative affect. Study 7 revealed that the capacity of nostalgia to bolster social bonds is not limited to the domain of romantic relationships. Relative to participants in the control condition, those in the nostalgia condition reported greater confidence in their ability to initiate interactions and relationships, disclose personal information, and provide emotional support to others.
Emotion theorists are unanimous in labeling nostalgia an emotion (Frijda, 1986; Johnson-Laird & Oatley, 1989; Kemper, 1987; Ortony, Clore, & Collins, 1988). We subscribe in particular to Johnson-Laird and Oatley’s (1989) view that nostalgia is a happiness-related emotion (Sedikides, Wildschut, Arndt, & Routledge, 2006). To be sure, there is compelling evidence that nostalgia is in a league with other positive emotions such as love (Izard, 1977), pride (Lewis, 1993) and joy (Ellsworth & Smith, 1988). Our findings indicate that, like love, nostalgia bolsters social bonds; that, like pride, nostalgia increases positive selfregard; and that, like joy, nostalgia generates positive affect. The classification of nostalgia as a happiness-related or positive emotion suggests various avenues for future research. For instance, is nostalgia associated with physical health? Although there is no direct evidence addressing this question, we can speculate based on some interesting findings. Danner, Snowdon, and Friesen (2001), for instance, coded the emotional content of brief autobiographical sketches written by Catholic nuns (ages 75–95) at the time they entered their convent in early adulthood. Early positive emotionality as expressed in these sketches was found to predict survival rates 60 years later. In another relevant study, Stone, Cox, Valdimarsdottir, Jandorf, and Neale (1987) used an experiencesampling methodology to examine the relation between daily mood and immunological changes. They found increased immunocompetence on days with high positive mood relative to days with low positive mood. Future research should harness the experience-sampling methodology to provide a window on the daily experience of nostalgia and its links to both psychological and physical well-being.
The Self People are motivated to protect and enhance the positivity of the self-concept (Sedikides & Gregg, 2003; Sedikides & Strube, 1997). Self-protection and self-enhancement mechanisms are typically activated when circumstances or events are perceived as self-threatening (Campbell & Sedikides, 1999; Sedikides, Green, & Pinter, 2004). Prior work on compensatory self-inflation (Baumeister & Jones, 1978; Greenberg & Pyszczynski, 1985) and self-affirmation (Steele, 1988) has revealed that, when people encounter self-threats, rather than countering directly the specific threat, they have the option of eliminating its effects by affirming essential, positive aspects of the self. We propose that nostalgia constitutes a benign mechanism through which people affirm valued aspects of the self. This suggests an interesting direction for future research. Given that efforts to protect and enhance the self often have undesirable consequences, such as reduced receptivity to critical feedback (Kumashiro & Sedikides, 2005), can nostalgia be used as a resource for responding to self-threats in a more open and constructive manner? Consistent with this possibility, recent evidence indicates that nostalgia attenuates the effect of mortality salience—a particularly potent self-threat (Pyszczynksi, Greenberg, Solomon, Arndt, & Schimel, 2004)— on death-thought accessibility (Routledge, Arndt, Sedikides, & Wildschut, 2006).
WILDSCHUT, SEDIKIDES, ARNDT, AND ROUTLEDGE
990 Relationships
Nostalgia as Compared With Other Past-Oriented Subjective States
In a resolute call to arms, Mikulincer and Shaver (2005) highlighted the need to identify “how various experiences and techniques, including psychotherapy, family therapy, skilled meditation, and participation in religious or charitable organizations, might enhance a person’s sense of [attachment] security” (p. 37). Our findings suggest that, by bolstering social bonds and rendering accessible positive relational knowledge structures, nostalgia offers an additional avenue to enhancing in-the-moment attachment security. The implications are far-reaching and manifold. Attachment security is associated with a host of desirable outcomes. For instance, research has shown that momentary or primed attachment security gives rise to greater compassion (Mikulincer et al., 2001) and altruism (Mikulincer, Shaver, Gillath, & Nitzberg, 2005). Can nostalgia, by virtue of its effect on attachment security, produce similar desirable outcomes?
Limitations Before generalizing from the findings, one must keep in mind that the samples consisted predominantly of college-age, British females. The question whether age-related changes in motivation have a bearing on nostalgia presents one suitable avenue for future research. The interaction between gender and culture in shaping nostalgia is another issue that deserves careful scrutiny. According to socioemotional selectivity theory (Carstensen, Isaacowitz, & Charles, 1999), with advancing age people come to view their life span as limited and shift attention from futureoriented, knowledge-related goals toward a desire to find purpose and meaning in life, to enjoy intimate friendships, and to be embedded in a social network. This raises two important issues pertaining to nostalgia. First, are such age-related changes in motivation reflected in the frequency and content of nostalgia? We would expect older (as compared with younger) adults to be more prone to nostalgia and more likely to give center stage to close others in their nostalgic reverie. The second issue is whether nostalgia acquires greater significance in old age. Although the problem of loneliness is not specific to old age (Ellaway, Wood, & MacIntyre, 1999), bereavements and declines in health status may render older adults particularly vulnerable to social isolation (Victor, Scambler, Bowling, & Bond, 2005), thus impairing the formation of intimate friendships and social networks they so highly value. Under these circumstances, nostalgia may play a vital role in reestablishing at least a symbolic connection with significant others (Batcho, 1998; Cavanaugh, 1989; Mills & Coleman, 1994). It seems plausible that British college students do not identify strongly with narrowly prescribed gender roles. Where we succeeded in recruiting sufficient male participants to perform a meaningful test of gender differences (Studies 2 and 7), no such differences were found. In cultural contexts that place a stronger emphasis on traditional gender roles, however, differences between females and males may well arise. In general, gender differences are shaped by culture (Hyde, 2003), and so it would be unwarranted to generalize our findings for gender to very different cultural settings.
Before closing, we should address one final issue. JohnsonLaird and Oatley (1989) assign nostalgia to the category of complex emotions, which, unlike basic emotions, arise from high-level cognitive processing and possess propositional content. This raises the question of how nostalgia differs from other processes— denoted with such words as “recall,” “recollection,” “reminiscence,” and “remembrance”—which also involve the cognitively demanding task of reconstructing the past. Davis (1977) proposed that “to merely remember the places of one’s youth is not the same as to feel nostalgia for them. Neither for that matter, does active reminiscence— however happy, benign or tortured its content— necessarily capture the subjective state characteristic of nostalgic feeling” (p. 418). We concur, but ultimately this is an issue that should be settled empirically. Our point of departure is that nostalgia shares with other past-oriented subjective states the involvement of high-level cognitive processing but can be distinguished from them, for instance in terms of its unique affective signature and its psychological functions (Castelnuovo-Tedesco, 1980; Cavanaugh, 1989).
Conclusion Nostalgia is a prevalent and fundamental human experience— one that serves a number of key psychological functions. As evidenced by the present findings, nostalgia may be uniquely positioned to offer integrative insights across several important facets of human functioning. We hope that this and future research will redress the paucity of knowledge regarding nostalgia and award it its proper place in the pantheon of emotions.
References Abelson, R. P., Kinder, D. R., Peters, M. D., & Fiske, S. T. (1982). Affective and semantic components in political person perception. Journal of Personality and Social Psychology, 42, 619 – 630. Aspinwall, L. G., & Taylor, S. E. (1997). A stitch in time: Self-regulation and proactive coping. Psychological Bulletin, 121, 417– 436. Baldwin, M. W., Keelan, J. P. R., Fehr, B., Enns, V., & Koh-Rangarajoo, E. (1996). Social-cognitive conceptualization of attachment working models: Availability and accessibility effects. Journal of Personality and Social Psychology, 71, 94 –109. Batcho, K. I. (1995). Nostalgia: A psychological perspective. Perceptual and Motor Skills, 80, 131–143. Batcho, K. I. (1998). Personal nostalgia, world view, memory, and emotionality. Perceptual and Motor Skills, 87, 411– 432. Baumeister, R. F., & Jones, E. E. (1978). When self-presentation is constrained by the target’s knowledge: Consistency and compensation. Journal of Personality and Social Psychology, 36, 608 – 618. 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. Baumeister, R. F., Stillwell, A., & Wotman, S. R. (1990). Victim and perpetrator accounts of interpersonal conflict: Autobiographical narratives about anger. Journal of Personality and Social Psychology, 59, 994 –1005. Baumeister, R. F., Wotman, S. R., & Stillwell, A. M. (1993). Unrequited love: On heartbreak, anger, guilt, scriptlessness, and humiliation. Journal of Personality and Social Psychology, 64, 377–394.
NOSTALGIA Best, J., & Nelson, E. E. (1985). Nostalgia and discontinuity: A test of the Davis hypothesis. Sociology and Social Research, 69, 221–233. Boden, J. M., & Baumeister, R. F. (1997). Repressive coping: Distraction using pleasant thoughts and memories. Journal of Personality and Social Psychology, 73, 45– 62. Bodenhausen, G. V., Sheppard, L. A., & Kramer, G. P. (1994). Negative affect and social judgment: The differential impact of anger and sadness. European Journal of Social Psychology, 24, 45– 62. Boym, S. (2002). The future of nostalgia. New York: Basic Books. Buhrmeister, D., Furman, W., Wittenberg, M. T., & Reis, H. T. (1988). Five domains of interpersonal competence in peer relationships. Journal of Personality and Social Psychology, 55, 991–1108. Campbell, K. W., & Sedikides, C. (1999). Self-threat magnifies the selfserving bias: A meta-analytic integration. Review of General Psychology, 3, 23– 43. Carstensen, L. L., Isaacowitz, D. M., & Charles, S. T. (1999). Taking time seriously: A theory of socioemotional selectivity. American Psychologist, 54, 165–181. Castelnuovo-Tedesco, P. (1980). Reminiscence and nostalgia: The pleasure and pain of remembering. In S. I. Greenspan & G. H. Pollack (Eds.), The course of life: Psychoanalytic contributions toward understanding personality development. Vol. III: Adulthood and the aging process (pp. 104 –118). Washington, DC: U.S. Government Printing Office. Cavanaugh, J. C. (1989). I have this feeling about everyday memory aging. Educational Gerontology, 15, 597– 605. Chaplin, S. (2000). The psychology of time and death. Ashland, OH: Sonnet Press. Chu, S., & Downes, J. J. (2000). Long live Proust: The odor-cued autobiographical memory bump. Cognition, 75, B41–B50. Colson, E. (1971). The social consequences of resettlement. Manchester, England: Manchester University Press. Cronbach, L. J., & Meehl, P. E. (1955). Construct validity in psychological tests. Psychological Bulletin, 52, 281–302. Danner, D. D., Snowdon, D. A., & Friesen, W. V. (2001). Positive emotions in early life and longevity: Findings from the nun study. Journal of Personality and Social Psychology, 80, 804 – 813. Davis, F. (1977). Nostalgia, identity, and the current nostalgia wave. Journal of Popular Culture, 11, 414 – 425. Davis, F. (1979). Yearning for yesterday: A sociology of nostalgia. New York: Free Press. Diener, E., Sandvik, E., & Pavot, W. (1991). Happiness is the frequency, not the intensity, of positive versus negative affect. In F. Strack (Ed.), Subjective well-being: An interdisciplinary perspective (pp. 119 –139). Oxford, England: Pergamon Press. Eagly, A. H., & Chaiken, S. (1993). The psychology of attitudes. Orlando, FL.: Harcourt, Brace. Ellaway, W. Wood, S., & MacIntyre, S. (1999). Someone to talk to? The role of loneliness as a factor in the frequency of GP consultations. British Journal of General Practice, 49, 363–367. Ellsworth, P. C., & Smith, C. A. (1988). Shades of joy: Patterns of appraisal differentiating pleasant emotions. Cognition and Emotion, 2, 301–331. Festinger, L., Schachter, S., & Back, K. (1950). Social pressures in informal groups: A study of a housing community. Palo Alto, CA: Stanford University Press. Fodor, N. (1950). Varieties of nostalgia. Psychoanalytic Review, 37, 25– 38. Forgas, J. P., & Ciarrochi, J. (2002). On managing moods: Evidence for the role of homeostatic cognitive strategies in affect regulation. Personality and Social Psychology Bulletin, 28, 336 –345. Fraley, R. C., Waller, N. G., & Brennan, K. A. (2000). An item response theory analysis of self-report measures of adult attachment. Journal of Personality and Social Psychology, 78, 350 –365. Frijda, N. H. (1986). The emotions. New York: Cambridge University Press.
991
Frost, I. (1938). Homesickness and immigrant psychoses. Journal of Mental Science, 84, 801– 847. Gabriel, Y. (1993). Organizational nostalgia: Reflections on “The Golden Age.” In S. Fineman (Ed.), Emotion in organizations (pp. 118 –141). London: Sage. Gardner, W. L., Pickett, C., & Brewer, M. B. (2000). Social exclusion and selective memory: How the need to belong influences memory for social events. Personality and Social Psychology Bulletin, 26, 486 – 496. Greenberg, J., & Pyszczynski, T. (1985). Compensatory self-inflation: A response to the threat to self-regard of public failure. Journal of Personality and Social Psychology, 49, 273–280. Hertz, D. G. (1990). Trauma and nostalgia: New aspects of the coping of aging holocaust survivors. Israeli Journal of Psychiatry and Related Sciences, 27, 189 –198. Hofer, J. (1934). Medical dissertation on nostalgia (C. K. Anspach, Trans.). Bulletin of the History of Medicine, 2, 376 –391. (Original work published 1688) Holak, S. L., & Havlena, W. J. (1992). Nostalgia: An exploratory study of themes and emotions in the nostalgic experience. Advances in Consumer Research, 19, 380 –386. Holak, S. L., & Havlena, W. J. (1998). Feelings, fantasies, and memories: An examination of the emotional components of nostalgia. Journal of Business Research, 42, 217–226. Homer. (1921). The odyssey (F. Caulfield, Trans.). London: G. Bell. Hyde, J. S. (2003). Half the human experience: The psychology of women. Boston, MA: Houghton Mifflin. Isen, A. M. (2004). Some perspectives on positive feelings and emotions: Positive affect facilitates thinking and problem solving. In A. S. R. Manstead, N. Frijda, & A. Fischer (Eds.), Feelings and emotions: The Amsterdam Symposium (pp. 263–281). New York: Cambridge University Press. Izard, C. E. (1977). Human emotions. New York: Plenum. Johnson-Laird, P. N., & Oatley, K. (1989). The language of emotions: An analysis of semantic field. Cognition and Emotion, 3, 81–123. Joormann, J., & Siemer, M. (2004). Memory accessibility, mood regulation, and dysphoria: Difficulties in repairing sad mood with happy memories? Journal of Abnormal Psychology, 113, 179 –188. Josephson, B. R., Singer, J. A., & Salovey, P. (1996). Mood regulation and memory: Repairing sad moods with happy memories. Cognition and Emotion, 10, 437– 444. Kaplan, H. A. (1987). The psychopathology of nostalgia. Psychoanalytic Review, 74, 465– 486. Kaplan, K. J. (1972). On the ambivalence-indifference problem in attitude theory and measurement: A suggested modification of the semantic differential technique. Psychological Bulletin, 77, 361–372. Kemper, T. (1987). How many emotions are there? Wedding the social and autonomic components. American Journal of Sociology, 93, 263–289. Kleiner, J. (1977). On nostalgia. In C. W. Socarides (Ed.), The world of emotions (pp. 471– 498). New York: International University Press. Kumashiro, M., & Sedikides, C. (2005). Taking on board liability-focused feedback: Close positive relationships as a self-bolstering resource. Psychological Science, 16, 732–739. Larsen, R. J., & Prizmic, Z. (2004). Affect regulation. In R. F. Baumeister & K. D. Vohs (Eds.), Handbook of self-regulation (pp. 40 – 61). New York: Guilford. Leary, M. R., Springer, C., Negel, L., Ansell, E., & Evans, K. (1998). The causes, phenomenology, and consequences of hurt feelings. Journal of Personality and Social Psychology, 74, 1225–1237. Lewis, M. (1993). Self-conscious emotions: Embarrassment, pride, shame, and guilt. In M. Lewis & J. M. Haviland (Eds.), Handbook of emotions (pp. 563–573). New York: Guilford. Lipkus, I. M., Green, J. D., Feaganes, J. R., & Sedikides, C. (2001). The relationship between attitudinal ambivalence and desire to quit smoking among college smokers. Journal of Applied Social Psychology, 31, 113–133.
992
WILDSCHUT, SEDIKIDES, ARNDT, AND ROUTLEDGE
Mackie, D. M., Devos, T., & Smith, E. R. (2000). Intergroup emotions: Explaining offensive action tendencies in an intergroup context. Journal of Personality and Social Psychology, 79, 602– 616. McAdams, D. P. (2001). The psychology of life stories. Review of General Psychology, 5, 100 –122. McAdams, D. P., Reynolds, J., Lewis, M., Patten, A. H., & Bowman, P. J. (2001). When bad things turn good and good things turn bad: Sequences of redemption and contamination in life narratives and their relation to psychosocial adaptation in midlife adults and in students. Personality and Social Psychology Bulletin, 27, 474 – 485. McCann, W. H. (1941). Nostalgia: A review of the literature. Psychological Bulletin, 38, 165–182. McFarland, C., & Buehler, R. (1997). Negative affective states and the motivated retrieval of positive life events: The role of affect acknowledgement. Journal of Personality and Social Psychology, 73, 200 –214. Mikulincer, M., Gillath, O., Halevy, V., Avihou, N., Avidan, S., & Eshkoli, N. (2001). Attachment theory and reactions to others’ needs: Evidence that activation of the sense of attachment security promotes empathic responses. Journal of Personality and Social Psychology, 81, 1205– 1224. Mikulincer, M., & Shaver, P. R. (2005). Attachment security, compassion, and altruism. Current Directions in Psychological Science, 14, 34 –38. Mikulincer, M., Shaver, P. R., Gillath, O., & Nitzberg, R. E. (2005). Attachment, caregiving, and altruism: Boosting attachment security increases compassion and helping. Journal of Personality and Social Psychology, 89, 817– 839. Mills, M. A., & Coleman, P. G. (1994). Nostalgic memories in dementia: A case study. International Journal of Aging and Human Development, 38, 203–219. The new Oxford dictionary of English. (1998). Oxford, England: Oxford University Press. Ortony, A., Clore, G. L., & Collins, A. (1988). The cognitive structure of emotions. Cambridge, England: Cambridge University Press. Parrott, W. G., & Sabini, J. (1990). Mood and memory under natural conditions: Evidence for mood incongruent recall. Journal of Personality and Social Psychology, 59, 321–336. Peters, R. (1985). Reflections on the origin and aim of nostalgia. Journal of Analytical Psychology, 30, 135–148. Pyszczynksi, T., Greenberg, J., Solomon, S., Arndt, J., & Schimel, J. (2004). Why do people need self-esteem? A theoretical and empirical review. Psychological Bulletin, 130, 435– 468. Rosen, G. (1975). Nostalgia: A “forgotten” psychological disorder. Psychological Medicine, 5, 340 –354. Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University Press. Routledge, C., Arndt, J., Sedikides, C., & Wildschut, T. (2006). A blast from the past: Nostalgia buffers the effect of mortality salience on death-thought accessibility. Manuscript submitted for publication. Russell, D., Peplau, L. A., & Cutrona, C. E. (1980). The revised UCLA Loneliness Scale: Concurrent and discriminant validity evidence. Journal of Personality and Social Psychology, 39, 472– 480. Rusting, C. L. (1998). Personality, mood, and cognitive processing of emotional information: Three conceptual frameworks. Psychological Bulletin, 124, 165–196. Rusting, C. L., & DeHart, T. (2000). Retrieving positive memories to regulate negative mood: Consequences for mood-congruent memory. Journal of Personality and Social Psychology, 78, 737–752. Schindler, R. M., & Holbrook, M. B. (2003). Nostalgia for early experience as a determinant of consumer preferences. Psychology & Marketing, 20, 275–302. Schwarz, N., & Clore, G. L. (1988). How do I feel about it? The information function of affective states. In K. Fiedler & J. Forgas (Eds.) Affect, cognition, and social behavior (pp. 44 – 62). Toronto, Canada: Hogrefe.
Sedikides, C. (1992). Changes in the valence of the self as a function of mood. Review of Personality and Social Psychology, 14, 271–311. Sedikides, C., Gaertner, L., & Toguchi, Y. (2003). Pan-cultural selfenhancement. Journal of Personality and Social Psychology, 84, 60 –79. Sedikides, C., & Green, J. D. (2001). Affective influences on the selfconcept: Qualifying the mood congruency principle. In J. P. Forgas (Ed.), The handbook of affect and social cognition (pp. 145–160). Mahwah, NJ: Erlbaum. Sedikides, C., Green, J. D., & Pinter, B. (2004). Self-protective memory. In D. R., Beike, J. M. Lampinen, & D. A. Behrend (Eds.), The self and memory (pp. 161–179). New York: Psychology Press. Sedikides, C., & Gregg. A. P. (2003). Portraits of the self. In M. A. Hogg & J. Cooper (Eds.), Sage handbook of social psychology (pp. 110 –138). London: Sage. Sedikides, C., & Strube, M. J. (1997). Self-evaluation: To thine own self be good, to thine own self be sure, to thine own self be true, and to thine own self be better. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 29, pp. 209 –269). New York: Academic Press. Sedikides, C., Wildschut, T., Arndt, J., & Routledge, C. D. (2006). Affect and the self. In J. P. Forgas (Ed.), Affect in social thinking and behavior: Frontiers in social psychology (pp. 197–215). New York: Psychology Press. Sedikides, C., Wildschut, T., & Baden, D. (2004). Nostalgia: Conceptual issues and existential functions. In J. Greenberg, S. Koole, & T. Pyszczynski (Eds.), Handbook of experimental existential psychology (pp. 200 –214). New York: Guilford. Shakespeare, W. (1996). The sonnets. Cambridge, England: Cambridge University Press. (Original work published 1609) Smith, S. M., & Petty, R. E. (1995). Personality moderators of mood congruency effects on cognition: The role of self-esteem and negative mood regulation. Journal of Personality and Social Psychology, 68, 1092–1107. Steele, C. M. (1988). The psychology of self-affirmation: Sustaining the integrity of the self. In L. Berkowitz (Ed.), Advances in experimental social psychology (pp. 261–302). Hillsdale, NJ: Erlbaum. Steele, C. M., Spencer, S. J., & Lynch, M. (1993). Self-image resilience and dissonance: The role of affirmational resources. Journal of Personality and Social Psychology, 64, 885– 896. Stern, B. B. (1992). Historical and personal nostalgia in advertising text: The fin de siecle effect. Journal of Advertising, 21, 11–22. Sternberg, R. J., & Grigorenko, E. L. (2001). The (mis)organization of psychology. APS Observer, 14, 18. Stone, A. A., Cox, D. S., Valdimarsdottir, H., Jandorf, L., & Neale, J. M. (1987). Evidence that secretory IgA antibody is associated with daily mood. Journal of Personality and Social Psychology, 52, 988 –993. Tangney, J. P. (1991). Moral affect: The good, the bad, and the ugly. Journal of Personality and Social Psychology, 61, 598 – 607. Thrash, T. M., & Elliot, A. J. (2003). Inspiration as a psychological construct. Journal of Personality and Social Psychology, 84, 871– 889. Thrash, T. M. & Elliot, A. J. (2004). Inspiration: Core characteristics, component processes, antecedents, and function. Journal of Personality and Social Psychology, 87, 957–973. Tiedens, L. Z., & Linton, S. (2001). Judgment under emotional certainty and uncertainty: The effects of specific emotions on information processing. Journal of Personality and Social Psychology, 81, 973–988. Van Tilburg, M. A. L., Vingerhoets, J. J. M., & van Heck, G. L. (1996). Homesickness: A review of the literature. Psychological Medicine, 26, 899 –912. Vaughan, D. (1986). Uncoupling. New York: Oxford University Press. Victor, C. A., Scambler, S. J., Bowling, A., & Bond, J. (2005). The prevalence of, and risk factors for, loneliness in later life: A survey of older people in Great Britain. Ageing & Society, 25, 357–375. Walker, W. R., Skowronski, J. J., & Thompson, C. P. (2003). Life is
NOSTALGIA pleasant—And memory helps to keep it that way! Review of General Psychology, 7, 203–210. Watson, D. Clark, L. A., & Tellegen, A. (1988). Development and validation of brief measures of positive and negative affect: The PANAS scales. Journal of Personality and Social Psychology, 55, 1063–1070. Watson, D., Wiese, D., Vaidya, J., & Tellegen, A. (1999). The two general activation systems of affect: Structural findings, evolutionary considerations, and psychobiological evidence. Journal of Personality and Social Psychology, 76, 820 – 838. Werman, D. S. (1977). Normal and pathological nostalgia. Journal of the American Psychoanalytic Association, 25, 387–398.
993
Williams, K. D. (1997). Social ostracism. In R. M. Kowalski (Ed.), Aversive interpersonal behaviors (pp. 133–170). New York: Plenum Press. Williams, K. D., & Sommers, K. L. (1997). Social ostracism by one’s coworkers: Does rejection lead to loafing or compensation? Personality and Social Psychology Bulletin, 23, 693–706.
Received November 3, 2005 Revision received May 19, 2006 Accepted May 22, 2006 䡲
Call for Nominations The Publications and Communications (P&C) Board has opened nominations for the editorships of Journal of Applied Psychology, Psychological Bulletin, Psychology of Addictive Behaviors, Journal of Personality and Social Psychology: Interpersonal Relations and Group Processes (IRGP), and Journal of Educational Psychology for the years 2009-2014. Sheldon Zedeck, PhD, Harris Cooper, PhD, Howard J. Shaffer, PhD, Charles S. Carver, PhD, and Karen R. Harris, PhD, respectively, are the incumbent editors. Candidates should be members of APA and should be available to start receiving manuscripts in early 2008 to prepare for issues published in 2009. Please note that the P&C Board encourages participation by members of underrepresented groups in the publication process and would particularly welcome such nominees. Self-nominations are also encouraged. Search chairs have been appointed as follows: • • • • •
Journal of Applied Psychology, William C. Howell, PhD and J Gilbert Benedict, PhD Psychological Bulletin, Mark Appelbaum, PhD and Valerie F. Reyna, PhD Psychology of Addictive Behaviors, Linda P. Spear, PhD and Robert G. Frank, PhD Journal of Personality and Social Psychology: IRGP, David C. Funder, PhD Journal of Educational Psychology, Peter A. Ornstein, PhD and Leah L. Light, PhD
Candidates should be nominated by accessing APA’s EditorQuest site on the Web. Using your Web browser, go to http://editorquest.apa.org. On the Home menu on the left, find “Guests”. Next, click on the link “Submit a Nomination,” enter your nominee’s information, and click “Submit.” Prepared statements of one page or less in support of a nominee can also be submitted by e-mail to Susan J.A. Harris, P&C Board Search Liaison, at
[email protected]. Deadline for accepting nominations is January 10, 2007, when reviews will begin.
NOSTALGIA pleasant—And memory helps to keep it that way! Review of General Psychology, 7, 203–210. Watson, D. Clark, L. A., & Tellegen, A. (1988). Development and validation of brief measures of positive and negative affect: The PANAS scales. Journal of Personality and Social Psychology, 55, 1063–1070. Watson, D., Wiese, D., Vaidya, J., & Tellegen, A. (1999). The two general activation systems of affect: Structural findings, evolutionary considerations, and psychobiological evidence. Journal of Personality and Social Psychology, 76, 820 – 838. Werman, D. S. (1977). Normal and pathological nostalgia. Journal of the American Psychoanalytic Association, 25, 387–398.
993
Williams, K. D. (1997). Social ostracism. In R. M. Kowalski (Ed.), Aversive interpersonal behaviors (pp. 133–170). New York: Plenum Press. Williams, K. D., & Sommers, K. L. (1997). Social ostracism by one’s coworkers: Does rejection lead to loafing or compensation? Personality and Social Psychology Bulletin, 23, 693–706.
Received November 3, 2005 Revision received May 19, 2006 Accepted May 22, 2006 䡲
Call for Nominations The Publications and Communications (P&C) Board has opened nominations for the editorships of Journal of Applied Psychology, Psychological Bulletin, Psychology of Addictive Behaviors, Journal of Personality and Social Psychology: Interpersonal Relations and Group Processes (IRGP), and Journal of Educational Psychology for the years 2009-2014. Sheldon Zedeck, PhD, Harris Cooper, PhD, Howard J. Shaffer, PhD, Charles S. Carver, PhD, and Karen R. Harris, PhD, respectively, are the incumbent editors. Candidates should be members of APA and should be available to start receiving manuscripts in early 2008 to prepare for issues published in 2009. Please note that the P&C Board encourages participation by members of underrepresented groups in the publication process and would particularly welcome such nominees. Self-nominations are also encouraged. Search chairs have been appointed as follows: • • • • •
Journal of Applied Psychology, William C. Howell, PhD and J Gilbert Benedict, PhD Psychological Bulletin, Mark Appelbaum, PhD and Valerie F. Reyna, PhD Psychology of Addictive Behaviors, Linda P. Spear, PhD and Robert G. Frank, PhD Journal of Personality and Social Psychology: IRGP, David C. Funder, PhD Journal of Educational Psychology, Peter A. Ornstein, PhD and Leah L. Light, PhD
Candidates should be nominated by accessing APA’s EditorQuest site on the Web. Using your Web browser, go to http://editorquest.apa.org. On the Home menu on the left, find “Guests”. Next, click on the link “Submit a Nomination,” enter your nominee’s information, and click “Submit.” Prepared statements of one page or less in support of a nominee can also be submitted by e-mail to Susan J.A. Harris, P&C Board Search Liaison, at
[email protected]. Deadline for accepting nominations is January 10, 2007, when reviews will begin.
Rumor Psychology SOCIAL AND ORGANIZATIONAL APPROACHES Nicholas DiFonzo and Prashant Bordia Contents
I
n Rumor Psychology, expert rumor researchers Nicholas DiFonzo and Prashant Bordia investigate how rumors start and spread, the accuracy of different types of rumor, and how rumors can be controlled, particularly given their propagation across media outlets and within organizations. Rumor is closely entwined with a host of social and organizational phenomena, including social cognition, attitude formation and maintenance, prejudice and stereotyping, interpersonal and intergroup relations, and organizational trust and communication. Organizational rumors, in contrast with natural disaster rumors, tend to be highly accurate, with accuracy being affected by cognitive, motivational, situational, group, and network factors. The authors describe how managers can most effectively manage and refute rumors and infer that employee trust in management inhibits rumor activity. This book comes at any interesting time given the sociopolitical Zeitgeist, making the study of rumor accuracy, transmission, and propagation a high priority for the international intelligence community. It will also be of interest to social psychologists, organizational psychologists, and researchers in organizational communication, organizational behavior, management, human resource administration, as well as managers who regularly ALSO AVAILABLE encounter rumors.
Introduction Chapter 1: Defining Rumor Chapter 2: Forms, Frequency, and Fallout of Rumors Chapter 3: Psychological Factors in Rumor Spread Chapter 4: Factors Associated With Belief in Rumor Chapter 5: Rumor as Sense Making Chapter 6: Rumor Accuracy Chapter 7: Mechanisms Facilitating Rumor Accuracy and Inaccuracy Chapter 8: Trust and Organizational Rumor Transmission Chapter 9: Rumor Management Chapter 10: Conclusions References
2007. 392 pages. Hardcover. List: $69.95 APA Member/Affiliate: $49.95 ISBN 1-59147-426-4 • Item # 4316079 ISBN-13: 978-1-59147-426-5
Social Problem Solving Theory, Research, and Training Edited By Edward C. Chang, Thomas J. D’zurilla, and Lawrence J. Sanna
Perspectivism in Social Psychology The Yin and Yang of Scientific Progress
2004. 276 pages. Hardcover. LIST: $39.95 • APA MEMBER/AFFILIATE: $34.95 ISBN 1-59147-147-8 • ITEM # 4317049 ISBN-13: 978-1-59147-147-9
Edited By John T. Jost, Mahzarin R. Banaji, and Deborah A. Prentice 2004. 338 pages. Hardcover. LIST: $59.95 • APA MEMBERS/AFFILIATES: $49.95 ISBN 1-59147-022-6 • ITEM # 4316009 ISBN-13: 978-1-59147-022-9
Ethnopolitical Warfare Causes, Consequences, and Possible Solutions Edited by Daniel Chirot and Martin E. P. Seligman 2001. 379 pages. Hardcover. LIST: $49.95 • APA MEMBERS/AFFILIATES: $39.95 ISBN 1-59798-737-8 • ITEM # 431649A ISBN-13: 978-1-55798-737-2
APA BOOKS ORDERING INFORMATION
800-374-2721 www.apa.org/books In Washington DC, call: 202-336-5510 TDD/TTY: 202-336-6123 Fax: 202-336-5502 In Europe, Africa, or the Middle East call: 44-1767-604972
AD0483
Dark Ages The Case for a Science of Human Behavior
New from The MIT Press
Lee McIntyre
now in paperback
now in paperback
“A compact, urgent, and brutally frank challenge to ideologues left and right to drop their resistance to learning scientific truths about human behavior and apply the scientific method to social science.” — Harvey A. Silverglate, civil liberties lawyer and writer, and coauthor of The Shadow University: The Betrayal of Liberty on America’s Campuses
Brain Fiction
Language Development and Learning to Read
192 pp. $24.95
Self-Deception and the Riddle of Confabulation
William Hirstein “An excellent starting point for thinking about the biological mechanisms that help us make sense of the world.” — American Scientist online Philosophical Pathology: Disorders in Mind series • A Bradford Book 288 pp., 16 illus. $18 paper
now in paperback
How the Mind Explains Behavior Folk Explanations, Meaning, and Social Interaction
now in paperback
The Genesis of Animal Play
The Scientific Study of How Language Development Affects Reading Skill
Diane McGuinness Why do some children learn to read easily and quickly while some don’t learn to read at all? A reading expert analyzes the scientific research that might provide answers. A Bradford Book 480 pp., 7 illus. $23 paper
Testing the Limits
Bertram F. Malle
Gordon M. Burghardt
“This is a significant contribution to psychology, and it will have lasting value.” — Bernard Weiner, University of California, Los Angeles
“This excellent book is the definitive contemporary treatment of play behavior.” — Science Books & Films
To order call 800-405-1619.
A Bradford Book 328 pp., 20 illus. $19 paper
A Bradford Book 488 pp., 46 illus. $28 paper
http://mitpress.mit.edu
Publication Manual
of the American Psychological Association, Fifth Edition With millions of copies sold, the Publication Manual of the American Psychological Association is the style manual of choice for writers and students in psychology, sociology, business, economics, nursing, social work, criminology, and other disciplines in which effective communication with words and data is fundamental. The Fifth E dition has been revised and updated to include The latest guidelines and examples for referencing electronic and on-line sources New and revised guidelines for transmitting papers electronically Simplified formatting guidelines for writers using up-to-date word-processing software Updates on copyright and permissions issues for writers An expanded and improved index for quick and easy access. New and experienced readers alike will find the Fifth Edition a complete and essential resource for writing, presenting, or publishing with clarity and persuasiveness. 2001. 439 PAGES.
Order today! Call 1-800-374-2721 or visit www.apa.org/books.
SOFTCOVER: List: $26.95 Item # 4200060 • ISBN: 1-55798-791-2 HARDCOVER: List: $39.95 Item # 4200061 • ISBN: 1-55798-790-4 LAY-FLAT SPIRAL BINDING: List: $33.95 Item # 4200063 • ISBN: 1-55798-810-2
ALSO OF INTEREST… visit www.apastyle.org for information about new software and other resources for use with the new Publication Manual, Fifth Edition. AD0148
Dark Ages The Case for a Science of Human Behavior
New from The MIT Press
Lee McIntyre
now in paperback
now in paperback
“A compact, urgent, and brutally frank challenge to ideologues left and right to drop their resistance to learning scientific truths about human behavior and apply the scientific method to social science.” — Harvey A. Silverglate, civil liberties lawyer and writer, and coauthor of The Shadow University: The Betrayal of Liberty on America’s Campuses
Brain Fiction
Language Development and Learning to Read
192 pp. $24.95
Self-Deception and the Riddle of Confabulation
William Hirstein “An excellent starting point for thinking about the biological mechanisms that help us make sense of the world.” — American Scientist online Philosophical Pathology: Disorders in Mind series • A Bradford Book 288 pp., 16 illus. $18 paper
now in paperback
How the Mind Explains Behavior Folk Explanations, Meaning, and Social Interaction
now in paperback
The Genesis of Animal Play
The Scientific Study of How Language Development Affects Reading Skill
Diane McGuinness Why do some children learn to read easily and quickly while some don’t learn to read at all? A reading expert analyzes the scientific research that might provide answers. A Bradford Book 480 pp., 7 illus. $23 paper
Testing the Limits
Bertram F. Malle
Gordon M. Burghardt
“This is a significant contribution to psychology, and it will have lasting value.” — Bernard Weiner, University of California, Los Angeles
“This excellent book is the definitive contemporary treatment of play behavior.” — Science Books & Films
To order call 800-405-1619.
A Bradford Book 328 pp., 20 illus. $19 paper
A Bradford Book 488 pp., 46 illus. $28 paper
http://mitpress.mit.edu
Publication Manual
of the American Psychological Association, Fifth Edition With millions of copies sold, the Publication Manual of the American Psychological Association is the style manual of choice for writers and students in psychology, sociology, business, economics, nursing, social work, criminology, and other disciplines in which effective communication with words and data is fundamental. The Fifth E dition has been revised and updated to include The latest guidelines and examples for referencing electronic and on-line sources New and revised guidelines for transmitting papers electronically Simplified formatting guidelines for writers using up-to-date word-processing software Updates on copyright and permissions issues for writers An expanded and improved index for quick and easy access. New and experienced readers alike will find the Fifth Edition a complete and essential resource for writing, presenting, or publishing with clarity and persuasiveness. 2001. 439 PAGES.
Order today! Call 1-800-374-2721 or visit www.apa.org/books.
SOFTCOVER: List: $26.95 Item # 4200060 • ISBN: 1-55798-791-2 HARDCOVER: List: $39.95 Item # 4200061 • ISBN: 1-55798-790-4 LAY-FLAT SPIRAL BINDING: List: $33.95 Item # 4200063 • ISBN: 1-55798-810-2
ALSO OF INTEREST… visit www.apastyle.org for information about new software and other resources for use with the new Publication Manual, Fifth Edition. AD0148
BEST SELLERS from the American Psychological Association
Handbook of Multimethod Measurement in Psychology
Emotion-Focused Therapy for Depression
MMPI-2: A Practitioner’s Guide
Edited by Michael Eid and Ed Diener
Leslie S. Greenberg, and Jeanne C. Watson
2006. 464 pages. Hardcover. List: $99.95 APA/Member Affiliate: $64.95 ISBN 1-59147-318-7 • Item # 4316063
2006. 344 pages. Hardcover. List: $59.95 APA Member/Affiliate: $49.95 ISBN 1-59147-280-6 • Item # 4317081
2006. 688 pages. Hardcover. List: $89.95 APA Member/Affiliate: $64.95 ISBN 1-59147-287-3 • Item # 4317086
Adlerian Therapy
Bullying Prevention
Theory and Practice
Creating a Positive School Climate and Developing Social Competence
Jon Carlson, Richard E. Watts, and Michael Maniacci
Pamela Orpinas and Arthur M. Horne
2006. 304 pages. Hardcover. List: $59.95 APA Member/Affiliate: $44.95 ISBN 1-59147-285-7 • Item # 4317084
2006. 293 pages. Hardcover. List: $59.95 APA Member/Affiliate: $49.95 ISBN 1-59147-282-2 • Item # 4317082
Constructive Divorce
Road Rage
Procedural Justice and Sociolegal Reform
Assessment and Treatment of the Angry, Aggressive Driver
Penelope Eileen Bryan
Tara E. Galovski, Loretta S. Malta, and Edward B. Blanchard
2006. 344 pages. Hardcover. List: $69.95 APA Member/Affiliate: $54.95 ISBN 1-59147-279-2 • Item # 4316056
2006. 256 pages. Hardcover. List: $59.95 APA Member/Affiliate: $44.95 ISBN 1-59147-305-5 • Item # 4317088
Your Practicum in Psychology A Guide for Maximizing Knowledge and Competence
The Ethical Practice of Psychology in Organizations
Edited by Janet R. Matthews and C. Eugene Walker
Second Edition
2006. 320 pages. Paperback. List: $49.95 APA Member/Affiliate: $39.95 ISBN 1-59147-328-4 • Item # 4313006
2006. 330 pages. Hardcover. List: $39.95 APA Member/Affiliate:$34.95 ISBN 1-59147-353-5 • Item # 4312006
Essentials of Clinical Hypnosis
Practical Ethics for Psychologists
An Evidence-Based Approach
A Positive Approach
Steven Jay Lynn and Irving Kirsch
Samuel J. Knapp and Leon Vandecreek
List: $69.95 APA Member/Affiliate: $49.95 ISBN 1-59147-344-6 • Item # 4317095
2006. 304 pages. Paperback. List: $49.95 APA Member/Affiliate: $39.95 ISBN 1-59147-326-8 • Item # 4312004
Edited by Rodney L. Lowman
Women and Victimization Contributing Factors, Interventions, and Implications
Ethical Practice in Small Communities
TK Logan, Robert Walker, Carol E. Jordan and Carl G. Leukefeld
Challenges and Rewards for Psychologists
2006. 408 pages. Hardcover. List: $79.95 APA Member/Affiliate: $49.95 ISBN 1-59147-316-0 • Item # 4316059
Janet A. Schank and Thomas M. Skovholt
Sex and Love in Intimate Relationships
2006. 184 pages. Paperback. List: $49.95 APA Member/Affiliate: $39.95 ISBN 1-59147-346-2 • Item # 4312005
Research and Evidence-Based Intervention Strategies
2006. 304 pages. Hardcover. List: $49.95 APA Member/Affiliate: $39.95 ISBN 1-59147-286-5 • Item # 4317085
Edited by John R. Lutzker
Examining Identity and Development in Lesbian, Gay, and Bisexual People
Edited by Allen M. Omoto and Howard S. Kurtzman 2006. 336 pages. Hardcover. List: $69.95 APA Member/Affiliate: $49.95 ISBN 1-59147-232-6 • Item # 4316052
Psychological Interventions in Childhood Chronic Illness Dennis Drotar 2006. 404 pages. Hardcover. List: $59.95 APA Member/Affiliate: $49.95 ISBN 1-59147-330-6 • Item # 4318027
Psychological Treatment of Chronic Illness A Biopsychosocial Therapy Approach
Len Sperry 2006. 311 pages. Hardcover. List: $59.95 APA Member/Affiliate: $49.95 ISBN 1-59147-354-3 • Item # 4317096
Chronic Depression Interpersonal Sources, Therapeutic Solutions
Jeremy W. Pettit and Thomas E. Joiner 2006. 232 pages. Hardcover. List: $59.95 APA Member/Affiliate: $44.95 ISBN 1-59147-306-3 • Item # 4317089
Therapeutic Alliances in Couple and Family Therapy An Empirically-Informed Guide to Practice
Myrna L. Friedlander, Valentín Escudero, and Laurie Heatherington 2006. 352 pages. Hardcover. List: $59.95 APA Member/Affiliate $49.95 ISBN 1-59147-331-4 • Item # 4317093
Couple Power Therapy Building Commitment, Cooperation, Communication, and Community in Relationships
Peter L. Sheras and Phyllis R. Koch-Sheras
Imagery in Psychotherapy
2006. 264 pages. Hardcover. List: $49.95 APA Member/Affiliate: $39.95 ISBN 1-59147-235-0 • Item # 4317074
Jerome L. Singer
Preventing Violence
Robert W. Firestone, Lisa A. Firestone, and Joyce Catlett
Sexual Orientation and Mental Health
Edited by James N. Butcher
2006. 320 pages. Hardcover. List: $69.95 APA Member/Affiliate: $49.95 ISBN 1-59147-342-X • Item # 4316067
Preventing Youth Violence in a Multicultural Society Edited by Nancy G. Guerra and Emilie Phillips Smith 2006. 304 pages. Hardcover. List: $69.95 APA Member/Affiliate: $49.95 ISBN 1-59147-327-6 • Item # 4316064
Reforming Punishment
2006. 280 pages. Hardcover. List: $69.95 APA Member/Affiliate: $49.95 ISBN 1-59147-333-0 • Item # 4316066
Psychological Limits to the Pain of Imprisonment
Emerging Adults in America
Craig Haney
Edited by Jeffrey Jensen Arnett and Jennifer Lynn Tanner
2006. 320 pages. Hardcover. List: $79.95 APA Member/Affiliate: $49.95 ISBN 1-59147-317-9 • Item # 4316061
Identity and Story Creating Self in Narrative
Edited by Dan P. McAdams, Ruthellen Josselson, and Amia Lieblich 2006. 357 pages. Hardcover. List: $59.59 APA Member/Affiliate: $49.95 ISBN 1-59147-356-X • Item # 4316071
Coming of Age in the 21st Century
2006. 376 pages. Hardcover. List: $79.95 APA Members/Affiliates: $49.95 ISBN 1-59147-329-2 • Item # 4317092
In the Room With Men A Casebook of Therapeutic Change
Edited by Matt Englar-Carlson and Mark A. Stevens 2006. 376 pages. Hardcover. List: $59.95 APA Member/Affiliate $49.95 ISBN 1-59147-332-2 • Item # 4317094 AD0457
To O r d e r : 8 0 0 - 374 -2721 • w w w. a p a .or g / b o o ks
Psychology and Economic Injustice PERSONAL, PROFESSIONAL, AND POLITICAL INTERSECTIONS Bernice Lott and Heather E. Bullock
Contents
S
ocial class is a powerful contributor to physical, social, and mental well-being. Psychology and Economic Injustice, which is part of a series on critical social issues addressed by psychologists empirically, politically, and in clinical practice, blends the personal experiences of feminist psychologists with empirical data. It argues persuasively for examining the relationship between economic disparity and gender, and for the need to propose and support policies to ensure positive social change. Authors Bernice Lott and Heather E. Bullock present narratives of their experiences as psychologists who grew up in working class and low-income families. They trace the relationship between their own development as feminist psychologists and their concern with social and economic justice. Drawing from diverse sources, they present data on social class and poverty, and tie these to social psychology and feminist perspectives. Empirical literature is described to illustrate the injustices faced by families in what is described as similar to racism in America—classism. The volume ends with recommendations for policy, education, further research, and citizen advocacy, while moving from psychology to state and federal issues. Researchers and practitioners in psychology, sociology, social work, or education, with an interest in social problems, gender issues, and multiculturalism, will find this to ALSO AVAILABLE be a noteworthy contribution to the field.
Chapter 1: Introduction: Social Class and Economic Injustice Chapter 2: Growing Up Poor and Middle Class: Heather Bullock’s Story Chapter 3: Working-Class Origins and Class-Conscious Awakening: Bernice Lott’s Story Chapter 4: Psychology, Social Class, and Resources for Human Welfare Chapter 5: The Psychology and Politics of Class Warfare Chapter 6: Professional Activism for Social Change and Economic Justice References Series: Psychology of Women 2007. 192 pages. Hardcover. List: $59.95 APA Member/Affiliate: $49.95 ISBN 1-59147-429-9 • Item # 4316082 ISBN-13: 978-1-59147-429-6
Defining Difference Race and Racism in the History of Psychology Edited by Andrew S. Winston
Featuring Females Feminist Analyses of Media
2004. 303 pages. Hardcover. LIST: $49.95 • APA MEMBER/AFFILIATE: $39.95 ISBN 1-59147-027-7 • ITEM # 4316011 ISBN-13: 978-1-59147-027-4
Edited By Ellen Cole and Jessica Henderson Daniel 2005. 231 pages. Hardcover. LIST: $69.95 • APA MEMBER/AFFILIATE: $49.95 ISBN 1-59147-278-4 • ITEM # 4316055 ISBN-13: 978-1-59147-278-0
Teaching and Social Justice Integrating Multicultural and Feminist Theories in the Classroom Edited by Carolyn Zerbe Enns and Ada L. Sinacore 2005. 257 pages. Hardcover. LIST: $49.95 • APA MEMBER/AFFILIATE: $39.95 ISBN 1-59147-167-2 • ITEM # 4318014 ISBN-13: 978-1-59147-167-7
APA BOOKS ORDERING INFORMATION
800-374-2721 www.apa.org/books In Washington DC, call: 202-336-5510 TDD/TTY: 202-336-6123 Fax: 202-336-5502 In Europe, Africa, or the Middle East call: 44-1767-604972
AD0482
30%#)!, 4)4,%3 vÀÊÌ
iÊiÀV>Ê*ÃÞV
}V>ÊÃÃV>Ì
ÜÜÜ°>«>°À}ÉLÃ !0! $ICTIONARY OF 0SYCHOLOGY
%VIDENCE "ASED 0SYCHOTHERAPY
%DITOR IN #HIEF 'ARY 2 6ANDEN"OS
7HERE 0RACTICE AND 2ESEARCH -EET
PAGES (ARDCOVER ,IST !0! -EMBER!FlLIATE )3". s )TEM
%DITED BY #AROL $ 'OODHEART !LAN % +AZDIN AND 2OBERT * 3TERNBERG PAGES (ARDCOVER ,IST !0! -EMBER!FlLIATE )3". s )TEM
2EVIEWING 3CIENTIlC 7ORKS IN 0SYCHOLOGY
%VIDENCE "ASED 0RACTICES IN -ENTAL (EALTH
%DITED "Y 2OBERT * 3TERNBERG PAGES 0APERBACK ,IST !0! -EMBER!FlLIATE )3". s )TEM
$EBATE AND $IALOGUE ON THE &UNDAMENTAL