Journal of Personality and Social Psychology, 2006, Vol. 91-01

Volume 91 Number 1 Published monthly by the American Psychological Association

July 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.html

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; Greg Long, Production Account Manager; Julie PalmerHoffman, Manuscript Editor; 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 2006 rates follow: Nonmember Individual: $421 Domestic, $464 Foreign, $491 Air Mail. Institutional: $1,249 Domestic, $1,340 Foreign, $1,367 Air Mail. APA Member: $202. 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.

APA DICTIONARY OF PSYCHOLOGY Editor-in-Chief: Gary R. VandenBos 2006. 1,008 pages. Hardcover. List: $49.95 • APA Member/Affiliate: $39.95 ISBN 1-59147-380-2 ISBN-13: 978-1-59147-380-0 Item # 4311007

BECOMING CULTURALLY ORIENTED Practical Advice for Psychologists and Educators Nadya A. Fouad and Patricia Arredondo 2007. 208 pages. Hardcover. List: $49.95 • APA Member/Affiliate: $39.95 ISBN 1-59147-424-8 ISBN-13: 978-1-59147-424-1 Item # 4317114

SPIRITUAL APPROACHES IN THE TREATMENT OF WOMEN WITH EATING DISORDERS P. Scott Richards, Randy K. Hardman, and Michael E. Berrett 2007. 312 pages. Hardcover. List: $59.95 • APA Member/Affiliate: $49.95 ISBN 1-59147-393-4 ISBN-13: 978-1-59147-393-0 Item # 4317103

GRADUATE STUDY IN PSYCHOLOGY, 2007 2007. 832 pages. Paperback. List: $24.95 • APA Member/Affiliate: $21.95 ISBN 1-59147-423-X ISBN-13: 978-1-59147-423-4 Item # 4270090

DISORDERS OF THE SELF A Personality-Guided Approach Marshall L. Silverstein 2007. 320 pages. Hardcover. List: $69.95 • APA Member/Affiliate: $49.95 ISBN 1-59147-430-2 ISBN-13: 978-1-59147-430-2 Item # 4317116

CHILD DEVELOPMENT AND SOCIAL POLICY Knowledge for Action

Second Edition Edited by Kathleen J. Bieschke, Ruperto M. Perez, and Kurt A. DeBord 2007. 464 pages. Hardcover. List: $79.95 • APA Member/Affiliate: $49.95 ISBN 1-59147-421-3 ISBN-13: 978-1-59147-421-0 Item # 4317113

INTERVENING IN CHILDREN’S LIVES An Ecological, Family-Centered Approach to Mental Health Care Thomas J. Dishion and Elizabeth A. Stormshak 2007. 320 pages. Hardcover. List: $69.95 • APA Member/Affiliate: $49.95 ISBN 1-59147-428-0 ISBN-13: 978-1-59147-428-9 Item # 4317115

PREVENTING YOUTH SUBSTANCE ABUSE Science-Based Programs for Children and Adolescents Edited by Patrick Tolan, José Szapocznik, and Soledad Sambrano 2007. 264 pages. Hardcover. List: $69.95 • APA Member/Affiliate: $49.95 ISBN 1-59147-307-1 ISBN-13: 978-1-59147-307-7 Item # 4316058

SCIENTIFIC JURY SELECTION Joel D. Lieberman and Bruce D. Sales 2007. 264 pages. Hardcover. List: $79.95 • APA Member/Affiliate: $49.95 ISBN 1-59147-427-2 ISBN-13: 978-1-59147-427-2 Item # 4316081

PRIMATE PERSPECTIVES ON BEHAVIOR AND COGNITION Edited by David A. Washburn 2007. 368 pages. Hardcover. List: $79.95 • APA Member/Affiliate: $49.95 ISBN 1-59147-422-1 ISBN-13: 978-1-59147-422-7 Item # 4318035

AD0456

Edited by J. Lawrence Aber, Sandra J. BishopJosef, Stephanie M. Jones, Kathryn Taaffe McLearn, and Deborah A. Phillips 2007. 352 pages. Hardcover. List: $79.95 • APA Member/Affiliate: $49.95 ISBN 1-59147-425-6 ISBN-13: 978-1-59147-425-8 Item # 4318036

HANDBOOK OF COUNSELING AND PSYCHOTHERAPY WITH LESBIAN, GAY, BISEXUAL, AND TRANSGENDER CLIENTS

1-800-374-2721 • WWW.APA.ORG/BOOKS

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.html July 2006 VOLUME 91 NUMBER 1

Copyright © 2006 by the American Psychological Association

Attitudes and Social Cognition 1

Alone but Feeling No Pain: Effects of Social Exclusion on Physical Pain Tolerance and Pain Threshold, Affective Forecasting, and Interpersonal Empathy C. Nathan DeWall and Roy F. Baumeister

16

The Single Category Implicit Association Test as a Measure of Implicit Social Cognition Andrew Karpinski and Ross B. Steinman

33

Thinking Within the Box: The Relational Processing Style Elicited by Counterfactual Mind-Sets Laura J. Kray, Adam D. Galinsky, and Elaine M. Wong

49

Self-Regulatory Processes Defend Against the Threat of Death: Effects of Self-Control Depletion and Trait Self-Control on Thoughts and Fears of Dying Matthew T. Gailliot, Brandon J. Schmeichel, and Roy F. Baumeister

Interpersonal Relations and Group Processes 63

Peacocks, Picasso, and Parental Investment: The Effects of Romantic Motives on Creativity Vladas Griskevicius, Robert B. Cialdini, and Douglas T. Kenrick

77

Navigating the Interdependence Dilemma: Attachment Goals and the Use of Communal Norms With Potential Close Others Jennifer A. Bartz and John E. Lydon

97

Intergroup Helping as Status Relations: Effects of Status Stability, Identification, and Type of Help on Receptivity to High-Status Group’s Help Arie Nadler and Samer Halabi

111

Information Quantity and Quality Affect the Realistic Accuracy of Personality Judgment Tera D. Letzring, Shannon M. Wells, and David C. Funder

124

Supplication and Appeasement in Conflict and Negotiation: The Interpersonal Effects of Disappointment, Worry, Guilt, and Regret Gerben A. Van Kleef, Carsten K. W. De Dreu, and Antony S. R. Manstead

Personality Processes and Individual Differences 143

Optimism in Close Relationships: How Seeing Things in a Positive Light Makes Them So Sanjay Srivastava, Kelly M. McGonigal, Jane M. Richards, Emily A. Butler, and James J. Gross

(contents continue)

154

Discrepancies Between Explicit and Implicit Self-Concepts: Consequences for Information Processing Pablo Brin˜ol, Richard E. Petty, and S. Christian Wheeler

171

Investigating the Dopaminergic Basis of Extraversion in Humans: A Multilevel Approach Jan Wacker, Mira-Lynn Chavanon, and Gerhard Stemmler

188

Possible Selves and Academic Outcomes: How and When Possible Selves Impel Action Daphna Oyserman, Deborah Bybee, and Kathy Terry

Other 15 ii 96 62

American Psychological Association Subscription Claims Information E-Mail Notification of Your Latest Issue Online! Instructions to Authors Subscription Order Form

ii

154

Discrepancies Between Explicit and Implicit Self-Concepts: Consequences for Information Processing Pablo Brin˜ol, Richard E. Petty, and S. Christian Wheeler

171

Investigating the Dopaminergic Basis of Extraversion in Humans: A Multilevel Approach Jan Wacker, Mira-Lynn Chavanon, and Gerhard Stemmler

188

Possible Selves and Academic Outcomes: How and When Possible Selves Impel Action Daphna Oyserman, Deborah Bybee, and Kathy Terry

Other 15 ii 96 62

American Psychological Association Subscription Claims Information E-Mail Notification of Your Latest Issue Online! Instructions to Authors 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 JACQUES-PHILIPPE LEYENS Catholic University of Louvain, Louvain-la-Neuve, Belgium ANTONY MANSTEAD Cardiff University, Cardiff, United Kingdom CYNTHIA L. PICKETT University of California, Davis JEFFRY A. SIMPSON University of Minnesota, Twin Cities Campus SCOTT TINDALE Loyola University Chicago JACQUIE D. VORAUER University of Manitoba, Winnipeg, Manitoba, Canada CONSULTING EDITORS DOMINIC ABRAMS University of Kent at Canterbury, Canterbury, England

XIMENA ARRIAGA Purdue University 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 B. ANNE BETTENCOURT University of Missouri—Columbia GERD BOHNER Universita¨t Bielefeld, Bielefeld, Germany NIALL BOLGER New York University

LORNE CAMPBELL University of Western Ontario, London, Ontario, Canada 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 DAVID DESTENO Northeastern University STEVE DRIGOTAS Johns Hopkins University ELISSA S. EPEL University of California, San Francisco

NYLA R. BRANSCOMBE University of Kansas

CHRIS AGNEW Purdue University

JONATHON D. BROWN University of Washington

VICTORIA ESSES University of Western Ontario, London, Ontario, Canada

ARTHUR ARON State University of New York at Stony Brook

RUPERT BROWN The University of Kent at Canterbury, Canterbury, England

BEVERLY FEHR University of Winnipeg, Winnipeg, Manitoba, Canada

(editors continue)

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 MARTIE G. HASSELTON University of California, Los Angeles S. ALEXANDER HASLAM University of Exeter, Exeter, United Kingdom 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—JESSICA LILLESAND

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

Alone but Feeling No Pain: Effects of Social Exclusion on Physical Pain Tolerance and Pain Threshold, Affective Forecasting, and Interpersonal Empathy C. Nathan DeWall and Roy F. Baumeister Florida State University Prior findings of emotional numbness (rather than distress) among socially excluded persons led the authors to investigate whether exclusion causes a far-reaching insensitivity to both physical and emotional pain. Experiments 1– 4 showed that receiving an ostensibly diagnostic forecast of a lonesome future life reduced sensitivity to physical pain, as indicated by both (higher) thresholds and tolerance. Exclusion also caused emotional insensitivity, as indicated by reductions in affective forecasting of joy or woe over a future football outcome (Experiment 3), as well as lesser empathizing with another person’s suffering from either romantic breakup (Experiment 4) or a broken leg (Experiment 5). The insensitivities to pain and emotion were highly intercorrelated. Keywords: social exclusion, rejection, emotion, affective forecasting, empathy

Social exclusion poses a serious threat to a person’s physical and psychological well-being, as indicated by higher rates of physical and mental illnesses among people who live alone as compared with people who have strong social networks (Argyle, 1987; Cacioppo, Hawkley, & Berntson, 2003; Kiecolt-Glaser et al., 1984; Lynch, 1979; Myers, 1992). These findings confirm that social exclusion has potentially drastic and negative effects on physical and psychological health, but they offer no explanation for the finding that people respond to social exclusion with emotional numbness (e.g., Twenge et al., 2001). Indeed, they underscore the paradox in the fact that people would respond to a potentially momentous event, such as social exclusion, with emotional numbness and detachment. The first purpose of the current investigation was to resolve the paradox as to why rejected people report emotional numbness, by identifying a possible mechanism by which people respond to social exclusion. We propose that certain interpersonal events, such as social rejection, activate the body’s pain response system and potentially alter how it registers physical and emotional pain (Eisenberger, Lieberman, & Williams, 2003). With regard to physical pain, social exclusion may disrupt the ability to respond to physical pain in the same manner as people who have not experienced social exclusion. This would lead to increases in both pain threshold (i.e., sensitivity to pain) and pain tolerance (i.e., withstanding greater pain). Hence, the first goal of this research (tested in Experiments 1– 4) was to demonstrate that socially excluded people become relatively numb to physical pain. The second goal of this work was to extend the idea of physical numbness to emotional functioning. As suggested by MacDonald and Leary (2005), if the body uses the same system to respond to physical injury and interpersonal injury, then physical pain and

People depend heavily on others for much of their physical and mental well-being. With no fangs, no claws, an extremely prolonged childhood phase of dependency and vulnerability, and other physical weaknesses, human beings are not well suited to living in isolation from others. Given the importance of acquiring and maintaining membership in social groups, it is therefore hardly surprising that people would react strongly to any threat of social exclusion. Multiple laboratory studies of social exclusion have found, however, that people respond to social exclusion in a seemingly detached and emotionally indifferent manner. Socially excluded people often report emotional states that do not differ significantly from participants in acceptance or control conditions (Baumeister, Twenge, & Nuss, 2002; Gardner, Pickett, & Brewer, 2000; Twenge, Baumeister, Tice, & Stucke, 2001; Twenge & Campbell, 2003; Twenge, Catanese, & Baumeister, 2002; Zadro, Williams, & Richardson, 2004).

C. Nathan DeWall and Roy F. Baumeister, Department of Psychology, Florida State University. We gratefully acknowledge support by National Institute of Mental Health Grant MH-65559. The research reported in this article was part of C. Nathan DeWall’s master’s thesis at Florida State University under the direction of Roy F. Baumeister. We graciously thank committee members Thomas Joiner, Jon K. Maner, and Dianne M. Tice for their helpful comments and suggestions. We also thank Carey Morewedge for helpful comments on an earlier version of this article. Correspondence concerning this article should be addressed to C. Nathan DeWall, Department of Psychology, Florida State University, Tallahassee, FL 32306-1270. E-mail: [email protected]

Journal of Personality and Social Psychology, 2006, Vol. 91, No. 1, 1–15 Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.1

1

2

DEWALL AND BAUMEISTER

interpersonal emotions may be linked—and just as the body goes numb to pain, it may also become less sensitive to emotion. If the emotional system ceases to function normally in the aftermath of social exclusion, people might show abnormalities not only in their emotional reactions to current events but also in their forecasts of emotional responses to future events. They might also lose their empathy for the physical suffering of another person. Last, and ironically, rejected people might even lose their empathy for the suffering of someone else who has also suffered rejection. Experiments 3–5 tested these predictions.

Social Exclusion and Emotional Responding: The Paradox of Detachment and Numbness A desire to form and maintain social bonds with others is a fundamental and pervasive desire among humans (Baumeister & Leary, 1995). Insofar as belongingness is a basic need rather than a want, people experience considerable difficulties when their need to belong is unsatisfied or frustrated. A lack of stable relationships also has detrimental effects on one’s health (Cacioppo, Hawkley, & Berntson, 2003; Hawkley, Burleson, Berntson, & Cacioppo, 2003; Lynch, 1979; see Uchino, Cacioppo, & Kiecolt-Glaser, 1996, for a review). One might assume and hope that excluded people would show adaptive responses of changing themselves so as to become more socially attractive or to take better care of themselves as individuals. The opposite is often found, however. Rejected people exhibit decreased intellectual functioning (Baumeister et al., 2002), become more aggressive toward others (Buckley, Winkel, & Leary, 2004; Twenge et al., 2001; Twenge & Campbell, 2003), are less willing to self-regulate (Baumeister, DeWall, Ciarocco, & Twenge, 2005), and engage in various self-defeating behaviors, such as risk taking and procrastination (Twenge et al., 2002). None of these responses seem likely to foster interpersonal acceptance or personal health and well-being. As already noted, socially excluded participants frequently report emotional states that do not differ significantly from socially accepted and control participants (Baumeister et al., 2002; Gardner et al., 2000; Twenge et al., 2002; Zadro et al., 2004). Even more surprisingly, when differences in emotion are found, these differences do not mediate the behavioral consequences of social exclusion. Even researchers who find that social exclusion increases emotional distress have not found that the distress mediates the behavioral consequences of social exclusion (Buckley et al., 2004; Williams, Cheung, & Choi, 2000). Given the fundamental nature of the need to belong, one might reasonably expect that real, potential, or imagined social exclusion would result in severe emotional reactions. It is therefore unclear why emotion plays such a minor role in explaining behavioral responses to social exclusion. One reason socially excluded people report feelings of emotional numbness may be that exclusion leads to a defensive state of cognitive deconstruction. The deconstructed state has been invoked to characterize presuicidal mental states (Baumeister, 1990), and it is marked by emotional numbness, an altered perception of time, thoughts of meaninglessness, lethargy, and avoidance of self-focused attention. Twenge, Catanese, and Baumeister (2003) showed that socially excluded people exhibited each of these symptoms of cognitive deconstruction. Social exclusion leads to behaviors that could preclude the possibility of gaining future

acceptance (e.g., aggression, self-defeating behavior), but the deconstructed state may offer rejected people a temporary reprieve from feeling the intense pain or distress that can accompany threats to belongingness. Although escaping from aversive emotional states and high self-awareness may be beneficial in the short-term, such behaviors are characteristic of severe psychopathology in clinical populations (Bancroft, Skirmshire, & Simkins, 1976; Hawton, Cole, O’Grady, & Osborn, 1982) and of a variety of selfdestructive behaviors in nonclinical samples (see Baumeister & Scher, 1988, for a review). The experiments reported in this article sought to link emotional unresponsiveness with insensitivity to physical pain, both of which could be natural defenses against one’s own suffering and could contribute to the deconstructed state. If people have normal or natural defenses that help them escape or minimize emotional distress, then social exclusion may temporarily impair the ability to experience emotions in a normal fashion. The theory proposed in the current article is that the emotion system temporarily ceases to function normally in response to social exclusion. The resulting emotional numbness could be linked to an insensitivity to physical pain, as the next section will explain.

Social and Physical Pain There is reason to believe that the pain of social exclusion shares many of the same neural and psychological mechanisms as experiences of physical pain. At a purely linguistic level, people frequently use words connoting physical pain when describing emotional responses to distressing events. For example, people may report feeling “hurt” or “crushed” following the dissolution of a meaningful relationship (Leary & Springer, 2000). MacDonald and Leary (2005) have suggested that the similarity between descriptions of social and physical pain extends beyond mere metaphor. Specifically, they proposed that social pain and physical pain operate via shared physiological mechanisms, including the anterior cingulate cortex, periaqueductal gray (PAG) brain structures, and the opioid and oxytocin neuroendocrine systems (see also Rossi, Pasternak, & Bodnar, 1993). Decades ago, Panksepp and colleagues proposed a link between social and physical pain (Herman & Panksepp, 1978; Panksepp, Herman, Conner, Bishop, & Scott, 1978; Panksepp, Vilberg, Bean, Coy, & Kastin, 1978). They suggested that as evolution prepared animals for increasing social interaction, instead of creating entirely new systems to react to social events, such as being rejected or excluded, it piggybacked these responses onto the existing systems that were hard-wired for responding to physical pain. Social events might therefore activate the body’s pain response system and potentially alter how it would register physical pain. Recent neuroscience and functional MRI research has shown that the anterior cingulate cortex (ACC) functions as a neural alarm system to warn people that factors in their environment threaten their goals (Bush, Luu, & Posner, 2000; Eisenberger & Lieberman, 2004; Eisenberger et al., 2003; Kimbrell et al., 1999; E. E. Nelson & Panksepp, 1998). Eisenberger et al. (2003) showed that the ACC plays a prominent role in the detection of threats to belongingness. In their functional MRI study, ostracized people showed activation of the dorsal ACC. Ostracized people also showed activation of the right ventral prefrontal cortex, which has

SOCIAL EXCLUSION AND PAIN

been previously linked to the regulation of physical pain distress and negative affect (Hariri, Bookheimer, & Mazziotta, 2000; Petrovic, Kalso, Petersson, & Ingvar, 2002). Thus, preliminary evidence suggests that Panksepp and colleagues (Herman & Panksepp, 1978; Panksepp, Herman, et al., 1978; Panksepp, Vilberg, et al., 1978) were correct in proposing that certain physiological processes responsible for detection and regulation of physical pain were co-opted to sense and respond to emotionally painful events, such as being rejected or excluded. Additional research has shown that the PAG brain structures, which receive input from the body’s injury detection system (nociceptive system) and ACC, are involved in the detection of physical pain and are implicated in animal bonding behavior (Craig & Dostrovsky, 1999). Panksepp (1998) demonstrated that activation of the PAG elicited separation distress cries from rats, and further evidence showed that lesions to the PAG led to reduced separation distress cries (Wiedenmayer, Goodwin, & Barr, 2000). Administration of the neuropeptide oxytocin and opioids such as morphine have also been shown to reduce separation distress cries in rats (Carden, Hernandez, & Hofer, 1996; Carden & Hofer, 1990; Insel & Winslow, 1991). These findings indicate that some physiological systems respond to both physical pain and socially distressing events. Most relevant to the current investigation, however, is evidence that separation from caregivers and isolation from conspecifics results in decreased sensitivity to physical pain, also known as analgesia. Several studies have shown that short-term isolation produces reduced pain sensitivity in rat pups (Kehoe & Blass, 1986a, 1986b; Naranjo & Fuentes, 1985; Spear, Enters, Aswad, & Louzan, 1985), mice (Konecka & Sroczynska, 1990), cows (Rushen, Boissy, Terlouw, & de Passille´, 1999), and chicks (Sufka & Hughes, 1990; Sufka & Weed, 1994; see MacDonald & Leary, 2005, for a review). Thus, threats to belongingness appear to activate neural mechanisms associated with physical pain and the regulation of pain in some nonhuman species (E. E. Nelson & Panksepp, 1998; Panksepp, 1998). Previous research has shown that social and physical pain share common physiological mechanisms in some animals, but research has not provided much evidence about whether a similar link exists in humans. MacDonald, Kingsbury, and Shaw (2005) showed that people high in rejection sensitivity responded to ostracism with decreased sensitivity to physical pain (pain threshold) on the cold-pressor task, but people low in rejection sensitivity did not show the same pattern of responding. If threats to belongingness activate basic neural mechanisms designed to regulate pain (Eisenberger et al., 2003), one would expect all ostracized participants to exhibit decreased sensitivity to physical pain, regardless of their propensity to anxiously expect and overreact to a threat to their sense of belonging (Downey & Feldman, 1996). Those preliminary findings were also mute about another important issue, namely, what relationship increased pain threshold and tolerance may have to emotional responding, to which we turn in the next section.

beneficial in terms of immediately reducing a person’s suffering, in the same way that physical injuries often create an analgesia that saves him or her from feeling acute, ongoing pain for as long as the injury lasts. If the neural systems responsible for detecting and regulating physical and emotional pain have a common physiological basis, then social exclusion should also influence how people respond to physically painful stimuli. Thus, excluded people should become numb to physical pain, and this physical numbness should be related to emotional insensitivity (see Figure 1). What would provide the best measures of emotional insensitivity? One method would be direct self-reports of emotion, which (sure enough) tend to indicate no emotional reaction (e.g., Baumeister et al., 2005). Self-reported emotion may not provide the best measure, however, because of social desirability biases and other factors that could influence responding. For example, excluded people might feel emotionally upset but refuse to admit it. To avoid these methodological pitfalls of direct self-reports of emotion, we measured two emotional phenomena in which people seemingly rely on their current state and emotional simulation (imagining how oneself would feel) to make judgments about events distant from the present. The first of these phenomena is affective forecasting; the other is empathy. If rejected people are replete with emotion but reluctant to admit their distress out of self-presentational concerns, they should be quite willing to predict strong future emotional reactions and to empathize with others’ suffering. But if their emotional systems have really ceased to function normally, excluded people should find it difficult to imagine having strong emotions in response to hypothetical future events, and they likewise might not empathize with another person’s distress. Affective forecasting refers to the ways in which people predict their future feelings (see Wilson & Gilbert, 2003, for a review). Research has shown consistently that people overestimate the duration and intensity of their emotional responses to future events. People overestimate how happy they will be after winning a date on a simulated dating game (Wilson, Wheatley, Kurtz, Dunn, & Gilbert, 2004), predict greater distress following a romantic breakup or denial for academic tenure than actually occurs (Gilbert, Pinel, Wilson, Blumberg, & Wheatley, 1998), and overrate what their quality of life will be if they receive an important medical procedure (Jepson, Loewenstein, & Ubel, 2001). Other research has shown that a person’s current emotional state can bias his or her predictions of future emotional responses and other decision-making processes (Gilbert, Gill, & Wilson, 2002; Loewenstein, O’Donoghue, & Rabin, in press; L. D. Nelson & Morrison, 2005; Van Boven & Loewenstein, 2003). Emotional Insensitivity Social Exclusion

Effects of Increased Pain Threshold and Pain Tolerance for Emotional Responding If people respond to social exclusion with emotional numbness, this temporary shutdown of the emotion system might be

3

Physical Insensitivity

Figure 1. Proposed model illustrating linked consequences of physical and emotional insensitivity following social exclusion.

DEWALL AND BAUMEISTER

4

If social exclusion hampers the capacity for people to respond to their own emotions, rejected people might show a different pattern of affective forecasting than other participants. Specifically, rejected participants might report less happiness when forecasting their emotional reaction to a positive event and less unhappiness when forecasting their emotional reaction to a negative event compared with participants who did not experience rejection. Another process that may illustrate the intimate link between social exclusion and emotional responding is the ability of rejected people to show empathy toward people in various forms of physical and psychological distress. Previous research has shown that rejected people tend to behave less prosocially than nonrejected people (Twenge, Baumeister, DeWall, Ciarocco, & Bartels, 2006), and work by Batson and colleagues has demonstrated that empathy plays a prominent role in shaping prosocial behavior (Batson, Klein, Highberger, & Shaw, 1995). It therefore seems plausible that rejected participants might feel reduced empathy toward another person’s suffering.

status (no feedback). The third control group (misfortune control), added in Experiment 2, told people that their futures would be marred by accidents. The purpose of this third control group was to rule out the possibility that the effects of social exclusion were simply due to receiving an unpleasant forecast about the future. Hence, the misfortune control feedback predicted bad outcomes that did not include social exclusion. We predicted that participants in the future alone condition would have higher pain thresholds and higher pain tolerance than participants in the three control conditions. The inclusion of both pain threshold and pain tolerance measures provided two separate chances to confirm or disconfirm our hypothesis that social exclusion leads to numbness to physical pain. We had no a priori reason to think that socially excluded participants would show increases in pain threshold but not in pain tolerance (or vice versa). If socially excluded participants showed increases in both pain threshold and pain tolerance, however, that would provide converging support for our hypothesis.

Present Research The present experiments tested the theory that social exclusion activates the body’s pain system and sets of responses that may reduce sensitivity to both physical and emotional pain. Our prediction had two parts. First, participants who experienced social exclusion would show increased pain threshold and pain tolerance compared with socially accepted and control participants. Second, this insensitivity to physical pain should be related to emotional insensitivity. Participants who experience social exclusion should find it difficult to predict strong emotional responses to hypothetical future events, and they should exhibit decreased empathy with others’ suffering, compared with participants who do not experience rejection. Thus, the current investigation sought to resolve the seeming paradox as to why people respond to such a threatening experience as rejection with emotional numbness. The first goal was to test the hypothesis that people become numb to physical pain in the aftermath of social exclusion. The second goal of this work was to demonstrate that physical numbness has implications for emotional responding, such as affective forecasting and interpersonal empathy.

Experiments 1 and 2: Not Feeling My Own Pain Experiments 1 and 2 tested the hypothesis that social exclusion reduces sensitivity to physical pain. Social exclusion was manipulated by having participants complete a personality test and giving some participants bogus feedback stating that their personality profile enabled the researchers to predict that they would most likely end up alone in life. This future alone condition has been used in past research to create a sense of social exclusion and impending isolation, because people take it to mean that something about their personalities will cause others to reject them (Twenge et al., 2001). There were three control groups. One of these groups received personality feedback stating that they had a personality type that would lead to a future filled with several meaningful and lasting relationships (future belonging). The second control group was given no feedback regarding their personality or the implications their personality may have for their future belongingness

Method Participants. Thirty-three right-handed undergraduates (24 women) participated in Experiment 1 for partial course credit. Experiment 2 had 30 participants (24 women). Of the participants in Experiment 1, 67% were White and 33% were a racial minority. Of the participants in Experiment 2, 83% were White and 17% were a racial minority. In these and subsequent studies, we restricted participation to people who were nonsmokers and right-handed because smoking has been shown to reduce pain sensitivity (Kanarek & Carrington, 2004; Pomerleau, Turk, & Fertig, 1984) and because left limbs show greater pain sensitivity than right limbs, regardless of hand of preference (Murray & Hagan, 1973). Participants were also required to not have ingested any sugared foods, alcoholic beverages, or pain medicine (including aspirin) for at least 8 hr prior to participation in the experiments (Kanarek & Carrington, 2004; Mercer & Holder, 1997). Materials and procedure. Participants arrived individually for a study ostensibly concerned with personality and physical sensitivity. After giving informed consent, participants in Experiment 1 but not 2 completed measures of rejection sensitivity (Downey & Feldman, 1996) and global self-esteem (Rosenberg, 1965). Baseline measurements of pain threshold and pain tolerance (two trials each) were then taken by using a pressure algometer (Type II; Somedic, Solletuna, Sweden). For pain threshold, participants were instructed to say “now” when they first felt pain due to the pressure increase. For pain tolerance, participants were instructed to say “stop” when the pain became too uncomfortable to continue. At this point, the experimenter immediately retracted the algometer. The digital display showed the value of pressure applied at the moment the algometer was retracted. The algometer was applied perpendicularly to the skin and lowered at a rate of approximately 5 kPA per second until pain threshold or tolerance was reached, as indicated by participants’ verbal report. All pain measurements were taken at the first dorsal interosseous muscle (i.e., behind the first knuckle of the index finger) of the participant’s dominant hand. The order of the pain tolerance and threshold measurements was counterbalanced across participants. To prevent habituation, there was a 1.5-min interval between all pain threshold and tolerance measurements (Orbach, Mikulincer, King, Cohen, & Stein, 1997). Participants then completed a brief demographic questionnaire and the Eysenck Personality Questionnaire (Eysenck & Eysenck, 1975). To bolster the credibility of the cover story, we gave participants accurate feedback

SOCIAL EXCLUSION AND PAIN regarding their extroversion score.1 Participants then received bogus feedback about the implications their extroversion score would have for their future belongingness. Following a procedure developed by Twenge et al. (2001), we randomly assigned participants in Experiment 1 to one of three social feedback conditions: future belongingness, future alone, and nofeedback control. For Experiment 2, a misfortune control group replaced the no-feedback control group. In the future belongingness condition, participants were given their correct score on extroversion and then told, as additional feedback, that they would have stable, rewarding relationships throughout life. In the (crucial) future alone condition, participants were given their correct extroversion score but then told that they would likely end up alone in life. They were told they might have friends now, but these would drift away after college, and once they were past the age at which people are constantly making new friends, they would spend more and more time alone. The experimenter said they might marry once or several times, but these would be short-lived. The no-feedback control condition gave neither an extroversion score nor a forecast about the future. The misfortune control condition included the extroversion score and a forecast that the participant would become increasingly accident prone in future years, with many injuries and hospital stays. Participants then completed the Brief Mood Introspection Scale (BMIS; Mayer & Gaschke, 1988). After completing the BMIS, measurements of pain threshold and tolerance (five trials each) were taken. When participants had completed the pain threshold and tolerance measurements, they were carefully and thoroughly debriefed. Care was taken to ensure that participants recognized that the feedback they had received was based on random assignment and had nothing to do with them, their personality, or any form they had completed during the experimental session. The experimenter also apologized for being misleading regarding the personality feedback. When the experimenter was certain that participants understood the true purpose of the experiment and the bogus nature of the personality feedback, participants were given partial course credit, were thanked for their time, and were dismissed.

Results We computed both the omnibus F statistic and performed a priori contrasts comparing socially excluded participants (i.e., future alone) with nonsocially excluded participants (i.e., future belongingness, no-feedback control, and misfortune control). Pain threshold and tolerance scores were compared by using baseline pain threshold and pain tolerance scores as a covariate. There were no main effects or interactions on pain threshold and tolerance (or any other dependent measure) involving participant gender or ethnicity in any other of the studies reported in this article. Hence, these variables will not be discussed further. Pain threshold. As predicted, socially excluded participants showed significantly higher pain thresholds than participants who were led to anticipate a future filled with social acceptance or future misfortune or those who received no personality feedback. Results from an analysis of covariance (ANCOVA) on pain threshold scores in Experiment 1 (using baseline score as covariate) indicated that there was significant variation between the three experimental groups, F(2, 30) ⫽ 17.85, p ⬍ .001. A 2⫺1⫺1 a priori contrast confirmed that future alone participants had higher pain thresholds compared with future belongingness and nofeedback control participants, F(1, 30) ⫽ 33.63, p ⬍ .001. For Experiment 2, the ANCOVA yielded the following: F(2, 27) ⫽ 10.84, p ⬍ .001; and the 2–1–1 contrast showed that future alone participants had substantially higher pain threshold scores com-

5

pared with future belonging and misfortune control participants, F(1, 27) ⫽ 21.35, p ⬍ .001. A one-cell t test confirmed that the difference between baseline and post-feedback pain threshold measurement among future alone participants was greater than zero in both Experiment 1, t(10) ⫽ ⫺4.83, p ⫽ .001, and Experiment 2, t(9) ⫽ ⫺3.14, p ⫽ .01. In Experiment 1, future belonging participants showed a significant decrease in pain threshold from baseline, t(10) ⫽ 4.16, p ⫽ .002, though this effect did not replicate in Experiment 2 (t ⬍ 1, ns). There was no change from baseline in the no-feedback control condition in Experiments 1 and 2 (both ts ⬍1, ns). These findings suggest that the increase in pain sensitivity was not merely relative to participants in the other two conditions but was an absolute change from baseline measurements. The means and standard deviations are presented in Table 1. Pain tolerance. In both studies, socially excluded participants showed significantly higher levels of pain tolerance than all other participants. Results from an ANCOVA on the pain tolerance scores indicated that there was significant variation between the three experimental groups in Experiment 1, F(2, 30) ⫽ 20.36, p ⬍ .001, and Experiment 2, F(2, 27) ⫽ 9.66, p ⫽ .001. A 2⫺1⫺1 a priori contrast confirmed that future alone participants had higher overall pain tolerance than both other conditions in Experiment 1, F(1, 30) ⫽ 40.60, p ⬍ .001, and Experiment 2, F(1, 26) ⫽ 19.21, p ⬍ .001. A one-cell t test confirmed that the difference between baseline and post-feedback pain tolerance measurement among future alone participants was greater than zero in Experiment 1, t(10) ⫽ ⫺4.94, p ⫽ .001, and again in Experiment 2, t(10) ⫽ ⫺2.91, p ⬍ .02, indicating that their tolerance was significantly higher after receiving the feedback than it had been beforehand. There was no change from baseline in the other conditions (both ts ⬍1, ns). Mood and emotion, rejection sensitivity, and self-esteem. To investigate the possible role of mood and emotion in shaping the observed effects, we tested for fluctuations in mood valence and arousal as a result of personality feedback. For each experiment, two one-way ANOVAs were carried out by using the Valence and Arousal subscales of the emotion measure (BMIS) as dependent measures. In Experiment 1, there was no significant variation between the three social feedback groups in terms of their reported mood valence, F(2, 30) ⫽ 2.21, p ⫽ .13. Pairwise comparisons revealed that future alone participants (M ⫽ 18.18, SD ⫽ 8.15) reported levels of emotional valence that did not differ significantly from future belonging participants (M ⫽ 20.73, SD ⫽ 8.46; F ⬍ 1, ns). No-feedback control participants (M ⫽ 11.73, SD ⫽ 13.54) reported more negative moods than future belonging participants (M ⫽ 20.73, SD ⫽ 8.46), F(1, 30) ⫽ 4.16, p ⫽ .05. No-feedback control participants (M ⫽ 11.73, SD ⫽ 13.54) also reported a somewhat less positive mood than future alone participants (M ⫽ 18.18, SD ⫽ 8.15), although this difference did not reach significance, F(1, 30) ⫽ 2.14, p ⫽ .15. There was also no significant variation between the three social feedback groups in terms of their reported mood arousal (F ⬍ 1, ns). Regression analyses revealed that neither self-esteem nor social sensitivity significantly moderated the effects of condition on pain threshold 1

It should be noted that individual differences on the Introversion– Extroversion Scale did not influence the results in any way.

DEWALL AND BAUMEISTER

6

Table 1 Pain Threshold and Pain Tolerance, Experiments 1– 4 Future alone Experiment and measure

Future belonging

Misfortune control

M

SD

M

SD

Pain threshold Pain tolerance

31.73 199.99

12.46 50.23

13.27 129.40

12.68 49.65

Pain threshold Pain tolerance

45.94 388.22

32.05 113.44

10.55 276.32

31.46 107.68

Pain threshold Pain tolerance

74.57 392.53

61.92 134.59

19.17 197.66

Pain threshold Pain tolerance

62.51 496.84

22.11 120.97

14.35 236.05

M

SD

Other control M

SD

19.01 133.20

12.17 49.88

61.81 136.34

12.52 216.58

61.87 133.93

21.05 120.81

10.24 244.47

22.22 109.89

1 2 14.41 271.94

31.89 114.82

3 4

Note. In all experiments, baseline pain threshold and pain tolerance scores were entered as covariates. Covariate values for each experiment were as follows: Experiment 1, Pain threshold ⫽ 20.35; Experiment 1, Pain tolerance ⫽ 163.83; Experiment 2, Pain threshold ⫽ 16.18; Experiment 2, Pain tolerance ⫽ 312.03; Experiment 3, Pain threshold ⫽ 23.02; Experiment 3, Pain tolerance ⫽ 269.02; Experiment 4, Pain threshold ⫽ 25.03; Experiment 4, Pain tolerance ⫽ 314.37.

or tolerance ( ps ⬎ .20). In Experiment 2, there was no significant variation between the three groups in either mood valence or mood arousal (both Fs ⬍ 1, ns). The means and standard deviations are presented in Table 2. Thus, the observed increases in pain tolerance and threshold among socially excluded participants were not due to differences in mood valence and arousal or individual differences in self-esteem and rejection sensitivity.

esis that people become less sensitive to physical pain as a result of having their need to belong thwarted. Participants who anticipated a lonely future showed greater tolerance and less sensitivity to physical pain than participants who experienced social acceptance, received no personality feedback, or received feedback forecasting future physical misfortunes. They also showed significantly less sensitivity to physical pain than they themselves had shown on the baseline measures. The estimated effect sizes for the increases in pain threshold and pain tolerance among socially excluded participants were quite large (Cohen, 1977), exceeding a full standard deviation in each case.

Discussion Social exclusion produced increases in both pain threshold and pain tolerance in both studies, consistent with the hypoth-

Table 2 Self-Reports of Emotion by Experimental Condition, Experiments 1–5 Future alone Experiment and measure

Future belonging

M

SD

M

SD

Mood Valence Mood Arousal

18.18 25.64

8.15 5.63

20.73 26.91

8.46 7.56

Mood Valence Mood Arousal

17.80 26.60

13.40 5.50

22.90 24.90

12.97 7.48

Mood Valence Mood Arousal

10.40 28.40

12.13 7.14

18.70 26.40

Mood Valence Mood Arousal

19.50 27.40

10.71 5.87

Negative Affect Positive Affect

12.55 22.98

3.52 5.74

Misfortune control M

SD

Other control M

SD

11.73 28.91

13.54 7.42

9.78 6.29

15.20 26.10

12.93 6.76

16.90 22.90

11.23 120.81

11.45 24.64

10.51 4.41

12.50 24.76

3.51 5.23

13.91 23.32

4.39 4.64

1 2 15.30 27.10

13.37 6.15

3 4 5

Note. Values represent mean scores on either the BMIS Mood Valence or Mood Arousal subscales (Mayer & Gaschke, 1988) or the PANAS Positive Affect and Negative Affect subscales (Watson et al., 1988). BMIS ⫽ Brief Mood Introspection Scale; PANAS ⫽ Positive and Negative Affect Schedule.

SOCIAL EXCLUSION AND PAIN

The results of Experiment 1 and 2 were not due to any differences in reported mood valence or mood arousal. Future alone participants did not report moods that differed significantly from those reported by future belonging or no-feedback control participants. Neither rejection sensitivity nor trait self-esteem played a moderating role in predicting the relationship between social exclusion and physical pain threshold and pain tolerance. Thus, a future diagnostic forecast of social exclusion led to substantial increases in pain threshold and pain tolerance compared with socially accepted and control participants.

Experiment 3: Not Feeling Future Pain Experiment 3 tested whether social exclusion would have implications for affective forecasting. Affective forecasting involves simulating emotional reactions to possible future events. Exclusion causes the emotion system to cease functioning properly, indeed muting and numbing emotional reactions. It is therefore possible that hypothetical emotional reactions to the future (i.e., imagining how one would feel) will also be numbed. Accordingly, rejected participants (who reported emotional states that did not differ from participants in the other conditions in Experiments 1 and 2) should make affective forecasts of relative emotional numbness compared with accepted and control participants. That is, rejected participants should report less happiness when forecasting their emotional reaction to a positive event and less sadness and distress when forecasting their emotional reaction to a negative event. Socially accepted and control participants, in contrast, should exhibit the previously documented impact bias, in which people overestimate their happiness and sadness to possible future events (e.g., Wilson et al., 2004). The current study also tested whether increases in pain threshold and pain tolerance were related to the predicted unemotional affective forecasts among socially excluded participants.

Method Participants. Thirty right-handed undergraduates (19 women) participated for partial course credit. Of the participants, 77% were White and 23% were a racial minority. The average age was 18.6 years. Materials and procedure. Participants arrived at the laboratory individually for an experiment ostensibly concerned with the relationship between personality, verbal and nonverbal behaviors, and physical sensitivity. After giving informed consent, participants completed baseline measurements of pain threshold and pain tolerance (two trials each). Participants then completed a personality test and were exposed to the same social exclusion manipulation used in Experiment 1. By random assignment, participants were told that they had a personality type in which they would likely end up alone (future alone) or that their personality type was diagnostic of a future filled with many rewarding and lasting relationships (future belonging). Another group was not given any feedback on their personality test (no-feedback control). After receiving their personality feedback, participants completed the BMIS mood and affective forecasting measures. The latter were modeled after previous research by Wilson and colleagues (Wilson, Wheatley, Meyers, Gilbert, & Axsom, 2000). Participants were reminded that the Florida State football team was scheduled to compete against its closest rival in approximately 2 months and were asked to predict what their level of overall happiness would be directly after a victory and after a loss. Participants also reported to what degree they considered themselves a Florida State University football fan and to what extent they cared whether

7

Florida State won this particular game. These two measures were highly correlated (r ⫽ .50, p ⫽ .005) and were combined to create a Florida State football “fan” index. Last, participants completed five additional measurements of pain threshold and tolerance. Participants were then debriefed, thanked for their time, given partial course credit, and dismissed.

Results Pain threshold and pain tolerance: Replication. As in Experiments 1 and 2, a future diagnostic forecast of social exclusion produced significant increases in pain threshold compared with participants who anticipated a future filled with meaningful relationships or who received no feedback on their personality test. An ANCOVA on the pain threshold scores, using baseline pain threshold scores as a covariate, revealed significant variation between the three experimental groups, F(2, 27) ⫽ 9.07, p ⫽ .001. A 2⫺1⫺1 a priori contrast also showed that future alone participants had significantly higher pain threshold scores than future belonging and no-feedback control participants, F(2, 27) ⫽ 17.96, p ⬍ .001. A one-cell t test confirmed that the difference between baseline and post-feedback pain threshold measurement among future alone participants was greater than zero, t(9) ⫽ ⫺2.50, p ⫽ .03, whereas there was no change from baseline in the other conditions (both ts ⬍1, ns). Socially excluded participants also showed increased pain tolerance compared with participants who experienced social acceptance and participants who received no feedback on a personality test. An ANCOVA on the pain tolerance scores revealed significant variation between the three experimental groups, F(2, 27) ⫽ 18.91, p ⬍ .001. A 2⫺1⫺1 a priori contrast confirmed that future alone participants had significantly higher pain tolerance scores than both future belonging and no-feedback control participants, F(2, 27) ⫽ 37.69, p ⬍ .001. A one-cell t test confirmed that the difference between baseline and post-feedback pain threshold measurement among future alone participants was greater than zero, t(9) ⫽ ⫺2.93, p ⫽ .02. There was no change from baseline in the other conditions (both ts ⬍ 1, ns). Thus, future alone participants showed substantial and absolute increases in both pain threshold and pain tolerance. The means and standard deviations are presented in Table 1. Affective forecasting. Replicating previous affective forecasting research, the results indicated that future belonging and nofeedback control participants predicted strong emotional responses to a Florida State victory and defeat. Most relevant to the current investigation, however, was the finding that participants in the future alone condition predicted relatively neutral emotional reactions to both football outcomes. A mixed-model analysis of variance (ANOVA) was conducted by using the personality feedback condition as a between-subjects factor and responses to the questions “Please predict what your overall happiness would be right after the game if Florida State beats Florida” and “Please predict what your overall happiness would be right after the game if Florida State loses to Florida” as two levels of a within-subjects factor. Results revealed a significant Personality Feedback ⫻ Differential Predicted Happiness interaction, F(2, 27) ⫽ 16.22, p ⬍ .001. A 2⫺1⫺1 a priori contrast confirmed that future alone participants (M ⫽ 4.70, SD ⫽ .99) predicted a significantly lower degree of happiness in response to Florida State defeating the University of Florida compared with both future belonging (M ⫽ 6.60, SD ⫽ 0.69) and no-feedback control participants (M ⫽ 6.70,

8

DEWALL AND BAUMEISTER

SD ⫽ .67), F(2, 27) ⫽ 41.42, p ⬍ .001. In addition, a 2⫺1⫺1 a priori contrast confirmed that future alone participants (M ⫽ 3.50, SD ⫽ 0.71) predicted greater happiness in response to Florida State being beaten by the University of Florida than both future belonging (M ⫽ 2.30, SD ⫽ 1.34) and no-feedback control (M ⫽ 1.90, SD ⫽ 1.10) participants, F(2, 27) ⫽ 11.20, p ⫽ .002. Thus, future alone participants predicted less happiness to a Florida State victory and less sadness to a Florida State defeat compared with no-feedback control and future belonging participants. Supplementary analyses were conducted to test whether the exclusion, acceptance, and control groups differed in the degree to which they perceived themselves as Florida State football fans and the degree to which they cared about the outcome of the upcoming football game. Results from an ANOVA revealed no significant variation between the three groups on the fan index, F(2, 27) ⫽ 2.29, p ⫽ .12. Pairwise comparisons revealed that future alone participants (M ⫽ 10.30, SD ⫽ 2.63) had scores on the fan index that were not significantly different from future belonging participants (M ⫽ 8.90, SD ⫽ 3.45), F(1, 27) ⫽ 1.04, p ⫽ .32, and not different from no-feedback control participants (M ⫽ 11.60, SD ⫽ 2.27), F(1, 27) ⫽ 1.40, p ⫽ .25. Future belonging participants (M ⫽ 8.90, SD ⫽ 3.45) had scores on the fan index that were marginally lower than those of the no-feedback participants (M ⫽ 11.60, SD ⫽ 2.27), F(1, 27) ⫽ 4.28, p ⬍ .06. Thus, the attenuated forecasts (in the future alone condition) of affective reactions to the future football game outcome were not due to participants feeling as less a fan or feeling less interested in the football game. Mood and emotion. To assess difference in mood, we conducted two one-way ANOVAs by using the Mood Valence and Mood Arousal subscales of the BMIS (Mayer & Gaschke, 1988) as dependent measures. Both ANOVAs revealed no significant variation between the three experimental groups (both Fs ⬍ 1.28, ps ⬎ .30; see Table 2). These results suggest that increases in pain threshold and pain tolerance following social exclusion were not the result of differences in mood. Link between physical and emotional insensitivity. To test the hypothesis that unemotional affective forecasts were related to increases in pain threshold and pain tolerance, we conducted a regression analysis in which the measures of physical insensitivity predicted measures of emotional insensitivity. For ease of analysis, we computed a differential predicted happiness index by subtracting scores to the question “Please predict what your overall happiness would be right after the game if Florida State beats Florida” from scores to the question “Please predict what your overall happiness would be right after the game if Florida State loses to Florida.” Pain threshold scores predicted differential predicted happiness index scores, t(27) ⫽ ⫺2.36, ␤ ⫽ ⫺.32, p ⬍ .03. Pain tolerance scores were also a significant predictor of differential predicted happiness scores, t(27) ⫽ ⫺2.68, ␤ ⫽ ⫺.31, p ⫽ .01. These findings provide evidence in support of the hypothesis that social exclusion has two linked consequences, namely, numbness to physiological pain and emotional numbness to possible future events.

Discussion Experiment 3 provided support for the hypothesis that responses to social exclusion operate via a shared psychological system that has related consequences for both physical and emotional pain. As

in Experiments 1 and 2, socially excluded participants exhibited increases in both pain threshold and pain tolerance compared with accepted and control participants. Excluded participants also reported emotional states that did not differ from accepted and control participants. It is difficult to ascertain whether excluded participants truly felt no emotion or whether they were simply refusing to admit that they were upset. If the latter were true, however, excluded participants should be willing to describe and predict their future emotional reactions, such as being happy if their university were to win an important football game 2 months in the future. The findings of Experiment 3 provide evidence that social exclusion causes the emotion system to cease functioning normally. Socially excluded participants not only failed to report much in the way of current emotion; they also predicted little in the way of a future emotional response to a major meaningful event. Excluded participants predicted less happiness when forecasting their emotional reaction to a positive event and greater happiness when forecasting their emotional reaction to a negative event, compared with participants in the other conditions. It would have been plausible for exclusion to cause people to identify either more or less strongly with Florida State University and its football team, but they did not. In any case, the failure of rejected people to predict much in the way of an emotional reaction to future football outcomes was not mediated by changes in their identification with Florida State University or expressed concern with the outcome of the football game. After hearing that they were likely to be alone later in their lives, they retained their (relatively strong and positive) identification with their university (i.e., they acknowledged their identity as Florida State University football fans who cared about the outcome of an important game), but they did not think they would feel much emotion about the outcome of its football game against a major rival. Thus, the results of Experiment 3 demonstrated that the body responds to social exclusion feedback with a kind of physical shock reaction that includes numbness and insensitivity to current physical pain and possible future emotional events (cf. Campbell, Baumeister, Dhavale, & Tice, 2003).

Experiment 4: Not Feeling Someone Else’s Pain If social exclusion causes the emotion system to cease functioning normally, one should be less capable of empathy, and so one’s reaction to another person’s suffering should be muted. Participants in Experiment 4 were asked to empathize with someone who was suffering over a recent romantic failure. The main predictions were that social exclusion would lead to less empathy toward the stimulus person who had suffered the romantic rejection and that this emotional insensitivity would be related to increases in both pain threshold and pain tolerance.

Method Participants. Thirty-one undergraduates (21 women) participated in this experiment in exchange for partial course credit. Right-handedness was stipulated, but one left-handed participant signed up and participated anyway. Of the participants, 71% were White and 29% were a racial minority. The average age was 18.5 years. Materials and procedure. The materials and procedure for Experiment 4 were similar to Experiments 1–3. Participants arrived at the laboratory individually for an experiment ostensibly concerned with the relationship

SOCIAL EXCLUSION AND PAIN between different aspects of personality, verbal and nonverbal performance, and physical sensitivity. After giving informed consent, participants completed baseline measurements of pain threshold and pain tolerance (two trials each) and the Eysenck Personality Questionnaire (Eysenck & Eysenck, 1975). By random assignment, participants received future alone feedback, future belonging feedback, or no feedback. After completing the BMIS mood measure, participants completed a measure of empathy. The experimenter explained that there was another experiment being conducted in the laboratory in which 1 participant writes an essay about something going on in his or her life and another participant reads the essay and answers some questions about it. Participants were told that the participant who was supposed to read and respond to the essay did not show up for the experiment. The experimenter explained that the participant would read and respond to the essay. Participants were then handed a manila folder that contained a handwritten essay (in male or female handwriting,2 to match the participant’s gender) and short questionnaire. The content of the essay was adapted from Batson et al. (1995). The essay read (in part): Two days ago I broke up with my (girlfriend) boyfriend. We’ve been going together since our junior year in high school and have been really close, and it’s been great being at FSU together. I thought (s)he felt the same, but things have changed. Now, (s)he wants to date other people. (S)He says (s)he still cares a lot about me, but (s)he doesn’t want to be tied down to just one person. I’ve been real down. It’s all I think about. My friends all tell me that I’ll meet other (girls)guys and they say that all I need is for something good to happen to cheer me up. I guess they’re right, but so far that hasn’t happened. Participants reported how sympathetic, warm, compassionate, softhearted, and tender they felt toward the author of the essay. These adjectives have been used in previous research to measure empathy (Batson, 1987, 1991; Batson et al., 1995). The internal reliability for the empathyrelated adjectives was good (Cronbach’s alpha ⫽ .92), and therefore an empathy index was created by summing responses to the five empathy adjectives (sympathetic, warm, compassionate, softhearted, and tender). When finished, participants placed the essay and the questionnaire in an envelope, sealed it, and handed it back to the experimenter. The experimenter then left the laboratory ostensibly to return the envelope to the experimenter for the other study. When the experimenter returned, the participant completed additional measurements of pain threshold and pain tolerance (five trials each). Participants were then thoroughly debriefed, thanked for their time, given partial course credit, and dismissed.

Results Pain threshold and pain tolerance: Replication. As in the previous three experiments, a future diagnostic forecast of social exclusion produced increases in pain threshold compared with participants who anticipated a future filled with social acceptance or who received no feedback on a personality test. An ANCOVA on the pain threshold scores revealed significant variation between the three experimental groups, F(2, 28) ⫽ 54.12, p ⬍ .001. A 2⫺1⫺1 a priori contrast confirmed that future alone participants had significantly higher pain threshold scores compared with future belonging and no-feedback control participants, F(2, 28) ⫽ 108.15, p ⬍ .001. A one-cell t test confirmed that the difference between baseline and post-feedback pain threshold measurement among future alone participants was significant, t(9) ⫽ ⫺5.99, p ⬍ .001, whereas there was no change from baseline in the other conditions (both ts ⬍1, ns). Socially excluded participants also demonstrated a significantly higher tolerance for pain than participants assigned to the other

9

experimental groups. An ANCOVA on the pain tolerance scores showed significant variation between the three experimental groups, F(2, 28) ⫽ 45.25, p ⬍ .001. A 2⫺1⫺1 a priori contrast confirmed that future alone participants had significantly higher pain tolerance scores than participants who did not experience social exclusion (i.e., future belonging and no-feedback control), F(2, 28) ⫽ 89.99, p ⬍ .001. A one-cell t test confirmed that the difference between baseline and post-feedback pain threshold measurement among future alone participants was significantly greater than zero, t(10) ⫽ ⫺6.69, p ⬍ .001. There was no change from baseline in the other conditions (both ts ⬍1, ns). The means and standard deviations are presented in Table 1. Empathy. Socially excluded participants showed less empathy than other participants toward the romantically rejected stimulus person. An ANOVA on the five-item empathic emotion index revealed significant variation between the three experimental groups, F(2, 28) ⫽ 10.10, p ⬍ .001. A 2⫺1⫺1 a priori contrast demonstrated that future alone participants reported feeling less sympathy, compassion, warmth, tenderness, and softheartedness toward the participant who had recently experienced relationship loss, compared with future belonging and no-feedback control participants, F(2, 28) ⫽ 19.67, p ⬍ .001. Pairwise comparisons using the ANOVA mean-square error showed that future alone participants (M ⫽ 32.80, SD ⫽ 7.66) had significantly lower scores on the empathy index compared with future belonging participants (M ⫽ 42.00, SD ⫽ 5.79), F(1, 28) ⫽ 12.37, p ⫽ .002, d ⫽ 1.35. Future alone participants (M ⫽ 32.80, SD ⫽ 7.66) also had lower scores on the empathy index compared with nofeedback control participants (M ⫽ 43.55, SD ⫽ 3.59), F(1, 28) ⫽ 17.67, p ⬍ .001, d ⫽ 1.79. Future belonging participants did not differ in their empathy scores compared with no-feedback control participants (F ⬍ 1, ns). Thus, future alone participants expressed less emotional responsiveness to another person who had experienced exclusion compared with future belonging and no-feedback control participants. Mood and emotion. Two one-way ANOVAs were conducting by using the Mood Valence and Mood Arousal subscales of the BMIS as dependent measures. Both ANOVAs revealed no significant variation between the three experimental groups (both Fs ⬍ 1.79, ps ⬎ .20; see Table 2). Link between physical and emotional insensitivity. To examine whether the lack of empathic responding among socially excluded participants was related to increased pain tolerance and pain threshold, a regression analysis was conducted in which pain threshold and pain tolerance scores predicted empathy scores. Results revealed that pain threshold scores were a significant predictor of empathy scores, t(28) ⫽ ⫺3.90, ␤ ⫽ ⫺.59, p ⫽ .001. Pain tolerance scores also predicted empathy scores, t(27) ⫽ ⫺5.24, ␤ ⫽ ⫺.61, p ⬍ .001. Thus, the increase in tolerance and threshold for physical pain following social exclusion was related to lack of empathic concern to others who had also experienced social exclusion. 2

The essay was handwritten by either a male or female research assistant. No participants expressed suspicion about the gender of the author of the essay. In fact, many participants commented during the debriefing that the handwriting was indicative of either a male or female author.

10

DEWALL AND BAUMEISTER

Discussion Experiment 4 provided further evidence that the emotion system temporarily ceases to function normally following social exclusion. As compared with socially accepted and no-feedback controls, socially excluded participants showed very little empathy toward another person who had suffered a recent romantic breakup. In fact, 1 participant in the future alone condition commented “tough shit” upon reading about the other participant’s romantic woes—a comment that clearly indicates an uncharitable and unsympathetic attitude toward another person’s heartbreak. The increases in pain threshold and pain tolerance among socially excluded participants were related to their lack of empathy toward the other participant. This supports the broad view that social exclusion causes both physical numbness and emotional insensitivity to others.

Experiment 5: I Can’t Feel Another Person’s Physical Pain One possible boundary condition for Experiment 4 is that it explored the effects of social exclusion on empathy toward someone else who had recently suffered romantic rejection. Although the essay used in Experiment 4 has been used in previous research to evoke general empathy (Batson et al., 1995), the lack of empathic responding among socially excluded participants may have been due to the author of the essay having also experienced a form of social exclusion. Excluded participants may have refused to empathize with another rejected person because the other person’s rejection may have reminded them of their own recent social exclusion. Therefore, Study 5 measured empathy toward someone who had suffered a broken leg. Exclusion was manipulated by having participants complete a vivid recall task in which they completed a brief autobiographical narrative recalling a time they experienced social rejection, social acceptance, or an unrelated event. This was meant to prime participants with thoughts of their relevant experience, which would in turn influence responding on a questionnaire designed to measure empathy. Previous research has shown that real and imagined events activate many of the same neural and psychological processes (Kosslyn et al., 1999; McGuire, Shah, & Murray, 1993), and other research has shown that the exclusion vivid recall task evokes responses similar to those found in manipulations of immediate rejection (Gardner et al., 2000; Pickett, Gardner, & Knowles, 2004). It was predicted that the current manipulation of rejection would produce the same emotional numbness as in excluded participants in Experiments 1– 4. The main prediction of Experiment 5 was that socially rejected participants would express less empathy toward the author of the essay compared with accepted and control participants.

Method Participants. One-hundred twenty-five participants (84 women) participated in this study in exchange for partial course credit. Results from 9 participants were discarded from all analyses because of incomplete questionnaire packets (thus, there were originally 134 participants). Of the participants, 74% were White and 26% were a racial minority. The average age was 18.5 years.

Materials and procedure. Participants arrived at a large classroom in groups of 10 –20 for a study ostensibly concerned with how people understand each other. After giving informed consent, participants completed a short questionnaire packet that contained instructions for an autobiographical narrative, a mood questionnaire, and additional materials aimed at measuring empathic responding. By random assignment, participants were assigned to one of three autobiographical narrative conditions: social rejection, social acceptance, and control. The instructions for the social rejection narrative read as follows: On the next pages, you will write an essay about a time when you experienced rejection or exclusion by others. Think of a time when you felt that others did not want to be in your company and when you did not feel a strong sense of belongingness with another person or group. Nearly everyone has experienced such an experience more than once; please choose an especially important and memorable event. The instructions urged participants to describe the rejection experience in as thorough detail as possible and to provide the “full story.” Other participants wrote about a childhood experience of social acceptance. The instructions for the social acceptance narrative mirrored the social rejection instructions and replaced only words that dealt directly with the experience of rejection or acceptance. Participants assigned to the control condition were instructed to write a detailed essay about one of the things he or she did the previous day. After participants completed writing their narrative, participants completed the Positive and Negative Affect Schedule (Watson, Clark, & Tellegen, 1988). Participants then read a short vignette that depicted a college student (of unknown gender) and was meant to evoke empathic responding from participants. The vignette read (in part): Two days ago I broke my leg playing intramural sports. I’ve been playing on the same intramural team for the past three years and I’m upset that my season has been cut short. I’m experiencing pain because of my injury. I’m also having a tough time getting around campus, as there are lots of hills and stairs that make it hard to use my crutches on. The parking people won’t let me get a handicapped permit because they said my injury was only temporary. I’ve been real down. It’s all I think about. Participants reported how sympathetic, warm, compassionate, softhearted, and tender they felt toward the author of the essay. The internal reliability of these five items was good (Cronbach’s alpha ⫽ .86), and therefore an empathy index was created by summing responses to these five items. When participants had finished the empathy rating task, they were debriefed, given partial course credit, thanked for their time, and dismissed.

Results Empathy. Participants who wrote about a time they experienced social rejection showed significantly less empathy toward the author compared with participants in the other two conditions. An ANOVA on the five-item empathic emotion index revealed significant variation between the three experimental groups, F(2, 122) ⫽ 6.72, p ⫽ .002. A 2⫺1⫺1 a priori contrast confirmed that socially rejected participants felt significantly less sympathy, warmth, compassion, tenderness, and softheartedness toward the author of the essay compared with both socially accepted and control participants, F(1, 122) ⫽ 12.79, p ⫽ .001. Pairwise comparisons using the ANOVA mean-square error showed that rejected participants (M ⫽ 26.33, SD ⫽ 9.42) empathized with the essay author significantly less than did accepted participants (M ⫽ 31.82, SD ⫽ 8.65), F(1, 122) ⫽ 7.53, p ⫽ .007, d ⫽ .61. In

SOCIAL EXCLUSION AND PAIN

addition, rejected participants (M ⫽ 26.33, SD ⫽ 9.42) showed less empathic concern toward the essay author compared with control participants (M ⫽ 32.96, SD ⫽ 8.43), F(1, 122) ⫽ 12.21, p ⫽ .001, d ⫽ .74. Socially accepted and control participants did not differ significantly in terms of how much they empathized with the essay author (F ⬍ 1, ns). Thus, rejected participants were quite unsympathetic toward a member of their peer group who had suffered a physically painful injury and was having difficulty adjusting to the changes brought about by the injury. These results suggest that rejection brings about a lack of empathic responding toward others, regardless of whether the person has experienced rejection or another traumatic and physically painful event. Mood and emotion. Two one-way ANOVAs were conducting by using the Positive Affect and Negative Affect subscales of the Positive and Negative Affect Schedule as dependent measures. There was no significant variation between the three experimental groups in terms of reported positive affect, F(2, 122) ⫽ 1.31, p ⫽ .27, or negative affect, F(2, 122) ⫽ 1.89, p ⫽ .16 (see Table 2).

Discussion Experiment 5 provided evidence that rejected participants expressed less empathy than accepted and control participants toward a person suffering physical pain and inconvenience from a broken leg. These findings provide converging evidence that the emotion system temporarily ceases to function normally following social exclusion, which leads to emotional numbness toward another person in physical pain. Thus, the excluded people’s lack of empathy found in Experiment 4 appears to be part of a broader pattern rather than being tied to the special case of refusing to empathize with someone else who has suffered social exclusion.

General Discussion Social exclusion threatens a fundamental human motivation for strong and stable relationships. Such a threat strikes at the core of human psychological and physical well-being. Social exclusion can have destructive effects on mental and physical well-being, but socially excluded people frequently report relatively numb emotional states. This paradox leaves open the mystery of why socially excluded people often respond to such a disturbing event as social exclusion in an emotionally numb manner. One possible explanation for this anomalous emotional responding following social exclusion is that the emotion system temporarily ceases functioning normally, leaving a person momentarily unable to respond to emotional events in a customary fashion. This should lead to numbness to physically painful stimuli and emotional numbness to current and possible future events. Across four experiments, socially excluded participants consistently showed both decreased sensitivity and increased tolerance to physical pain, compared with socially accepted and control participants and compared with their own baseline responses. The estimated effect size of the difference in pain threshold and pain tolerance between socially excluded participants and participants in the other conditions was consistently large, exceeding the criteria typically used to describe large effects (Cohen, 1977). Thus, participants responded to the social exclusion feedback by growing increasingly numb to physical pain stimuli, and these increases in pain threshold and pain tolerance were large in comparison to

11

participants who experienced social acceptance, received a negatively valenced future diagnostic forecast, or received no feedback on their personality test. The second goal of the current investigation was to examine whether insensitivity to physical pain might be linked to emotional numbness following social exclusion. In all five experiments, excluded participants reported emotional states that did not differ significantly from socially accepted and control participants. Admittedly, these results suffer from the uncertainties that accompany the use of self-reports of emotional responding, especially the idea that excluded participants might be reluctant to admit that they are upset over the experimental feedback. But excluded participants should be willing and able to predict future emotional reactions or empathize with others. The results of Experiments 3–5 showed that excluded participants failed to make the standard predictions about their emotional reactions to future events and failed to empathize with others’ suffering. Social exclusion feedback caused people to predict less happiness over a future football victory and less sadness over a possible defeat. Social exclusion also led participants to become less empathic toward another person who had just suffered a relationship loss, which as a form of social exclusion might have elicited high empathy from participants who themselves had also experienced exclusion. And recalling a past experience of social exclusion made people less empathic toward a student who was suffering over a broken leg (which was unrelated to social exclusion). These findings suggest that social exclusion produces a sweeping shutdown of the emotional system. Thus, the findings from the current experiments indicate that the body responds to social exclusion in much the same manner as it responds to physical injury. Exclusion may well produce a biochemical reaction that leads to temporary numbness to physical pain. This physical numbness is also linked to emotional responding, which is demonstrated in emotional insensitivity. The findings of Experiment 3 make a novel contribution to the affective forecasting literature by identifying a moderating variable to the previously documented pattern of people overestimating the impact of future emotional events on their lives (e.g., Wilson & Gilbert, 2003). Exclusion temporarily disrupted the capacity for the emotion system to function normally, and this led excluded participants to predict a significantly less emotional response to both future positive and negative events than participants who had experienced social acceptance or than control participants. This finding is potentially intriguing in that socially excluded people made affective forecasts that were likely more realistic than participants in the other conditions (D. T. Gilbert, personal communication, December 16, 2004), though we did not collect data on participants’ actual emotional responses to the game’s outcome. Regardless of accuracy, our finding that decreased sensitivity to pain was closely linked to affective forecasts of indifference provides further reason to think that the physical pain and social emotion systems may share a common physiological basis (or at least are intimately linked). In sum, the results of the current studies provide evidence of an immediate physical numbing effect in the aftermath of social exclusion, which is related to the degree of emotional numbness that socially excluded people exhibit. These findings help to resolve the paradox in the rejection literature by showing that one possible reason people report relative emotional numbness to social exclusion is that they experience a physiological reaction

12

DEWALL AND BAUMEISTER

that creates numbness to physically and emotionally painful stimuli. It is not likely that these physical numbing effects are long lasting. Instead, physical numbness likely functions as an immediate buffer that wards off potentially threatening aspects of one’s current surroundings. The possible long-term effects of repeated or chronic social rejection remain to be explored.

Limitations and Alternative Explanations The five experiments reported in this article provide consistent evidence that social exclusion produces increases in pain threshold and pain tolerance and that this physical numbness to pain is related to emotional insensitivity. Despite the consistency of these effects, however, several alternative explanations warrant further consideration. A first possibility is that social exclusion simply constitutes a form of bad news. From this perspective, receiving any negatively valenced feedback should temporarily cause the emotion system to cease functioning normally, resulting in numbness to painful physical stimuli. The results of our misfortune control condition in Experiment 2 provide evidence that is inconsistent with this perspective. Participants assigned to that condition received a negatively valenced future diagnostic forecast that was unrelated to their future belongingness status, namely, that they would become accident prone later in life. Despite the relative negativity of this diagnostic forecast, misfortune control participants responded to painful stimuli in a manner that was most similar to participants who experienced social acceptance or who received no personality feedback. A second possibility is that the methods used for assessing pain threshold assessed something other than pain threshold, such as a general ability to report tactile stimulation. If the pain threshold measures were tapping something other than physical numbness to pain, the pain threshold scores in the current study should not be associated with measures of tolerance for physical pain. Against this line of reasoning, Time 2 pain threshold scores (controlling for baseline measurement) were strongly correlated with pain tolerance scores (controlling for baseline measurement). The correlations between pain threshold and pain tolerance were consistently large and were reduced significantly when controlling for social exclusion condition. The correlations were as follows: Experiment 1, r(29) ⫽ .43, p ⬍ .02; controlling for exclusion condition, r(30) ⫽ ⫺.20, p ⫽ .30; Experiment 2, r(29) ⫽ .82, p ⬍ .001; controlling for exclusion condition, r(27) ⫽ .67, p ⬍ .001; Experiment 3, r(26) ⫽ .46, p ⫽ .01; controlling for exclusion condition, r(25) ⫽ ⫺.06, p ⫽ .77; Experiment 4, r(27) ⫽ .85, p ⬍ .001; controlling for exclusion condition, r(26) ⫽ .22, p ⫽ .27. Thus, it appears that our measures of pain threshold were measuring one component of physical pain sensation as a result of the social exclusion manipulation, namely, sensitivity to physical pain. Another possible limitation in the current studies is whether there is any relationship between chronic levels of pain sensitivity and emotional sensitivity not induced by the social exclusion manipulation. To examine this possibility, we predicted Time 2 pain sensitivity scores from the emotional sensitivity measures in Experiments 3 and 4, while simultaneously controlling for social exclusion condition and baseline pain sensitivity scores. In Experiment 3, results revealed no significant relationship between Time 2 pain threshold and tolerance scores and scores on the differential predicted happiness index (␤s ⫽ .08 and .15, respectively; both

ps ⬎ .24). Similar results emerged in Experiment 4. After controlling for baseline pain sensitivity measurement and social exclusion condition, we found that the relationship between Time 2 pain threshold and tolerance scores and empathy scores was nonsignificant (␤s ⫽ ⫺.18 and ⫺.16, respectively; both ps ⬎ .14). Thus, these findings indicate that the changes induced by condition in pain sensitivity were directly related to the differences in emotional sensitivity as a result of the social exclusion manipulation and that within each condition there were not any strong and consistent relationships between the physical and emotional sensitivity measures after controlling for baseline measurements and social exclusion condition. A final explanation is that increased pain tolerance and pain threshold among socially excluded participants were due to increased self-regulation instead of a temporary shutdown in the emotion system. Pain tolerance is often used as a measure of self-regulation (Hilgard et al., 1974), and participants may have responded to social exclusion with increased self-regulation as a means of improving their moods. The results of Experiments 1–5 provided consistent and converging evidence that socially excluded participants did not report moods that were significantly different from socially accepted or control participants, which casts doubt on the likelihood that excluded participants were trying to improve their moods. The present data cannot easily rule out the idea that reduced sensitivity and increased tolerance to pain reflects increased self-regulation in response to social exclusion. Past work has clearly shown the opposite pattern, however, namely, that social exclusion leads to poor self-regulation (Baumeister et al., 2005; DeWall, Baumeister, & Vohs, 2004). If the current results could be explained in terms of effective self-regulation, excluded participants would have also probably shown high levels of empathy toward others’ suffering. This was, however, not the case.

Concluding Remarks The purpose of the current investigation was to resolve the seeming paradox that people respond to social exclusion with emotional numbness. The need to belong is a fundamental and pervasive human motivation, and the behavioral consequences of social exclusion are often quite large. It was therefore a mystery as to why people fail to show strong emotional responses in reaction to such a seemingly momentous event as social exclusion. The apparent solution to the mystery, as suggested by the present findings, is that the body responds to social exclusion in much the same way (and possibly by means of the same physiological mechanisms) as it responds to painful physical injury. Physical sensation and emotion reactivity become dulled and insensitive. This may temporarily reduce suffering and enable the individual to begin to cope without feeling acutely distressed each moment. Reduced suffering and potentially better coping would of course be adaptive responses, but past work has suggested that the immediate behavioral effects of rejection are often far from beneficial (e.g., Baumeister et al., 2005; Twenge et al., 2001, 2002, 2003), even extending to behaviors that would seemingly reduce chances of gaining future acceptance. From the perspective of the present findings, one possible implication is that people use their emotional faculties to help understand their world. Emotional numb-

SOCIAL EXCLUSION AND PAIN

ness may therefore impair the person’s ability to relate to others and function effectively in a complex social world. The present results showed that excluded people were seemingly unable to furnish the typical (and presumably socially adaptive) patterns of affective forecasting and empathy. Most likely there are other cognitive tasks that social animals routinely perform with the aid of their emotion systems and that, accordingly, may abruptly be lost when social exclusion temporarily causes the emotional system to cease functioning normally. There was no sign that people realize how much they are affected by rejection. When asked to forecast their future reactions, our excluded participants did not say they were uncertain as to how they would feel. Rather, they engaged in self-reflection and told us they would not have a strong emotional response to whether their football team won or lost. Recent work has increasingly shown that some emotional capabilities are essential for good decision making and other adaptive behavioral patterns (e.g., Bechara, Damasio, Damasio, & Lee, 1999; Loewenstein, Weber, Hsee, & Welch, 2001). If normal emotional functioning is essential for effective decision making and behaviors, then when the emotion system ceases to function normally, the person may be given faulty information with which to make decisions and interact with others, which may have potentially costly results. The emerging image of the socially rejected person is thus one of someone who is emotionally and hence socially impaired— but may not realize it. Without a functioning emotion system, the person may not anticipate how his or her actions will affect others or even the person’s own future outcomes. Without recognizing this impairment, the person may even act on those faulty intuitions. In that way, the short-term relief brought by the emotional numbness caused by social exclusion may also bring destructive social consequences.

References Argyle, M. (1987). The psychology of happiness. London: Methuen. Bancroft, J., Skirmshire, A., & Simkins, S. (1976). The reasons people give for taking overdoses. British Journal of Psychiatry, 128, 538 –548. Batson, C. D. (1987). Prosocial motivation: Is it ever truly altruistic? In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 20, pp. 65–122). New York: Academic Press. Batson, C. D. (1991). The altruism question: Toward a social psychological answer. Hillsdale, NJ: Erlbaum. Batson, C. D., Klein, T. R., Highberger, L., & Shaw, L. L. (1995). Immorality from empathy-induced altruism: When compassion and justice conflict. Journal of Personality and Social Psychology, 68, 1042– 1054. Baumeister, R. F. (1990). Suicide as escape from the self. Psychological Review, 97, 90 –113. Baumeister, R. F., DeWall, C. N., Ciaracco, N. J., & Twenge, J. M. (2005). Social exclusion impairs self-regulation. Journal of Personality and Social Psychology, 88, 589 – 604. 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., & Scher, S. J. (1988). Self-defeating behavior patterns among normal individuals: Review and analysis of common selfdestructive tendencies. Psychological Bulletin, 104, 3–22. Baumeister, R. F., Twenge, J. M., & Nuss, C. K. (2002). Effects of social exclusion on cognitive processes: Anticipated aloneness reduces intelligent thought. Journal of Personality and Social Psychology, 83, 817– 827.

13

Bechara, A., Damasio, H., Damasio, A., & Lee, G. P. (1999). Different contributions of the human amygdala ventromedial prefrontal cortex to decision-making. Journal of Neuroscience, 19, 5473–5481. Buckley, K. E., Winkel, R. E., & Leary, M. R. (2004). Reactions to acceptance and rejection: Effects of level and sequence of relational evaluation. Journal of Experimental Social Psychology, 40, 14 –28. Bush, G., Luu, P., & Posner, M. I. (2000). Cognitive and emotional influences in anterior cingulate cortex. Trends in Cognitive Sciences, 4, 215–222. Cacioppo, J. T., Hawkley, L. C., & Berntson, G. G. (2003). The anatomy of loneliness. Current Directions in Psychological Science, 12, 71–74. Campbell, W. K., Baumeister, R. F., Dhavale, D., & Tice, D. M. (2003). Responding to major threats to self-esteem: A preliminary, narrative study of ego-shock. Journal of Social and Clinical Psychology, 22, 79 –96. Carden, S. E., Hernandez, N., & Hofer, M. A. (1996). The isolation and companion comfort responses of 7- and 3-day-old rat pups are modulated by drugs active at the opioid receptor. Behavioral Neuroscience, 110, 324 –330. Carden, S. E., & Hofer, M. A. (1990). Socially mediated reduction of isolation distress in rat pups is blocked by naltrexone but not by Ro 15–1788. Behavioral Neuroscience, 104, 457– 463. Cohen, J. (1977). Statistical power analysis for the behavioral sciences. New York: Academic Press. Craig, A. D., & Dostrovsky, J. O. (1999). Medulla to thalamus. In P. Wall & R. Melzack (Eds.), Textbook of pain (pp. 183–214). New York: Churchill Livingstone. DeWall, C. N., Baumeister, R. F., & Vohs, K. D. (2004). Recovering from rejection: Undoing the self-regulation deficits stemming from social exclusion. Manuscript submitted for publication. Downey, G., & Feldman, S. I. (1996). Implications for rejection sensitivity for intimate relationships. Journal of Personality and Social Psychology, 70, 1327–1343. Eisenberger, N. I., & Lieberman, M. D. (2004). Why rejection hurts: The neurocognitive overlap between physical and social pain. Trends in Cognitive Sciences, 8, 294 –300. 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. Eysenck, H. J., & Eysenck, S. B. G. (1975). Manual of the Eysenck Personality Questionnaire. San Diego, CA: EDITS. Gardner, W. L., Pickett, C. L., & 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. Gilbert, D. T., Gill, M. J., & Wilson, T. D. (2002). The future is now: Temporal correction in affective forecasting. Organization Behavior and Human Decision Processes, 88, 430 – 444. Gilbert, D. T., Pinel, E. C., Wilson, T. D., Blumberg, S. J., & Wheatley, T. (1998). Immune neglect: A source of durability bias in affective forecasting. Journal of Personality and Social Psychology, 75, 617– 638. Hariri, A. R., Bookheimer, S. Y., & Mazziotta, J. C. (2000). Modulating emotional responses: Effects of a neocortical network on the limbic system. Neuroreport, 11, 43– 48. Hawkley, L. C., Burleson, M. H., Berntson, G. G., & Cacioppo, J. T. (2003). Loneliness in everyday life: Cardiovascular activity, psychosocial context, and health behaviors. Journal of Personality and Social Psychology, 85, 105–120. Hawton, K., Cole, D., O’Grady, J., & Osborn, M. (1982). Motivational aspects of deliberate self-poisoning in adolescents. British Journal of Psychiatry, 141, 286 –291. Herman, B. H., & Panksepp, J. (1978). Effects of morphine and naloxone on separation distress and approach attachment: Evidence for opiate mediation of social affect. Pharmacology Biochemistry & Behavior, 9, 213–220.

14

DEWALL AND BAUMEISTER

Hilgard, E. R., Ruch, J. C., Lange, A. F., Lenox, J. R., Morgan, A. H., & Sachs, L. B. (1974). The psychophysics of cold pressor pain and its modification through hypnotic suggestion. American Journal of Psychology, 87, 17–31. Insel, T. R., & Winslow, J. T. (1991). Central administration of oxytocin modulates the infant rat’s response to social isolation. European Journal of Pharmacology, 203, 149 –152. Jepson, C., Loewenstein, G., & Ubel, P. (2001). Actual versus estimated differences in quality of life before and after renal transplant. Unpublished manuscript, Department of Social and Decision Sciences, Carnegie Mellon University. Kanarek, R. B., & Carrington, C. (2004). Sucrose consumption enhances the analgesic effects of cigarette smoking in male and female smokers. Psychopharmacology, 173, 57– 63. Kehoe, P., & Blass, E. M. (1986a). Behaviorally functional opioid systems in infant rats: II. Evidence for pharmacological, physiological, and psychological mediation of pain and stress. Behavioral Neuroscience, 100, 624 – 630. Kehoe, P., & Blass, E. M. (1986b). Opioid-mediation of separation distress in 10-day-old rats: Reversal of stress with maternal stimuli. Developmental Psychobiology, 19, 385–398. Kiecolt-Glaser, J. K., Ricker, D., George, J., Messick, G., Speicher, C. E., Garner, W., & Glaser, R. (1984). Urinary cortisol levels, cellular immunocompetency, and loneliness in psychiatric inpatients. Psychosomatic Medicine, 46, 15–24. Kimbrell, T. A., George, M. S., Parekh, P. I., Ketter, T. A., Podell, D. M., Danielson, A. L., et al. (1999). Regional brain activity during transient self-induced anxiety and anger in healthy adults. Biological Psychiatry, 46, 454 – 465. Konecka, A. M., & Sroczynska, I. (1990). Stressors and pain sensitivity in CFW mice: Role of opioid peptides. Archives Internationales de Physiologie et de Biochimie, 98, 245–252. Kosslyn, S. M., Pascual-Leone, A., Felician, O., Camposano, S., Keenan, J. P., Thompson, W. L., et al. (1999, May 7). The role of area 17 in visual imagery: Convergent evidence from PET and rTMS. Science, 284, 167–170. Leary, M. R., & Springer, C. A. (2000). Hurt feelings: The neglected emotion. In R. Kowalski (Ed.), Behaving badly: Aversive behaviors in interpersonal transgression (pp. 151–175). Washington, DC: American Psychological Association. Loewenstein, G., O’Donoghue, T., & Rabin, M. (in press). Projection bias in predicting future utility. Quarterly Journal of Economics. Loewenstein, G. F., Weber, E. U., Hsee, C. K., & Welch, N. (2001). Risk as feelings. Psychological Bulletin, 127, 267–286. Lynch, J. J. (1979). The broken heart: The medical consequences of loneliness. New York: Basic Books. MacDonald, G., Kingsbury, R., & Shaw, S. (2005). Adding insult to injury: Social pain theory and response to social exclusion. In K. Williams, J. Forgas, & W. von Hippel (Eds.), The social outcast: Ostracism, social exclusion, rejection, and bullying (pp. 77–90). New York: Psychology Press. MacDonald, G., & Leary, M. R. (2005). Why does social exclusion hurt? The relationship between social and physical pain. Psychological Bulletin, 131, 202–223. Mayer, J. D., & Gaschke, Y. N. (1988). The experience and metaexperience of mood. Journal of Personality and Social Psychology, 55, 102–111. McGuire, P. K., Shah, G. M. S., & Murray, R. M. (1993). Increased blood flow in Broca’s area during auditory hallucinations in schizophrenia. Lancet, 342, 703–706. Mercer, M. E., & Holder, M. D. (1997). Antinociceptive effects of palatable sweet ingesta on human responsivity to pressure pain. Physiology & Behavior, 61, 311–318. Murray, F. S., & Hagan, B. C. (1973). Pain threshold and tolerance of

hands and feet. Journal of Comparative and Physiological Psychology, 84, 639 – 643. Myers, D. (1992). The pursuit of happiness. New York: Morrow. Naranjo, J. R., & Fuentes, J. A. (1985). Association between hypoalgesia and hypertension in rats after short-term isolation. Neuropharmacology, 24, 167–171. Nelson, E. E., & Panksepp, J. (1998). Brain substrates of infant–mother attachment: Contributions of opioids, oxytocin, and norepinephrine. Neuroscience & Biobehavioral Reviews, 22, 437– 452. Nelson, L. D., & Morrison, E. L. (2005). The symptoms of resource scarcity: Judgments of food and finances influence preferences for potential partners. Psychological Science, 16, 167–173. Orbach, I., Mikulincer, M., King, R., Cohen, D., & Stein, D. (1997). Thresholds and tolerance of physical pain in suicidal and nonsuicidal patients. Journal of Consulting and Clinical Psychology, 65, 646 – 652. Panksepp, J. (1998). Affective neuroscience: The foundations of human and animal emotions. London: Oxford University Press. Panksepp, J., Herman, B. H., Conner, R., Bishop, P., & Scott, J. P. (1978). The biology of social attachments: Opiates alleviate separation distress. Biological Psychiatry, 13, 607– 618. Panksepp, J., Vilberg, T., Bean, N. J., Coy, D. H., & Kastin, A. J. (1978). Reduction of distress vocalization in chicks by opiate-like peptides. Brain Research Bulletin, 3, 663– 667. Petrovic, P., Kalso, E., Petersson, K. M., & Ingvar, M. (2002). Placebo and opioid analgesia—imaging a shared neural network. Neuroimage, 16, 1142–1150. 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. Pomerleau, O. F., Turk, D. C., & Fertig, J. B. (1984). The effects of cigarette smoking on pain and anxiety. Addictive Behaviors, 9, 265–271. Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University Press. Rossi, G. C., Pasternak, G. W., & Bodnar, R. J. (1993). Synergistic brainstem interactions for morphine analgesia. Brain Research, 624, 171–180. Rushen, J., Boissy, A., Terlouw, E. M., & de Passille´, A. M. (1999). Opioid peptides and behavioral and physiological responses of dairy cows to social isolation in unfamiliar surroundings. Journal of Animal Science, 77, 2918 –2924. Spear, L. P., Enters, E. K., Aswad, M. A., & Louzan, M. (1985). Drug and environmentally induced manipulations of the opiate and serotonergic systems alter nociception in neonatal rat pups. Behavioral and Neural Biology, 44, 1–22. Sufka, K. J., & Hughes, R. A. (1990). Dose and temporal parameters of morphine-induced hyperalgesia in domestic fowl. Physiology & Behavior, 47, 385–387. Sufka, K. J., & Weed, N. C. (1994). Construct validation of behavioral indices of isolation stress and inflammatory nociception in young domestic fowl. Physiology & Behavior, 55, 741–746. Twenge, J. M., Baumeister, R. F., DeWall, C. N., Ciarocco, N. J., & Bartels, J. M. (2006). Social exclusion reduces prosocial behavior. Unpublished manuscript. Twenge, J. M., Baumeister, R. F., Tice, D. M., & Stucke, T. S. (2001). If you can’t join them, beat them: Effects of social exclusion on aggressive behavior. Journal of Personality and Social Psychology, 81, 1058 –1069. Twenge, J. M., & Campbell, W. K. (2003). “Isn’t it fun to get the respect that we’re going to deserve?” Narcissism, social rejection, and aggression. Personality and Social Psychology Bulletin, 29, 261–272. Twenge, J. M., Catanese, K. R., & Baumeister, R. F. (2002). Social exclusion causes self-defeating behavior. Journal of Personality and Social Psychology, 83, 606 – 615. Twenge, J. M., Catanese, K. R., & Baumeister, R. F. (2003). Social exclusion and the deconstructed state: Time perception, meaningless-

SOCIAL EXCLUSION AND PAIN ness, lethargy, lack of emotion, and self-awareness. Journal of Personality and Social Psychology, 85, 409 – 423. 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. Van Boven, L., & Loewenstein, G. (2003). Social projection of transient drive states. Personality and Social Psychology Bulletin, 29, 1159 – 1168. 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. Wiedenmayer, C. P., Goodwin, G. A., & Barr, G. A. (2000). The effect of periaqueductal gray lesions on responses to age-specific threats in infant rats. Developmental Brain Research, 120, 191–198. 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.

15

Wilson, T. D., & Gilbert, D. T. (2003). Affective forecasting. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 35, pp. 345– 411). San Diego, CA: Academic Press. Wilson, T. D., Wheatley, T., Kurtz, J., Dunn, E., & Gilbert, D. T. (2004). When to fire: Anticipatory versus post-event reconstrual of uncontrollable events. Personality and Social Psychology Bulletin, 30, 340 –351. 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. Zadro, L., Williams, K. D., & Richardson, R. (2004). How low can you go? Ostracism by a computer is sufficient to lower self-reported levels of belonging, control, self-esteem, and meaningful existence. Journal of Experimental Social Psychology, 40, 560 –567.

Received February 24, 2005 Revision received August 15, 2005 Accepted August 25, 2005 䡲

Journal of Personality and Social Psychology 2006, Vol. 91, No. 1, 111–123

Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.111

Information Quantity and Quality Affect the Realistic Accuracy of Personality Judgment Tera D. Letzring

Shannon M. Wells

University of California, Riverside

Sanford Systems, Incorporated

David C. Funder University of California, Riverside Triads of unacquainted college students interacted in 1 of 5 experimental conditions that manipulated information quantity (amount of information) and information quality (relevance of information to personality), and they then made judgments of each others’ personalities. To determine accuracy, the authors compared the ratings of each judge to a broad-based accuracy criterion composed of personality ratings from 3 types of knowledgeable informants (the self, real-life acquaintances, and clinicianinterviewers). Results supported the hypothesis that information quantity and quality would be positively related to objective knowledge about the targets and realistic accuracy. Interjudge consensus and self– other agreement followed a similar pattern. These findings are consistent with expectations based on models of the process of accurate judgment (D. C. Funder, 1995, 1999) and consensus (D. A. Kenny, 1994). Keywords: personality judgment, judgment accuracy, realistic accuracy, information quantity, information quality

more accurate judgments of personality. The effect of information quantity has also been referred to as the acquaintanceship effect (Colvin & Funder, 1991; Kenny, 1991), because as people are acquainted longer they are assumed to acquire more information about each other. The concept of information quality is more complex, but at its core refers to the likelihood that even when information quantity is held constant, different contexts of acquaintanceship might vary in the degree to which personalityrelevant information becomes available in them. For example, a context in which everybody acts the same would yield only lowquality information for personality judgment because information about individual differences in personality would not be revealed, whereas a context in which behavior is relatively free to vary would yield higher quality information because people would be more likely to behave in a way that reflects their personalities (Snyder & Ickes, 1985). It is also possible that different kinds of information—for example, thoughts and feelings as opposed to hobbies and activities (Anderson, 1984)—are differentially informative about personality. When more personality-relevant information is available, the accuracy of judgments of individual differences in personality is expected to be higher. Information quantity and quality are two aspects of good information, one of four factors proposed to be related to judgment accuracy (Funder, 1999). It is difficult to examine information quantity and quality independently of each other in natural settings of acquaintanceship, because people are likely to emit a broader range of cues to personality and share more personality-relevant information with acquaintances they have known longer. For example, according to social penetration theory (Altman & Taylor, 1973), people share information about a larger number of the aspects of their person-

Every day, people make personality judgments of other people. The person making the judgments (the judge) has known some targets for several years and has seen them in many different contexts (e.g., a sibling or a spouse), whereas other targets are people the judge has just met and has seen only in a relatively limited context (e.g., a new coworker, a bank teller). Not surprisingly, these judgments vary in how accurate they are, perhaps because they are made on the basis of varying degrees of acquaintance and information that vary in how relevant they are to personality. These two aspects of the information available to a judge are referred to as information quantity, the sheer amount of information that is available, and information quality, the degree to which the available information is relevant to personality (Blackman & Funder, 1998; Funder, 1999). The concept of information quantity is straightforward in that it assumes that judges who have access to more information about the personality of the target will make

Tera D. Letzring and David C. Funder, Department of Psychology, University of California, Riverside; Shannon M. Wells, Sanford Systems, Incorporated, Lake Elsinore, California (now Key Data Systems). This article is based, in part, on the doctoral dissertation by Shannon M. Wells titled “The Effect of Information Quantity and Quality on the Accuracy of Personality Judgments.” Some of these findings were also presented at the annual meeting of the Society for Personality and Social Psychology as part of a symposium titled “The Accuracy of Interpersonal Judgment.” Data gathering for this article was supported by National Institute of Mental Health Grant MH42427 to David C. Funder. Correspondence concerning this article should be addressed to Tera D. Letzring, who is now at the Department of Psychology, Idaho State University, Pocatello, ID 83209. E-mail: [email protected] 111

LETZRING, WELLS, AND FUNDER

112

ality and share more personal information within these aspects as a relationship progresses, which suggests that information increases in both quantity and quality along with acquaintance. In the current study, the quantity and quality of the information available to judges were experimentally and independently manipulated so that the effects of these two factors on the accuracy of personality judgment could be examined, in order to achieve two goals. The first goal was to further examine the effects of information quantity on judgment accuracy by (a) testing for the acquaintanceship effect in a new context and across different levels of information quantity and by (b) examining the effects of information quantity on a relatively new construct, realistic accuracy, in addition to the more traditional constructs of consensus and self– other agreement. The second goal of the present research was to directly examine the effects of experimentally manipulated information quality on realistic accuracy, consensus, and self– other agreement, which is a more distinctive contribution, as this is one of the first empirical studies to examine these effects.

Indicators of Judgment Accuracy Three indicators of judgment accuracy were assessed: realistic accuracy, consensus, and self– other agreement.

Realistic Accuracy Realistic accuracy refers to a hypothetical construct representing the level of agreement between a personality judgment and what a target is really like (Funder, 1995). This construct cannot be directly measured by any single personality or behavior rating, as any single rating is highly uncertain as an indicator of what a person is really like. Instead, the ideal of realistic accuracy can be approached to the degree that multiple methods of measurement are used and combined to form a broad-based accuracy criterion for each target of judgment. The assumption is that this broadbased criterion is closer to what the target is really like than is any single rating, because random errors in the ratings should cancel each other out as more ratings are combined. In the current study, we formed an accuracy criterion by using a composite rating based on judgments from real-life acquaintances, psychologists who interviewed the targets, and the self. The difficulty of the criterion problem has led many researchers to shy away from studying the accuracy of judgments of real people and to instead turn their attention to models of the cognitive processing of artificial stimuli (Funder, 1987, 1995). As research progresses, it behooves investigators to begin to respond to the criterion problem instead of continuing to bypass it. The ultimate goal should be an appropriate, broad-based accuracy criterion, which would ideally derive from multiple modes of assessment, including the target’s self-perspective, ratings from knowledgeable informants, clinical judgments, behavioral measures, life outcome data, and perhaps even biological information such as hormone levels and functional MRI images (Funder, 1995). We propose that research that seeks to go beyond single operationalizations of the accuracy of personality judgment should refer to the outcome as realistic accuracy in order to distinguish this construct from other constructs that measure some other singular form of agreement or accuracy in personality judgment.

Consensus Consensus can be measured more directly than realistic accuracy and is more commonly studied. Consensus is the level of agreement between the personality judgments rendered by two or more people about another person. Consensus can be used as an indicator of accuracy, but it is important to keep in mind that consensus is not necessarily related to what the person is actually like, because many people could agree about the personality of someone else, and all could be wrong, as is the case when inaccurate stereotypes are used to make judgments.

Self–Other Agreement Self– other agreement can also be directly measured and has been used frequently in previous research (Ambady, Hallahan, & Rosenthal, 1995; Blackman & Funder, 1998). It refers to the level of agreement between judgments made about another person and that person’s self-judgments. The self-judgments are often implicitly assumed to reflect reality, and the judgments made by others are assumed to be accurate to the degree that they are related to the self-judgments. However, self– other agreement does not guarantee highly accurate judgments, because the self-rating will not adequately represent one’s real personality if people are unwilling or unable to provide accurate judgments of themselves (Hofstee, 1994; Kolar, Funder, & Colvin, 1996). Despite this complication of using self– other agreement to measure accuracy, many investigations refer to self– other agreement simply as accuracy (Ambady et al., 1995; Bernieri, Zuckerman, Koestner, & Rosenthal, 1994; Blackman & Funder, 1998). However, in the current article the term accuracy is used more specifically to describe realistic accuracy so that realistic accuracy can be differentiated from consensus and self– other agreement. We report findings concerning all three indicators of judgment accuracy for two reasons. First, realistic accuracy, although related to consensus and self– other agreement, is far from synonymous with either of these operationalizations and could potentially yield different results. Second, consensus and self– other agreement have been used in many prior studies (e.g., Anderson, 1984; Blackman & Funder, 1998; Kenny, Albright, Malloy, & Kashy, 1994), and our presentation of these constructs allows comparison with the existing literature.

Methods of Personality Judgment Realistic Accuracy Model The Realistic Accuracy Model (RAM; Funder, 1995, 1999) describes an interpersonal and cognitive process that results in accurate personality judgment when all four of its stages are successfully completed. First, the target of the judgment must display cues or behaviors that are relevant to the characteristic being judged in such a way and in contexts that are available to the judge. Then, the judge must detect these cues and correctly use them to make a judgment. RAM assumes that the four stages combine in a multiplicative fashion so that a failure at any of the four stages will make accuracy impossible. For example, cues can only be detected if they are available. Four factors are thought to influence the judgment process by affecting one or more stages of RAM (Funder, 1995): the good

INFORMATION AND PERSONALITY JUDGMENT

judge (Allport, 1937; Kolar, 1995; Taft, 1955; Vernon, 1933; Vogt & Colvin, 2003), good target (Colvin, 1993), good trait (Borkenau & Liebler, 1993; Funder & Dobroth, 1987; John & Robins, 1993; Norman & Goldberg, 1966), and good information (Blackman & Funder, 1998; Funder & Colvin, 1988; Funder, Kolar, & Blackman, 1995). Two aspects of good information, information quantity and information quality, are the focus of the current article. A prediction regarding the relations between information quantity and quality and all three indicators of accuracy can be generated on the basis of RAM. Even though RAM is a model for accuracy, it can also be used to make predictions about consensus if one assumes that personality is something real, in which case two accurate judgments of personality must agree with each other, and therefore consensus can be expected to be high at sufficiently high levels of accuracy. A linear increase between information and realistic accuracy, consensus, and self– other agreement was used as the prediction based on RAM. This is a conservative prediction because it is not possible to determine the degree to which the manipulated levels of information quantity and quality are equally spaced and, therefore, whether the increases in accuracy should also be equally spaced.

Weighted-Average Model The Weighted-Average Model (WAM; Kenny, 1991, 1994) offers a detailed description of the basis of consensus in personality judgment. The model predicts that consensus will not increase with acquaintanceship when judges see completely overlapping behaviors of the target and interpret what they see in exactly the same way (Kenny, 1991). However, when these assumptions are replaced by the more realistic expectations that different judges will interpret behaviors in different ways and will not detect exactly the same behaviors of the target, WAM predicts that consensus will increase rapidly at low levels of acquaintance and remain about the same across higher levels of acquaintance. This reasoning led us to predict that consensus would increase between low and medium levels of information quantity and quality but would be the same between medium and high levels of information quantity and quality. The prediction based on WAM was tested only for consensus because WAM is explicitly formulated as a model of the process of consensus, and it has been suggested that the relations between different indicators of accuracy and acquaintanceship are not necessarily the same (Kenny, 1991). For example, Kenny (1991) pointed out that accuracy can increase with acquaintance even as consensus stays the same, as is the case when all judges become more accurate but continue to agree with each other to the same degree. It was not feasible to determine whether the prediction based on RAM was superior to the prediction based on WAM, as the two predictions were highly correlated.1 However, it was still interesting to determine how well the obtained data fit each prediction.

Past Findings and Rationale for Current Experiment Information quantity has received a fair amount of research attention, and although the findings have been mixed, the main conclusion has been that, all other things being equal, to know someone longer is to know him or her better. Information quality

113

has received less attention, although the available evidence suggests that not all information is created equal and that some kinds of information are more likely to be related to judgment accuracy than others (e.g., Anderson, 1984).

Information Quantity A fair amount of research has examined the relationships between information quantity and two aspects of personality judgment, consensus and self– other agreement. To our knowledge, research has not previously been published on the relationship between information quantity and realistic accuracy. Past findings on consensus and self– other agreement have been mixed. In between-subjects designs, in which different participants are involved at each level of acquaintance, findings concerning consensus have both supported and not supported the acquaintanceship effect. Blackman and Funder (1998) experimentally manipulated level of acquaintance by having some people observe a target interacting for 5 or 10 min and other people observe a target for 25 or 30 min, and they found that consensus did not increase with acquaintance. However, researchers have found that consensus is higher among real-life acquaintances than among relative strangers—that is, people who viewed dyadic interactions for either 5 min (Funder & Colvin, 1988) or 25 or 30 min (Blackman & Funder, 1998). Furthermore, a meta-analysis suggests that for agreeableness, conscientiousness, neuroticism, and openness to experience, consensus is greater among long-term acquaintances than among people who were unacquainted or had interacted only once (Kenny et al., 1994). In general, higher levels of consensus have been associated with higher acquaintance in studies with between-subjects designs in which there was a relatively large difference in the amount of available information between the high and low acquaintance groups, but not when there was a relatively small difference in acquaintance. Between-subjects designs have generally found support for the acquaintanceship effect when self– other agreement is the outcome. For example, self– other agreement was higher among people who watched videotaped unstructured dyadic interactions for 25 to 30 min versus 5 or 10 min (Blackman & Funder, 1998), people who had been roommates for more than 10 months versus less than 10 months (Bernieri at al., 1994), people who indicated they knew the target of judgment extremely well versus not at all (Paunonen, 1989), and real-life acquaintances versus people who had watched 5-min unstructured dyadic interactions (Funder et al., 1995). In between-subjects designs, information quantity appears to have a positive linear relationship with self– other agreement across all levels of acquaintance. However, a complication common to between-subjects designs of consensus and self– other agreement is that real-life acquaintances have self-selected to know a person for longer, which introduces the possibility that confounds such as similarity to the target and liking of the target influence accuracy in addition to length of acquaintance. This is why, in the current study, we have experimentally manipulated information quantity. In within-subjects research designs in which the same participants are involved at each level of acquaintance, previous findings With contrast weights of –1, 0, ⫹1 for RAM and –2, ⫹1, ⫹1 for WAM, the correlation between the predictions was .87. 1

114

LETZRING, WELLS, AND FUNDER

suggest that as acquaintance increases, consensus generally remains constant, whereas self– other agreement generally increases. Consensus has been shown to stay about the same as acquaintance increases across a variety of samples, situations, and lengths of time, including previously unacquainted students who participated in 30-min group interactions on 4 consecutive days (Park & Judd, 1989), students who participated in weekly 20-min group meetings for 7 weeks (Paulhus & Bruce, 1992), and students living near each other in a dorm from 2 to 33 weeks after the start of a semester (Park, Kraus, & Ryan, 1997). Furthermore, a metaanalysis suggested that consensus did not increase among people who interacted for between 8 min and 2 hr in both laboratory and naturalistic settings (Kenny et al., 1994). On the other hand, an examination of self– other agreement provided evidence that self– other agreement generally increased among students who participated in weekly 20-min group meetings for 7 weeks (Paulhus & Bruce, 1992). Also, findings by Borkenau and colleagues (Borkenau, Mauer, Riemann, Spinath, & Angleitner, 2004) suggest that acquaintance and accuracy are positively related when accuracy is operationalized as the level of agreement with ratings from the target, acquaintances of the target, experimenters, and a confederate who interacted with the target; and when acquaintance is operationalized as the number of behavioral episodes observed and rated by different judges that were included in a composite judgment. One possible explanation for the inconsistent findings concerning the acquaintanceship effect is that consensus was used as the outcome, which is not necessarily the appropriate way to approach this issue if one wishes to determine the extent to which judgments become more accurate with increased acquaintance. It is possible that the ratings of the judges in the studies mentioned previously (Kenny et al., 1994; Park & Judd, 1989; Park et al., 1997; Paulhus & Bruce, 1992) did become more accurate across time, but this change might not be reflected in consensus if the ratings continued to have the same level of similarity with each other (Blackman & Funder, 1998). Another possible limitation of past research is that researchers have simply assumed that people who have known each other longer have also acquired more information about each other, which is not necessarily true. One way to determine whether people in longer relationships have actually acquired more personality-relevant information is to directly measure the amount of objective information they know about the target. In the current study, this concern was addressed by asking participants a series of factual questions about their interaction partners that sampled from the kinds of information people might learn in an initial interaction. This sample of relatively objective questions was used as a manipulation check to determine whether participants who interact for a longer period of time acquire more information about each other. If the manipulation is successful, then people who interact longer will have acquired more of this information, concomitant to picking up more information relevant to personality, and therefore we expected to find a moderate positive relation between information acquisition and realistic accuracy, consensus, and self– other agreement. The information quantity analyses are a conceptual replication of a study by Blackman and Funder (1998). The current project also went beyond Blackman and Funder’s study and other previous research in several important ways. First, participants in Blackman

and Funder’s experiment were exposed to targets via videotaped two-person interactions, whereas participants in the current study actually interacted with each other in three-person groups before making personality judgments. This method may have different implications for the way judges detect and use information when making personality judgments. For example, judges watching videotapes have no need to be concerned with how they are perceived or what they will say next, whereas judges who are involved in an interaction are likely to be concerned with these and other issues. More important, many researchers, including Blackman and Funder, have used self– other agreement as a proxy for accuracy, whereas in the current study, we went beyond self– other agreement and examined a construct closer to the ideal of realistic accuracy by using a broad-based accuracy criterion.

Information Quality Although a considerable amount of research has focused on the relationship between information quantity and personality judgment, surprisingly little research has focused on an arguably even more important aspect of information—information quality, or the personality relevance of the information that becomes available within a given period of time. Intuitively, it seems apparent that not all kinds of information contribute equally to the achievement of an accurate judgment, as it is possible to be acquainted with someone for only a short time but to know him or her very well, or to be acquainted with someone for years and still know little about that person. From a research perspective, the first step in examining information quality is to determine the extent to which it is related to indicators of accuracy, including realistic accuracy, consensus, and self– other agreement, which was one goal of the current study. Anderson’s (1984) groundbreaking study supported the idea that information quality is related to self– other agreement. Anderson found that observing an interview in which questions about thoughts and feelings were discussed yielded higher self– other agreement of personality ratings than did observing an interview in which hobbies and activities were discussed. One conclusion that can be based on this finding is that thoughts and feelings are higher quality information than hobbies and activities, and this is why the judges who learned about thoughts and feelings were able to agree more closely with the target’s self-judgments of personality than were the judges who learned about hobbies and activities. Another aspect that might influence the quality of the information that is likely to be revealed is the situation’s “strength” (Snyder & Ickes, 1985). “Strong” situations limit the range of behavior that people display because they include explicit rules or evoke implicit norms to which people generally adhere. At the other extreme, “weak” situations allow for considerable behavioral variation because there are few rules or norms for typical behavior. Therefore, in comparison with a strong situation, a weak situation should allow for more behavioral variation, the availability of higher quality individuating information, the acquisition of more objective information, and greater judgment accuracy. The present study was designed to begin the investigation of what occurs when people interact in situations with different levels of information quality. A strong situation was created experimentally by giving participants the specific task of answering a long set of trivia questions. Two weaker situations were created by telling

INFORMATION AND PERSONALITY JUDGMENT

participants to talk about whatever they would like or to get to know one another as well as possible. We predicted that more personality-relevant information would be revealed in the two weak situations than in the strong situation and that realistic accuracy, self– other agreement, and consensus would be higher in the two weak situations. We also predicted that more personalityrelevant information would be available in the weak situation in which participants were told to get to know each other than in the weak situation in which participants were told to talk about whatever they would like, because the participants would be more likely to reveal or ask about personality-relevant information as they attempted to get to know each other than when they were simply trying to pass the time by talking about whatever they liked.

115

relevant information and to achieve higher levels of realistic accuracy, consensus, and self– other agreement than participants who were simply instructed to talk about whatever they would like. Therefore, on the basis of RAM, we predicted a linear relationship between information quality and all three indictors of accuracy. We could also make a prediction based on WAM. Even though the model does not directly consider information quality, we have defined information quality as the availability of more personalityrelevant information with time held constant, so that the acquaintanceship parameter of WAM should increase with information quality. Therefore, on the basis of WAM, we predicted that consensus would increase between the low- and medium-quality conditions and remain about the same between the medium- and high-quality conditions.

Research Questions With the theoretical basis of both the RAM (Funder, 1995) and the WAM (Kenny, 1991) in mind, in the current study we examined how quantity and quality of available information affect realistic accuracy, consensus, and self– other agreement. Hypothesis 1: Participants who interact for longer periods of time will achieve higher levels of realistic accuracy, consensus, and self– other agreement than people who interact for shorter periods of time. To examine this issue, we compared personality judgments following three-person unstructured interactions of three lengths (minimal interaction, 50 min, and 3 hr). On the basis of RAM (Funder, 1995, 1999), we expected that all three indicators of personality judgment—realistic accuracy, consensus, and self– other agreement—would increase linearly across these experimental conditions. On the basis of WAM (Kenny, 1991), we expected consensus to increase between the low- and medium-quantity conditions and to stay about the same between the medium- and high-quantity conditions. Thus, the predicted relationships between information quantity and consensus differ slightly depending on which model is used, and both predictions were tested. Hypothesis 2: Participants who interact in situations in which more personality-relevant information is likely to be available will achieve higher levels of realistic accuracy, consensus, and self– other agreement than participants who interact in situations in which less personality-relevant information is likely to be available. In one of the first experimental investigations to directly examine this issue, we compared the accuracy of personality judgments made following one of three 50-min interactions that vary in the amount of personality-relevant information likely to be available in them. The low-quality condition was a strong situation that allowed for little behavioral variation and was expected to elicit the least amount of personality-relevant information and therefore the lowest levels of realistic accuracy, consensus, and self– other agreement. The medium- and high-quality conditions were both relatively weak situations that were expected to elicit more behavioral variation and therefore higher levels of all indicators of accuracy than the strong situation. Between the two weak situations, participants who were given the objective to get to know each other were expected to share a broader range of personality-

Method The current study was a between-subjects experimental design in which each participant interacted in one of five conditions. The first independent variable, information quantity, had three levels: low (minimal information condition), medium (short unstructured condition), and high (long unstructured condition). The second independent variable, information quality, also had three levels: low (trivia condition), medium (short unstructured condition), and high (get to know condition). The effects of each independent variable were examined for three dependent variables: realistic accuracy (agreement between personality judgments and a broad-based accuracy criterion), self– other agreement (agreement between judgments and the target’s self-ratings), and consensus (agreement between two judges about a target).

Participants A total of 506 undergraduate students participated in the Riverside Accuracy Project—Phase 2 (RAP–II) and were paid $10 per hour for their time. A core group of 180 target participants (90 men and 90 women) were recruited via announcements made in psychology classes and fliers placed on bulletin boards advertising “research on personality.” These participants were asked to recruit two close acquaintances (for a total of 326 acquaintance informants) to provide personality descriptions of themselves and the target participants. The ethnic breakdown of the target participants was 38% Asian American, 20% Hispanic, 14% Caucasian, 12% African American, and 16% other or not specified. There were three groups of each possible gender composition (female–female–female [FFF], male–male– male [MMM], female–female–male [FFM], and female–male–male [FMM]) within each experimental condition, and all of the gender compositions were analyzed together because of the balanced nature of the design and because the number of participants within each gender composition would be too small to find reliable results if analyzed separately. This is the second article to come out of the extensive RAP–II data set, and the analyses do not overlap with the previous project, which focused on the constructs of ego-control and ego-resiliency (see Letzring, Block, & Funder, 2005), or with future planned projects.

Materials California Adult Q (CAQ) set. The CAQ (Block, 1961, as modified for use by nonprofessionals by Bem & Funder, 1978) consists of 100 carefully formulated descriptive statements about personality (e.g., is critical/skeptical, is personally charming, is cheerful). In the Q-item rating, each item is responded to using a 9-point Likert-type scale ranging from 1 (extremely uncharacteristic) to 9 (extremely characteristic). Participants rated themselves and their group interaction partners using the Q-item rating, and acquaintance informants also rated the target participants using the Q-item

116

LETZRING, WELLS, AND FUNDER

rating. Participants rated both group interaction partners at the same time for each item of the CAQ, which implies that for each item the interaction partners were directly compared with each other at the time of the judgment. We expected this procedure to help the judges make individuating and thoughtful ratings for each item and both targets. Clinicians rated the target participants using the traditional Q-sort method, in which the 100 items are placed into a forced-choice, quasi-normal distribution so that each category (1 through 9) contains a predetermined number of items. Information fact sheet. An information fact sheet was constructed for RAP–II that consisted of open-ended questions that were designed to sample from the wide array of information that could become available during the experimental interactions. The information fact sheet was used as a manipulation check to determine whether participants in higher quantity and quality interactions actually acquired more information about their interaction partners than did participants in lower quantity or quality interactions. Questions included information about age, place of birth, political affiliation, future goals, family makeup, and so forth. No attempt was made to assess the information that participants picked up in any comprehensive fashion, or even to focus specifically on personalityrelevant information (because techniques for either goal are presently unavailable). Rather, we aimed simply to obtain a small sample of the information that one person might learn about another during a casual interaction, with the assumption that scores on this measure would correlate with the amount of information, including personality-relevant information, acquired overall. The information fact sheet was completed by the target participants about their group interaction partners immediately following the ratings of personality and was completed on a later occasion by each participant about him- or herself. Each question was scored correct (1 point) if the answer was equivalent to the answer given by the self; half (.5 points) if the answer was similar, but not equivalent, to the answer given by the self; or incorrect (0 points) if the answer was different from the answer given by the self or if there was no response. Each information fact sheet was scored by two independent judges, and, as the decisions about the correctness of an answer were relatively objective, any discrepancies in scoring were reconciled by a third judge. The reliability of the 19 items was high enough to justify using the sum as a score representing information acquisition (Cronbach’s alpha ⫽ .77).

Procedures Overview. Target participants came to the lab on three separate occasions and also completed three packets of self-report questionnaires outside of the lab, only some of which were used in the current analyses. During the first session, participants interacted in three-person groups in one of five experimental conditions and then made personality judgments of their two interaction partners. During a subsequent session, each participant was interviewed by a clinical psychologist who described the personality of the participant after the conclusion of the interview. Additionally, acquaintance informants completed questionnaires regarding the target participant with whom they were acquainted. Presession. At the presession, participants were given the first takehome packet of questionnaires and were scheduled into one of the five experimental conditions. The times of the conditions were established in advance, and the participants chose a session that fit with their schedules. Although technically this procedure was only quasi-random, it did assign participants to conditions independently of any aspect of the experimental design, and we have no reason to believe that participants in the various conditions differed from each other in any way that might have affected the outcome of the study. Additionally, each participant was shown a folder containing pictures of the other participants in the same group to determine whether he or she had ever seen the other participants. If the participant recognized another participant in the same group, the participant was assigned to a different group. This procedure ensured that none of the

participants were acquainted with their interaction partners prior to the experimental group interaction. Experimental conditions. Target participants interacted in the laboratory in one of five experimental conditions designed to vary in the quantity and quality of interpersonal information likely to become available in them. Each of the groups consisted of three previously unacquainted target participants in one of four possible gender compositions: all female (FFF), all male (MMM), 2 females and 1 male (FFM), and 1 female and 2 males (FMM). An equal number of the four gender compositions was assigned to each experimental condition. A total sample size of 180 allowed for a balanced design with 3 groups of each gender composition per condition, for a total of 12 groups per condition.2 In all conditions, participants were seated in a room at a round table and given verbal instructions. At the conclusion of the interaction, participants rated their two partners using three measures, which included the rating form of the CAQ and the information fact sheet. Minimal information condition. Participants rated the other two participants in the room immediately following a short set of instructions explaining that they were studying first impressions and to do the best they could describing their first impressions of the other two people in the room. They were asked not to speak to each other but were in each others’ physical presence for as long as it took to complete the ratings (less than 1 hr). This situation provided an empirical estimate of the baseline of realistic accuracy, consensus, and self– other agreement and was used as the low-level condition for information quantity. Trivia quiz condition. Participants were presented with a packet containing 380 trivia-type questions, each of which had a single correct answer. They were informed that they would have 50 min to work jointly through the packet of questions and arrive unanimously at what they believed to be the correct answers. This context was designed to be a strong situation that would leave relatively little room for extraneous commentary or the disclosure of personality-relevant information, as the group would spend the entire interaction time responding to the trivia questions. This condition was used as the low-level condition for information quality. Short unstructured condition. Participants were told that they could talk about anything they liked for the next 50 min. No attempt was made to direct or suggest what should be done over the course of the interaction. This context provided a weak situation in that participants were able to do and say whatever they liked and to exhibit individual differences in social behavior, which should have included at least some personality-relevant information on which judgments could be based. This condition was used as the mid-level condition for both information quantity and quality. Long unstructured condition. Participants were told that they could talk about anything they liked for the next 3 hr, and as in the short unstructured condition, no attempt was made to direct or suggest what should be done over the course of the interaction. The experimenter returned to the room halfway through the interaction to give participants a short break, at which time snacks were provided and a restroom break was offered. As in the short unstructured condition, this context provided a weak situation with very little structure and should have included at least some personality-relevant information on which judgments could be based. This condition was used as the high-level condition for information quantity. Get to know (GTK) condition. Participants were told that their task for the next 50 min was to get to know each other as well as possible and to learn as much as they could about what type of person each of them was.

2

On completion of data collection, it was found that 2 target participants were inadvertently included in the study twice. For both participants, the second condition that they were in was different from the first. Analyses were done with these groups included and excluded. The differences between these two sets of findings were within rounding error. To maintain a fully balanced design, we report results from the data of all 60 groups.

INFORMATION AND PERSONALITY JUDGMENT This context provided a weak situation that was expected to yield a wide range of behaviors relevant to personality, and the objective of getting to know each other was expected to guide participants to probe for and to remember information related to personality. This condition was used as the high-level condition for information quality. Life history interview. Participants were individually interviewed by one of four professionally trained (i.e., MSW, MA in counseling, or PhD in clinical psychology) interviewers who had experience with a college population. With consent of the participants, all interviews were videotaped. The clinicians conducted a 1-hr semistructured life history interview adapted from a protocol used for many years by the Institute of Personality Assessment and Research (Craik et al., 2002). The protocol used in the current study was adapted to better apply to college students and sought to capture a broad range of personality-relevant information without explicitly asking about sensitive topics and risky behaviors. Each interview started with the clinician asking the participant, “Tell me something about yourself,” and then covered a broad range of topics, including college and academic experiences, future plans, interpersonal relationships, and childhood and family history. In conclusion, each participant was asked to describe “a defining event in [their] life that had a significant impact on or changed [their] life in some way.”3 Following the interview, the clinician completed a Q-sort description of the target participant. Some of the interviews were later viewed by a second clinician, and in these cases the second clinician also described the participant’s personality using the CAQ.4 When two ratings of the participant were available, a composite score was computed for each item and used in subsequent analyses. The average interclinician profile agreement was r ⫽ .50 (SD ⫽ 0.17),5 and the coefficient of internal consistency based on a composite of the two raters was .67. Acquaintance–informant ratings. Participants were asked to provide the names and contact information of the two people who knew them best at the university. Acquaintances came into the lab and provided descriptions of the target participant by whom they had been identified, using the rating format of the CAQ. The average profile agreement for the 100 items as rated by two acquaintances was r ⫽ .40 (SD ⫽ 0.19).6 The average coefficient of internal consistency, or the dependability of the profile based on a composite of the two acquaintances, was .57.

Analyses Realistic accuracy criterion. Many researchers have used self– other agreement as a proxy for accuracy, but in the current project we strived to go beyond this simple operationalization by comparing judgments to a broader based criterion composed of ratings by three types of knowledgeable informants: the self, two acquaintances nominated by the target, and a professionally trained clinician-interviewer. The 100 items of the CAQ were rated by each type of rater, and these ratings were used to compute a composite score for each item. First, when ratings from two acquaintances or two clinicians were available, these ratings were averaged.7 Then, a simple average for each item was computed across the three ratings from the self, acquaintances, and clinician. The construct of realistic accuracy is very new, so there is not an established level of reliability for determining the adequacy of the accuracy composite. However, on the basis of the magnitudes of correlations that are reported in the self– other agreement literature (Bernieri et al., 1994; Funder, 1980), the mean alpha reliability of the accuracy composite seemed high enough to justify the use of these average ratings (M ⫽ 0.42, SD ⫽ 0.13, range ⫽ 0.08 – 0.72).8 Computation of realistic accuracy. Profile analysis allows for an examination of judgments in regard to a target’s overall personality by the use of the entire set of 100 CAQ items in a single analysis. The computation of a profile correlation simply involves correlating two sets of ratings for the same target across the 100 items of the CAQ. First, to examine realistic accuracy, we correlated the ratings by each judge of each interaction partner with the accuracy composite for that interaction partner, resulting in a profile correlation that represented the relation between the ratings of

117

Figure 1. Computation of profile realistic accuracy scores. Note that profile correlations were computed for each of the three judges in each experimental group, and these scores were Z transformed and averaged to compute scores for each group; the scores of all 12 groups in each type of experimental interaction were averaged to determine the accuracy for the corresponding interaction type. CAQ ⫽ California Adult Q.

the judge and an approximation of what the target is really like. This procedure yielded six scores for each group (two scores for each of three judges). These scores could be transformed using Fisher’s r-to-z transformation and averaged so that each group had a single realistic accuracy score (see Figure 1). Computation of consensus. To examine consensus, we correlated the ratings provided by the two interaction partners of each target with each other, resulting in a profile correlation that represented the degree of similarity in the partner’s ratings of that target. This analysis yielded three consensus scores for each group (one for each target), which could be Z transformed and averaged so that each group had one consensus score. Computation of self– other agreement. Finally, to examine self– other agreement, we correlated the ratings of each interaction partner with the partner’s self-ratings, resulting in a profile correlation that represented the relation between the ratings of the judge and the target’s self-ratings. As in realistic accuracy, this procedure yielded six self– other agreement scores

3

The full protocol for the clinical interview is available on the Riverside Accuracy Project’s Web site at http://www.rap.ucr.edu/interview.htm. 4 When the participant scheduled for a live interview was unable to keep the appointment, clinicians were given the option to observe and rate a randomly chosen, previously recorded interview. In this manner, 47 interviews were observed and rated by a second clinician. 5 This number can be compared with 30 random pairings of ratings by clinicians who described different targets, which resulted in an average profile correlation of .29. 6 This number can be compared with 30 random pairings of ratings by acquaintances who described different targets, which resulted in an average profile correlation of .24. 7 When ratings from only one acquaintance or clinician were available, then there was no need to compute an average before averaging the ratings with the other sources. 8 On the basis of correlations among types of raters, the accuracy criterion is appropriate to use for computing profile correlations: self– acquaintance (M ⫽ 0.48, SD ⫽ 0.17), self– clinician (M ⫽ 0.39, SD ⫽ 0.14), acquaintance– clinician (M ⫽ 0.37, SD ⫽ 0.19); Spearman–Brown for three raters ⫽ .68.

118

LETZRING, WELLS, AND FUNDER

per group, and these scores could be Z transformed and averaged so that each group had a single self– other agreement score. Components of accuracy scores. Cronbach (1955) was one of the first to point out that profile correlations may be confounded with several other elements besides differential accuracy, including elevation, differential elevation, and stereotype accuracy, which result from the use of response sets or a reliance on stereotypes. The effects of response sets were not of concern for between-groups comparisons in the current experiment because participants were randomly assigned to one of five experimental conditions, and therefore, participants with various response sets were equally distributed among the conditions. The effects of stereotype accuracy also were not of concern because stereotype accuracy should have equivalent effects on the magnitude of the profile correlations across all conditions, and we were interested in the differences between conditions and not in the absolute magnitude of accuracy. For this reason, we used the raw profile correlations in our cross-group comparisons, as did Blackman and Funder (1998) when they conducted similar analyses. However, it is true that as a result of stereotype accuracy, the baseline correlation between two sets of ratings was not expected to be zero, as some items of the CAQ are generally rated higher than others, regardless of the target (Blackman & Funder, 1998). To determine baseline profile correlations within the current data, we computed the correlations among 30 random pairs of ratings. For realistic accuracy, the profile correlations were computed between the ratings of one target and the accuracy composite for a different target, which resulted in a baseline profile correlation of .36. For consensus, the profile correlations were computed between two interaction partners of different targets, which resulted in a baseline profile correlation of .25. For self– other agreement, the profile correlations were computed between the ratings of one target and the self-ratings of a different target, which resulted in a baseline profile correlation of .28. These baseline correlations may at first seem somewhat high, but when one realizes that all participants in the sample were college students and that, in general, college students are similar to each other in many ways, it makes sense that even random pairings resulted in profile correlations of moderate magnitude. Profile correlations greater than these baseline correlations indicate differential accuracy (Blackman & Funder, 1998). Contrast analysis. Contrast analysis was used to test focused questions by assigning weights to experimental conditions that reflect a prediction concerning the structure of the data (Rosenthal, Rosnow, & Rubin, 2000). Once contrast weights were assigned to each condition, it was possible to determine the extent to which the data were in line with the prediction. When a t test was computed for a contrast analysis,9 the degrees of freedom were the number of participants minus the number of groups.10 One-tailed p values that are associated with the tcontrast are reported. It was also possible to determine how much variance in the group means was accounted for by the prediction by correlating the group means with the contrast weights and squaring the obtained correlation, which resulted in an r2alerting. A large r2alerting provides support for the prediction. Furthermore, it was possible to test whether the leftover variance, or the variance not accounted for by the prediction, was significant by computing a tnoncontrast.11 A nonsignificant tnoncontrast also lends support to the a priori prediction because little variance is left over that is not accounted for by the prediction. On the basis of RAM, we predicted that there would be linear relationships among information quantity and quality and realistic accuracy, consensus, and self– other agreement, for which the appropriate contrast weights are ⫺1 (low information), 0 (medium information), and ⫹1 (high information). On the basis of WAM, we predicted a slightly different relationship between information quantity and quality and consensus, which reflects an increase at low levels of acquaintance and then a leveling off at higher levels of acquaintance, for which the appropriate contrast weights are ⫺2 (low information), ⫹1 (medium information), and ⫹1 (high information).

Post hoc t tests. Contrast analyses are informative regarding how well the obtained data fit an a priori prediction about how the data should be ordered, but they do not provide information regarding which pairs of groups are reliably different from each other. Therefore, post hoc independent-samples t tests were used to determine which pairs of experimental conditions had reliably different levels of realistic accuracy, consensus, and self– other agreement. No adjustments were made to the p values associated with these tests, so the conservative reader may wish to adjust the p values to account for the number of comparisons in each analysis.

Results The focus of our analyses was on realistic accuracy, as this project was one of the first to explore the construct and the first, to our knowledge, to examine the relationships between realistic accuracy and experimentally manipulated information quantity and quality. We also provided evidence concerning our predictions for consensus and self– other agreement. Before examining the effects of information quantity and quality on judgment accuracy, we first determined whether participants in longer and higher quality interactions did indeed acquire more information (on the basis of a sample of information possibly available in such interactions), as a manipulation check. Then, we used contrast analysis (Rosenthal et al., 2000) to examine how well the changes in accuracy across experimental conditions fit the predictions based on RAM for realistic accuracy, consensus, and self– other agreement and based on WAM for consensus only. Finally, the degree to which information acquisition was related to realistic accuracy, consensus, and self– other agreement, regardless of the type of group the participants interacted in, was examined. There are several possible ways to analyze these data. We used traditional statistical procedures to examine differences at the group level by conducting a between-groups analysis to determine how well the obtained data fit the predictions. Our predictions were concerned with properties of the context, not of the individual, and a group-level analysis reflects this orientation. Although none of the basic conclusions of this study change when data are examined at different levels of analysis, an advantage of analyzing data at the level of the group is that it allowed us to perform all analyses in a consistent manner (at the same level of analysis).

Information Acquisition To determine whether participants in interactions of greater quantity and quality acquired more information about each other, 9

The formula used to compute tcontrast was

冑

2 ralerting ⫻ SSbetween . MSwithin

The formula used to compute rcontrast was

冑 10

2 contrast 2 contrast

t

t

⫹ df

.

In the current analyses, there were 12 participants in each of the 3 groups, so there were [(12)(3) – 3] ⫽ 33 degrees of freedom for each contrast analysis. SSbetween ⫺ SScontrast 11 tnoncontrast ⫽ . MSwithin

冑

INFORMATION AND PERSONALITY JUDGMENT

we compared the scores achieved on the information fact sheet across experimental conditions. This can be considered a manipulation check that determines the degree to which participants in longer and higher quality interactions actually learned more about each other. Information quantity. A good fit was found between amount of acquired information and a linear increase across the three levels of information quantity: minimal information (M ⫽ 2.80, SD ⫽ 1.10); short unstructured (M ⫽ 5.17, SD ⫽ 1.09); long unstructured (M ⫽ 7.00, SD ⫽ 1.58); and rcontrast ⫽ .81, tcontrast(33) ⫽ 8.02, p ⬍ . 0001 (see Figure 2). Furthermore, on the basis of independent-samples t tests, we found that the scores of all groups differed significantly from each other: minimal information and short unstructured, t(22) ⫽ 5.29, p ⬍ .0001; minimal information and long unstructured, t(22) ⫽ 7.54, p ⬍ .0001; and short unstructured and long unstructured, t(22) ⫽ 3.29, p ⫽ .003. Information quality. A good fit was also found between the amount of acquired information and a linear increase across the three levels of information quality: trivia (M ⫽ 1.65, SD ⫽ 1.04); short unstructured (M ⫽ 5.17, SD ⫽ 1.09); GTK (M ⫽ 5.87, SD ⫽ 1.57); and rcontrast ⫽ .82, tcontrast(33) ⫽ 8.22, p ⬍ .0001 (see Figure 3). Participants in both the short unstructured and GTK conditions acquired reliably more information than did participants in the trivia condition, t(22) ⫽ 8.08, p ⬍ .0001, and t(22) ⫽ 7.75, p ⬍ .0001, respectively, although information acquisition did not differ reliably between the short unstructured and GTK conditions, t(22) ⫽ 1.26, p ⫽ .22. These findings concerning a small sample of the available information imply that participants in longer and higher quality interactions acquired more information about each other. The next analyses addressed the question of whether personality judgments were also more accurate in the longer and higher quality interactions.

119

Figure 3. Mean information acquisition scores for experimental conditions used in the information quality analyses. GTK ⫽ get to know condition.

Information quantity. To test the prediction that judges would be more accurate after interacting for longer periods of time, we examined personality judgments at three levels of information quantity: low (minimal information–no interaction), medium (short unstructured–50-min interaction), and high (long unstructured–3-hr interaction). First, profile correlations were computed for each group, as described in the Analyses section. The

profile correlations were transformed using Fisher’s r-to-z transformation, and then contrast analyses (Rosenthal et al., 2000) were used to determine the degree to which the profile correlations followed the predictions based on RAM and WAM. As predicted, we found a good fit between realistic accuracy scores and the prediction of a positive linear relationship with information quantity: minimal information (mean r ⫽ .29, SD ⫽ 0.20); short unstructured (mean r ⫽ .45, SD ⫽ 0.14); long unstructured (mean r ⫽ .47, SD ⫽ 0.09); and rcontrast ⫽ .53, tcontrast(33) ⫽ 3.56, p ⫽ .000058 (see Figure 4). The linear prediction accounted for 86% of the variance in the group means, and the variance not accounted for by the prediction did not reach the .05 level of significance, tnoncontrast(33) ⫽ 1.41, p ⫽ .08. When the groups were compared with each other, realistic accuracy was reliably higher in both the short unstructured and long unstructured conditions than in the minimal information condition, t(22) ⫽ 2.57, p ⫽ .02, and t(22) ⫽ 3.46, p ⫽ .002, respectively, but accuracy in the long unstructured condition was not reliably higher than accuracy in the short unstructured condition, t(22) ⫽ 0.74, p ⫽ .47. Information quantity also had a good fit to the prediction of a positive linear relationship with consensus: minimal information (mean r ⫽ .25, SD ⫽ 0.18); short unstructured (mean r ⫽ .41, SD ⫽ 0.20); long unstructured (mean r ⫽ .44, SD ⫽ 0.17); and rcontrast ⫽ .44, tcontrast(33) ⫽ 2.83, p ⫽ .004. The linear prediction accounted for 86% of the variance in the group means, and the

Figure 2. Mean information acquisition scores for experimental conditions used in the information quantity analyses.

Figure 4. Mean realistic accuracy scores for experimental conditions used in the information quantity analyses. Note that the random baseline profile correlation was .36.

Personality Judgment

120

LETZRING, WELLS, AND FUNDER

variance not accounted for by the prediction was not significant, tnoncontrast(33) ⫽ 1.14, p ⫽ .13. Judges in the short unstructured and long unstructured conditions had reliably higher consensus than judges in the minimal information condition, t(22) ⫽ 2.33, p ⫽ .03, and t(22) ⫽ 2.98, p ⫽ .007, respectively, although consensus did not differ for judges in the short unstructured and long unstructured conditions, t(22) ⫽ 0.42, p ⫽ .68. The consensus data were also in line with the prediction based on WAM, in which an increase was predicted between the lowand medium-quantity conditions but not between the medium- and high-quantity conditions, rcontrast ⫽ .48, tcontrast(33) ⫽ 3.04, p ⫽ .0023. The nonlinear prediction accounted for 98% of the variance in the group means, and the variance not accounted for by the prediction was not significant, tnoncontrast(33) ⫽ 1.18, p ⫽ .12. Finally, information quantity had a good fit to the prediction of a positive linear increase with self– other agreement: minimal information (mean r ⫽ .21, SD ⫽ 0.19); short unstructured (mean r ⫽ .36, SD ⫽ 0.12); long unstructured (mean r ⫽ .39, SD ⫽ 0.12); rcontrast ⫽ .49, tcontrast(33) ⫽ 3.24, p ⫽ .0014. The linear prediction accounted for 86% of the variance in the group means, and the variance not accounted for by the prediction was not significant, tnoncontrast(33) ⫽ 1.29, p ⫽ .10. Judges in the short unstructured and long unstructured conditions had reliably higher self– other agreement than judges in the minimal information condition, t(22) ⫽ 2.58, p ⫽ .02, and t(22) ⫽ 3.03, p ⫽ .006, respectively, although self– other agreement did not differ for judges in the short unstructured and long unstructured conditions, t(22) ⫽ 0.72, p ⫽ .55. Note that for all indicators of accuracy, the mean profile correlations for the minimal information condition were approximately equal to or smaller than the random baseline correlations of .36 for realistic accuracy, t(11) ⫽ 0.31, p ⫽ .31; .25 for consensus, t(11) ⫽ 0.11, p ⫽ .91; and .28 for self– other agreement, t(11) ⫽ 1.29, p ⫽ .22; but the mean profile correlations were all higher than the random baseline correlations for the short unstructured condition: realistic accuracy, t(11) ⫽ 2.94, p ⫽ .01; consensus, t(11) ⫽ 3.22, p ⫽ .008; and self– other agreement, t(11) ⫽ 2.76, p ⫽ .02; and for the long unstructured condition: realistic accuracy, t(11) ⫽ 6.31, p ⬍ .0001; consensus, t(11) ⫽ 4.44, p ⫽ .001; and self– other agreement, t(11) ⫽ 3.53, p ⫽ .005. This pattern suggests that even though realistic accuracy, consensus, and self– other agreement can reach moderate strengths without any interaction, this strength is at or below chance levels. It is only when people are allowed to interact that differential realistic accuracy, consensus, and self– other agreement are achieved. Information quality. Recall that information quality refers to the relevance of information to personality. To test our prediction that judgments would be more accurate when participants interacted in higher quality situations, we had participants interact in three experimental conditions in which information quantity was held constant while information quality was manipulated. We again used contrast weights reflecting the prediction of a linear relationship between information quality and realistic accuracy, consensus, and self– other agreement to test the prediction based on RAM and a nonlinear relationship to test the prediction for consensus based on WAM. As predicted, we found that realistic accuracy had a good fit to the prediction of a positive linear relationship with information quality: trivia (mean r ⫽ .36, SD ⫽ 0.12); short unstructured

(mean r ⫽ .45, SD ⫽ 0.12); GTK (mean r ⫽ .49, SD ⫽ 0.16); and rcontrast ⫽ .43, tcontrast(33) ⫽ 2.71, p ⫽ .0053 (see Figure 5). The linear prediction accounted for 97% of the variance in the group means, and the variance not accounted for by the prediction was not significant, tnoncontrast(33) ⫽ 0.50, p ⫽ .31. Realistic accuracy was reliably higher in the GTK condition than in the trivia condition, t(22) ⫽ 2.76, p ⫽ .01, and the difference between the trivia and short unstructured conditions approached the conventional level of significance, t(22) ⫽ 1.94, p ⫽ .07. However, the GTK condition did not differ reliably from the short unstructured condition, t(22) ⫽ 0.88, p ⫽ .39. A good fit to the prediction of a linear increase was also found with consensus: trivia (mean r ⫽ .29, SD ⫽ 0.17); short unstructured (mean r ⫽ .41, SD ⫽ 0.20); GTK (mean r ⫽ .41, SD ⫽ 0.10); and rcontrast ⫽ .33, tcontrast(33) ⫽ 1.99, p ⫽ .028. The linear prediction accounted for 74% of the variance in the group means, and the variance not accounted for by the prediction was not significant, tnoncontrast(33) ⫽ 1.19, p ⫽ .12. Judges in the GTK condition had reliably higher consensus than judges in the trivia condition, t(22) ⫽ 2.34, p ⫽ .03; the difference between consensus in the short unstructured and trivia conditions approached significance, t(22) ⫽ 1.78, p ⫽ .09; and consensus did not differ for judges in the short unstructured and GTK conditions, t(22) ⫽ 0.04, p ⫽ .97. The consensus data were also in line with the prediction based on WAM, rcontrast ⫽ .37, tcontrast(33) ⫽ 2.31, p ⫽ .014. The nonlinear prediction accounted for 99% of the variance in the group means, and the variance not accounted for by the prediction was not significant, tnoncontrast(33) ⫽ 0.001, p ⫽ .50. Finally, a good fit to the prediction of a linear increase was found with self– other agreement: trivia (mean r ⫽ .27, SD ⫽ 0.09); short unstructured (mean r ⫽ .36, SD ⫽ 0.12); GTK (mean r ⫽ .38, SD ⫽ 0.16); rcontrast ⫽ .39, tcontrast(33) ⫽ 2.40, p ⫽ .011. The linear prediction accounted for 91% of the variance in the group means, and the variance not accounted for by the prediction was not significant, tnoncontrast ⫽ 0.75, p ⫽ .23. Judges in the short unstructured and GTK conditions had reliably higher self– other agreement than judges in the trivia condition, t(22) ⫽ 2.16, p ⫽ .04, and t(22) ⫽ 2.33, p ⫽ .03, respectively, although self– other agreement did not differ for judges in the short unstructured and GTK conditions, t(22) ⫽ 0.49, p ⫽ .63. Note that again, the mean profile correlations for the trivia condition were approximately equal to or smaller than the random baseline correlations of .36 for realistic accuracy, t(11) ⫽ 0.48,

Figure 5. Mean realistic accuracy scores for experimental conditions used in the information quality analyses. Note that the random baseline profile correlation was .36. GTK ⫽ get to know condition.

INFORMATION AND PERSONALITY JUDGMENT

p ⫽ .64; .25 for consensus, t(11) ⫽ 1.09, p ⫽ .30; and .28 for self– other agreement, t(11) ⫽ 0.03, p ⫽ .98, but the mean profile correlations were all higher than the random baseline correlations for the short unstructured condition: realistic accuracy, t(11) ⫽ 2.94, p ⫽ .01; consensus, t(11) ⫽ 3.22, p ⫽ .008; and self– other agreement, t(11) ⫽ 2.76, p ⫽ .02; and for the GTK condition: realistic accuracy, t(11) ⫽ 3.84, p ⫽ .003; consensus, t(11) ⫽ 6.45, p ⬍ .0001; and self– other agreement, t(11) ⫽ 2.73, p ⫽ .02. This pattern suggests that not just any context of interaction or observation will afford differential judgment but that differential realistic accuracy, consensus, and self– other agreement are only achieved in the conditions that allow for the exchange of moderate- to high-quality information.

Relation Between Information Acquisition and Accuracy We could also examine the degree to which information acquisition is related to realistic accuracy, self– other agreement, and consensus across all groups. RAM predicts that people who acquire more information that is relevant to personality will make more accurate judgments of personality, and it is reasonable to assume that participants who acquired more of the sample of information asked about on the information fact sheet also acquired more personality-relevant information, broadly speaking. If this is true, then there should be a positive relationship between scores on the information fact sheet and realistic accuracy, self– other agreement, and consensus, regardless of the type of experimental condition in which the participants interacted.12 This prediction can be tested by correlating the average information fact sheet score of each group with the average realistic accuracy, consensus, and self– other agreement scores of each group across all five experimental conditions. As predicted, moderate positive relationships were found between information acquisition and realistic accuracy (r ⫽ .31, p ⫽ .02), self– other agreement (r ⫽ .30, p ⫽ .02), and consensus (r ⫽ .30, p ⫽ .02), indicating that regardless of information quantity or quality, groups that acquired more of the objective information that was sampled with the information fact sheet also judged their partners with higher average levels of accuracy, consensus, and self– other agreement.

Discussion The preceding analyses provide tests of the relationships between two aspects of good information and the accuracy of personality judgment. Experimentally manipulated information quantity, or length of acquaintance, was positively related to realistic accuracy, consensus, and self– other agreement. Additionally, information quality was manipulated while information quantity was held constant, and support was found for the prediction that realistic accuracy, consensus, and self– other agreement are higher in situations in which personality-relevant information is more likely to be available.

Information Quantity RAM predicts that accuracy will increase in relation to information quantity (Funder, 1995), but previous research has only tested this prediction using self– other agreement, which is a limited definition of accuracy. The current research confirms past

121

findings that self– other agreement increases with acquaintance (Bernieri et al., 1994; Blackman & Funder, 1998; Funder et al., 1995; Paulhus & Bruce, 1992; Paunonen, 1989) and provides one more piece of evidence in favor of the existence of a positive relationship between level of acquaintance and self– other agreement. In addition to confirming past findings, the current research takes an additional step in the examination of the acquaintanceship effect by using a broad-based accuracy criterion based on the ratings of several people who knew the target well. Such a criterion allows for a more precise test of the acquaintanceship effect than was possible in past work. In line with our prediction, realistic accuracy was positively related to information quantity. Both RAM and WAM can be used to make predictions concerning consensus. RAM predicts that realistic accuracy will increase across all levels of acquaintance, and because people who are highly accurate about something real must also be in agreement, RAM predicts that consensus will also increase with acquaintance (Funder, 1995). WAM is specifically aimed at explaining consensus, and our prediction was that consensus would increase at low levels of acquaintance and then stay about the same even as acquaintance continued to increase (Kenny, 1991). These predictions are fairly similar, and we found support for both predictions. It is not possible to determine if one prediction is superior to the other, for a couple of reasons. First, the predictions themselves are highly similar, and so any data should fit both predictions to about the same degree. Second, it was not possible to determine whether the experimentally manipulated difference between the low- and medium-information quantity conditions was equal to the difference between the medium and high conditions and, therefore, whether a linear or nonlinear prediction was more appropriate. However, it is obvious from the means of the conditions that the larger increase was between the low and medium groups, but it is not possible to determine whether the findings support the prediction based on WAM better than the prediction based on RAM. Overall, these findings support the existence of the acquaintanceship effect and the role of information quantity in accurate judgment. This interpretation was bolstered by the additional finding that participants learned more facts about each other in the longer interactions and therefore had a larger quantity of information available to them when making judgments of personality.

Information Quality To our knowledge, the current article provides the first direct evidence of the positive relationships between information quality and realistic accuracy, consensus, and self– other agreement. Information quality was manipulated by varying situational strength and instructions that provided an objective for the interaction. For 12

Some astute readers may wonder why we did not perform a mediational analysis with amount of information acquired as a mediator of the relations of information quantity and quality with accuracy. We did not compute these analyses because the information fact sheet was designed to be a manipulation check and is only a sample of some relatively objective information that is likely to become available during an initial interaction, and it is not a comprehensive measure of the amount of personalityrelevant information acquired.

122

LETZRING, WELLS, AND FUNDER

the low-quality interaction, participants interacted in a strong situation with specific instructions to complete a packet of trivia-type questions. We expected participants in this condition to exhibit little behavioral variation and to not share information about thoughts and feelings, and therefore only limited amounts of personality-relevant information would become available. For the medium- and high-quality interactions, participants interacted in relatively unstructured situations that allowed for a good deal of behavioral variation and larger amounts of personality-relevant information to become available. The difference in these two conditions was the instructions informing the participants about the objective of the interaction. Participants in the medium-quality interaction were not provided with an objective, whereas participants in the high-quality interaction were given the specific objective of getting to know each other as well as possible. Participants in this latter condition were expected to reveal the most personality-relevant information and to pay the best attention to the interaction partners in an attempt to get to know them well, which should have increased accuracy by affecting the availability and detection stages of RAM. Results confirmed these expectations, with realistic accuracy, consensus, and self– other agreement all increasing, along with factual knowledge, across the three conditions of information quality. However, the present data do not allow us to determine whether these effects should be attributed to what cognitive psychologists would call the encoding stage or the decoding stage. It may be that more relevant information was available to be detected (encoding), that the detected information was remembered and utilized better (decoding), or both. As noted in the Results section, even though the data showed a good fit to a linear prediction, most of the increase in the indicators of accuracy was between the low- and medium-quality conditions, and the medium- and high-quality conditions were quite similar. One explanation for this finding is that people naturally try to get to know each other in an initial interaction, so even though we explicitly instructed participants in the high-quality condition to get to know each other and simply told participants in the mediumquality condition to talk about whatever they liked, they may have had similar interactions in which they learned approximately the same amounts of personality-relevant information that could be used when making judgments of personality. If this is the case, then information quality may not have been very different between our medium- and high-quality conditions, which would explain why realistic accuracy, consensus, and self– other agreement were also not very different in these conditions.

Limitations and Future Directions As in any study, our experimental design has limitations that may affect interpretation. First, the construct of realistic accuracy is a philosophical ideal, and any measurement of it must necessarily be an approximation. We attempted to approach realistic accuracy by having several knowledgeable people, including expert judges, describe the personalities of the target participants. This procedure was not likely to tell us exactly what the target is like, but it was likely to bring us closer to reality than would ratings provided by any single informant, even the self. In future research, we would hope to see criteria for realistic accuracy expanded even further to include direct behavioral measurements, important life outcomes, and perhaps even biological markers.

The present results involving realistic accuracy, consensus, and self– other agreement were in general quite similar to each other. This finding could be interpreted as evidence that realistic accuracy cannot be meaningfully distinguished from the simpler and more traditional operationalizations or that constructing multifaceted criteria is not worth the time and effort. We caution against both conclusions. First, realistic accuracy is a hypothetical construct that is in principle quite different from any specific criterion for or measurement of it, and on theoretical grounds it is important to be clear about the distinction. Second, it is easy to imagine circumstances in which self– other agreement, consensus, and realistic accuracy would diverge (such as when an absentminded job applicant tries to appear conscientious), and at our present state of knowledge it is wise to be alert to this possibility and to develop tools to detect such differentiation between constructs when it happens. Even though the contrast analyses suggest that there are moderate relationships between information quantity and quality and indicators of accuracy, this effect was largely driven by differences between the low conditions (minimal information and trivia) and the medium and high conditions (short unstructured, GTK, and long unstructured), with the medium and high conditions being only slightly different, albeit consistently in the predicted direction. The pattern of findings would have been even more convincing had there been a larger difference in realistic accuracy between the medium- and high-level conditions. We propose that the present evidence supports the role of quantity and quality in accuracy but that our highest conditions may not have increased acquaintance enough or provided enough motivation to show a substantial increase in indicators of accuracy beyond the medium levels. Future research should experiment further with techniques for providing participants large amounts of information or information that is highly relevant to personality. Another reasonable next step would be to increase the external validity of our conclusions through a project in which interactions take place in real social contexts (vs. in the lab, as was the case in the current experiment). There are many situations in which previously unacquainted groups of people interact and get to know each other over a period of time, such as 1st-year freshmen in college dorms, participants in volunteer corps who live and/or work together (i.e., Peace Corps, Jesuit Volunteer Corps), and older adults who move into retirement communities. Information quantity could be examined in such situations by obtaining judgments about others in the judge’s living community at several stages of acquaintance. Information quality could be examined by obtaining information about the quality of interactions from both the target and the judge, perhaps by asking for reports of types of topics most often discussed and amount of time spent talking versus engaging in other activities. Such a project is likely to be informative about how information quantity and quality are related to realistic accuracy in situations and contexts outside of the laboratory, as well as to how these two aspects of information are related to each other and personality judgment. A third direction for future research is to determine the actual behavioral events during an interaction that are related to the achieved levels of realistic accuracy, self– other agreement, and consensus. Behavioral coding of the experimental interactions might increase our understanding of why information quantity and quality are related to accuracy, self– other agreement, and consensus.

INFORMATION AND PERSONALITY JUDGMENT

Conclusion The current findings support several aspects of RAM, including the implications of information for the relevance, availability, and detection stages that make accurate judgment possible, along with specific predictions of WAM concerning consensus. Our main findings are that judgments of personality are more likely to achieve higher levels of realistic accuracy, consensus, and self– other agreement when judges have interacted with targets for longer periods of time or in situations that allow for or encourage people to reveal personality-relevant information.

References Allport, G. W. (1937). The ability to judge people. In G. W. Allport, Personality: A psychological interpretation (pp. 499 –522). New York: Holt. Altman, I., & Taylor, D. A. (1973). Social penetration: The development of interpersonal relationships. New York: Holt, Rinehart & Winston. Ambady, N., Hallahan, M., & Rosenthal, R. (1995). On judging and being judged accurately in zero-acquaintance situations. Journal of Personality and Social Psychology, 69, 518 –529. Anderson, S. M. (1984). Self-knowledge and social inference: II. The diagnosticity of cognitive/affective and behavioral data. Journal of Personality and Social Psychology, 46, 294 –307. Bem, D. J., & Funder, D. C. (1978). Predicting more of the people more of the time: Assessing the personality of situations. Psychological Review, 85, 485–501. Bernieri, F. J., Zuckerman, M., Koestner, R., & Rosenthal, R. (1994). Measuring person perception accuracy: Another look at self– other agreement. Personality and Social Psychology Bulletin, 20, 367–378. Blackman, M. C., & Funder, D. C. (1998). The effect of information on consensus and accuracy in personality judgment. Journal of Experimental Social Psychology, 34, 164 –181. Block, J. (1961). The Q-sort method in personality assessment and psychiatric research. Palo Alto, CA: Consulting Psychologists Press. Borkenau, P., & Liebler, A. (1993). Consensus and self– other agreement for trait inferences from minimal information. Journal of Personality, 61, 477– 496. Borkenau, P., Mauer, N., Riemann, R., Spinath, F. M., & Angleitner, A. (2004). Thin slices of behavior as cues of personality and intelligence. Journal of Personality and Social Psychology, 86, 599 – 614. Colvin, R. C. (1993). “Judgeable” people: Personality, behavior, and competing explanations. Journal of Personality and Social Psychology, 64, 861– 873. Colvin, R. C., & Funder, D. C. (1991). Predicting personality and behavior: A boundary on the acquaintanceship effect. Journal of Personality and Social Psychology, 60, 884 – 894. Craik, K. H., Ware, A. P., Kamp, J., O’Reilly, C., III, Staw, B., & Zedeck, S. (2002). Explorations of construct validity in a combined managerial and personality assessment programme. Journal of Occupational and Organizational Psychology, 75, 171–193. Cronbach, L. J. (1955). Processes affecting scores on “understanding of others” and “assumed similarity.” Psychological Bulletin, 52, 177–193. Funder, D. C. (1980). On seeing ourselves as others see us: Self– other agreement and discrepancy in personality ratings. Journal of Personality, 48, 473– 493. Funder, D. C. (1987). Errors and mistakes: Evaluating the accuracy of social judgment. Psychological Bulletin, 101, 75–90. Funder, D. C. (1995). On the accuracy of personality judgment: A realistic approach. Psychological Review, 102, 652– 670. Funder, D. C. (1999). Personality judgment: A realistic approach to person perception. San Diego, CA: Academic Press.

123

Funder, D. C., & Colvin, C. R. (1988). Friends and strangers: Acquaintanceship, agreement, and the accuracy of personality judgment. Journal of Personality and Social Psychology, 55, 149 –158. Funder, D. C., & Dobroth, K. M. (1987). Differences between traits: Properties associated with interjudge agreement. Journal of Personality and Social Psychology, 52, 409 – 418. Funder, D. C., Kolar, D. C., & Blackman, M. C. (1995). Agreement among judges of personality: Interpersonal relations, similarity, and acquaintanceship. Journal of Personality and Social Psychology, 69, 656 – 672. Hofstee, W. K. B. (1994). Who should own the definition of personality? European Journal of Personality, 8, 149 –162. 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, 41, 521–551. Kenny, D. A. (1991). A general model of consensus and accuracy in interpersonal perception. Psychological Review, 98, 155–163. Kenny, D. A. (1994). Interpersonal perception. New York: Guilford Press. Kenny, D. A., Albright, L., Malloy, T. E., & Kashy, D. A. (1994). Consensus in interpersonal perception: Acquaintance and the Big Five. Psychological Bulletin, 116, 245–258. Kolar, D. W. (1995). Individual differences in the ability to accurately judge the personality characteristics of others. Unpublished doctoral dissertation, University of California, Riverside. Kolar, D. W., Funder, D. C., & Colvin, C. R. (1996). Comparing the accuracy of personality judgments by the self and knowledgeable others. Journal of Personality, 64, 311–337. Letzring, T. D., Block, J., & Funder, D. C. (2005). Ego-control and ego-resiliency: Generalization of self-report scales based on personality descriptions from acquaintances, clinicians, and the self. Journal of Research in Personality, 39, 395– 422. Norman, W. T., & Goldberg, L. R. (1966). Raters, ratees, and randomness in personality structure. Journal of Personality and Social Psychology, 4, 681– 691. Park, B., & Judd, J. M. (1989). Agreement on initial impressions: Differences due to perceivers, trait dimensions, and target behaviors. Journal of Personality and Social Psychology, 56, 493–505. Park, B., Kraus, S., & Ryan, C. S. (1997). Longitudinal changes in consensus as a function of acquaintance and agreement in liking. Journal of Personality and Social Psychology, 72, 604 – 616. Paulhus, D. L., & Bruce, M. N. (1992). The effect of acquaintanceship on the validity of personality impressions: A longitudinal study. Journal of Personality and Social Psychology, 63, 816 – 824. Paunonen, S. V. (1989). Consensus in personality judgment: Moderating effects of target–rater acquaintanceship and behavior observability. Journal of Personality and Social Psychology, 56, 823– 833. Rosenthal, R., Rosnow, R. L., & Rubin, D. B. (2000). Contrasts and effect sizes in behavioral research: A correlational approach. New York: Cambridge University Press. Snyder, M., & Ickes, W. (1985). Personality and social behavior. In G. Lindzey & E. Aronson (Eds.), Handbook of social psychology (3rd ed., Vol. 2, pp. 883–947). New York: Random House. Taft, R. (1955). The ability to judge people. Psychological Bulletin, 52, 1–23. Vernon, P. E. (1933). Some characteristics of the good judge of personality. Journal of Social Psychology, 4, 42–57. Vogt, D. S., & Colvin, C. R. (2003). Interpersonal orientation and the accuracy of personality judgments. Journal of Personality, 71, 267–295.

Received August 26, 2004 Revision received July 22, 2005 Accepted December 14, 2005 䡲

Journal of Personality and Social Psychology 2006, Vol. 91, No. 1, 124 –142

Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.124

Supplication and Appeasement in Conflict and Negotiation: The Interpersonal Effects of Disappointment, Worry, Guilt, and Regret Gerben A. Van Kleef and Carsten K. W. De Dreu

Antony S. R. Manstead

University of Amsterdam

Cardiff University

This study examined the social effects of emotions related to supplication and appeasement in conflict and negotiation. In a computer-simulated negotiation, participants in Experiment 1 were confronted with a disappointed or worried opponent (supplication), with a guilty or regretful opponent (appeasement), or with a nonemotional opponent (control). Compared with controls, participants conceded more when the other experienced supplication emotions and conceded less when the other experienced appeasement emotions (especially guilt). Experiment 2 replicated the effects of disappointment and guilt and showed that they are moderated by the perceiver’s dispositional trust: Negotiators high in trust conceded more to a disappointed counterpart than to a happy one, but those with low trust were unaffected. In Experiment 3, trust was manipulated through information about the other’s personality (cooperative vs. competitive), and a similar moderation was obtained. Keywords: conflict, negotiation, emotion, supplication, appeasement

duration (Barry, 1999; Oatley & Jenkins, 1996), and intentional— directed at an object, person, or event (Frijda, 1993). In this article, we use the term emotion in the sense intended above, whereas affect is used as a superordinate construct that encompasses both moods and emotions (Barry & Oliver, 1996). Research on emotion in conflict and negotiation can be roughly divided into two categories: studies of intrapersonal effects and studies of interpersonal effects. In the 1980s and 1990s, researchers focused mostly on the intrapersonal effects of moods and emotions, investigating the influence of a negotiator’s emotional state on his or her own cognitions and behavior. For example, positive affect has been shown to increase concession making (Baron, 1990), stimulate creative problem solving (Isen, Daubman, & Nowicki, 1987), increase joint gains (Allred, Mallozzi, Matsui, & Raia, 1997; Carnevale & Isen, 1986), increase preferences for cooperation (Baron, Fortin, Frei, Hauver, & Shack, 1990), reduce the use of contentious tactics (Carnevale & Isen, 1986), and increase the use of cooperative negotiation strategies (Forgas, 1998). Conversely, negative affect has been shown to decrease initial offers (Baron et al., 1990), decrease joint gains (Allred et al., 1997), promote the rejection of ultimatum offers (Pillutla & Murnighan, 1996), increase the use of competitive strategies (Forgas, 1998), and decrease the desire to work together in the future (Allred et al., 1997). Recently, scientific interest in the role of affect in conflict and negotiation has shifted away from the intrapersonal effects of moods and emotions. Recognizing that negotiation is a social phenomenon—negotiators’ emotions influence not only themselves, but also their counterparts—several scholars have emphasized the importance of the interpersonal or social effects of emotions in negotiations (Adler, Rosen, & Silverstein, 1998; Barry, Fulmer, & Van Kleef, 2004; Davidson & Greenhalgh, 1999; Morris & Keltner, 2000; Thompson, Medvec, Seiden, & Kopelman, 2001; Van Kleef, De Dreu, & Manstead, 2004a, b). The basic premise is that emotions have important social functions and

Social interactions can produce conflict at all levels of society. One of the most common and constructive ways of resolving such conflicts and conducting social and economic exchange is by means of negotiation. Negotiation can be defined as a discussion between two or more parties aimed at resolving a perceived divergence of interests (Pruitt & Carnevale, 1993). People may negotiate with a car dealer when buying a new car, work groups may negotiate the allocation of organizational resources, and parents may negotiate with their children about how to spend the holidays. Emotions are inherent to negotiation and social conflict (Davidson & Greenhalgh, 1999) and are crucial to understanding how individuals behave within bargaining situations (Barry, 1999). So far, empirical research on emotion in conflict and negotiation has focused almost exclusively on the effects of anger and happiness. In this article, we focus on the social effects of emotions related to supplication (e.g., disappointment, worry) and appeasement (e.g., guilt, regret) in negotiation, examining the ways in which negotiators respond to their opponent’s emotions.

Emotions in Conflict and Negotiation There are multiple definitions of emotion, most of which point to three distinct features of emotion: physiological reactions, action tendencies, and subjective experience (Lazarus, 1991). Emotions differ from moods in that they are discrete (Russell & Feldman Barrett, 1999), of relatively high intensity and short

Gerben A. Van Kleef and Carsten K. W. De Dreu, Department of Psychology, University of Amsterdam, Amsterdam, the Netherlands; Antony S. R. Manstead, School of Psychology, Cardiff University, Cardiff, United Kingdom. Correspondence concerning this article should be addressed to Gerben A. Van Kleef, Department of Psychology, University of Amsterdam, Roetersstraat 15, 1018 WB Amsterdam, the Netherlands. E-mail: [email protected] 124

SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS

consequences (Frijda & Mesquita, 1994; Keltner & Haidt, 1999; Oatley & Jenkins, 1992; Parkinson, 1996). Most notable is that emotions convey information (Carver & Scheier, 1990), for instance about how one feels about things (Ekman, 1993), about one’s social intentions (Fridlund, 1994), and about one’s orientation toward other people (Knutson, 1996). In this way, one’s emotions may influence not only one’s own behavior, but also the behavior of others (Levenson, 1994). In line with this social functions perspective, recent research has demonstrated the pervasive interpersonal effects of anger and happiness in negotiations. In a computer-mediated negotiation task with a simulated opponent, Van Kleef et al. (2004a) provided participants with information about the opponent’s emotional state (angry, happy, or no emotion) at three time points during the negotiation. They found that participants with an angry opponent placed lower demands and made larger concessions than did participants with a nonemotional opponent, whereas participants with a happy opponent placed higher demands and made smaller concessions. Sinaceur and Tiedens (in press) examined the effects of anger and happiness using a different paradigm and obtained similar results. In face-to-face negotiations, they instructed one negotiator in each dyad to show either anger or happiness. In keeping with the results obtained by Van Kleef et al. (2004a), Sinaceur and Tiedens found that participants conceded more to an angry than to a happy counterpart. So results from research using different procedures (i.e., computer mediated and face to face) point to the social impact of anger and happiness on negotiation behavior. Given the pervasive interpersonal effects of anger and happiness on negotiation behavior, it is worth considering whether other emotions have the potential to affect negotiation behavior. According to the social functions perspective, emotions convey information that is likely to influence other people’s behavior. For example, Van Kleef et al. (2004a) demonstrated that negotiators concede more to angry counterparts than to happy ones because anger signals high limits, whereas happiness signals low limits. Obviously, emotions can also convey other important information, the strategic implications and interpretation of which are likely to depend in part on observers’ appraisals of the cause of the emotion. For instance, in a negotiation the distribution of resources can be perceived as fair or unfair (Bazerman, Curhan, Moore, & Valley, 2000; Hegtvedt & Killian, 1999). Research on distributive justice has shown that fair distributions give rise to positive emotions, whereas unfair distributions give rise to negative emotions (De Dreu, Lualhati, & McCusker, 1994; Loewenstein, Thompson, & Bazerman, 1989). Individuals who receive what they expected tend to experience happiness or satisfaction (Hegtvedt, 1990; Messick & Sentis, 1979; Sprecher, 1992). By contrast, individuals who feel that they are getting too much or too little are likely to experience negative emotions. Emotional reactions that are likely to occur in individuals who feel underrewarded include disappointment, sadness, depression, anger, and resentment (Hegtvedt & Killian, 1999), whereas those who feel overrewarded can be expected to experience guilt (Hegtvedt & Killian, 1999; Homans, 1974) or regret (van Dijk & Zeelenberg, 2002; Zeelenberg, van der Pligt, & Manstead, 1998). The purpose of the present research was to investigate the interpersonal effects of emotions that may arise as a result of the appraisal that one has taken too much or received too little.

125

Specifically, we focused on the interpersonal effects of disappointment, worry, guilt, and regret on demands and concessions in negotiations. Below we review research pertaining to these emotions, and we build on this research to advance a number of hypotheses regarding the interpersonal effects of these emotions on negotiation behavior.

When One Feels One Has Taken Too Much: Appeasement Emotions such as guilt, shame, embarrassment, and interpersonal regret serve an appeasement function (Baumeister, Stillwell, & Heatherton, 1994; Keltner & Buswell, 1997). Of these emotions, guilt is perhaps the most extensively researched. It entails a feeling of self-reproach resulting from the belief that one has done something wrong (R. H. Smith, Webster, Parrott, & Eyre, 2002). Baumeister et al. (1994) noted that “from an interpersonal perspective, the prototypical cause of guilt would be the infliction of harm, loss, or distress on a relationship partner” (p. 245). The experience of guilt is typically rooted in an interpersonal context and often (although not necessarily) arises as a result of a perceived transgression, in particular when the individual feels that he or she has violated some expectation or norm (Leith & Baumeister, 1998; H. B. Lewis, 1971; Tangney, 1990, 1995, 1999). Guilt is closely linked with reactions such as regret, selfreproach, repentance, and remorse (R. H. Smith et al., 2002) and tends to produce outwardly focused behaviors aimed at reducing the damage caused by one’s behavior (Barrett, 1995; Tangney, 1995). Guilt is associated with perspective-taking, interpersonal sensitivity, and improved relationship outcomes (Baumeister et al., 1994; Leith & Baumeister, 1998). People experiencing guilt tend to engage in behaviors aimed at repairing the social relationship (Baumeister et al., 1994; M. Lewis, 2000). Research has shown that transgressions and concomitant guilt increase subsequent helping, compliance, and cooperation on the part of the transgressor (Carlsmith & Gross, 1969; Freedman, Wallington, & Bless, 1967; Ketelaar & Au, 2003). Other research has found that guilt motivates people to apologize and to make reparations or amends (Friedman, 1985; Hoffman, 1982; H. B. Lewis, 1971). Furthermore, when the transgression has an interpersonal character, guilt motivates people to compensate the victim (Berscheid & Walster, 1967; Wallace & Sadalla, 1966). Thus, if the transgressor displays guilt, the victim may see this as an implicit commitment to rectify the transgression by making amends and as a promise of better treatment in the future (Baumeister et al., 1994; Manstead, 1991). In an interpersonal context, social transgressions can also cause feelings of regret (Zeelenberg et al., 1998). When the regret is interpersonal in nature—that is, when one regrets a behavior that has inflicted harm on another person (rather than on oneself)—it shares a number of characteristics with guilt (Berndsen, van der Pligt, Doosje, & Manstead, 2004; Roseman, Wiest, & Swartz, 1994). Because regret is an aversive state, people are motivated to avoid it and, once they experience it, to take action to undo it (Zeelenberg & Beattie, 1997; Zeelenberg, van Dijk, Manstead, & van der Pligt, 2000). Gilovich and Medvec (1994, 1995) refer to this undoing as “behavioral repair work” or “ameliorative behavior.” In the case of interpersonal regret, this repair work typically takes the form of apologizing to the person who has been affected by the transgression (Steiner, 2000; Zeelenberg et al., 1998).

126

VAN KLEEF, DE DREU, AND MANSTEAD

By extrapolating these findings to the domain of conflict and negotiation, we can advance the following general predictions. First, a negotiator who deals with an opponent who appears to experience emotions of appeasement (e.g., guilt or regret) is likely to expect the opponent to make up for his or her previous “transgression” (e.g., tough demands) by making a concession. Following this line of reasoning, a negotiator faced with an opponent who seems to be guilty or regretful can be expected to stand firm and wait for the other to make a concession. Second, emotions of appeasement are associated with interpersonal sensitivity and the willingness to appreciate another person’s perspective, and they share a number of relationship-repairing qualities (Baumeister et al., 1994; Leith & Baumeister, 1998; Steiner, 2000). Guilt and regret can therefore be expected to signal an outward focus and a concern for the other and to have a beneficial effect on the interpersonal relationship.

When One Feels One Has Received Too Little: Supplication Emotions such as sadness, disappointment, fear, and worry serve a supplication function (M. S. Clark, Pataki, & Carver, 1996). These emotions communicate dependency and a need for support (Eisenberg, 2000; Kennedy-Moore & Watson, 2001), and they evoke empathy and helping behavior (M. S. Clark et al., 1996; Eisenberg, 2000; Hill, Weary, & Williams, 1986; Parrott, 1993). Sadness, for example, communicates to the self and to others that one is in need of help (Tomkins, 1963). Expressions of sadness have been demonstrated to increase perceptions of neediness and dependency (M. A. Clark & Taraban, 1991) and to evoke helping behavior in both children (Barnett, Howard, Melton, & Dino, 1982) and adults (M. S. Clark, Ouellette, Powell, & Milberg, 1987; Yee & Greenberg, 1998). In a similar vein, crying serves a helpseeking function (Labott, Martin, Eason, & Berkey, 1991). For example, Cornelius (1984) demonstrated that (involuntary) crying was an effective means of eliciting a positive and desired change in other people’s behavior, in this case a shift from conflict to support. Similar effects on helping behavior have been observed for expressions of worry and fear. Like sadness, worry and fear communicate a need for assistance, and they elicit sympathetic and supportive responses in others (Eisenberg, 2000; Kennedy-Moore & Watson, 2001). For example, a study of reactions to crime victims by Yee and Greenberg (1998) revealed that fear on the part of the victims influences observers’ appraisals of need and increases the inclination to help, especially if the observer and the victim are in a communal rather than an exchange relationship. By the same token, employees who display fear are likely to evoke helpful, supportive responses from coworkers (Coˆte´, 2005). In contrast to the interpersonal effects of sadness and, to a lesser degree, fear and worry, the interpersonal effects of disappointment have not received much research attention. Carver and Scheier (1990) suggested that disappointment signals that progress toward a goal is below expectations. In other words, disappointment arises when a desired outcome is not achieved (Bell, 1985; Frijda, 1986; van Dijk & van der Pligt, 1997), and as such, it is highly relevant to negotiation. Research on the effects of disappointment is sparse, and it has focused predominantly on intrapersonal consequences (e.g., engaging in behaviors aimed at minimizing future disap-

pointment; Bell, 1985; Loomes & Sugden, 1986; see Zeelenberg et al., 2000, for an overview). Timmers, Fischer, and Manstead (1998) reported evidence suggesting that under particular circumstances, people may deliberately express disappointment to change the behavior of a target person in a direction that would be beneficial for the expresser. We argue that, at the interpersonal level, disappointment is likely to have effects similar to those of other distress-related emotions (e.g., fear, worry, sadness), which have generally been shown to facilitate prosocial behavior aimed at easing the other’s pain (Barnett, King, & Howard, 1979; Batson, 1987; Eisenberg, Fabes, Miller, et al., 1989; Fabes, Eisenberg, Karbon, Troyer, & Switzer, 1994; Morris & Keltner, 2000). On the basis of the above considerations, it can be predicted that a negotiator who is confronted with an opponent who appears to experience emotions of supplication (e.g., disappointment or worry) will try to relieve the other’s pain by making concessions. Further, emotions of distress are associated with self-focus and egoistic motivations (Eisenberg, Fabes, Schaller, & Miller, 1989). By reverse analogy with previous research suggesting that appeasement emotions enhance interpersonal relationships because they signal interpersonal sensitivity and concern (Baumeister et al., 1994; Leith & Baumeister, 1998), we argue that supplication emotions lead to less positive impressions because they signal an inward focus and preoccupation with the self.

Experiment 1 The purpose of Experiment 1 was to investigate the interpersonal effects of emotions related to supplication and appeasement in negotiations. We predicted that participants with an opponent who experiences disappointment or worry (i.e., emotions of supplication) would make smaller demands than would participants with a nonemotional opponent (Hypothesis 1a), whereas negotiators with an opponent who experiences guilt or regret (i.e., emotions of appeasement) would make larger demands (Hypothesis 1b). We also expected that participants with a guilty or regretful opponent would perceive the opponent as more interpersonally sensitive than would participants with a disappointed or worried opponent (Hypothesis 2a) and that they would develop a more favorable impression of the opponent (Hypothesis 2b). Finally, we explored whether the effects of the opponent’s emotion on participants’ impressions of the opponent are mediated by their perception of the opponent’s interpersonal sensitivity (Hypothesis 2c).

Method Participants and Experimental Design A total of 84 male and female undergraduate students at the University of Amsterdam participated either in partial fulfillment of a course requirement or for monetary compensation (EUR 7, approximately US $8). The experimental design included the opponent’s emotion (disappointment vs. worry vs. guilt vs. regret vs. no emotion) as the independent variable and demand level as the main dependent variable. Participants were randomly assigned to the conditions, and the experimenters were blind to this assignment.

Procedure For each session, 6 to 8 participants were invited to the laboratory. On arrival, participants were welcomed to the experiment and seated in sep-

SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS arate cubicles in front of a computer. From that point onward all instructions, questionnaires, and experimental tasks were presented on the computer screen. To facilitate the manipulation of the opponent’s emotion (see below), we led participants to believe that the purpose of the study was to find out how knowledge about one’s opponent’s intentions affects negotiation processes in a situation in which the negotiating parties cannot see each other. They were then told that they would engage in a computermediated negotiation with another participant (whose behavior was in fact simulated by the computer). Negotiation task. The negotiation task was one used previously by Van Kleef et al. (2004a, b; see also De Dreu & Van Lange, 1995; Hilty & Carnevale, 1993). The task captures the main characteristics of real-life negotiation (i.e., multiple issues differing in utility to the negotiator, information about one’s own payoffs only, and the typical offer– counteroffer sequence). In the current version, participants learned that they would be assigned the role of buyer or seller of a consignment of mobile phones and that their objective was to negotiate the price, the warranty period, and the duration of the service contract of the phones. They were then presented with a payoff chart (see Appendix A) that showed which outcomes were most favorable to them, and they learned that their objective was to earn as many points as possible. Level 9 on price ($110) yielded 0 points, and Level 1 ($150) yielded 400 points (i.e., increments of 50 points per level). For warranty period, Level 9 (9 months) yielded 0 points, and Level 1 (1 month) yielded 120 points (i.e., increments of 15 points per level). Finally, for duration of service contract, Level 9 (9 months) yielded 0 points, and Level 1 (1 month) yielded 240 points (i.e., increments of 30 points per level). Participants were told, “You can see that the best deal for you is 1–1–1, for a total outcome of 760 points (400 ⫹ 120 ⫹ 240).” The corresponding payoff table for the other party was not displayed, and participants were told only that it differed from their own. To enhance involvement in the negotiation task, we informed participants that points would be converted to lottery tickets at the end of the experiment and that the more points earned, the more lottery tickets one would obtain and the greater would be one’s chance of winning a EUR 50 (approximately US $64) prize. To emphasize the mixed-motive nature of the negotiation, we told participants that only those who reached an agreement would participate in the lottery. Thus, there were incentives both to earn as many points as possible and to reach an agreement. After a short pause during which the computer supposedly assigned buyer and seller roles to the participants, all participants were assigned the role of seller. They were told that the buyer (i.e., the opponent) would make the first offer and that the negotiation would continue until an agreement was reached or time ran out. Just before the negotiation started, participants learned that an additional goal of the study was to examine the effects of having versus not having information about the opposing negotiator’s intentions. They read that the computer had randomly determined that they would receive information about the intentions of the opponent and that the opponent would not receive information about their intentions. After these instructions, the negotiation started, and the buyer (i.e., the computer) made a first offer. Over the negotiation rounds the buyer proposed the following levels of agreement (for price–warranty–service): 8 –7– 8 (Round 1), 8 –7–7 (Round 2), 8 – 6 –7 (Round 3), 7– 6 –7 (Round 4), 7– 6 – 6 (Round 5), and 6 – 6 – 6 (Round 6). Past research has shown that this preprogrammed strategy has face validity and is seen as intermediate in cooperativeness and competitiveness (De Dreu & Van Lange, 1995). A demand by the participant was accepted if it equaled or exceeded the offer the computer was about to make in the next round. If no agreement was reached by the sixth round, the negotiation was interrupted (see De Dreu & Van Lange, 1995). Following the procedure of a study by Tripp and Sondak (1992), we excluded from the sample participants who reached agreement before Round 6 (n ⫽ 8) to allow for repeated-measures analyses. (However, retaining these participants yielded a similar pattern of results.)

127

Manipulation of the opponent’s emotion. We chose to manipulate the opponent’s emotion in the context of a computer-mediated negotiation in which parties could not see each other and communicated via computers (see De Dreu & Van Kleef, 2004; De Dreu & Van Lange, 1995; Hilty & Carnevale, 1993). Participants were led to believe that the purpose of the study was to find out how knowledge about one’s opponent’s intentions affects negotiation processes and outcomes. After the first, third, and fifth negotiation rounds, participants received information about “the intentions of the buyer,” which contained the manipulation of the buyer’s emotion. Participants had to wait for about a minute and a half while the buyer was supposedly asked to reveal what he or she intended to offer in the next round, and why. After this short wait, participants received what appeared to be the buyer’s answer. This was presented in a separate box in a different font and contained some minor typing errors in order to enhance experimental realism. The buyer’s intentions were held constant across conditions and contained the buyer’s intended offer for the next round. That is, after Round 1 the buyer wrote, “I think I will offer 8 –7–7,” which would indeed be the buyer’s next offer. The buyer’s intention information also contained an emotional statement that constituted the experimental manipulation. The emotion statements were pretested in a pilot study involving 64 psychology students, none of whom participated in the main study. We tested seven statements designed to reflect disappointment, six statements designed to reflect worry, seven statements designed to reflect interpersonal regret, and seven statements designed to reflect guilt. The statements were pretested using a within-participants design. All participants rated a selection of 13 or 14 out of the total of 27 statements, the order of statements being randomized across participants. The statements were distributed in such a way that each was rated by half of the participants. For each statement, participants were asked to indicate on a 7-point scale how comprehensible they found it (1 ⫽ very incomprehensible, 7 ⫽ very comprehensible) and to what extent they felt it reflected disappointment, worry, regret, and guilt (1 ⫽ not at all, 7 ⫽ to a great extent). We then selected the statements that had the highest scores on the emotion they were supposed to reflect and the lowest scores on the emotions that they were not supposed to reflect, provided that they did not differ with respect to their comprehensibility. We selected three statements for each emotion. All selected statements were rated higher on the emotion they were supposed to express than on the emotions they were not supposed to express according to paired-samples t tests (5.09 ⬍ ts ⬍ 18.29, all ps ⬍ .01). Further, one-sample t tests showed that there was a significant effect of all statements on the rating of the corresponding emotion (13.80 ⬍ ts ⬍ 18.32, all ps ⬍ .01). Finally, paired-samples t tests revealed that the statements did not differ with respect to comprehensibility (all ts ⬍ 1.12, ns). After the first negotiation round, participants in the disappointed opponent condition received the following information: “I am pretty disappointed about this,” followed by the intention statement “I think I will offer 8 –7–7,” which was the same for all conditions. In the worried opponent condition, participants read “This worries me quite a lot”; in the guilty opponent condition, participants read “I feel guilty for not having conceded more”; in the regret condition, participants read “I am sorry that I haven’t conceded more.” In the control condition participants received only the intention statement. After the third and fifth negotiation rounds, participants in the experimental conditions again received an emotional statement and an intention, whereas those in the control condition simply received the intention. Appendix B lists all statements used in the experiment. Dependent measures. The main dependent variable was participants’ level of demand in Rounds 1 to 6. In addition, participants completed a postnegotiation questionnaire that contained manipulation checks and items designed to measure impressions of the opponent. To check the adequacy of the emotion manipulation, we asked participants to indicate on a 7-point scale how disappointed, worried, guilty, and regretful they thought their opponent had been during the negotiation. Perceptions of the

VAN KLEEF, DE DREU, AND MANSTEAD

128

opponent’s disappointment were measured by three items (e.g., “The buyer appeared to be disappointed during the negotiation,” 1 ⫽ totally disagree to 7 ⫽ totally agree), which were combined into a single index of perceived disappointment (␣ ⫽ .83). Perception of the opponent’s worry was measured using five items (e.g., “The buyer appeared to be worried during the negotiation”), which were averaged into an index of perceived worry (␣ ⫽ .88). Perception of guilt was measured by three items (e.g., “The buyer appeared to feel guilty during the negotiation”), which were combined into an index of perceived guilt (␣ ⫽ .93). Finally, perceptions of the opponent’s regret were measured by three items (e.g., “The buyer appeared to feel sorry during the negotiation”), which were averaged into a scale of perceived regret (␣ ⫽ .91). Impression of the opponent was assessed with seven items (“The buyer strikes me as a sympathetic person”; “During the negotiation, the buyer made a hostile impression,” reverse scored; “The buyer made a cooperative impression”; “The buyer made a friendly impression”; “The buyer made a competitive impression,” reverse scored; “The buyer made a stubborn impression,” reverse scored; “I have developed a positive impression of the buyer”; 1 ⫽ totally disagree to 7 ⫽ totally agree). The seven items were combined into a measure of impression of the opponent (␣ ⫽ .79). We also included three items measuring participants’ perceptions of the opponent’s interpersonal sensitivity and perspective taking (“During the negotiation, the buyer was self-centered,” reverse scored; “During the negotiation, the buyer took my interests into consideration”; “During the negotiation, the buyer appeared to be preoccupied with him- or herself,” reverse scored). These three items were averaged into an index of the opponent’s interpersonal sensitivity (␣ ⫽ .82).

Results Treatment of the Data The offers made by participants in each round were transformed into an index revealing the negotiator’s total level of demand for each negotiation round (i.e., the number of points demanded in that round, summed across the three negotiation issues of price, warranty, and service; see Appendix A).

intended emotion than for the other emotions (2.27 ⬍ ts ⬍ 11.73, all ps ⬍ .05).

Demand Level Demand level in Rounds 1 to 6 was analyzed using a mixedmodel ANOVA with the opponent’s emotion (disappointment vs. worry vs. guilt vs. regret vs. no emotion) as a between-participants variable and negotiation round (1 to 6) as a repeated-measures variable. Unsurprisingly, this analysis revealed a main effect of round, F(5, 355) ⫽ 286.28, p ⬍ .01 (partial ␩2 ⫽ .80), indicating that demands declined from Round 1 (M ⫽ 677, SD ⫽ 63) to Round 6 (M ⫽ 484, SD ⫽ 93). More important, there was a significant effect of the opponent’s emotion on average demands, F(4, 71) ⫽ 7.76, p ⬍ .01 (partial ␩2 ⫽ .30), indicating that participants’ demands were influenced by their adversaries’ emotions (disappointment: M ⫽ 533, SD ⫽ 59; worry: M ⫽ 513, SD ⫽ 76; guilt: M ⫽ 633, SD ⫽ 79; regret: M ⫽ 592, SD ⫽ 46; control: M ⫽ 569, SD ⫽ 56). Finally, the main effects of the opponent’s emotion and negotiation round were qualified by a significant two-way interaction, F(20, 355) ⫽ 5.21, p ⬍ .01 (partial ␩2 ⫽ .23). As can be seen in Figure 1, the influence of the opponent’s emotion became more apparent as the negotiation progressed. In Round 1, there were no differences among any of the conditions. This is hardly surprising, given that the emotion manipulation began only after this round. From the second round onward, however, the different conditions started to diverge, the effect becoming stronger after each consecutive round. Therefore, we decided to use demands in Round 6 for our specific hypothesis tests. Hypotheses 1a and 1b were tested with planned comparisons. The means and standard deviations as well as the specific contrasts

Manipulation Check A 5 (opponent’s emotion: disappointment, worry, guilt, regret, no emotion) ⫻ 4 (perception of the opponent’s emotion: disappointment, worry, guilt, regret) analysis of variance (ANOVA) with repeated measures on the last factor revealed a significant interaction between the opponent’s emotion and participants’ perceptions of the opponent’s emotion, F(12, 213) ⫽ 31.77, p ⬍ .01 (partial ␩2 ⫽ .59). Post hoc tests showed that participants in the disappointed opponent condition rated the opponent as more disappointed (M ⫽ 5.96, SD ⫽ 1.25) than did participants in all the other conditions (2.84 ⬍ Ms ⬍ 4.95, 0.96 ⬍ SDs ⬍ 1.29). Similarly, participants with a worried opponent rated the opponent as more worried (M ⫽ 5.94, SD ⫽ 0.85) than did participants in the other conditions (2.54 ⬍ Ms ⬍ 4.79, 0.83 ⬍ SDs ⬍ 1.68), and participants with a guilty opponent rated the opponent as more guilty (M ⫽ 5.85, SD ⫽ 1.39) than did those in the other conditions (1.85 ⬍ Ms ⬍ 4.39, 0.82 ⬍ SDs ⬍ 1.85). Finally, participants with a regretful opponent rated the other as more regretful (M ⫽ 5.03, SD ⫽ 1.60) than did participants in all the other conditions (1.98 ⬍ Ms ⬍ 2.77, 0.68 ⬍ SDs ⬍ 1.04) except the guilty opponent condition (M ⫽ 4.81, SD ⫽ 1.72). Furthermore, paired-samples t tests showed that ratings within each condition were higher for the

Figure 1. Demand level as a function of the opponent’s emotion and negotiation round in Experiment 1.

SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS

that were computed to test these hypotheses are shown in Table 1. Four contrasts were computed. First, we tested whether participants made smaller demands to opponents who experienced supplication emotions (i.e., disappointment and worry) than to nonemotional opponents (Contrast 1). Second, we examined whether participants responded differentially to disappointed versus worried counterparts in terms of their demands (Contrast 2). Third, we examined whether participants made larger demands to opponents engaging in appeasement (i.e., guilt or regret) than to nonemotional opponents (Contrast 3). Fourth, we tested whether participants responded differentially to guilty versus regretful counterparts (Contrast 4). For Hypotheses 1a and 1b to be supported, Contrasts 1 and 3 should be significant (participants should make lower demands to opponents showing signs of supplication and higher demands to opponents showing signs of appeasement). Contrasts 2 and 4 speak to the question of whether the two supplication emotions (disappointment and worry) have similar effects on demands, and whether the two appeasement emotions (guilt and regret) have similar effects on demands. The latter two contrasts were computed for exploratory purposes. In accordance with Hypothesis 1a, Contrast 1 was significant. Participants made smaller demands to a disappointed or worried opponent than to a nonemotional opponent, t(71) ⫽ 2.89, p ⬍ .01 (partial ␩2 ⫽ .15). Contrast 2 was nonsignificant, indicating that participants did not make differential demands to disappointed versus worried counterparts, t(71) ⬍ 1, ns. In support of Hypothesis 1b, Contrast 3 was significant—participants made larger demands to a guilty or regretful opponent than to a nonemotional one, t(71) ⫽ 2.10, p ⬍ .05 (partial ␩2 ⫽ .08). Unexpectedly, Contrast 4 was also significant, indicating that participants made smaller demands to regretful than to guilty counterparts, t(71) ⫽ 2.02, p ⬍ .05 (partial ␩2 ⫽ .07). These findings clearly support the prediction that negotiators concede more to opponents who experience supplication emotions than to opponents who experience appeasement emotions. Although the supplication emotions disappointment and worry had similar effects on behavior, the effects of the appeasement emotions guilt and regret differed.

Ratings of Interpersonal Sensitivity and Impression of the Opponent An ANOVA showed that participants’ ratings of the opponent’s interpersonal sensitivity were influenced by the opponent’s emotion, F(4, 71) ⫽ 5.62, p ⬍ .01 (partial ␩2 ⫽ .24). As predicted

129

(Hypothesis 2a), planned comparisons revealed that participants with a guilty or regretful opponent judged the opponent to be more interpersonally sensitive (M ⫽ 4.31, SD ⫽ 1.36 and M ⫽ 4.33, SD ⫽ 1.39, respectively) than did participants with a worried or disappointed opponent (M ⫽ 3.15, SD ⫽ 1.05 and M ⫽ 2.69, SD ⫽ 1.14, respectively), t(71) ⫽ 4.37, p ⬍ .002 (partial ␩2 ⫽ .21). Participants with a nonemotional opponent occupied an intermediate position (M ⫽ 3.86, SD ⫽ 1.15) that differed from the disappointment and worry conditions, t(71) ⫽ 2.76, p ⬍ .01 (partial ␩2 ⫽ .15), but not from the guilt and regret conditions, t(71) ⫽ 1.22, ns. We also found a significant effect of emotion on participants’ impressions of the opponent, F(4, 71) ⫽ 4.87, p ⬍ .002 (partial ␩2 ⫽ .22). Consistent with Hypothesis 2b, planned comparisons showed that participants developed more favorable impressions of opponents who experienced guilt or regret (M ⫽ 4.48, SD ⫽ 0.99 and M ⫽ 4.24, SD ⫽ 0.87, respectively) than of opponents who experienced disappointment or worry (M ⫽ 3.19, SD ⫽ 1.07 and M ⫽ 3.78, SD ⫽ 0.75, respectively), t(71) ⫽ 3.64, p ⬍ .01 (partial ␩2 ⫽ .16). Again, participants with a nonemotional opponent occupied an intermediate position (M ⫽ 4.02, SD ⫽ 0.79), differing from the disappointment and worry conditions, t(71) ⫽ 2.14, p ⬍ .04 (partial ␩2 ⫽ .09), but not from the guilt and regret conditions, t(71) ⫽ 1.28, ns. To test the idea that participants developed more favorable impressions of opponents who engaged in appeasement than of opponents who engaged in supplication because they perceived the former to be more interpersonally sensitive than the latter (Hypothesis 2c), we conducted mediated regression analyses (see Baron & Kenny, 1986). To this end, the opponent’s emotion was dummy coded (0 for worry and disappointment and 1 for guilt and regret). This dummy variable had a significant effect on participants’ impressions of the opponent (the dependent variable; ␤ ⫽ .45, p ⬍ .01) and on participants’ judgments of the opponent’s interpersonal sensitivity (the mediator; ␤ ⫽ .51, p ⬍ .01). When both variables were simultaneously entered into the regression, a significant effect of interpersonal sensitivity on impression emerged (␤ ⫽ .69, p ⬍ .01), and the originally significant effect of the dummy variable on impression was reduced to nonsignificance (␤ ⫽ .10, ns). The reduction of the direct path from the opponent’s emotion to participants’ impressions of the opponent was significant according to a Sobel test (Z ⫽ 3.74, p ⬍ .01; see Kenny, Kashy, & Bolger, 1998, and for an updated version of the formula,

Table 1 Means, Standard Deviations (in Parentheses), and Contrasts Computed to Test Hypothesis 1 (Experiment 1) Condition Demand and contrast Demand Contrast Contrast Contrast Contrast

in Round 6 1* 2 3* 4*

* p ⬍ .05.

Disappointment

Worry

Guilt

Regret

No emotion

435 (63) 1 1 0 0

422 (85) 1 ⫺1 0 0

573 (100) 0 0 1 1

527 (59) 0 0 1 ⫺1

492 (70) ⫺2 0 ⫺2 0

VAN KLEEF, DE DREU, AND MANSTEAD

130

see Kenny’s Web site at http://users.rcn.com/dakenny/mediate. htm). These findings suggest that the effect of the opponent’s emotion on impressions is indeed mediated by judgments of interpersonal sensitivity, although it should be noted that both constructs were measured at the same time.

Discussion The results of Experiment 1 are largely supportive of our predictions. In line with Hypothesis 1, participants whose opponents experienced emotions of supplication (i.e., disappointment or worry) made smaller demands than did participants whose opponents experienced emotions of appeasement (i.e., guilt or regret). Negotiators who are faced with an opponent who experiences emotions of appeasement appear to anticipate behavioral repair in the form of concessions and stand firm. Conversely, negotiators who deal with a counterpart who experiences emotions of supplication respond to the other’s discomfort by making concessions. Thus, supplication and appeasement emotions have unique and theoretically meaningful effects on negotiation behavior. Further, and consistent with Hypotheses 2a to 2c, participants whose opponents showed signs of appeasement developed a more positive impression of the opponent than did those whose opponents engaged in supplication, because the former were perceived as more interpersonally sensitive than the latter. This finding is in line with the presumed relationship-enhancing functions of expressing guilt and regret. One finding of Experiment 1 was unanticipated. On the basis of the literature reviewed in the introduction, we expected guilt and regret to have similar effects on demands. However, the present findings suggest that guilt has a stronger effect on demands. A possible explanation could be that expressions of regret are more ambiguous. Expressing regret for not having conceded more could indicate regret for hurting someone else or regret for not having been more strategic and self-interested. Expressions of guilt are more unequivocal in that they necessarily imply that one feels bad about one’s behavior vis-a`-vis someone else. Another issue concerns the emotional statements that were used to manipulate guilt and regret. In previous work (Van Kleef et al., 2004a, b), we used rather straightforward expressions to manipulate anger and happiness (e.g., “This offer makes me really angry/ happy”). Although similar expressions could easily be used to express disappointment and worry (e.g., “I am very disappointed”), we felt that expressions of guilt and regret would appear implausible unless they were accompanied by some kind of justification. We therefore decided to add a brief explanation to the first guilt and regret statements for why the opponent felt guilty or regretful (e.g., “I feel guilty for not having conceded more”). However, in addition to making the emotion statement more credible, this also inadvertently introduced an implicit intention to concede. Despite the fact that only the first of three guilt and regret statements was accompanied by such an implicit intention, we cannot be sure that guilt and regret were driving the effects. This issue is dealt with in Experiment 1b.

Experiment 1b The goal of Experiment 1b was to identify the unique interpersonal effects of guilt on demands by contrasting the guilt manip-

ulation of Experiment 1 with a “clean” manipulation that did not include a justification for why the guilt was experienced. Both guilt conditions were compared with a control condition. We expected that participants in both guilt conditions would make smaller concessions in the course of the negotiation than would participants in the control condition and that the two guilt conditions would not differ from each other. Extending the line of argument presented earlier, we further predicted that participants in both guilt conditions would expect larger concessions from the opponent than would participants in the control condition (Hypothesis 3).

Method Participants and Design Eighty-four undergraduate students at the University of Amsterdam were assigned to one of three conditions: guilt with implicit intention to concede, guilt without implicit intention to concede, or no emotion. In return for participation, they received course credit or EUR 5 (approximately $6).

Procedure The procedure was similar to the one used in Experiment 1, except for the manipulation of guilt. In the guilt with intention condition, we used the same guilt statements as in Experiment 1. Thus, after the first negotiation round, participants received the following statement from the opponent (see Appendix B): “I feel guilty for not having conceded more, I think I will offer 8 –7–7.” In the guilt without intention condition, we omitted the explanatory part. Thus, after the first round, participants simply read “I feel guilty, I think I will offer 8 –7–7.” Participants who reached agreement before Round 6 (n ⫽ 7) were excluded from the analyses (although retaining these participants yielded similar results).

Dependent Measures In addition to recording demands, we included a brief questionnaire to measure participants’ expectations of the opponent’s future concessions and to check the emotion manipulation. After Round 6, participants read that the negotiation would be temporarily interrupted for some questions and that the negotiation would resume later. Expectations regarding the opponent’s future concessions were then measured by three items (“I expect that the opponent will make large concessions in the next rounds”, “I expect that the opponent will take a cooperative stance”, “I expect that the opponent will be conciliatory”; 1 ⫽ totally disagree to 9 ⫽ totally agree; ␣ ⫽ .85). The manipulation of guilt was checked using the same scale as in Experiment 1 (␣ ⫽ .89).

Results Manipulation Check An ANOVA revealed a significant effect of the opponent’s emotion on participants’ perceptions of the opponent’s emotion, F(2, 74) ⫽ 37.99, p ⬍ .01 (partial ␩2 ⫽ .51). Planned comparisons showed that participants in the guilt with intention (M ⫽ 6.36, SD ⫽ 1.77) and guilt without intention (M ⫽ 6.26, SD ⫽ 2.01) conditions rated the other as more guilty than did participants in the no-emotion condition (M ⫽ 3.02, SD ⫽ 1.18), t(74) ⫽ 8.72, p ⬍ .01 (partial ␩2 ⫽ .51). The two guilt conditions did not differ from each other, t(74) ⫽ .21, ns.

SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS

Demand Level As in Experiment 1, an ANOVA revealed a significant main effect of round, F(5, 370) ⫽ 128.76, p ⬍ .01 (partial ␩2 ⫽ .76), indicating that demands declined from Round 1 (M ⫽ 641, SD ⫽ 72) to Round 6 (M ⫽ 489, SD ⫽ 100). More important, we found a significant effect of the opponent’s emotion on average demands, F(2, 74) ⫽ 4.53, p ⬍ .015 (partial ␩2 ⫽ .08), indicating that participants’ demands were influenced by the opponent’s emotion (guilt with intention: M ⫽ 575, SD ⫽ 78; guilt without intention: M ⫽ 574, SD ⫽ 88; control: M ⫽ 521, SD ⫽ 66). These main effects were qualified by a significant interaction between emotion and negotiation round, F(10, 370) ⫽ 4.64, p ⬍ .01 (partial ␩2 ⫽ .14). As in Experiment 1, the influence of the opponent’s emotion became more apparent as the negotiation progressed. In Round 1, there were no differences among any of the conditions. From the third round onward, the different conditions started to diverge, the effect becoming stronger after each round. Planned comparisons of demands in Round 6 revealed that both guilt conditions (guilt with intention: M ⫽ 524, SD ⫽ 88; guilt without intention: M ⫽ 536, SD ⫽ 107) differed significantly from the control condition (M ⫽ 430, SD ⫽ 70), t(74) ⫽ 4.93, p ⬍ .01 (partial ␩2 ⫽ .19), and that they did not differ significantly from each other, t(74) ⫽ .45, ns.

Expectations of Opponent’s Future Concessions An ANOVA yielded an effect of emotion on participants’ expectations of the opponent’s future concessions, F(2, 74) ⫽ 6.35, p ⬍ .01 (partial ␩2 ⫽ .14). Planned comparisons revealed that participants in both guilt conditions expected more cooperation from the opponent (guilt with intention: M ⫽ 5.45, SD ⫽ 1.19; guilt without intention: M ⫽ 5.10, SD ⫽ 1.28) than did participants in the control condition (M ⫽ 4.28, SD ⫽ 1.31), t(74) ⫽ 3.45, p ⬍ .01 (partial ␩2 ⫽ .06). Again, the guilt conditions did not differ from each other, t(74) ⫽ .82, ns.

Discussion These results show that the interpersonal effects of guilt as observed in Experiment 1 also obtain when no justification for the experienced guilt is provided or, in other words, no implicit intention to concede is mentioned. In line with our theoretical model, participants in both guilt conditions expected the opponent to make large concessions, supporting Hypothesis 3. Furthermore, participants in both guilt conditions made smaller concessions than did participants in the control condition, providing additional support for Hypothesis 1b.

Experiment 2 The results thus far support the central proposition of the present research: In a negotiation, emotions of appeasement elicit high demands from one’s adversary, whereas emotions of supplication elicit low demands. The objective of Experiment 2 was twofold. The first objective was to replicate and extend the findings of Experiment 1 by investigating the potential moderating role of interpersonal trust. Trust plays an important role in negotiations, and it is essential to the resolution of mixed-motive conflict (Lindskold, 1978). It can be defined as “a psychological state comprising the intention to accept vulnerability based upon posi-

131

tive expectations of the intentions or behaviors of another” (Rousseau, Sitkin, Burt, & Camerer, 1998, p. 395). Among other things, level of trust has been shown to affect the exchange of information regarding preferences and priorities in negotiations (Pruitt & Kimmel, 1977), the willingness to cooperate (Yamagishi, 1986), the attainment of integrative solutions (De Dreu, Giebels, & Van de Vliert, 1998; Kimmel, Pruitt, Magenau, Konar-Goldband, & Carnevale, 1980), and the desire for future interaction (Naquin & Paulson, 2003; for a review, see Ross & LaCroix, 1996). Aside from its specific effects in the negotiation arena, trust has a more generalized influence on the trustworthiness individuals ascribe to others and on their willingness to accept information as sincere and accurate (Parks, Henager, & Scamahorn, 1996; Rotter, 1980). This has important consequences for negotiators’ reactions to their opponent’s emotions. Although the effects of trust on responses to other people’s emotions have not been the explicit focus of research attention, some indirect evidence exists for the moderating role of trust. For instance, research has shown that the interpersonal consequences of expressions of distress depend on how receivers interpret such expressions (Kennedy-Moore & Watson, 2001). Sometimes recipients respond to these expressions in less than helpful ways because they misread the expresser’s intent (L. F. Clark, 1993). When observers perceive that the expression is untruthful or manipulative, they are less likely to respond positively. On the other hand, when they view the expressions as trustworthy or at least benign, they are more likely to respond with social helping (Kennedy-Moore & Watson, 2001). In other words, individuals’ reactions to others’ emotions are likely to be moderated by interpersonal trust. Variations in trust may be rooted in individual differences or stem from characteristics of the situation (Yamagishi, 1986). The present experiment is concerned with the potential moderating role of individual differences in interpersonal trust. The aforementioned research suggests that a negotiator’s decision to adapt his or her demands to the opponent’s emotions will be influenced by the focal negotiator’s level of interpersonal trust. Negotiators with high levels of trust can be expected to adapt their demands to the opponent’s emotional state, whereas those with low trust may not respond differentially to the other’s emotions. To test this idea, we manipulated the opponent’s emotion (guilt vs. disappointment) and measured participants’ level of dispositional trust. We predicted that individuals high in trust would make higher demands to a guilty opponent (i.e., an opponent engaging in appeasement) than to a disappointed opponent (i.e., an opponent engaging in supplication). By contrast, we predicted that individuals low in dispositional trust would not respond differentially to their counterpart’s emotions (Hypothesis 4). We further hypothesized that, compared with high trust participants, low trust participants would be more suspicious about the opponent’s emotion and more likely to discount the other’s emotions rather than incorporate them into their negotiation strategy (Hypothesis 5). The second objective of Experiment 2 was to shed more light on the process underlying the effects of supplication and appeasement emotions on demands. According to the social functions perspective that was outlined in the introduction, appeasement emotions such as guilt signal that behavioral repair can be anticipated, whereas supplication emotions such as disappointment signal that the other needs help. It seems reasonable to assume that such considerations would affect participants’ negotiation objectives

VAN KLEEF, DE DREU, AND MANSTEAD

132

and strategy. In this context it is useful to consider the role of the negotiator’s goal (Kelley, Beckman, & Fischer, 1967; Pruitt & Carnevale, 1993; Siegel & Fouraker, 1960; Zetik & Stuhlmacher, 2002). Among other things, higher goals tend to produce larger demands and greater resistance to concession making (Holmes, Throop, & Strickland, 1971; Kelley et al., 1967; D. L. Smith, Pruitt, & Carnevale, 1982; Yukl, 1974a, b). A negotiator who expects to receive compensation in the form of large concessions (e.g., in the case of a guilty opponent) can be expected to adopt higher goals than a negotiator who does not anticipate large concessions (e.g., in the case of a disappointed opponent). Thus, it may be that negotiators concede more to a disappointed opponent than to a guilty one because the other’s guilt raises their goals, whereas disappointment lowers them. To examine this possibility, we measured participants’ goals and hypothesized that participants with a guilty opponent would adopt higher goals than would those with a disappointed opponent (Hypothesis 6). We further predicted that negotiators’ tendency to make higher demands to guilty opponents than to disappointed ones would be mediated by the ambitiousness of their goals (Hypothesis 7). Finally, we were interested to see whether guilt and disappointment are indeed interpreted to mean that someone has received too much versus too little, as suggested by the literature reviewed in the introduction. This issue was addressed in an exploratory fashion.

Method Participants and Experimental Design Participants were 154 undergraduate students at the University of Amsterdam. They took part in the experiment for course credit or for monetary compensation (EUR 7, approximately U.S.$8). The experimental design included the opponent’s emotion (disappointment vs. guilt) and the participant’s dispositional trust (high vs. low) as the independent variables and demand level as the main dependent variable. Participants were randomly assigned to the conditions with the use of a double-blind procedure.

Procedure The procedure was identical to the one used in Experiment 1, with one major exception. In addition to manipulating the opponent’s emotion, we also measured the participant’s level of dispositional trust (see below). As in Experiment 1, participants who reached agreement before Round 6 (n ⫽ 6) were excluded from the analyses (although retaining these participants yielded a similar pattern of results). Assessment of trust. Trust was assessed with Yamagishi’s (1986) trust scale, which comprises the following five items: “Most people tell a lie if they can benefit from doing so”, “Those devoted to unselfish causes are often exploited by others”, “Many people do not cooperate because they pursue only their own interests”, “Most people are basically honest”, and “There will be more people who will not work if the social security system is developed further” (1 ⫽ totally disagree to 7 ⫽ totally agree). The internal consistency of the scale was good (␣ ⫽ .77). Following past research (Parks & Hulbert, 1995), a median split (Mdn ⫽ 4.63) was performed to classify participants as high or low in trust, yielding 35 to 40 participants per condition. Ratings on the trust scale did not differ across emotion conditions (disappointment: M ⫽ 4.68, SD ⫽ 0.61; guilt: M ⫽ 4.65, SD ⫽ 0.63), F(1, 146) ⬍ 1, ns. After the assessment of trust, participants completed a 10-min filler task (consisting of a number of unrelated scales) to reduce the likelihood of carryover effects from the trust assessment to the negotiation task. After the filler task, participants pro-

ceeded to the negotiation task, which was identical to the one used in Experiment 1. Dependent measures. In addition to recording demands, we measured participants’ goals, their interpretation of the other’s emotions, suspicion regarding the trustworthiness of the opponent and his or her emotions, and discounting of the other’s emotions. We used the same manipulation checks as in Experiment 1 (disappointment: ␣ ⫽ .85; guilt: ␣ ⫽ .88). Participants’ goals were measured with six items, two for each issue (“On which level of [price/warranty/service] do you strive to reach an agreement?” and “On which level of [price/warranty/service] do you hope to reach an agreement?”). Responses could range from 1 (indicating an extremely low goal) to 9 (indicating an extremely high goal; see Appendix A). The items were averaged into a single index of participant’s goal (␣ ⫽ .79). Interpretation of the opponent’s emotion was assessed by two semantic differentials, both of which were introduced by the question, “How do you interpret the emotions the buyer expressed during the negotiation?” The two items participants responded to were “The buyer has received too much/the buyer has received too little” and “The buyer has offered me too much/the buyer has offered me too little” (both measured on a 7-point scale). Because the correlation between the items was modest (r ⫽ ⫺.39), we analyzed them separately. Suspicion was measured with three items (“During the negotiation I was suspicious”; “I distrusted the information I received from the buyer”; “The information I received from the buyer made me suspicious”; 1 ⫽ totally disagree, 7 ⫽ totally agree), which were combined into a single index (␣ ⫽ .79). Discounting of the opponent’s emotion was also measured by three items (“During the negotiation I did not take the information about the buyer into account”; “During the negotiation, I paid serious attention to the information I received about the buyer,” reverse scored; “I ignored the information I received from the buyer”; 1 ⫽ totally disagree, 7 ⫽ totally agree), which were also averaged into a single index (␣ ⫽ .67).

Results Manipulation Check A 2 (emotion of the opponent: disappointment vs. guilt) ⫻ 2 (dispositional trust: high vs. low) ⫻ 2 (perception of the opponent’s emotion: disappointed vs. guilty) ANOVA with repeated measures on the last factor revealed a significant interaction between the opponent’s emotion and participants’ perceptions of the opponent’s emotion, F(1, 144) ⫽ 300.54, p ⬍ .01 (partial ␩2 ⫽ .67). Participants with a disappointed opponent rated the opponent as more disappointed (M ⫽ 5.25, SD ⫽ 1.42) than did those with a guilty opponent (M ⫽ 3.29, SD ⫽ 1.16). Similarly, participants with a guilty opponent rated the other as more guilty (M ⫽ 4.56, SD ⫽ 1.54) than did those with a disappointed opponent (M ⫽ 2.21, SD ⫽ 0.76). Paired-sample t tests showed that participants in the disappointed-opponent condition rated the opponent as more disappointed than guilty (M ⫽ 5.25 vs. M ⫽ 2.21), t(72) ⫽ 15.84, p ⬍ .01, and that participants in the guilty-opponent condition rated the opponent as more guilty than disappointed (M ⫽ 4.56 vs. M ⫽ 3.29), t(74) ⫽ 7.79, p ⬍ .01. There were no effects of trust.

Demand Level Demands in Rounds 1 to 6 were submitted to a 2 (opponent’s emotion: disappointment vs. guilt) ⫻ 2 (participant’s trust: high vs. low) mixed-model ANOVA with the opponent’s emotion and the participant’s trust as between-participants variables and demands in Rounds 1 to 6 as a repeated-measures variable. As in Experi-

SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS

ment 1, we first describe lower order effects, and we then turn to the hypothesized interaction. The anticipated effect of round was significant, F(5, 720) ⫽ 413.82, p ⬍ .01 (partial ␩2 ⫽ .74), showing that demands declined from Round 1 (M ⫽ 643, SD ⫽ 77) to Round 6 (M ⫽ 466, SD ⫽ 91). Results further revealed a main effect of the opponent’s emotion on average demand, F(1, 144) ⫽ 10.30, p ⬍ .002 (partial ␩2 ⫽ .07), indicating that participants’ demands were influenced by their adversaries’ emotions (disappointment: M ⫽ 521, SD ⫽ 75; guilt: M ⫽ 560, SD ⫽ 84). These effects were qualified by an Emotion ⫻ Round interaction, F(5, 720) ⫽ 8.48, p ⬍ .01 (partial ␩2 ⫽ .06), which showed that the effect of the opponent’s emotion on participants’ demands increased from Round 1 (disappointment: M ⫽ 641, SD ⫽ 77; guilt: M ⫽ 646, SD ⫽ 78) to Round 6 (disappointment: M ⫽ 439, SD ⫽ 80; guilt: M ⫽ 491, SD ⫽ 93). In line with Hypothesis 4, there was a significant interaction between opponent’s emotion and participant’s trust, F(1, 144) ⫽ 7.67, p ⬍ .01 (partial ␩2 ⫽ .05). As predicted, simple effects analysis revealed that high trust participants made higher demands to a guilty opponent (M ⫽ 586, SD ⫽ 66) than to a disappointed one (M ⫽ 510, SD ⫽ 66), F(1, 144) ⫽ 16.96, p ⬍ .01 (partial ␩2 ⫽ .26), whereas those low in trust did not respond differentially to their counterpart’s emotions (guilt: M ⫽ 537, SD ⫽ 92; disappointment: M ⫽ 531, SD ⫽ 82), F(1, 144) ⬍ 1, ns. This interaction was qualified by a significant three-way interaction between emotion, trust, and round, F(5, 720) ⫽ 2.59, p ⬍ .03 (partial ␩2 ⫽ .08), indicating that the interactive effects of emotion and trust became more apparent as the negotiation progressed (see Figure 2). Simple effects analysis revealed a significant Emotion ⫻ Round interaction for high trusters, F(5, 720) ⫽ 8.07, p ⬍ .01 (partial ␩2 ⫽ .16), but not for low trusters, F(5, 720) ⫽ 1.74, ns. As can be seen from Figure 2, participants with high levels of trust made larger concessions in the course of the negotiation when the opponent experienced disappointment than when the opponent experienced guilt, whereas participants with low levels of trust were not responsive to the opponent’s emotional state.

Interpretation of the Opponent’s Emotions ANOVAs yielded significant main effects of emotion on the two items tapping the participant’s interpretation of the opponent’s

133

emotion. Participants who negotiated with a guilty opponent interpreted the other’s emotion as signifying that the other had obtained too much (M ⫽ 3.71, SD ⫽ 0.97), whereas those who dealt with a disappointed opponent took the other’s emotion to mean that the other had received too little (M ⫽ 4.47, SD ⫽ 1.17), F(1, 144) ⫽ 19.33, p ⬍ .01 (partial ␩2 ⫽ .11). Conversely, participants with a guilty opponent interpreted the other’s emotion as a signal that the other had offered too little (M ⫽ 4.88, SD ⫽ 1.09), whereas those with a disappointed partner were more likely to believe that the other’s emotion revealed that he or she had offered too much (M ⫽ 4.49, SD ⫽ 1.04), F(1, 144) ⫽ 5.28, p ⬍ .025 (partial ␩2 ⫽ .04). There was no significant effect of trust and no interaction.

Suspicion and Discounting of the Opponent’s Emotions According to Hypothesis 5, low trusters would be more suspicious and more likely than high trusters to discount their counterpart’s emotion. In line with this prediction, an ANOVA showed a tendency for participants low in trust to be more suspicious (M ⫽ 3.92, SD ⫽ 1.20) regarding the opponent’s emotions than were participants high in trust (M ⫽ 3.50, SD ⫽ 1.31), F(1, 144) ⫽ 2.79, p ⬍ .10 (partial ␩2 ⫽ .02). Low trusters were also more likely to discount the opponent’s emotion than were high trusters (M ⫽ 3.14, SD ⫽ 1.23 vs. M ⫽ 2.71, SD ⫽ 0.99), F(1, 144) ⫽ 5.14, p ⬍ .025 (partial ␩2 ⫽ .04). There were no effects of emotion and no interaction (Fs ⬍ 1, ns).

Participants’ Goals In line with Hypothesis 6, participants’ goals were influenced by the opponent’s emotion. Participants who dealt with a guilty opponent reported higher goals (M ⫽ 5.04, SD ⫽ 0.79) than did those who dealt with a disappointed opponent (M ⫽ 5.38, SD ⫽ 0.79), F(1, 144) ⫽ 7.73, p ⬍ .01, partial ␩2 ⫽ .05 (recall that low numbers correspond to high goals; see Appendix A). This effect was moderated by the participant’s level of trust, F(1, 144) ⫽ 4.63, p ⬍ .04 (partial ␩2 ⫽ .03). Means and standard deviations pertaining to this interaction are shown in Table 2. Simple effects analyses revealed that high trusters adapted their goals to the

Figure 2. Demand level as a function of the opponent’s emotion, participants’ trust, and negotiation round in Experiment 2.

VAN KLEEF, DE DREU, AND MANSTEAD

134

Table 2 Participants’ Goals and Average Demands as a Function of the Opponent’s Emotion and the Participant’s Trust (Experiment 2) Participant’s trust Low Condition

High

Disappointment

Guilt

Disappointment

Guilt

4.88a (0.86) 531a (82)

4.81a (0.70) 537a (92)

5.07a (0.67) 510a (66)

4.45b (0.82) 586b (66)

Goal Demand

Note. Participants’ goals ranged from 1 (indicating a high goal) to 9 (indicating a low goal). Demand refers to participants’ average demands in Rounds 1 to 6. Standard deviations are in parentheses. Means within the low and high trust conditions with a different subscript differ at p ⬍ .05 according to simple-effects analyses.

opponent’s emotion, F(1, 144) ⫽ 11.54, p ⬍ .01 (partial ␩2 ⫽ .15), whereas low trusters did not, F(1, 144) ⬍ 1, ns.

Mediation Analysis To investigate whether the interactive effects of the opponent’s emotion and the participant’s trust on demands were mediated by the participant’s goals (as predicted in Hypothesis 7), we performed mediated regression analyses, following Baron and Kenny’s (1986) procedure. In Step 1 we entered emotion, trust, and their interaction as the independent variables and demands as the dependent variable. This produced a significant main effect of emotion (␤ ⫽ .24, p ⬍ .003) and an interaction between emotion and trust (␤ ⫽ .22, p ⬍ .01). In Step 2, we used the same independent variables to predict the participant’s goals. This, too, yielded a significant main effect of emotion (␤ ⫽ .21, p ⬍ .01) and a significant interaction (␤ ⫽ .17, p ⬍ .04). In Step 3, we simultaneously entered emotion and goals to predict demands. This yielded a significant effect of goals on demands (␤ ⫽ .61, p ⬍ .01) and reduced the formerly significant Emotion ⫻ Trust interaction to nonsignificance (␤ ⫽ .11, ns). A Sobel test indicated that the reduction of the direct path from the Emotion ⫻ Trust interaction to demands was significant (Z ⫽ 2.10, p ⬍ .04; see Kenny et al., 1998). In support of Hypothesis 7, these results show that the interactive effect of the opponent’s emotion and the participant’s trust on demands is fully mediated by the participant’s goals.

sheds some light on the processes underlying the effects of guilt and disappointment on demands. A mediation analysis revealed that negotiators with high levels of trust made smaller demands to a disappointed opponent than to a guilty one because the other’s disappointment led them to lower their goals, whereas the other’s guilt led them to raise their goals. Finally, results pertaining to participants’ interpretation of their counterparts’ emotions are compatible with the framework that was outlined in the introduction: Guilt is interpreted to mean that the other has claimed too much, whereas disappointment is taken as a signal that the other has received too little. The fact that trust was measured rather than manipulated might be regarded as a limitation of Experiment 2. Although the results are consistent with our predictions, we cannot rule out the possibility that they are caused by some unknown third variable that was not taken into account. Another shortcoming of Experiment 2 is that there was no nonemotional control condition. Therefore, we cannot draw firm conclusions as to whether disappointment led participants to place lower demands, whether guilt led them to place higher demands, or both. Although Experiment 1 demonstrated that guilt and disappointment elicited higher and lower demands, respectively, compared with a nonemotional control condition, we felt that a replication of this effect would increase confidence in its robustness. We therefore conducted a third experiment in which trust was manipulated (through expectations about the opponent’s personality) and a control condition was included.

Discussion The results of Experiment 2 corroborate our hypotheses. As predicted, negotiators’ reactions to their opponent’s emotions are moderated by trust. Negotiators with high levels of dispositional trust responded with high demands to an opponent who appeared to experience guilt and with low demands to an opponent who appeared to experience disappointment. By contrast, negotiators with low trust did not respond differentially to the opponent’s emotions. These findings replicate and extend those of Experiment 1 by showing that negotiators are more likely to act on their opponent’s emotions to the extent that they trust the opponent and see his or her emotions as trustworthy and reliable. We further found that individuals with low trust were more suspicious regarding the trustworthiness of the other’s emotions and were more likely to discount them rather than take them into account when designing their negotiation strategy. Furthermore, Experiment 2

Experiment 3 Experiment 2 showed that individual differences in dispositional trust moderate the interpersonal effects of supplication and appeasement emotions in negotiations. As noted earlier, variations in trust may also arise from features of the situation and/or the other party (Yamagishi, 1986). Among other things, trust depends on a negotiator’s expectations about the other’s cooperation or competition. Indeed, trust has been defined as the expectation that the other will cooperate (Pruitt & Kimmel, 1977). Because trusting another person is more risky to the extent that he or she can be expected to take advantage of and exploit the trust (Ross & LaCroix, 1996), people exhibit greater levels of trust in relation to others whom they expect to be cooperative than others whom they expect to be competitive (De Cremer, Snyder, & Dewitte, 2001;

SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS

De Dreu et al., 1998; Deutsch, 1960; Kee & Knox, 1970; Loomis, 1959; Pruitt & Kimmel, 1977; Rempel, Holmes, & Zanna, 1985; for an overview of bases of trust in negotiation, see Ross & LaCroix, 1996). On the basis of this research and the results of Experiment 2, we predicted that negotiators would make higher demands to a guilty opponent than to a disappointed one, but only if they believed that the opponent was cooperative. If the opponent was expected to be competitive, the trustworthiness of his or her emotions should be decreased, and hence negotiators should not adapt their demands to the other’s emotion (Hypothesis 8).

Method Participants and Experimental Design Ninety-four male and female undergraduate students at the University of Amsterdam participated for course credit or monetary compensation (EUR 7, approximately US $8). The experimental design included the opponent’s emotion (disappointment vs. guilt vs. no emotion) and the opponent’s personality (cooperative vs. competitive) as the independent variables and demand level as the main dependent variable. Participants were randomly assigned to conditions, and the experimenters were blind to this assignment.

Procedure The procedure was the same as in Experiment 2, except that trust was manipulated rather than measured. As in the preceding experiments, participants who reached agreement before Round 6 (n ⫽ 2) were excluded from the analyses (although including these participants in the analyses yielded a similar pattern of findings). Manipulation of trust. Trust was manipulated by varying participants’ expectations regarding the opponent’s cooperation versus competition with a procedure that has been used successfully in past research (see De Dreu & Van Kleef, 2004; Steinel & De Dreu, 2004; Van Kleef & De Dreu, 2002). At the beginning of the experiment, participants were asked to complete a (fake) “Personality Test.” This questionnaire was described as measuring collaboration skills and contained 20 items having to do with cooperation in daily life (e.g., “In the bus, I vacate my seat for older people”; “I enjoy working with other people”; “Love and respect are more important than status and money”; “Winning is everything”; “I like situations in which it is me against someone else”). Participants were asked to indicate their agreement on 5-point Likert scales (1 ⫽ strongly disagree to 5 ⫽ strongly agree). After completion of this personality test, participants received the same instructions as in Experiments 1 and 2. They then learned that on the basis of the personality test, each participant had been classified as either “cooperative” or “competitive” and that some participants would receive this information about their opponent, whereas others would not. Next, participants read that the computer had selected them to receive their opponent’s personality information, but that the opponent would not receive their information. In the cooperative opponent condition, the outcome of the opponent’s personality test was presented on the screen with answers allegedly given by the opponent suggesting that he or she was very cooperative. These answers were accompanied by a “general test result,” which indicated that the other could best be classified as cooperative. In the competitive opponent condition, participants were shown answers suggesting that the opponent was very competitive, and the general test result showed that the other could best be classified as competitive. Dependent measures. The emotion check was the same as in Experiment 2. To check participants’ expectations regarding the opponent’s cooperative versus competitive orientation, we asked participants to rate the opponent’s personality on five 7-point semantic differential scales (e.g., “cooperative– competitive,” “compliant– bossy,” “dominant– docile,” re-

135

verse scored). These items were averaged into a composite index of judgment of the opponent’s cooperativeness versus competitiveness, ranging from 1 ⫽ cooperative to 7 ⫽ competitive (␣ ⫽ .90). Additionally, we assessed participants’ perceptions of the opponent’s trustworthiness using three items (“The buyer is very trustworthy–not very trustworthy,” reverse scored; “The buyer is very unreliable–very reliable”; “The buyer is very honest–very dishonest”). These items were combined into an index of the opponent’s trustworthiness, ranging from 1 ⫽ not very trustworthy to 7 ⫽ very trustworthy (␣ ⫽ .84).

Results Manipulation Checks Opponent’s emotion. As in the previous experiments, an ANOVA showed the predicted interaction between the opponent’s emotion and the participants’ perception of the opponent’s emotion, F(2, 86) ⫽ 168.30, p ⬍ .01 (partial ␩2 ⫽ .80). Participants in the disappointed opponent condition rated the opponent as significantly more disappointed (M ⫽ 6.00, SD ⫽ 0.93) than did participants in the guilty or nonemotional opponent conditions (M ⫽ 3.91, SD ⫽ 1.16; M ⫽ 2.81, SD ⫽ 1.00). Similarly, participants with a guilty opponent rated the opponent as more guilty (M ⫽ 5.67, SD ⫽ 0.85) than did participants in the other two conditions (disappointment: M ⫽ 2.22, SD ⫽ 1.10; no emotion: M ⫽ 2.57, SD ⫽ 0.89). Furthermore, paired-samples t tests showed that ratings within the emotion conditions were higher for the intended emotion than for the other emotion (8.13 ⬍ ts ⬍ 15.82; both ps ⬍ .01). The ratings did not differ in the control condition. Opponent’s cooperation/competition and trustworthiness. An ANOVA revealed a significant effect of the manipulation of the opponent’s orientation on participants’ judgments of the opponent’s cooperativeness versus competitiveness. Participants in the cooperative opponent conditions rated the opponent as significantly more cooperative (M ⫽ 3.67, SD ⫽ 1.08) than did those in the competitive opponent conditions (M ⫽ 4.77, SD ⫽ 1.29), F(1, 86) ⫽ 19.63, p ⬍ .01 (partial ␩2 ⫽ .19). There was no main effect of the opponent’s emotion and no interaction. We also obtained a significant main effect of the opponent’s orientation on perceptions of the opponent’s trustworthiness, F(1, 86) ⫽ 18.69, p ⬍ .01 (partial ␩2 ⫽ .18). Participants who learned that the opponent had a cooperative orientation judged the other as more trustworthy (M ⫽ 5.14, SD ⫽ 0.98) than did those who learned that the other had a competitive orientation (M ⫽ 4.17, SD ⫽ 1.23).

Demand Level Demands were submitted to a 3 (opponent’s emotion: disappointment vs. guilt vs. no emotion) ⫻ 2 (opponent’s personality: cooperative vs. competitive) mixed-model ANOVA with the opponent’s emotion and personality as between-participants variables and demands in Rounds 1 to 6 as a repeated-measures variable. A main effect of round indicated that participants’ demands declined over time (Round 1: M ⫽ 643, SD ⫽ 89; Round 6: M ⫽ 457, SD ⫽ 103), F(5, 430) ⫽ 143.22, p ⬍ .01 (partial ␩2 ⫽ .63). Furthermore, we found a main effect of emotion on average demands, F(2, 86) ⫽ 4.41, p ⬍ .02 (partial ␩2 ⫽ .10). Planned comparison results were consistent with those of the previous experiments, indicating that participants with a disappointed opponent made lower demands (M ⫽ 505, SD ⫽ 88) than did

VAN KLEEF, DE DREU, AND MANSTEAD

136

participants with a guilty opponent (M ⫽ 556, SD ⫽ 94), t(89) ⫽ 1.88, p ⬍ .05 (partial ␩2 ⫽ .05). Participants with a nonemotional opponent occupied an intermediate position (M ⫽ 541, SD ⫽ 70) that did not differ significantly from the other two conditions (ts ⬍ 1.50, ns). We also found a significant interaction between the opponent’s emotion and the opponent’s orientation, F(2, 86) ⫽ 4.74, p ⬍ .01 (partial ␩2 ⫽ .10). In line with our expectations, simple effects analysis revealed a significant effect of the opponent’s emotion in the cooperative opponent condition, indicating that average demands were lower for participants who negotiated with a disappointed opponent (M ⫽ 475, SD ⫽ 52) than for those who dealt with a guilty or nonemotional opponent (M ⫽ 603, SD ⫽ 81 and M ⫽ 558, SD ⫽ 77, respectively), F(2, 86) ⫽ 7.73, p ⬍ .01 (partial ␩2 ⫽ .37). In the competitive opponent condition, by contrast, there was no effect of the opponent’s emotion on demands (523 ⬍ M ⬍ 525), F(2, 86) ⫽ 0.01, ns. Finally, we found a significant three-way interaction between opponent’s emotion, opponent’s orientation, and negotiation round, F(10, 430) ⫽ 3.84, p ⬍ .025, partial ␩2 ⫽ .08 (see Figure 3). Results of simple effects analysis were consistent with Hypothesis 8, revealing a significant Emotion ⫻ Round interaction in the cooperative opponent condition, F(10, 430) ⫽ 5.62, p ⬍ .01 (partial ␩2 ⫽ .22), but not in the competitive opponent condition, F(10, 430) ⫽ .31, ns. Planned comparisons of demands in Round 6 revealed that participants with a cooperative opponent made higher demands when the opponent appeared to feel guilty than when the opponent expressed no emotion (M ⫽ 553, SD ⫽ 130 vs. M ⫽ 472, SD ⫽ 69), t(39) ⫽ 2.24, p ⬍ .03 (partial ␩2 ⫽ .06), and that they made lower demands when the other appeared to be disappointed (M ⫽ 393, SD ⫽ 59), t(39) ⫽ 3.06, p ⬍ .01 (partial ␩2 ⫽ .08). There were no differences in the competitive opponent condition (439 ⬍ Ms ⬍ 449), ts ⬍ 1.0, ns.

Discussion The results of Experiment 3 replicate those of Experiment 2 using a situational manipulation of trust. Participants who thought that the opponent had a cooperative orientation (i.e., high trust) conceded more to a disappointed opponent than to a nonemotional one, and they tended to concede less to a guilty opponent. By contrast, participants who believed the other to be competitive

(i.e., low trust) did not respond differentially to the other’s emotions. Thus, the results of this experiment are consistent with the findings reported in Experiment 2, and this suggests that trust is an important prerequisite for the interpersonal effects of disappointment and guilt to obtain. Further, the inclusion of a nonemotional control condition allows for specific conclusions regarding the respective effects of disappointment and guilt. In line with the results of Experiment 1, the present data show that participants who expected cooperation made significantly smaller demands to a disappointed opponent than to a nonemotional one, whereas they made significantly larger demands to a guilty counterpart.

General Discussion The results of the present experiments support our predictions. Negotiators whose opponents appeared to experience emotions of appeasement (i.e., guilt or regret) developed a positive impression of their opponents, but they were nonconciliatory in the level of their demands. By contrast, participants whose opponents experienced emotions of supplication (i.e., disappointment or worry) rated their opponents less positively, but they made larger concessions in the course of the negotiation (Experiment 1). Experiment 1b showed that negotiators with a guilty opponent expected to receive larger concessions from the other than did those with a nonemotional opponent. Also in line with our theoretical framework were results showing that the interpersonal effects of guilt and disappointment on demands were mediated by the focal negotiator’s goals. Negotiators with a disappointed opponent lowered their goals and made smaller demands, whereas those with a guilty opponent raised their goals and made larger demands. The data further showed that participants with a guilty opponent believed that the other had claimed too much and offered too little, whereas those with a disappointed counterpart believed that the other had received too little and offered too much. This, too, is consistent with our theoretical framework. Apparently, emotions of supplication signal that one is in need of compensation, which may lead others to lower their goals and make concessions. Conversely, emotions of appeasement appear to signal that one is willing to compensate one’s counterpart, which may lead others to increase their goals and stand firm. The findings of Experiments 1 and 2 point to the pervasive effects of emotions related to supplication and appeasement on

Figure 3. Demand level as a function of the opponent’s emotion, the opponent’s orientation, and negotiation round in Experiment 3.

SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS

negotiation behavior and impressions, and the results of these experiments shed some light on the processes underlying these effects. In addition, the present research has identified an important moderator of the interpersonal effects of appeasement and supplication emotions on demands and concessions: interpersonal trust. Experiment 2 showed that only negotiators with high levels of trust adapted their goals and demands to their opponent’s emotion. Participants with low trust reported more suspicion regarding the opponent’s emotions, were more likely to discount them, and did not adapt their goals and demands to the other’s emotion. In Experiment 3, trust was manipulated by varying participants’ expectations regarding the opponent’s cooperative versus competitive orientation, and a similar effect was found. Participants who expected a cooperative opponent had high trust and made larger concessions to a disappointed opponent and smaller concessions to a guilty opponent. By contrast, participants who thought that the opponent had a competitive orientation had low trust and did not respond differentially to the opponent’s disappointment versus guilt in terms of their demands and concessions. Together, the results of these experiments strongly support the idea that emotions of appeasement and supplication have the potential to influence negotiation behavior at the interpersonal level. This conclusion has interesting implications for research on conflict and negotiation and for our understanding of the social consequences of emotions. Below we discuss these implications as well as address some of the strengths and limitations of our approach. We conclude by outlining some avenues for future research.

Implications and Contributions In exploring the interpersonal effects of appeasement and supplication emotions in negotiations, the present work makes a number of important contributions. First, prior research on emotions in negotiation has focused almost exclusively on the role of general positive versus negative affect or, in a few cases, discrete emotions such as anger, happiness, and compassion (Allred et al., 1997; Van Kleef et al., 2004a, b). The present research extends this line of inquiry by examining the effects of guilt, regret, disappointment, and worry. Our findings demonstrate that these emotions, like anger and happiness, can have a powerful impact on negotiation behavior. This underscores the recent acknowledgment of the importance of considering affective phenomena in conflict and negotiation and indicates that scholarly attention should not remain limited to the role of anger and happiness. Second, the current research indicates that the role of emotion in negotiation cannot be understood by merely classifying emotions as positive or negative. Previous research has explored the interpersonal effects of anger and happiness in negotiations, demonstrating that negotiators concede more to angry opponents than to happy ones (Van Kleef et al., 2004a, b). Although it is tempting to explain this finding in terms of a positivity–negativity dimension, the present results suggest that it would be unwise to do so. The emotions that were investigated in this research— guilt, regret, worry, and disappointment—are all negative in valence, yet they have quite different effects on behavior: The effects of guilt and regret were opposite to those of worry and disappointment. A more fruitful approach, then, is to adopt a social–functional perspective on emotion, which assumes that emotions have distinct social

137

functions and consequences (Frijda & Mesquita, 1994; Keltner & Haidt, 1999; Morris & Keltner, 2000). This conclusion points to the need for more research on the effects of discrete emotions rather than nonspecific positive versus negative affect. Third, the present findings contribute to a more thorough understanding of the social consequences of expressing guilt, regret, disappointment, and worry. Although all of these emotions have been the subject of research, most of this research has focused on the intrapersonal effects of these emotions on the individual’s cognitions and behavior. For example, research on guilt has addressed the question of how the experience of guilt influences the guilty party’s behavior and thereby his or her relationship to the interaction partner. Thus it has been shown that guilt motivates people to make apologies and amends and to compensate the other for one’s transgression (see Baumeister et al., 1994, for an overview of this research). Although this research contains an interpersonal component (guilt may affect interpersonal relations), the major focus has been on the impact of guilt on the individual’s own motivations and behavior. Our findings move beyond the intrapersonal effects of guilt by showing that interaction partners anticipate compensation from the guilty party by making high demands and small concessions. In a similar vein, previous research on regret has established that regret instigates a desire to undo one’s actions (Gilovich & Medvec, 1994, 1995; Zeelenberg et al., 2000). The present research suggests that regret may not only affect one’s own behavior but also that of others, although the effects of regret appear to be weaker than those of guilt. A possible explanation for this lies in the ambiguous nature of regret. An opponent who regrets his or her behavior may do so because he or she has harmed the other or because he or she should have been more strategic and self-interested. The present findings also increase our understanding of the interpersonal effects of worry and disappointment. As is the case with guilt and regret, most previous research on disappointment has adopted an intrapersonal approach, investigating for instance how the experience of disappointment motivates people to minimize future disappointment (see Zeelenberg et al., 2000, for an overview). Our research shows that disappointment can also influence behavior at the interpersonal level—it appears to be effective in eliciting concessions. The same holds for expressions of worry. These findings are consistent with prior research on distress-related emotions such as sadness, which have been shown to facilitate prosocial behavior (Barnett et al., 1979; Batson, 1987; Eisenberg, Fabes, Miller, et al., 1989; Fabes et al., 1994; Morris & Keltner, 2000). Apparently, various emotions related to distress and supplication have broadly comparable effects on others’ behavior. Altogether, it appears that discrete emotions have distinct and predictable effects in negotiations, which can be conceptualized and understood in terms of the information they provide. For example, guilt (and to a lesser degree regret) informs the adversary that one has taken too much, and it signals that one is willing to compensate for this. Disappointment and worry, on the other hand, inform the other that one has received less than expected and signal that one is in need of compensation. Because the information conveyed by these and other emotions is similar across situations, we believe that parallel effects are to be expected in other domains of social interaction.

138

VAN KLEEF, DE DREU, AND MANSTEAD

The identification of trust as a moderator of the interpersonal effects of supplication and appeasement emotions on negotiation behavior constitutes a fourth contribution of the present work. Two different operationalizations of trust were used. In Experiment 2 we measured individual differences in trust; in Experiment 3 we manipulated trust by providing information about the opponent’s personality. Although both methods have been shown in prior research to constitute valid operationalizations of trust (De Dreu & Van Kleef, 2004; Steinel & De Dreu, 2004; Yamagishi, 1986), they differ in a number of respects. Measures of dispositional trust tap directly into the generalized tendency of individuals to trust others, to believe in their honesty, and to accept what they say and do as true and righteous (or not). By contrast, the method that was used in Experiment 3 varied levels of state trust by manipulating characteristics of the opponent, rather than assessing a characteristic of the focal individual. Despite these differences, the two methods similarly affected participants’ perceptions of the trustworthiness of the opponent and of his or her emotional expressions, and both meaningfully moderated the interpersonal effects of the other’s emotions. Previous research has demonstrated the important role of information processing motivation in determining the interpersonal effects of anger and happiness on demands and concessions (Van Kleef et al., 2004b). Negotiators concede more to angry opponents than to happy ones, but only if they are sufficiently motivated to consider the emotions of the other and to think about their implications for their own goal attainment. The present research shows that trust has a similar moderating effect on the interpersonal effects of supplication and appeasement emotions. Negotiators tend to give in when the opponent experiences emotions of supplication, but they stand firm when the opponent experiences emotions of appeasement. However, this holds only for negotiators with high levels of trust; those with low trust do not adapt their demands to the other’s emotion. The current findings point to an interesting dilemma facing negotiators who anticipate future interaction. Expressions of disappointment or worry (supplication) can elicit concessions from others, but they may also contribute to a negative impression. By contrast, expressions of guilt and regret (appeasement) may serve to engender a more positive impression, but they may lead others to stand firm and resist concession making. Thus, on the one hand, negotiators may be motivated to express guilt or regret strategically to make a good impression and to induce or maintain a positive interpersonal relationship, but this would be at the expense of their personal negotiation outcomes. On the other hand, they may choose to display disappointment or worry to get their opponents to comply with their wishes, thereby inadvertently spoiling the interpersonal climate.

Limitations and Suggestions for Future Research There are some limitations to our findings. First, there was no face-to-face interaction. The primary purpose of this research was to enhance our knowledge and understanding of the interpersonal effects of discrete emotions in negotiation by generating and testing new hypotheses about the effects of supplication and appeasement emotions. In doing this, we made an explicit decision to maintain as much experimental control as possible, and we chose to use a computer-mediated negotiation paradigm to permit a

carefully controlled manipulation of the opponent’s emotion. As a result, some caution is needed when generalizing the results. At the very least, our findings pertain to computer-mediated negotiations. Given the pervasiveness of negotiation as a form of social interaction and the increasing popularity of information technologies as a communication medium, the question of how individuals react to each other’s emotions in computer-mediated communication is itself of great theoretical and practical importance (McGrath & Hollingshead, 1994; McKersie & Fonstad, 1997; Moore, Kurtzberg, Thompson, & Morris, 1999). However, considering that our paradigm has previously yielded results that have also been found in face-to-face settings (Sinaceur & Tiedens, in press) we have no reason to suspect that our findings are restricted to the domain of computer-mediated interaction. Future research could shed more light on this issue by investigating the extent to which the interpersonal effects of emotions relating to supplication and appeasement generalize across settings. Another issue concerns the “cognitive” nature of the emotion manipulation that was used in the present experiments. The fact that we used verbal manipulations of emotion raises the question of whether our findings generalize to settings in which emotions are communicated in a different manner (e.g., nonverbally). One could argue that the effects would be different if people were presented with behavioral rather than cognitive emotion cues. This possibility cannot be ruled out on the basis of the present data. However, previous research on anger and happiness in negotiations has documented similar effects regardless of whether a verbal (Van Kleef et al., 2004a, b) or nonverbal (Sinaceur & Tiedens, in press) manipulation was used. We therefore have no reason to doubt the generalizability of our findings. However, future research is needed to explore this issue in greater depth. A final issue concerns the way in which the information about the opponent’s emotions and personality was presented. We told participants that the computer had randomly determined that they would receive information about the intentions or personality of their counterpart, but that the other would not receive such information about them. A possible downside of this procedure is that it created an asymmetric situation in that participants believed that they had more information about the opponent than vice versa. For one thing, knowledge of this fact may have given participants a sense of power over the opponent. Given that previous research has found that high-power negotiators are less susceptible to their counterpart’s emotions than are low-power negotiators (Sinaceur & Tiedens, in press; Van Kleef, De Dreu, Pietroni, & Manstead, in press), this asymmetry is likely to have contributed to a more conservative test of our hypotheses. However, we cannot rule out the possibility that the informational asymmetry may have influenced our findings in one way or another. Future research might explore possible differences in responses to the other’s emotions as a function of the distribution of information. After a decade of research on the intrapersonal effects of moods and, occasionally, of emotions on the negotiator’s cognition and behavior, there has been a recent upsurge of interest in the social effects of discrete emotions in negotiations. This research has begun to document the interpersonal effects of emotions relevant to negotiation and conflict resolution. Although the results are promising, many questions remain unanswered. How does the expectation of future interaction with the same partner influence the interpersonal effects of different emotions on negotiation be-

SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS

havior? Are there other variables, besides trust and information processing, that moderate these effects? What happens when more than one negotiator expresses certain emotions? What factors determine whose emotions will have the bigger impact? How long can a negotiator effectively continue to use anger or disappointment as a means of eliciting concessions? Do the effects of these emotions on demands wear off or even reverse in the long run? The investigation of these and other questions will allow the study of emotion in social conflict to continue to advance.

Conclusion The present research investigated the interpersonal effects of guilt, regret, disappointment, and worry in negotiations. The results showed that negotiators tend to make larger concessions to opponents who experience disappointment or worry (supplication emotions) than to nonemotional opponents, whereas they make smaller concessions to adversaries who experience guilt or regret (appeasement emotions). This effect was shown to be mediated by the focal negotiator’s goals (negotiators with a guilty opponent adopted higher goals than did those with a disappointed opponent) and moderated by interpersonal trust—negotiators with high levels of trust adapted their demands to their counterpart’s emotion, but those with low trust did not. These findings point to the pervasive interpersonal effects of emotions in negotiations, and they stress the need for more research on the role of emotion in conflict and negotiation. Such research promises to enhance our understanding of the negotiation process, of the factors that facilitate or hinder constructive conflict resolution, and of the social consequences of emotions in general.

References Adler, R. S., Rosen, B., & Silverstein, E. M. (1998). Emotions in negotiation: How to manage fear and anger. Negotiation Journal, 14, 161–180. Allred, K. G., Mallozzi, J. S., Matsui, F., & Raia, C. P. (1997). The influence of anger and compassion on negotiation performance. Organizational Behavior and Human Decision Processes, 70, 175–187. Barnett, M. A., Howard, J. A., Melton, E. M., & Dino, G. A. (1982). Effect of inducing sadness about self or other on helping behavior in high- and low-empathic children. Child Development, 53, 920 –923. Barnett, M. A., King, L. W., & Howard, J. A. (1979). Inducing affect about self or other: Effects on generosity in children. Developmental Psychology, 15, 164 –167. Baron, R. A. (1990). Environmentally induced positive affect: Its impact on self-efficacy, task performance, negotiation, and conflict. Journal of Applied Social Psychology, 20, 368 –384. Baron, R. A., Fortin, S. P., Frei, R. L., Hauver, L. A., & Shack, M. L. (1990). Reducing organizational conflict: The role of socially-induced positive affect. International Journal of Conflict Management, 1, 133– 152. 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. Barrett, K. C. (1995). A functionalist approach to shame and guilt. In J. P. Tangney & K. W. Fischer (Eds.), Self-conscious emotions: The psychology of shame, guilt, embarrassment, and pride (pp. 25– 63). New York: Guilford Press. Barry, B. (1999). The tactical use of emotion in negotiation. Research on Negotiation in Organizations, 7, 93–121. Barry, B., Fulmer, I. S., & Van Kleef, G. A. (2004). I laughed, I cried, I

139

settled: The role of emotion in negotiation. In M. J. Gelfand & J. M. Brett (Eds.), The handbook of negotiation and culture (pp. 71–94). Stanford, CA: Stanford University Press. Barry, B., & Oliver, R. L. (1996). Affect in dyadic negotiation: A model and propositions. Organizational Behavior and Human Decision Processes, 67, 127–143. Batson, C. D. (1987). Prosocial motivation: Is it ever truly altruistic? In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 20, pp. 65–122). New York: Academic Press. Baumeister, R. F., Stillwell, A. M., & Heatherton, T. F. (1994). Guilt: An interpersonal approach. Psychological Bulletin, 115, 243–267. Bazerman, M. H., Curhan, J. R., Moore, D. A., & Valley, K. L. (2000). Annual Review of Psychology, 51, 279 –314. Bell, D. E. (1985). Disappointment in decision making under uncertainty. Operations Research, 33, 1–27. Berndsen, M., van der Pligt, J., Doosje, B., & Manstead, A. S. R. (2004). Guilt and regret: The determining role of interpersonal and intrapersonal harm. Cognition and Emotion, 18, 55–70. Berscheid, E., & Walster, E. (1967). When does a harm-doer compensate a victim? Journal of Personality and Social Psychology, 6, 435– 441. Carlsmith, J. M., & Gross, A. (1969). Some effects of guilt on compliance. Journal of Personality and Social Psychology, 11, 232–239. Carnevale, P. J., & Isen, A. M. (1986). The influence of positive affect and visual access on the discovery of integrative solutions in bilateral negotiation. Organizational Behavior and Human Decision Processes, 37, 1–13. Carver, C. S., & Scheier, M. F. (1990). Origins and functions of positive and negative affect: A control-process view. Psychological Review, 97, 19 –35. Clark, L. F. (1993). Stress and the cognitive– conversational benefits of social interaction. Journal of Social and Clinical Psychology, 12, 25–55. Clark, M. A., & Taraban, C. B. (1991). Reactions to and willingness to express emotion in two types of relationships. Journal of Experimental Social Psychology, 27, 324 –336. Clark, M. S., Ouellette, R., Powell, M. C., & Milberg, S. (1987). Recipient’s mood, relationship type, and helping. Journal of Personality and Social Psychology, 53, 94 –103. Clark, M. S., Pataki, S. P., & Carver, V. H. (1996). Some thoughts and findings on self-presentation of emotions in relationships. In G. J. O. Fletcher & J. Fitness (Eds.), Knowledge structures in close relationships: A social psychological approach (pp. 247–274). Mahwah, NJ: Erlbaum. Cornelius, R. R. (1984). A rule model of emotional expression. In C. Z. Malatesta & C. E. Izard (Eds.), Emotion in adult development (pp. 213–233). Beverly Hills, CA: Sage. Coˆte´, S. (2005). A social interaction model of the effects of emotion regulation on work strain. Academy of Management Review, 30, 509 – 530. Davidson, M. N., & Greenhalgh, L. (1999). The role of emotion in negotiation: The impact of anger and race. Research on Negotiation in Organizations, 7, 3–26. De Cremer, D., Snyder, M., & Dewitte, S. (2001). ‘The less I trust, the less I contribute (or not?)’: The effects of trust, accountability and selfmonitoring in social dilemmas. European Journal of Social Psychology, 31, 93–107. De Dreu, C. K. W., Giebels, E., & Van de Vliert, E. (1998). Social motives and trust in integrative negotiation: The disruptive effects of punitive capability. Journal of Applied Psychology, 83, 408 – 422. De Dreu, C. K. W., Lualhati, J. C., & McCusker, C. (1994). Effects of gain/loss frames on satisfaction with self/other outcome-differences. European Journal of Social Psychology, 24, 497–510. De Dreu, C. K. W., & Van Kleef, G. A. (2004). The influence of power on the information search, impression formation, and demands in negotiation. Journal of Experimental Social Psychology, 40, 303–319.

140

VAN KLEEF, DE DREU, AND MANSTEAD

De Dreu, C. K. W., & Van Lange, P. A. M. (1995). The impact of social value orientations on negotiator cognition and behavior. Personality and Social Psychology Bulletin, 21, 1178 –1188. Deutsch, M. (1960). The effect of motivational orientation upon trust and suspicion. Human Relations, 13, 123–139. Eisenberg, N. (2000). Emotion, regulation, and moral development. Annual Review of Psychology, 51, 665– 697. Eisenberg, N., Fabes, R. A., Miller, P. A., Fultz, J., Mathy, R. M., Shell, R., & Reno, R. R. (1989). The relations of sympathy and personal distress to prosocial behavior: A multimethod study. Journal of Personality and Social Psychology, 57, 55– 66. Eisenberg, N., Fabes, R. A., Schaller, M., & Miller, P. (1989). Sympathy and personal distress: Development, gender differences, and interrelations of indexes. In N. Eisenberg (Ed.), Empathy and related emotional responses (pp. 107–126). San Francisco: Jossey-Bass. Ekman, P. (1993). Facial expression and emotion. American Psychologist, 48, 384 –392. Fabes, R. A., Eisenberg, N., Karbon, M., Troyer, D., & Switzer, G. (1994). The relations of children’s emotion regulation to their vicarious emotional responses and comforting behaviors. Child Development, 65, 1678 –1693. Forgas, J. P. (1998). On feeling good and getting your way: Mood effects on negotiator cognition and behavior. Journal of Personality and Social Psychology, 74, 565–577. Freedman, J. L., Wallington, S., & Bless, E. (1967). Compliance without pressure: The effect of guilt. Journal of Personality and Social Psychology, 7, 117–124. Fridlund, A. J. (1994). Human facial expression: An evolutionary view. San Diego, CA: Academic Press. Friedman, M. (1985). Toward a reconceptualization of guilt. Contemporary Psychoanalysis, 21, 501–547. Frijda, N. H. (1986). The emotions. Cambridge, England: Cambridge University Press. Frijda, N. H. (1993). Moods, emotion episodes, and emotions. In M. Lewis & J. M. Haviland (Eds.), Handbook of emotions (pp. 381– 403). New York: Guilford Press. Frijda, N. H., & Mesquita, B. (1994). The social roles and functions of emotions. In S. Kitayama & H. S. Markus (Eds.), Emotion and culture: Empirical studies of mutual influence (pp. 51– 87). Washington, DC: American Psychological Association. Gilovich, T., & Medvec, V. H. (1994). The temporal pattern to the experience of regret. Journal of Personality and Social Psychology, 67, 357–365. Gilovich, T., & Medvec, V. H. (1995). The experience of regret: What, when, and why. Psychological Review, 102, 379 –395. Hegtvedt, K. A. (1990). The effects of relationship structure on emotional responses to inequity. Social Psychology Quarterly, 53, 214 –228. Hegtvedt, K. A., & Killian, C. (1999). Fairness and emotions: Reactions to the process and outcomes of negotiations. Social Forces, 78, 269 –302. Hill, M. G., Weary, G., & Williams, J. (1986). Depression: A selfpresentation formulation. In R. Baumeister (Ed.), Public self and private self (pp. 213–240). New York: Springer-Verlag. Hilty, J. A., & Carnevale, P. J. (1993). Black-hat/white-hat strategy in bilateral negotiation. Organizational Behavior and Human Decision Processes, 55, 444 – 469. Hoffman, M. L. (1982). Development of prosocial motivation: Empathy and guilt. In N. Eisenberg (Ed.), The development of prosocial behavior (pp. 281–313). New York: Academic Press. Holmes, J. G., Throop, W. F., & Strickland, L. H. (1971). The effects of prenegotiation expectations on the distributive bargaining process. Journal of Experimental Social Psychology, 7, 582–599. Homans, G. C. (1974). Social behavior: Its elementary forms. Oxford, England: Harcourt Brace Jovanovich. 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. Kee, H. W., & Knox, R. E. (1970). Conceptual and methodological considerations in the study of trust and suspicion. Journal of Conflict Resolution, 14, 357–366. Kelley, H. H., Beckman, L. L., & Fischer, C. S. (1967). Negotiating the division of a reward under incomplete information. Journal of Experimental Social Psychology, 3, 361–398. Keltner, D., & Buswell, B. N. (1997). Embarrassment: Its distinct form and appeasement functions. Psychological Bulletin, 122, 250 –270. Keltner, D., & Haidt, J. (1999). Social functions of emotions at four levels of analysis. Cognition and Emotion, 13, 505–521. Kennedy-Moore, E., & Watson, J. C. (2001). How and when does emotional expression help? Review of General Psychology, 5, 187–212. Kenny, D. A., Kashy, D. A., & Bolger, N. (1998). Data analysis in social psychology. In D. T. Gilbert, S. T. Fiske, & A. Lindzey (Eds.), The handbook of social psychology (Vol. 1, pp. 233–265). Boston: McGraw-Hill. Ketelaar, T., & Au, W. T. (2003). The effects of feelings of guilt on the behaviour of uncooperative individuals in repeated social bargaining games: An affect-as-information interpretation of the role of emotion in social interaction. Cognition and Emotion, 17, 429 – 453. Kimmel, M. J., Pruitt, D. G., Magenau, J. M., Konar-Goldband, E., & Carnevale, P. J. D. (1980). Effects of trust, aspiration, and gender on negotiation tactics. Journal of Personality and Social Psychology, 38, 9 –22. Knutson, B. (1996). Facial expressions of emotion influence interpersonal trait inferences. Journal of Nonverbal Behavior, 20, 165–182. Labott, S. M., Martin, R. B., Eason, P. S., & Berkey, E. Y. (1991). Social reactions to the expression of emotion. Cognition and Emotion, 5, 397– 417. Lazarus, R. S. (1991). Emotion and adaptation. Oxford, England: Oxford University Press. Leith, K. P., & Baumeister, R. F. (1998). Empathy, shame, guilt, and narratives of interpersonal conflicts: Guilt-prone people are better at perspective taking. Journal of Personality and Social Psychology, 66, 1–37. Levenson, R. W. (1994). Human emotion: A functional view. In P. Ekman & R. J. Davidson (Eds.), The nature of emotion: Fundamental questions (pp. 123–126). New York: Oxford University Press. Lewis, H. B. (1971). Shame and guilt in neurosis. New York: International Universities Press. Lewis, M. (2000). Self-conscious emotions: Embarrassment, pride, shame, and guilt. In M. Lewis & J. M. Haviland-Jones (Eds.), Handbook of emotions (2nd ed., pp. 623– 636). New York: Guilford Press. Lindskold, S. (1978). Trust development, the GRIT proposal, and the effects of conciliatory acts on conflict and cooperation. Psychological Bulletin, 85, 772–793. Loewenstein, G. F., Thompson, L., & Bazerman, M. H. (1989). Social utility and decision making in interpersonal contexts. Journal of Personality and Social Psychology, 57, 426 – 441. Loomes, G., & Sugden, R. (1986). Disappointment and dynamic inconsistency in choice under uncertainty. Review of Economic Studies, 53, 271–282. Loomis, J. L. (1959). Communication, the development of trust, and cooperative behavior. Human Relations, 12, 305–315. Manstead, A. S. R. (1991). Emotion in social life. Cognition and Emotion, 5, 353–362. McGrath, J. E., & Hollingshead, A. B. (1994). Groups interacting with technology: Ideas, evidence, issues, and an agenda. Thousand Oaks, CA: Sage. McKersie, R. B., & Fonstad, N. O. (1997). Teaching negotiation theory and skills over the Internet. Negotiation Journal, 13, 363–368.

SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS Messick, D. M., & Sentis, K. P. (1979). Fairness and preference. Journal of Experimental Social Psychology, 15, 418 – 434. Moore, D. A., Kurtzberg, T. R., Thompson, L. L., & Morris, M. W. (1999). Long and short routes to success in electronically mediated negotiations: Group affiliations and good vibrations. Organizational Behavior and Human Decision Processes, 77, 22– 43. Morris, M. W., & Keltner, D. (2000). How emotions work: An analysis of the social functions of emotional expression in negotiations. Research in Organizational Behavior, 22, 1–50. Naquin, C. E., & Paulson, G. D. (2003). Online bargaining and interpersonal trust. Journal of Applied Psychology, 88, 113–120. Oatley, K., & Jenkins, J. M. (1992). Human emotions: Functions and dysfunctions. Annual Review of Psychology, 43, 55– 85. Oatley, K., & Jenkins, J. M. (1996). Understanding emotions. Cambridge, MA: Blackwell. Parkinson, B. (1996). Emotions are social. British Journal of Psychology, 87, 663– 683. Parks, C. D., Henager, R. F., & Scamahorn, S. D. (1996). Trust and messages of intent in social dilemmas. Journal of Conflict Resolution, 40, 134 –151. Parks, C. D., & Hulbert, L. G. (1995). High and low trusters’ responses to fear in a payoff matrix. Journal of Conflict Resolution, 39, 718 –730. Parrott, W. G. (1993). Beyond hedonism: Motives for inhibiting good moods and for maintaining bad moods. In D. M. Wegner & J. W. Pennebaker (Eds.), Handbook of mental control (pp. 278 –305). Englewood Cliffs, NJ: Prentice Hall. Pillutla, M. M., & Murnighan, J. K. (1996). Unfairness, anger, and spite: Emotional rejections of ultimatum offers. Organizational Behavior and Human Decision Processes, 68, 208 –224. Pruitt, D. G., & Carnevale, P. J. (1993). Negotiation in social conflict. Buckingham, England: Open University Press. Pruitt, D. G., & Kimmel, M. J. (1977). Twenty years of experimental gaming: Critique, synthesis, and suggestions for the future. Annual Review of Psychology, 28, 363–392. Rempel, J. K., Holmes, J. G., & Zanna, M. P. (1985). Trust in close relationships. Journal of Personality and Social Psychology, 49, 95–112. Roseman, I. J., Wiest, C., & Swartz, T. S. (1994). Phenomenology, behaviors, and goals differentiate discrete emotions. Journal of Personality and Social Psychology, 67, 206 –221. Ross, W., & LaCroix, J. (1996). Multiple meanings of trust in negotiation theory and research: A literature review and integrative model. International Journal of Conflict Management, 7, 314 –360. Rotter, J. B. (1980). Interpersonal trust, trustworthiness, and gullibility. American Psychologist, 35, 1–7. Rousseau, D. M., Sitkin, S. B., Burt, R. S., & Camerer, C. (1998). Not so different after all: A cross-discipline view of trust. Academy of Management Review, 23, 393– 404. Russell, J. A., & Feldman Barrett, L. (1999). Core affect, prototypical emotional episodes, and other things called emotion: Dissecting the elephant. Journal of Personality and Social Psychology, 76, 805– 819. Siegel, S., & Fouraker, L. E. (1960). Bargaining and group decision making: Experiments in bilateral monopoly. New York: McGraw-Hill. Sinaceur, M., & Tiedens, L. Z. (in press). Get mad and get more than even: When and why anger expression is effective in negotiations. Journal of Experimental Social Psychology. Smith, D. L., Pruitt, D. G., & Carnevale, P. J. (1982). Matching and mismatching: The effect of own limit, other’s toughness, and time pressure on concession rate in negotiation. Journal of Personality and Social Psychology, 42, 876 – 883. Smith, R. H., Webster, J. M., Parrott, W. G., & Eyre, H. L. (2002). The role of public exposure in moral and nonmoral shame and guilt. Journal of Personality and Social Psychology, 83, 138 –159. Sprecher, S. (1992). How men and women expect to feel and behave in

141

response to inequity in close relations. Social Psychology Quarterly, 55, 57– 69. Steinel, W., & De Dreu, C. K. W. (2004). Social motives and strategic misrepresentation in social decision making. Journal of Personality and Social Psychology, 86, 419 – 434. Steiner, C. (2000). Apology: The transactional analysis of a fundamental exchange. Transactional Analysis Journal, 30, 145–149. Tangney, J. P. (1990). Assessing individual differences in proneness to shame and guilt: Development of the self-conscious affect and attribution inventory. Journal of Personality and Social Psychology, 59, 102– 111. Tangney, J. P. (1995). Shame and guilt in interpersonal relations. In J. P. Tangney & K. W. Fischer (Eds.), Self-conscious emotions: The psychology of shame, guilt, embarrassment, and pride (pp. 114 –139). New York: Guilford Press. Tangney, J. P. (1999). The self-conscious emotions: Shame, guilt, embarrassment, and pride. In T. Dalgleish & M. Power (Eds.), Handbook of cognition and emotion (pp. 541–568). New York: Wiley. Thompson, L., Medvec, V. H., Seiden, V., & Kopelman, S. (2001). Poker face, smiley face, and rant ‘n’ rave: Myths and realities about emotion in negotiation. In M. A. Hogg & R. S. Tindale (Eds.), Blackwell handbook of social psychology: Group processes (pp. 139 –163). Malden, MA: Blackwell. Timmers, M., Fischer, A. H., & Manstead, A. S. R. (1998). Gender differences in motives for regulating emotions. Personality and Social Psychology Bulletin, 24, 974 –985. Tomkins, S. S. (1963). Affect, imagery, consciousness. Vol. 2: The negative affects. New York: Springer. Tripp, T. M., & Sondak, H. (1992). An evaluation of dependent variables in experimental negotiation studies: Impasse rates and Pareto efficiency. Organizational Behavior and Human Decision Processes, 51, 273–295. van Dijk, W. W., & van der Pligt, J. (1997). The impact of probability and magnitude of outcome on disappointment and elation. Organizational Behavior and Human Decision Processes, 69, 277–284. van Dijk, W. W., & Zeelenberg, M. (2002). Investigating the appraisal patterns of regret and disappointment. Motivation and Emotion, 26, 321–331. Van Kleef, G. A., & De Dreu, C. K. W. (2002). Social value orientation and impression formation: A test of two competing hypotheses about information search in negotiation. International Journal of Conflict Management, 13, 59 –77. Van Kleef, G. A., De Dreu, C. K. W., & Manstead, A. S. R. (2004a). The interpersonal effects of anger and happiness in negotiations. Journal of Personality and Social Psychology, 86, 57–76. Van Kleef, G. A., De Dreu, C. K. W., & Manstead, A. S. R. (2004b). The interpersonal effects of emotions in negotiations: A motivated information processing approach. Journal of Personality and Social Psychology, 87, 510 –528. Van Kleef, G. A., De Dreu, C. K. W., Pietroni, D., & Manstead, A. S. R. (in press). Power and emotion in negotiation: Power moderates the interpersonal effects of anger and happiness on concession making. European Journal of Social Psychology. Wallace, J., & Sadalla, E. (1966). Behavioral consequences of transgression: I. The effects of social recognition. Journal of Experimental Research in Personality, 1, 187–194. Yamagishi, T. (1986). The provision of a sanctioning system as a public good. Journal of Personality and Social Psychology, 50, 110 –116. Yee, J. L., & Greenberg, M. S. (1998). Reactions to crime victims: Effects of victims’ emotional state and type of friendship. Journal of Social and Clinical Psychology, 17, 209 –226. Yukl, G. A. (1974a). Effects of situational variables and opponent concessions on a bargainer’s perception, aspirations, and concessions. Journal of Personality and Social Psychology, 29, 227–236. Yukl, G. A. (1974b). Effects of the opponent’s initial offer, concession

VAN KLEEF, DE DREU, AND MANSTEAD

142

magnitude, and concession frequency on bargaining behavior. Journal of Personality and Social Psychology, 30, 323–335. Zeelenberg, M., & Beattie, J. (1997). Consequences of regret aversion 2: Additional evidence for effects of feedback on decision making. Organizational Behavior and Human Decision Processes, 72, 63–78. Zeelenberg, M., van der Pligt, J., & Manstead, A. S. R. (1998). Undoing regret on Dutch television: Apologizing for interpersonal regrets involv-

ing actions or inactions. Personality and Social Psychology Bulletin, 24, 1113–1119. Zeelenberg, M., van Dijk, W. W., Manstead, A. S. R., & van der Pligt, J. (2000). On bad decisions and disconfirmed expectancies: The psychology of regret and disappointment. Cognition and Emotion, 14, 521–541. Zetik, D. C., & Stuhlmacher, A. F. (2002). Goal setting and negotiation performance: A meta-analysis. Group Processes and Intergroup Relations, 5, 35–52.

Appendix A

Appendix B

Participants’ Payoff Chart

Statements Used for the Manipulation of the Opponent’s Emotion

Price of phones Level

Price

Payoff

1 2 3 4 5 6 7 8 9

$150 $145 $140 $135 $130 $125 $120 $115 $110

400 350 300 250 200 150 100 50 0

Warranty period Warranty 1 2 3 4 5 6 7 8 9

month months months months months months months months months

Payoff 120 105 90 75 60 45 30 15 0

Service contract Service 1 2 3 4 5 6 7 8 9

month months months months months months months months months

Payoff 240 210 180 150 120 90 60 30 0

Opponent’s emotion

Statement After Round 1

Disappointment Worry Guilt Regret No emotion

Note. Prices in Euros were converted to U.S. dollars and rounded to the nearest U.S.$5.

I am pretty disappointed about this, I think I will offer 8–7–7 This worries me quite a lot, I think I will offer 8–7–7 I feel guilty for not having conceded more, I think I will offer 8–7–7 I feel sorry that I haven’t conceded more, I think I will offer 8–7–7 I think I will offer 8–7–7 After Round 3

Disappointment Worry Guilt Regret No emotion

This is going awry, I am very disappointed. I am going to offer 7–6–7 This is going awry, I am very worried. I am going to offer 7–6–7 This is going awry, I feel pretty guilty. I am going to offer 7–6–7 This is going awry, I regret it. I am going to offer 7–6–7 I am going to offer 7–6–7 After Round 5

Disappointment Worry Guilt Regret No emotion

I am going to offer disappointed I am going to offer worried I am going to offer guilty I am going to offer sorry I am going to offer

6–6–6, because I am really 6–6–6, because I am really 6–6–6, because I feel very 6–6–6, because I am very 6–6–6

Note. Statements were pretested and have been translated from Dutch. The typing errors deliberately included in the original versions are not shown.

Received April 17, 2004 Revision received May 31, 2005 Accepted December 14, 2005 䡲

PERSONALITY PROCESSES AND INDIVIDUAL DIFFERENCES

Optimism in Close Relationships: How Seeing Things in a Positive Light Makes Them So Sanjay Srivastava

Kelly M. McGonigal

University of Oregon

Stanford University

Jane M. Richards

Emily A. Butler

University of Texas at Austin

University of Arizona

James J. Gross Stanford University Does expecting positive outcomes— especially in important life domains such as relationships—make these positive outcomes more likely? In a longitudinal study of dating couples, the authors tested whether optimists (who have a cognitive disposition to expect positive outcomes) and their romantic partners are more satisfied in their relationships, and if so, whether this is due to optimists perceiving greater support from their partners. In cross-sectional analyses, both optimists and their partners indicated greater relationship satisfaction, an effect that was mediated by optimists’ greater perceived support. When the couples engaged in a conflict conversation, optimists and their partners saw each other as engaging more constructively during the conflict, which in turn led both partners to feel that the conflict was better resolved 1 week later. In a 1-year follow-up, men’s optimism predicted relationship status. Effects of optimism were mediated by the optimists’ perceived support, which appears to promote a variety of beneficial processes in romantic relationships. Keywords: optimism, relationship satisfaction, perceived support, close relationships

affected by the cognitive dispositions of the individuals involved. As Coleridge might say, lovers’ eyes are in their minds. In this article, we present an investigation of the consequences of one particular cognitive disposition, namely optimism, within romantic relationships. Is optimism associated with happier and longer lasting romantic relationships? To answer this question, our research was designed to test two related hypotheses. First, we tested the hypothesis that optimists and their partners would have relationships that are more satisfying, are characterized by better conflict resolution, and are longer lasting.1 Second, we tested the hypothesis that the reason why optimists have better relationship outcomes is that they perceive their partners as more supportive. We tested these hypotheses in cross-sectional analyses of couples’

I have heard of reasons manifold Why Love needs be blind, But this the best of all I hold— His eyes are in his mind. —Samuel Taylor Coleridge (1811)

Individuals’ perceptions of the social world are more than just objective reports of an external reality. Social perceptions are shaped in the mind of the perceiver, a fact that can have very real consequences for social life. Romantic relationships, in particular, have long been observed by poets and writers to be substantially

Sanjay Srivastava, Department of Psychology, University of Oregon; Kelly M. McGonigal and James J. Gross, Department of Psychology, Stanford University; Jane M. Richards, Department of Psychology, University of Texas at Austin; Emily A. Butler, Department of Psychology, University of Arizona. This research was supported by National Institutes of Health Grant R01 M58147 awarded to James J. Gross. Kelly M. McGonigal was supported by a National Science Foundation Graduate Research Fellowship. Correspondence concerning this article should be addressed to Sanjay Srivastava, Department of Psychology, 1227 University of Oregon, Eugene, OR 97403-1227. E-mail: [email protected]

1

Optimism and pessimism can be conceptualized several different ways: as opposite poles of a single dimension, as two distinct dimensions, or as discrete categories. In this article we treat optimism both conceptually and empirically as a single, bipolar dimension, an approach that was supported by analyses of the data. To avoid cumbersome language, we have used the term optimists in this article as a shorthand, meaning in effect, “individuals who score higher in optimism, relative to those who score lower.” It is not our intention to suggest that optimists are a discrete category.

Journal of Personality and Social Psychology, 2006, Vol. 91, No. 1, 143–153 Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.143

143

144

SRIVASTAVA, MCGONIGAL, RICHARDS, BUTLER, AND GROSS

reports about their relationships, in analyses of how couples responded to a conflict conversation, and in a 1-year follow-up of relationship dissolution.

Optimism, Perceived Support, and Social Functioning Optimism is defined as the cognitive disposition to expect favorable outcomes (Scheier & Carver, 1985). A substantial body of research has linked optimism to effective coping and to positive mental and physical health outcomes (e.g., Scheier, Carver, & Bridges, 2001; Taylor, Kemeny, Reed, Bower, & Gruenewald, 2000). A smaller number of studies have also shown that optimism leads to better social functioning. For example, optimism is associated with lower social alienation (Scheier & Carver, 1985) and with longer lasting friendships (Geers, Reilley, & Dember, 1998). In romantic relationships, two prior studies have suggested that optimism about a particular relationship predicts greater satisfaction in that relationship and reduced likelihood of relationship dissolution (Helgeson, 1994; Murray & Holmes, 1997), although the mechanisms explaining such a relation were not directly tested. Why might optimists have more positive experiences in relationships? As a cognitive disposition, optimism should influence how individuals attend to and interpret others’ behaviors and intentions. We propose that within a close relationship, this cognitive disposition may manifest itself as perceived support, the belief that one’s partner is able and willing to provide support if necessary (Murray & Holmes, 1997). Perceived support could in turn have a number of benefits: It has been shown to lead individuals to feel that their relationship facilitates their personal and collective goals (Brunstein, Dangelmayer, & Schultheiss, 1996; Kaplan & Maddux, 2002), and it may buffer against stress and negative affect in relationships (Dehle, Larsen, & Landers, 2001). This latter effect may be particularly important in close relationships. Research on perceived support indicates that, like optimism, it is moderately stable over time (Sarason, Sarason, & Shearin, 1986), and it appears to be something more than simply a direct reflection of others’ actual supportive behaviors (Barrera, 1986; Belsher & Costello, 1991; Newcomb, 1990). Yet despite the agreement among many researchers that perceived support is influenced by personality variables, Lakey, McCabe, Fisicaro, and Drew (1996) wrote that “surprisingly, there has been very little research on the personality factors that predict the development of perceived support” (p. 1278). Among personality factors that might promote perceived support, optimism seems to be a likely candidate. Perceived support is associated with positive biases in evaluating and remembering supportive behaviors in specific interactions and relationships (Lakey et al., 1996; Lakey & Cassady, 1990; Pierce, Sarason, & Sarason, 1992). Furthermore, the proposed mechanisms of perceived support—positive affect, coping self-efficacy, and adaptive coping—are all robustly associated with optimism (Chang, 2001; Cozarelli, 1993; Scheier et al., 2001). Optimists are better liked by others, which may reinforce their expectations about how others will treat them (Carver, Kus, & Scheier, 1994). In relationships, we expect that optimists would be more likely to perceive others’ behaviors as supportive and to respond accordingly. A few studies have offered some evidence directly linking optimism to perceived support. Associations between optimism and perceived support have been found among air crash rescue

workers (Dougall, Hyman, Hayward, McFeeley, & Baum, 2001), bereaved men (Park & Folkman, 1997), and college students (Sarason, Levine, Basham, & Sarason, 1983). In a longitudinal investigation, Brissette, Scheier, and Carver (2002) investigated the relationship between optimism and perceived social support. In a sample of incoming college students, optimism was associated with concurrent reports of perceived support and number of close friendships at the beginning of college and with increases in perceived support over the course of the semester. The increases in perceived support mediated the effect of optimism on depression, though not the effect of optimism on stress. Brissette and colleagues’ findings are important and suggestive, but they were not able to examine relational outcomes such as relationship satisfaction or conflict resolution; their study also did not examine the effects of an individual’s optimism on relationship partners.

The Present Study The available evidence suggests that optimism is associated with positive outcomes in relationships in general, possibly as a result of processes that promote and maintain perceived support. Our particular interest was in examining these processes in the context of close relationships. Optimism and perceived support are often studied in terms of their consequences for social life in general; an examination of close relationships offers several distinct opportunities to complement this research. For researchers who study close relationships, studying optimism and perceived support can potentially provide insights into the cognitive processes that maintain security and closeness between partners. For researchers who study optimism, close relationships are an important life domain for which optimism may have meaningful consequences. Studies of perceived support also suggest that there may be important processes taking place in the context of dyadic relationships that could be missed in broad-bandwidth studies of social life. Although individuals do differ in their general tendency to perceive all others as supportive, perceived support also draws substantially on relationship-specific perceptions (Lakey et al., 1996). That is, individuals form distinct judgments about the supportiveness of other individuals, above and beyond their broad judgments about others in general. Although much research on social support has focused broadly on social networks, this finding suggests that it is also important to examine the consequences of perceived support in the context of specific relationships. In developing our questions and hypotheses, we organized our investigation around two guiding questions. First, what consequences, if any, does optimism have for satisfaction in close relationships, both for the optimist and for the optimist’s partner? Second, does perceived support explain the relation between optimism and relationship satisfaction? Because of the complexity of the research design, we present the findings in three parts (see Table 1). Part 1 examines the crosssectional relations among both partners’ optimism, perceived support, and relationship satisfaction at Time 1. Part 2 reports a closer examination of how the couples reacted to conflict (Time 2) and how well they felt the conflict was resolved 1 week later (Time 3). Part 3 examines an objective outcome, relationship maintenance versus dissolution, 1 year later (Time 4).

OPTIMISM IN CLOSE RELATIONSHIPS

145

Table 1 Overview of Design, Procedures, and Measures Timepoint (time since Time 1)

Procedure

Major measure

Part 1 Time 1

Questionnaire packet

Optimism, perceived support, relationship satisfaction

Part 2 Time 2 (1 week) Time 3 (2 weeks)

Laboratory-based conflict conversation Post-conflict follow-up

Positive conflict engagement Conflict resolution

Part 3 Time 4 (1 year)

One-year follow-up

Part 1: Optimism and Perceived Support in a Dating Relationship In Part 1, we examined partners’ reports regarding their dating relationship in general. We hypothesized that optimism would be associated with greater relationship satisfaction. Furthermore, we expected not only that optimists themselves would report greater relationship satisfaction than would pessimists but also that the partners of optimists would report greater relationship satisfaction than the partners of pessimists. Such an effect would indicate that the positive relational consequences of optimism are not just “in the head” of the optimists. We further hypothesized that the effects of optimism would be mediated by optimists’ tendency to perceive their partners as supportive in the relationship. To rule out possible confounds, we conducted several additional analyses. One possible confounding variable was partner investment: Perhaps optimists attract more supportive partners, in which case an effect of optimism on perceived support could simply reflect an accurate appraisal rather than a perceptual disposition. Thus, we also obtained reports from each partner of offered support in the relationship to use as control variables. If optimists have a global tendency to see their partners as supportive, that relation should be independent of the actual amount of support offered by their partners. Finally, some studies have suggested that optimism may be correlated with the personality traits of Neuroticism, Extraversion, or self-esteem (see Scheier et al., 2001). Thus, we conducted additional control analyses to ensure that the effects of optimism were independent of these other dimensions of individual differences, as well as the individuals’ ages, the length of the relationship, and whether the partners were living together.

Method

Relationship status

At least one member of each couple was an undergraduate recruited from one of three northern California universities. Couples were exclusive and had been dating for at least 6 months at the start of the study, with a median relationship length of 16 months; 12% of couples were cohabiting. Participant ages ranged from 18 to 25 years, with a mean age of 20.4 years. The ethnic and racial composition of this sample was 2.1% African American, 23.8% Asian, 56.3% Caucasian, 14.6% Latino/Hispanic, 0.8% Native American, and 2.5% other. Participants were paid $15/hr for their participation.

Measures Optimism. The Life Orientation Test (LOT; Scheier & Carver, 1985) is an eight-item self-report measure of general outcome expectancies. Sample items include “In uncertain times, I usually expect the best” and the reverse-coded item “If something can go wrong for me, it will.” Responses range from 1 (strongly disagree) to 5 (strongly agree). We rescaled scores of all individual difference measures to percent of maximum possible (POMP) metric, which sets the theoretical range of a scale from 0 to 100. POMP scoring is a linear transformation of raw scores and thus does not affect standardized analyses, but it can aid in interpretation of raw scores by putting them on an intuitive metric (Cohen, Cohen, Aiken, & West, 1999). Actual scores on the LOT, in POMP metric, ranged from 22 to 100; means and standard deviations for the LOT and other major variables are reported in Table 2. Alpha reliability coefficients were .80 for men and .86 for women, and factor analysis indicated a unidimensional structure. All of our data analyses controlled for possible confounding due to partner similarity on optimism. However, it is worth noting that the correlation between partners’ optimism was r ⫽ .12, p ⫽ .22. In other words, there was not a strong or reliable tendency for optimists to be partnered with other optimists. Perceived support. To assess perceived support in the dating relationship, we used the Maintenance Questionnaire (MQ; Stafford & Canary, 1991). Participants rated 24 statements concerning their partner’s behaviors on a scale from 1 (strongly disagree) to 7 (strongly agree). The MQ has five subscales that cover a broad range of supportive behaviors: (a) positivity (e.g., “Does not criticize me”), (b) openness (e.g., “Encourages me to

Procedure and Participants We examined data from a study of dating couples assessed at multiple time points over a 1-year period (see Table 1). For the analyses presented in this article as Parts 1 and 2, we included couples from the original sample who completed all measures at Times 1, 2, and 3 (but not necessarily Time 4); this left us with 108 couples (N ⫽ 216) for the present report.2 In Part 1 we analyze data from Time 1, when participants completed measures of personality, social support, and the dating relationship.

2

We compared the 108 couples included in this report with the 12 couples who did not return at Times 2 or 3. Analyses indicated no differences on optimism for the men or women of these couples (rs ⬍ .07, ps ⬎ .52); however, the men in the 108 included couples were somewhat higher in perceived support (r ⫽ .26, p ⫽ .004) and higher in relationship satisfaction (r ⫽ .19, p ⫽ .04). The women in these couples did not differ significantly on those dimensions (rs ⬍ .13, ps ⬎ .16).

SRIVASTAVA, MCGONIGAL, RICHARDS, BUTLER, AND GROSS

146

Table 2 Correlations, Means, and Standard Deviations for Main Measures from Parts 1 and 2 Variable

1. MLOT

2. MMQ

3. MCSS

4. MCONV

5. MRES

6. FLOT

7. FMQ

8. FCSS

9. FCONV

10. FRES

1. MLOT 2. MMQ 3. MCSS 4. MCONV 5. MRES 6. FLOT 7. FMQ 8. FCSS 9. FCONV 10. FRES M SD

— .36 .32 .19 .21 .12 .09 .15 .16 .23 66.2 15.9

— .68 .44 .42 .21 .41 .29 .46 .25 81.4 12.1

— .49 .40 .26 .50 .36 .41 .26 80.4 16.11

— .46 .23 .51 .30 .71 .40 0.0 0.7

— .26 .33 .29 .49 .48 5.0 2.5

— .26 .27 .30 .32 66.8 17.8

— .62 .49 .23 79.1 12.8

— .31 .32 82.1 16.4

— .43 0.0 0.7

— 5.5 2.5

Note. N ⫽ 108 couples. Effect sizes greater than .20 are significant at p ⬍ .05. The first letter in the variable name indicates gender (M ⫽ male, F ⫽ female). LOT ⫽ Life Orientation Test; MQ ⫽ Maintenance Questionnaire; CSS ⫽ Couple Satisfaction Survey; CONV ⫽ positive engagement in conflict conversation (average of z-scored self-reports and partner-reports); RES ⫽ resolution of conflict.

disclose my thoughts and feelings to him/her”), (c) assurances (e.g., “Stresses his or her commitment to me”), (d) social network (e.g., “Focuses on common friends and affiliations”), and (e) sharing tasks (e.g., “Helps equally with tasks that need to be done”). The five subscales were all positively correlated (mean r ⫽ .38, ranging from .20 to .58), so we averaged the five scales and converted to POMP metric to create a global measure of perceived support from the dating partner. Scores ranged from 35 to 100. Alphas (computed at the item level) were .91 for men and .92 for women. Relationship satisfaction. To measure relationship satisfaction, we used the Couple Satisfaction Scale (CSS; Cowan & Cowan, 1990). The CSS includes eight items that are rated on scale from 1 (very dissatisfied) to 5 (very satisfied). A sample item is, “In general, how do you feel about the closeness and distance in your relationship with your partner now?” Whereas the MQ, our measure of perceived support, asks members of couples to report what their partners do, the CSS asks individuals how they feel about the relationship. CSS scores, computed in POMP metric, ranged from 9 to 100. Alphas for the CSS were .89 for men and .89 for women. Control measure: Offered support. We used a subset of 10 items from the Investment Scale (IS; Lund, 1985), which asks the participant to rate “how much you feel you have invested in your relationship in each of the following ways” on a scale from 1 (not invested) to 7 (very invested). Items were selected to match the subscales of the MQ, for example, “Trying to encourage and support your partner” (positivity), “Telling your partner your true feelings about the relationship” (openness), “Integrating your partner into your family” (social network), “Making formal agreements about your relationship, such as deciding to go steady, get engaged, or get married” (assurances), and “Doing favors for or helping your partner, such as lending money or doing errands” (tasks). The items were summed and converted to POMP metric to create a global self-report measure of offered support. Scores ranged from 39 to 100; means were 76.8 (SD ⫽ 13.2) for men and 76.7 (SD ⫽ 11.1) for women. Alphas were .80 for men and .72 for women. Control measures: Extraversion, Neuroticism, self-esteem, and demographics. Extraversion and Neuroticism were measured with eight-item scales from the Big Five Inventory (John & Srivastava, 1999). Alphas for Extraversion were .89 for men and .88 for women; alphas for Neuroticism were .77 for men and .82 for women. Self-esteem was measured with the 10-item Rosenberg Self-Esteem Scale (Rosenberg, 1965); alphas were .88 for men and .90 for women. We also measured each partner’s age, how long the couple had been together, and whether they were cohabiting. Discriminant validity among optimism and relational measures. Conceptually, the measures of optimism, perceived support, offered support, and relationship satisfaction are all supposed to measure different things.

However, it was important to establish discriminant validity; a possible counterhypothesis was that the measures simply reflected a general relational positivity factor. To test this counterhypothesis, we ran a confirmatory factor analysis in which all four of the men’s measures loaded on a latent “men’s positivity” factor, all of the women’s measures loaded on a latent “women’s positivity” factor, and the men’s and women’s factors were allowed to correlate.3 The analysis showed that the counterhypothesis did not fit the data, ␹2(19, N ⫽ 108) ⫽ 52.9, p ⬍ .001; normed fit index (NFI) ⫽ .80; root-mean-square error of approximation (RMSEA) ⫽ .13. Analyses of reduced sets of variables, created by eliminating optimism or offered support, did not show substantially better fit.

Results and Discussion For our analyses we were interested in estimating both withinperson and between-persons effects—for example, how an individual’s optimism relates to his or her own relationship satisfaction (a within-person effect) and to his or her partner’s relationship satisfaction (a between-persons effect). Both of these kinds of questions are addressed by the actor–partner interdependence model (APIM; Kashy & Kenny, 1997), a data analysis procedure for dyads. The APIM was also designed to deal with the violations of statistical independence associated with dyadic data. Thus, we adopted the APIM as our basic data-analytic strategy. The APIM estimates two kinds of effects: actor effects and partner effects. Actor effects are within-person effects: They represent the influence of an individual’s level of a predictor variable on that individual’s level of an outcome variable. Partner effects are between-person effects: They represent the influence of an individual’s level of a predictor on that individual’s partner’s level of the outcome variable. APIM estimates also control for confounding due to partner similarity. The APIM is rooted in regression (Kashy & Kenny, 1997). As with regression, it is possible to extend the APIM to include moderators, control variables, and mediators. We had a substantive interest in taking advantage of all of these possibilities. One important question was whether gender moderated the actor and partner effects. In the APIM, actor and partner effects are aggre3

We thank an anonymous reviewer for suggesting this analysis.

OPTIMISM IN CLOSE RELATIONSHIPS

gated across both members of the couple. When members of couples are distinguishable on some variable—such as gender, in the case of our heterosexual dating couples—it is possible to ask whether actor and partner effects are moderated by gender. All of the analyses we report were tested for moderator effects of gender. Unless reported otherwise, such effects were not significant, and thus results apply to both men and women.4 The basic APIM can also be elaborated to test models with multiple predictors (for control analyses) or with mediated paths. Shrout and Bolger (2003) reported that more sensitive tests of mediation can be conducted by using bootstrap analyses, as compared with other methods. Thus, we ran our analyses in Amos 4.01 (Arbuckle, 1999), which can conduct bootstrap analyses.

Do Optimists and Their Partners Report Greater Relationship Satisfaction? We expected that optimists and their partners would experience their relationships as more satisfying. To test this hypothesis, we performed an APIM analysis using optimism to predict relationship satisfaction. The results indicated that optimists reported greater relationship satisfaction. The standardized actor effect was .27, p ⬍ .001, with a 95% confidence interval (CI) ranging from .15 to .38. (The p values and CIs reported for all APIM analyses are bias-corrected values from bootstrap analyses.) Furthermore, optimists’ partners also reported greater relationship satisfaction, indicating that the positive relational consequences of optimism were not just “in the head” of the optimists: standardized partner effect ⫽ .18, p ⫽ .006, 95% CI ⫽ (.06, .30).

Does Perceived Support Mediate Relations Between Optimism and Relationship Satisfaction? Having established that optimism was related to relationship satisfaction, we then tested whether this relation was mediated by perceived support. Following Shrout and Bolger’s (2003) procedure (the logic of which is modeled on Baron & Kenny, 1986), this required four further steps. Each step must produce a significant result to proceed to the next. First, we tested whether optimism predicts perceived support. Second, we tested whether perceived support predicts relationship satisfaction when controlling for optimism. Third, we tested the mediated paths from optimism via perceived support to relationship satisfaction; a significant bootstrap test would support mediation. This bootstrap test is a more powerful replacement for the Sobel test used in conventional mediation analysis. Fourth, we tested the direct paths from optimism to relationship satisfaction when controlling for perceived support; this last step would indicate whether mediation was partial or complete. Did optimism predict perceived support? The results indicated that optimists perceived greater support from their partners: actor effect ⫽ .29, p ⬍ .001, 95% CI ⫽ (.17, .41). Optimists’ partners had marginally higher levels of perceived support: partner effect ⫽ .12, p ⫽ .07, 95% CI ⫽ (–.01, .24). Did perceived support predict relationship satisfaction? The effect of perceived support on an actor’s own relationship satisfaction was substantial: actor effect ⫽ .58, p ⫽ .001, 95% CI ⫽ (.44, .70). Individuals who perceived greater support also had more satisfied partners: partner effect ⫽ .16, p ⫽ .003, 95% CI ⫽ (.07, .27).

147

Were the mediated paths significant? The bootstrap tests indicated that the actor effect of optimism on relationship satisfaction, reported earlier, was significantly mediated by perceived support: mediated actor effect ⫽ .18, p ⫽ .001, 95% CI ⫽ (.09, .27). Likewise, the effect of optimism on a partner’s relationship satisfaction was also significantly mediated by the optimist’s perceived support: mediated partner effect ⫽ .10, p ⫽ .003, 95% CI ⫽ (.03, .18). Did the direct effects indicate full or partial mediation? If the direct effect of optimism on an actor’s own relationship satisfaction was still significant, that would indicate partial (rather than full) mediation of the actor effect. This effect was not significant: direct actor effect ⫽ .10, p ⫽ .15, 95% CI ⫽ (–.03, .24). Nor was the direct partner effect significant: direct partner effect ⫽ .07, p ⫽ .17, 95% CI ⫽ (–.03, .18). Thus, the analyses indicated the effects of an individual’s optimism on both the individual’s own relationship satisfaction and on a partner’s satisfaction could be explained by the optimist’s perceived support.

Control Analyses To ensure that the effect of optimism on global perceived support was not a result of optimists attracting more supportive partners, we conducted an APIM analysis testing the effect of optimism on perceived support while controlling for offered support. The effects of optimism were virtually unchanged: actor effect ⫽ .28, p ⫽ .001, 95% CI ⫽ (.16, .40); partner effect ⫽ .10, p ⫽ .13, 95% CI ⫽ (–.02, .22). Thus, optimists’ perceptions of their partners’ supportiveness could not be explained away by them attracting genuinely more supportive partners. We also wanted to ensure that the effects of optimism on relationship satisfaction were specific to optimism rather than being attributable to related traits. To test this, we conducted APIM analyses with covariates, controlling for individual differences in Extraversion, Neuroticism, and self-esteem, as well as both partners’ ages, the length of the relationship, and whether the couple was living together; covariates were tested one by one because of concerns about multicollinearity. Pitted against each covariate, optimism always was a significant predictor; furthermore, no covariate had a significant effect on relationship satisfaction after controlling for optimism (all absolute effects ⬍ .12; all ps ⬎ .16). Thus, we felt fairly confident that the effects of optimism on relationship satisfaction were not confounded with broader personality traits, with self-esteem, or with the demographic and background variables we examined. Part 1 thus shows that the romantic relationships of optimists are characterized by greater relationship satisfaction than the relationships of those who are less optimistic. The mediation analyses suggested that optimists’ general tendency to see their partners as supportive mediated these positive relationship outcomes. Not only did optimists report greater relationship satisfaction, but so did their (not necessarily optimistic) partners, suggesting that the 4

The APIM can be specified as a path model with equality constraints between members of the dyad; in this study, the APIM was specified by setting men’s parameter estimates equal to women’s. The unconstrained or “saturated” model produces separate parameter estimates for men and for women. Thus, the chi-square test of model fit (which compares the constrained model to a saturated model) is, in the present context, a test of moderation by gender.

148

SRIVASTAVA, MCGONIGAL, RICHARDS, BUTLER, AND GROSS

positive relationship outcomes are not merely a Pollyanna-like fantasy of the optimists. Because Part 1 was based on crosssectional data, however, the ordering of variables in the mediational model was based on conceptual considerations rather than on the design of the study. In Part 2, we adopted a design in which the temporal structure of the design gave us a stronger basis to evaluate whether the relationship benefits of optimism are explained by perceived support. Part 1 focused on global perceptions and feelings about the relationship. In Part 2, we wanted to move beyond this global level of analyses and take a closer look at the role of optimism in relationship processes. To do this, we brought the same couples from Part 1 into the laboratory and facilitated a conflict conversation. We assessed whether optimists and their partners perceived each other as supportive during the conflict, and whether that perception of support contributed to both partners’ reports of how well the conflict was resolved 1 week later.

Part 2: The Conflict Conversation In dating relationships, a common stressor is disagreement between partners, such as disagreement about finances or time spent together. How members of a couple perceive and react to disagreements can be important for the health of the relationship (Bradbury & Fincham, 1990; Carstensen, Gottman, & Levenson, 1995; Gable, Reis, & Downey, 2003). In Part 2, we examined how the couples in our study responded to conflict by bringing them into the laboratory to have a conversation about the most stressful area of current disagreement in their relationship. Immediately after the interaction, we asked each member of the couple to report how positively and constructively they had engaged in the conflict, and how positively and constructively their partner had engaged. One week later, we asked each member of the couple how well they had resolved the conflict as a couple. In measuring positive conflict engagement, we believed that it was critical to take advantage of the participants’ position as informants within their relationships to tell us about how effective they were in addressing the conflict together. In essence, we were interested in the shared reality of the relationship—the couple’s joint construal of how effectively they mutually dealt with the conflict. We hypothesized that optimists and their partners would both see the conflict as better resolved 1 week after their conversation. We also hypothesized that this resolution would be explained, at least in part, by a shared perception that both partners engaged positively and constructively during the conversation. These hypotheses were brought together in a double-mediation model. In Part 1, we found that optimists had a global tendency to perceive their partners as supportive. Now, we anticipated that this global tendency toward perceived support would be manifested in the more specific context of the conversation through positive engagement and that this positive engagement would be recognized by both partners. More positive engagement in the conflict would, in turn, lead to a better resolution to the conflict in the eyes of both partners. In examining whether optimists and their partners reported better conflict resolution, we considered the alternative explanation that optimists’ relationships might be characterized by relatively low-intensity conflicts. That is, optimists might appear to be good at resolving conflict, but only because their conflicts are relatively easy to resolve. To address this possibility, we also

examined the participants’ ratings of how intensely they disagree about various topics in their relationship, including the one discussed; this measure was taken before the conversation took place.

Method Procedure The conflict conversation procedure was modeled after that used by Carstensen et al. (1995).5 On arrival at the laboratory at Time 2, a female experimenter gave participants an overview of the study. Participants were told that the study was about “how couples talk to each other about important conflicts or areas of disagreement in the relationship.” Thus, they would need to talk to each other for 10 min and complete questionnaires concerning their reactions to the conversation. Both members of the couple had separately reported on their area of greatest current disagreement in the Time 1 questionnaire set using the Couple Problem Inventory (Gottman, Markman, & Notarius, 1977). In this questionnaire, participants indicated how much they disagree with their partner in a number of preestablished areas (money, jealousy, recreation, etc.) and also had the opportunity to list additional areas of disagreement. After rating disagreement across all areas, participants then filled in a response to the question, “Which is currently the greatest area of disagreement in your relationship?” (additional questions asked for the second and third greatest areas). Prior to the Time 2 session, the experimenter randomly selected either the male’s or female’s area of greatest disagreement. The experimenter then raised the topic and asked each partner to describe (a) more specifically how the problem area was relevant to their relationship, (b) the last time this problem came up between them, (c) his or her emotions surrounding this specific incident, and (d) why he or she experienced these emotions. If the partners’ responses indicated that the topic was not likely to be appropriate for the experiment, the experimenter selected a different area of disagreement. Disagreement in the area chosen for discussion, rated by participants on a scale from 0 (don’t disagree at all) to 100 (disagree very much), averaged 60 for men and 65 for women. The couples discussed the topic for 10 min. At the end of the conversation, participants completed questionnaires about positive conflict resolution behaviors. One week later (at Time 3), participants returned to the laboratory to complete questionnaires about the conflict topic and the conversation they had in the laboratory. Participants rated the degree to which the conflict had been resolved since the conversation.

Measures Intensity of disagreements. On the Couple Problem Inventory, each partner rated the intensity of their disagreement in each potential area of disagreement, using a scale from 0 (don’t disagree at all) to 100 (disagree very much). We computed disagreement scores by averaging ratings across the 13 preestablished areas and, if applicable, the 1 or 2 additional areas identified by the participant (for men, M ⫽ 23, SD ⫽ 14; for women, M ⫽ 23, SD ⫽ 13). Alpha reliabilities of the disagreement composite, computed across the 13 preselected topics that every participant rated, were .81 for men and .74 for women. Positive engagement in conflict. After the conversation, participants reported the extent to which they and their partners engaged in positive or

5

The data presented in this article were originally collected as part of an experimental study of emotions and physiology. For the purposes of the experimental study, physiological measurements were taken in the Time 2 laboratory session, and couples were randomly assigned to have one member suppress or reappraise his or her emotions or to a control condition (Richards, Butler, & Gross, 2003). Our present focus is on individual differences rather than the effects of the experimental manipulation, and none of the effects reported in this article interacted with experimental condition.

OPTIMISM IN CLOSE RELATIONSHIPS supportive behaviors during the conflict conversation. Sample items include, “During the conversation, to what extent were you [was your partner] a good listener?” “During the conversation, to what extent did you [your partner] try to understand your partner’s [your] point of view?” and “During the conversation, to what extent did you [your partner] criticize your partner [you]?” (reverse scored). A total of 17 items were rated on a scale from 0 (none/not at all) to 10 (a great deal/extremely). We averaged the items to create composites. Self-reports ranged from 3.8 to 9.7; partner reports ranged from 3.0 to 9.4. Means (and SDs) were as follows: men’s reports of women, 6.8 (1.3); women’s reports of men, 6.9 (1.3); men’s self-reports, 6.9 (1.2); women’s self-reports, 6.9 (1.2). Alphas were .87 for men’s reports of women, .86 for women’s reports of men, .83 for men’s reports of their own behavior, and .83 for women’s reports of their own behavior. To simplify the analyses, we created a positive engagement variable for each individual that aggregated across data sources. That is, the positive engagement variable for men was an average of the men’s self-reports with women’s reports of their male partners, and vice versa to create an aggregate for women; variables were converted to z scores before averaging. These aggregates were justified by the substantial (though not perfect) agreement between self-reports and partner reports: agreement between men’s self-reports and women’s partner reports, indexed as an alpha coefficient, was .64; agreement between women’s self-reports and men’s partner reports was .53. To make sure that we were measuring the shared reality of the relationship and not merely the positive biases of optimists, we also examined the reports that relied on a single data source (either self or partner) and attempted to replicate all of the analyses with the singlereporter variables. In the model run with partner reports, optimists’ own positive engagement would be reported by their partners, and thus the actor effects would be immune from any positive perceptual bias “in the heads” of optimists. In the model run with individuals’ construals of their own behavior, the optimists’ partners’ behavior would be reported by the partners rather than the optimists; thus the partner effects would be untainted by optimists’ internal biases. If these analyses replicated the findings with the aggregated variables, that would ensure that the effects reflected the shared reality of the relationship rather than the idiosyncratic views of one person. Conflict resolution. Two items, rated on a scale from 0 (none/not at all) to 10 (a great deal/extremely), were used to assess both partners’ feelings about how well the conflict was resolved 1 week after the conversation: “At this point, to what extent is the conflict you talked about in your previous session resolved?” and “At this point, to what extent have you and your partner moved in the right direction to resolve the conflict you talked about in your previous session?” These two items were averaged to create conflict resolution scores. Actual scores covered the full range of the scale. Alphas were .80 for men and .81 for women. Controls. As in Part 1, we examined Extraversion, Neuroticism, selfesteem, both partners’ ages, length of relationship, and cohabitation status as control variables. We also included a measure of relationship satisfaction (the CSS from Part 1) to rule out the possibility that participants were simply saying good things about how they resolved conflict because they were generally satisfied with their relationships.

Results and Discussion Did Optimists Have Less Intense Disagreements? We examined the effect of optimism on both individuals’ ratings of the intensity of disagreement in their relationships. Optimists and their partners described their disagreements as somewhat less intense: actor effect ⫽ ⫺.15, p ⫽ .02, 95% CI ⫽ (–.29, ⫺.02); partner effect ⫽ ⫺.16, p ⫽ .01, 95% CI ⫽ (–.30, ⫺.03). Thus, we included intensity of disagreement as a control variable when testing the effects of optimism on conflict resolution.

149

Did Optimists and Their Partners See the Conflict as Better Resolved? Were optimists’ relationships characterized by better positive conflict resolution, as perceived by both partners? In an APIM analysis entering optimism and intensity of disagreement simultaneously to predict conflict resolution, the actor effect of optimism was .17, p ⫽ .018, 95% CI ⫽ (.05, .31), and the partner effect of optimism was .15, p ⫽ .02, 95% CI ⫽ (.02, .28). Individuals who rated their disagreements as relatively intense did report poorer conflict resolution, actor effect ⫽ ⫺.20, p ⫽ .01, 95% CI ⫽ (⫺.35, ⫺.04); the partner effect was not significant ( p ⫽ .13). From this analysis it can be concluded that both optimists and their partners agreed that their conflicts had reached a more satisfactory resolution 1 week later and that this effect could not be explained away by baseline differences in the intensity of their disagreements.

Do Perceived Support and Positive Engagement Mediate the Benefits of Optimism? We hypothesized that optimists’ global perceived support would promote positive engagement in the conflict conversation, as recognized by both partners, and that this would explain the effects of optimism on achieving a more satisfactory conflict resolution. Part 1 already demonstrated the effect of optimism on global perceived support; here we present evidence testing the remaining elements of the double-mediation hypothesis. Did perceived support promote positive engagement? We ran APIMs testing the effects of global perceived support on the aggregated positive engagement measure. The analyses showed that individuals with greater perceived support were seen as engaging more positively in the conflict: actor effect ⫽ .32, p ⫽ .004, 95% CI ⫽ (.22, .42). In follow-up analyses in which we analyzed the self-reports and partner reports separately, this effect was significant regardless of whose reports of positive engagement we analyzed: Individuals with higher global perceived support saw themselves as engaging more positively in the conflict, and their partners saw them that way as well. The analyses also showed that an individual’s perceived support predicted the partner’s positive engagement: partner effect ⫽ .35, p ⫽ .001, 95% CI ⫽ (.26, .45). Again, the follow-up analyses indicated that this effect was significant with both data sources: Individuals who were high in global perceived support saw their partners as engaging more positively in the conflict, and their partners shared that perception. In an additional follow-up analysis that included intensity of disagreement as a control variable, perceived support still had significant actor and partner effects on positive engagement. Intensity of disagreement did not have significant actor or partner effects in this analysis ( ps ⬎ .25). Did positive engagement predict better resolution 1 week later? Individuals who engaged more positively in the conflict conversation reported better conflict resolution 1 week later: actor effect ⫽ .27, p ⫽ .002, 95% CI ⫽ (.13, .41). Their partners also saw the conflict as better resolved: partner effect ⫽ .26, p ⫽ .002, 95% CI ⫽ (.11, .40). These analyses held up regardless of the data source for the positive engagement variable. Was the effect of optimism on conflict resolution mediated by perceived support and positive conflict resolution? We tested for mediation by evaluating whether the mediated paths from opti-

SRIVASTAVA, MCGONIGAL, RICHARDS, BUTLER, AND GROSS

150

mism, through perceived support, through positive engagement, to conflict resolution were significant. The compound mediated path from optimism to the optimist’s own report of conflict resolution was significant: mediated actor effect ⫽ .07, p ⫽ .001, 95% CI ⫽ (.03, .12). The compound mediated path from optimism to the partner’s conflict resolution was also significant: mediated partner effect ⫽ .07, p ⫽ .001, 95% CI ⫽ (.04, .12). This analysis supported the mediation hypothesis. Was the mediation full or partial? In the double-mediation model, there are six ways that direct effects could have “bypassed” the mediated pathways. Optimism could have had a direct effect on an actor’s own conflict resolution or on a partner’s conflict resolution that was not mediated by perceived support or positive engagement. Optimism could have had direct actor or partner effects on positive engagement that were not mediated by perceived support. Additionally, perceived support could have had direct actor or partner effects on conflict resolution that were not mediated by positive engagement. To test these various possibilities together, we took advantage of Amos’s model-comparison capabilities to test models with full and partial mediation. In the full-mediation model, depicted in Figure 1, we allowed only effects from optimism to perceived support, perceived support to positive engagement, and positive engagement to conflict resolution. (As noted earlier, we tested for moderating effects of gender by constraining men’s and women’s paths to be equal; thus a ⫽ a’, b ⫽ b’, and so on. No gender moderation was found, so we report the results of the analysis with equality constraints.) The second model, called the partial-mediation model, was a less-restricted model that added all of the previously described indirect paths to the model depicted in Figure 1. On its own, the full-mediation model was a good fit to the data: ␹2(22, N ⫽ 108) ⫽ 25.1, p ⫽ .29, NFI ⫽ .90, RMSEA ⫽ .036. However, compared with the partial-mediation model, the fullmediation model’s fit was slightly worse: ⌬␹2(6, N ⫽ 108) ⫽ 12.7, p ⫽ .05. When we examined the individual paths in the partialmediation model, we found that all of the effects specified in the full-mediation model were still significant. In addition, however, optimism had a direct effect on a partner’s conflict resolution (i.e., its effect was partially but not wholly explained by the mediating variables): direct partner effect ⫽ .12, p ⫽ .04, 95% CI ⫽ (.01, .26).

Control Analyses We separately analyzed each link in the double-mediation model controlling for Extraversion, Neuroticism, self-esteem, part-

ners’ ages, length of relationship, cohabitation status, relationship satisfaction, intensity of disagreement, and which partner’s topic was (randomly) selected by the experimenter. All of the links in the double-mediation model remained significant when each of these control variables was included. Part 2 showed that optimism was associated not just with global relationship satisfaction but also with how well both partners perceived their engagement and resolution of a significant area of conflict in a relationship. This effect seemed to be partially driven by optimists’ tendency to perceive their partners as supportive, which not only led optimists to engage more positively in discussing the conflict (according to both optimists and their partners) but also elicited more positive engagement from their partners as well. Our emphasis on the participants’ reports of engagement in the conflict discussion and resolution of the conflict allowed us to gain valuable insights into their relationships. By asking both partners for their assessments of their own engagement, their partners’ engagement, and the conflict resolution, we were able to assess the shared social reality of these intimate relationships. Nevertheless, if optimism affects the shared reality of a relationship, then at some point that shared reality might affect outcomes that are objectively verifiable. For Part 3, we examined what is literally the ultimate relationship outcome: relationship dissolution.

Part 3: The 1-Year Follow-Up Parts 1 and 2 demonstrated that optimism was associated with relationship satisfaction and subjective conflict resolution, largely owing to the association between optimism and the perception of greater social support. These findings suggest that optimism influences relationship processes relevant to relationship maintenance and survival. In Part 3, we examined whether the dating couples were still together 1 year after the initial phases of the study. Because dating relationships are not as enduring as marriages, it was reasonable to expect that enough relationships would have ended that we could test for effects of optimism on relationship longevity. We hypothesized that optimism would be associated with relationship status at the 1-year follow-up and that this effect would be mediated by perceived support.

Method We attempted to contact all original participants via e-mail 1 year after their participation in Part 2. Because data collection for Part 2 spanned several months, we contacted couples within 1 week of the 1-year anniversary of their participation in Part 2. If neither member of a couple

Figure 1. The full-mediation model for Part 2. Error variances are not shown; men’s and women’s error variances for the same measure were allowed to covary.

OPTIMISM IN CLOSE RELATIONSHIPS responded within that week, we contacted them a second and third time, via e-mail and phone. Through this procedure, we were able to obtain relationship status information from at least one member of 101 (94%) of the couples. Analyses showed that members of responding couples and nonresponding couples did not differ significantly on measures of optimism or relationship satisfaction. We asked all participants whether they were still in an exclusive dating relationship with their partner. Couples responding yes were coded as still together at 1 year (1), and couples reporting no were coded as having broken up (0). In the responding sample, 67 couples (66%) were still together at the 1-year follow-up, and 34 had broken up.

Results and Discussion Did Optimism Predict Relationship Status at 1 Year? We hypothesized that greater optimism would be associated with a higher probability of being together at a 1-year follow-up. To test this hypothesis, we performed a logistic regression with couple as the unit of analysis, using both male and female optimism to predict the couple’s 1-year status. Greater male optimism predicted relationship survival, B ⫽ 0.03, Wald ⫽ 7.53, p ⫽ .006, but female optimism did not predict relationship survival, B ⫽ 0.01, Wald ⫽ 0.48, p ⫽ .49. To illustrate this effect, we split the couples into two groups according to the male optimism median and examined survival for each group. We found that 75% of couples with men at or above the median were still together at 1 year, contrasted with 54% of couples with men below the median.

151

satisfaction was a significant predictor of relationship longevity ( ps ⬎ .42). The effect of men’s perceived support was in the expected direction but was not statistically significant ( p ⫽ .10). Given the sample size, it should perhaps not be surprising that in a model with six moderately intercorrelated predictors, the predicted effect was only marginally significant. Part 3 showed that optimism is associated with an important social outcome: relationship survival. Intriguingly, this analysis showed a sex difference, with men’s optimism being the important predictor; this was in contrast to the other analyses, which indicated that men’s and women’s optimism did not have different effects on perceived support, relationship satisfaction, or conflict resolution processes. Why might male optimism play a more important role than female optimism in predicting relationship survival? One possible explanation has to do with the nature of men’s and women’s social support networks. A number of studies have suggested that men tend to rely more heavily on romantic partners for social support, whereas women tend to draw on a wider network of family and friends (e.g., Taylor et al., 2000; Voss, Markiewicz, & Doyle, 1999; Walen & Lachman, 2000). Thus, the tendency of male pessimists to perceive their partners as less supportive might be especially likely to produce shorter lived relationships, because for men, negative perceptions of their partners would implicate their entire support system and give them greater incentive to terminate the relationship.

General Discussion Did Perceived Support Mediate the Effects of Optimism on Relationship Longevity? To evaluate whether perceived support mediated the effect of optimism on relationship longevity, we added male and female perceived support as predictors in the logistic regression model. In this second model, male perceived support was a significant predictor, B ⫽ .06, Wald ⫽ 6.21, p ⫽ .01, the effect of male optimism was reduced, B ⫽ .02, Wald ⫽ 3.28, p ⫽ .07, and female optimism and perceived support were not significant predictors ( ps ⬎ .75). We evaluated the reduction in the effect of male optimism by computing a bootstrap confidence interval for the difference between the male optimism effect in the first and second models; the confidence interval did not include a null effect, 95% CI ⫽ (.001, .024), consistent with mediation. The effects of men’s optimism and perceived support remained significant when controlling for both partners’ Extraversion, Neuroticism, and self-esteem. Analyses controlling for relationship satisfaction were consistent with our main conclusions, though they yielded somewhat more complicated results. In a logistic regression in which men’s and women’s relationship satisfaction were the only predictors, there was a significant effect of men’s relationship satisfaction, B ⫽ .04, Wald ⫽ 7.59, p ⫽ .006, but not of women’s relationship satisfaction, B ⫽ ⫺.01, Wald ⫽ 0.27, p ⫽ .61. In a regression that included both optimism and relationship satisfaction, there were significant effects both of men’s optimism ( p ⫽ .03) and men’s relationship satisfaction ( p ⫽ .03), indicating that men’s relationship satisfaction did not account for the effect of men’s optimism on longevity. When we ran the full mediational model with a control for relationship satisfaction, the results were in the generally expected direction but were not as clear as without the controls. In this analysis, neither men’s nor women’s relationship

In a longitudinal study of dating couples, we found that optimism was associated with better relationship outcomes in a number of domains. Part 1 found that optimists and their partners both experienced greater overall relationship satisfaction; Part 2 found that optimists and their partners saw themselves and each other as engaging more positively in the conflict and as reaching a better resolution; and Part 3 found that the relationships of male optimists lasted longer than the relationships of male pessimists. Furthermore, all of the relationship consequences of optimism were mediated by optimists’ tendency to perceive their partners as supportive.

How Does Perceived Support Affect the Relational Environment? Why should optimism be an asset in close relationships? This study provided some insight into why optimism may lead to more satisfying and longer lasting relationships by identifying perceived support as a mediator. Perceived support was hypothesized to be an important relational mediator because it creates a more adaptive relational environment. We believe that perceived support probably helps relationships in a variety of ways. First, optimists’ tendency to perceive their partners as supportive may act as a buffer against negative attributions. Relationships in which individuals attribute their partners’ negative behaviors to global, stable, voluntary dispositions rather than narrow and temporary inclinations tend to be marked by lower relationship satisfaction and other maladaptive outcomes (Bradbury & Fincham, 1990). Optimists may attribute specific instances of unsupportive or ambiguous behavior to temporary and situationally limited states. Second, optimists’ positive views of their partners may prevent or interrupt

152

SRIVASTAVA, MCGONIGAL, RICHARDS, BUTLER, AND GROSS

cycles of negative reciprocity by refocusing optimists’ attention on the constructive things that their partners do and say, instead of on their partners’ negative affect (Gottman, 1998). Third, optimists may be better at acting as a “secure base” (Bowlby, 1988; Hazan & Shaver, 1987), providing their partners with a reliable source of support. As a result, optimists and their partners may be more satisfied because they feel that their relationship helps them pursue their personal goals (Brunstein et al., 1996; Kaplan & Maddux, 2002).

Optimism, Shared Reality, and Positive Illusions Did optimists and their partners benefit from positive illusions (Taylor & Brown, 1988)? In conceptualizing this study, we have sidestepped questions about accuracy and bias, instead focusing on perceptions of support and on both partners’ assessment of positive engagement and conflict resolution. Partners’ beliefs and perceptions of themselves and each other define the shared reality of a relationship, a reality that is important independent of any objective analysis of accuracy (Gable et al., 2003). In support of the value of such an approach, Part 3 suggests that shared reality can have very real consequences for the long-term success of a relationship. How might we apply a different perspective to our findings? Drawing on a positive illusions perspective, we could interpret the results in Part 1 as stemming from positive illusions that optimists hold about their relationships. In Part 2, we could conclude that such illusions drive optimists to practice and elicit “objectively” better conflict-related behavior; or alternatively, perhaps both partners share an illusion about how they handle conflict. We believe that the former interpretation is more compatible with other findings about positive illusions in close relationships (e.g., Murray & Holmes, 1997), though the present data cannot speak strongly to questions of illusion or accuracy. Relatedly, perceived support may function as a self-fulfilling prophecy: By virtue of optimists’ general tendency to see their partners as supportive, they may elicit actual support from their partners. That would explain why optimists report using social support as a coping strategy in general (Scheier, Carver, & Bridges, 1994) but not in response to specific, everyday stressors (Aspinwall & Taylor, 1992; Brissette et al., 2002): Optimists expect to receive support from others, but they do not directly ask for it.

Limitations and Future Directions Throughout this article, we have adopted the perspective that optimism leads to perceived support, which leads to positive relationship outcomes. This was based in large part on theoretical considerations: We ordered the constructs from optimism, the most general and broad-based construct, to perceived support, which is more domain-specific but still fairly broad as an individual difference (Sarason et al., 1986), to outcomes relevant to a specific relationship (Part 1) or specific events within that relationship (Parts 2 and 3), which are the most contextual. There is also a temporal logic to the order of precedence: Optimism is a meaningful construct even for an individual who has no experience or beliefs about close relationships, and general perceptions of support can preexist specific experiences in relationships. In Part 2, this temporal ordering was reflected in the design of the study and

was strengthened by an analysis controlling for Time 1 levels of relationship satisfaction. In Part 3, relationship dissolution obviously is an outcome that temporally follows all other characteristics of the relationship. The strengths of our naturalistic approach are balanced against the limitations of a nonexperimental design, however, and we acknowledge that this sequencing is not airtight. It could be argued, for example, that perceptions of support are a consequence, rather than a cause, of higher quality relationships (Metts, Geist, & Gray, 1994). From this perspective, individuals may form their perceptions of their partner’s supportiveness based on some other aspect of the relationship. We partially addressed this concern by controlling for relationship satisfaction and other relationship characteristics in Part 2, but we cannot fully rule out the possibility that some other, unmeasured feature of the relationship acted as a third variable. Our reliance on partners’ reports about the conflict conversation in Part 2 might be regarded as a double-edged sword. As implied earlier, this approach gave us insight into the shared reality of the relationship, and the results showed that both self- and partner reports from both members of the couple led to the same conclusions about conversation processes. In fact, if an “objective” observer failed to corroborate the positive conversation processes evident in optimists’ relationships, we might have cause to suspect the observer rather than the couple. Nevertheless, it would be interesting in future research to examine more objectively the specific processes that we believe are being promoted by optimists: adaptive attributions, interruption of cycles of negative reciprocity, and use of the relationship as a secure base. Such research would elucidate the mechanisms that link optimists’ positive expectations to the fulfillment of these expectations.

References Arbuckle, J. L. (1999). Amos 4.0. Chicago: SmallWaters Corporation. Aspinwall, L. G., & Taylor, S. E. (1992). Modeling cognitive adaptation: A longitudinal investigation of the impact of individual differences and coping on college adjustment and performance. Journal of Personality and Social Psychology, 63, 989 –1003. 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. Barrera, M. Jr. (1986). Distinctions between social support concepts, measures, and models. American Journal of Community Psychology, 14, 413– 455. Belsher, G., & Costello, C. G. (1991). Do confidants of depressed women provide less social support than confidants of nondepressed women? Journal of Abnormal Psychology, 100, 516 –525. Bowlby, J. (1988). A secure base: Parent– child attachment and healthy human development. New York: Basic Books. Bradbury, T. N., & Fincham, F. D. (1990). Attributions in marriage: Review and critique. Psychological Bulletin, 107, 3–33. Brissette, I., Scheier, M. F., & Carver, C. S. (2002). The role of optimism in social network development, coping, and psychological adjustment during a life transition. Journal of Personality and Social Psychology, 82, 102–111. Brunstein, J. C., Dangelmayer, G., & Schultheiss, O. C. (1996). Personal goals and social support in close relationships: Effects on relationship mood and marital satisfaction. Journal of Personality and Social Psychology, 71, 1006 –1019. Carstensen, L. L., Gottman, J. M., & Levenson, R. W. (1995). Emotional behavior in long-term marriage. Psychology and Aging, 10, 140 –149.

OPTIMISM IN CLOSE RELATIONSHIPS Carver, C. S., Kus, L. A., & Scheier, M. F. (1994). Effects of good versus bad mood and optimistic versus pessimistic outlook on social acceptance versus rejection. Journal of Social and Clinical Psychology, 13, 138 – 151. Chang, E. C. (2001). Optimism and pessimism: Implications for theory, research, and practice. Washington, DC: American Psychological Association. Cohen, P., Cohen, J., Aiken, L. S., & West, S. G. (1999). The problem of units and the circumstance for POMP. Multivariate Behavioral Research, 34, 315–346. Coleridge, S. T. (1881). Reason for love’s blindness. In Bartlett’s familiar quotations. Retrieved from http://www.bartleby.com/100/ Cowan, C. P., & Cowan, P. A. (1990). Couple Communication Questionnaire. Unpublished questionnaire, University of California. Cozarelli, C. (1993). Personality and self-efficacy as predictors of coping with abortion. Journal of Personality and Social Psychology, 65, 1224 – 1236. Dehle, C., Larsen, D., & Landers, J. E. (2001). Social support in marriage. American Journal of Family Therapy, 29, 307–324. Dougall, A. L., Hyman, K. B., Hayward, M. C., McFeeley, S., & Baum, A. (2001). Optimism and traumatic stress: The importance of social support and coping. Journal of Applied Social Psychology, 31, 223–245. 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. Geers, A. L., Reilley, S. P., & Dember, W. N. (1998). Optimism, pessimism, and friendship. Current Psychology: Developmental, Learning, Personality, Social, 17, 3–19. Gottman, J. M. (1998). Psychology and the study of the marital processes. Annual Review of Psychology, 49, 169 –197. Gottman, J. M., Markman, H., & Notarius, C. (1977). The topography of marital conflict: A sequential analysis of verbal and nonverbal behavior. Journal of Marriage and the Family, 39, 461– 477. Hazan, C., & Shaver, P. (1987). Romantic love conceptualized as an attachment process. Journal of Personality and Social Psychology, 52, 511–524. Helgeson, V. S. (1994). The effects of self-beliefs and relationship beliefs on adjustment to a relationship stressor. Personal Relationships, 1, 241–258. 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: Theory and research (2nd ed., pp. 102–138). New York: Guilford Press. Kaplan, M., & Maddux, J. E. (2002). Goals and marital satisfaction: Perceived support for personal goals and collective efficacy for collective goals. Journal of Social and Clinical Psychology, 21, 157–164. Kashy, D. A., & Kenny, D. A. (1997). The analysis of data from dyads and groups. In H. Reis & C. M. Judd (Eds.), Handbook of research methods in social psychology (pp. 451– 477). New York: Cambridge University Press. Lakey, B., & Cassady, P. B. (1990). Cognitive processes in perceived support. Journal of Personality and Social Psychology, 59, 337–348. 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. Lund, M. (1985). The development of investment and commitment scales for predicting continuity of personal relationships. Journal of Social and Personal Relationships, 2, 3–23. Metts, S., Geist, P., & Gray, J. L. (1994). The role of relationship characteristics in the provision and effectiveness of supportive messages

153

among nursing professionals. In B. R. Burleson, T. L. Albrecht, & I. G. Sarason (Eds.), Communication of social support (pp. 113–135). Thousand Oaks, CA: Sage. Murray, S. L., & Holmes, J. G. (1997). A leap of faith? Positive illusions in romantic relationships. Personality and Social Psychology Bulletin, 23, 586 – 604. Newcomb, M. D. (1990). What structural equation modeling can tell us about social support. In B. R. Sarason, I. G. Sarason, & G. R. Pierce (Eds.), Social support: An interactionist view (pp. 26 – 63). New York: Wiley. Park, C. L., & Folkman, S. (1997). Stability and change in psychosocial resources during caregiving and bereavement in partners of men with AIDS. Journal of Personality, 65, 421– 447. Pierce, G. R., Sarason, B. R., & Sarason, I. G. (1992). General and specific support expectations and stress as predictors of perceived supportiveness: An experimental study. Journal of Personality and Social Psychology, 63, 297–307. Richards, J. M., Butler, E. A., & Gross, J. J. (2003). Emotion regulation in romantic relationships: The cognitive consequences of concealing feelings. Journal of Social and Personal Relationships, 20, 599 – 620. Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University Press. Sarason, I. G., Levine, H. M., Basham, R. B., & Sarason, B. R. (1983). Assessing social support: The Social Support Questionnaire. Journal of Personality and Social Psychology, 44, 127–139. Sarason, I. G., Sarason, B. R., Shearin, E. N. (1986). Social support as an individual difference variable: Its stability, origins, and relational aspects. Journal of Personality and Social Psychology, 50, 845– 855. Scheier, M. F., & Carver, C. S. (1985). Optimism, coping, and health: Assessment and implications of generalized outcome expectancies. Health Psychology, 4, 219 –247. Scheier, M. F., Carver, C. S., & Bridges, M. W. (1994). Distinguishing optimism from neuroticism (and trait anxiety, self-mastery, and selfesteem): A reevaluation of the Life Orientation Test. Journal of Personality and Social Psychology, 67, 1063–1078. Scheier, M. F., Carver, C. S., & Bridges, M. W. (2001). Optimism, pessimism, and psychological well-being. In E. C. Chang (Ed.), Optimism and pessimism: Implications for theory, research, and practice (pp. 189 –216). Washington, DC: American Psychological Association. Shrout, P. E., & Bolger, N. (2003). Mediation in experimental and nonexperimental studies: New procedures and recommendations. Psychological Methods, 7, 422– 445. Stafford, L., & Canary, D. J. (1991). Maintenance strategies and romantic relationship type, gender and relational characteristics. Journal of Social and Personal Relationships, 8, 217–242. Taylor, S. E., & Brown, J. D. (1988). Illusion and well-being. A social psychological perspective on mental health. Psychological Bulletin, 103, 193–210. Taylor, S. E., Kemeny, M. E., Reed, G. M., Bower, J. E., & Gruenewald, T. L. (2000). Psychological resources, positive illusions, and health. American Psychologist, 55, 99 –109. Voss, K., Markiewicz, D., Doyle, A. B. (1999). Friendship, marriage and self-esteem. Journal of Social and Personal Relationships, 16, 103–122. Walen, H. R., & Lachman, M. E. (2000). Social support and strain from partner, family, and friends: Costs and benefits for men and women in adulthood. Journal of Social and Personal Relationships, 17, 5–30.

Received August 11, 2003 Revision received February 6, 2006 Accepted February 7, 2006 䡲

SOCIAL EXCLUSION AND PAIN ness, lethargy, lack of emotion, and self-awareness. Journal of Personality and Social Psychology, 85, 409 – 423. 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. Van Boven, L., & Loewenstein, G. (2003). Social projection of transient drive states. Personality and Social Psychology Bulletin, 29, 1159 – 1168. 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. Wiedenmayer, C. P., Goodwin, G. A., & Barr, G. A. (2000). The effect of periaqueductal gray lesions on responses to age-specific threats in infant rats. Developmental Brain Research, 120, 191–198. 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.

15

Wilson, T. D., & Gilbert, D. T. (2003). Affective forecasting. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 35, pp. 345– 411). San Diego, CA: Academic Press. Wilson, T. D., Wheatley, T., Kurtz, J., Dunn, E., & Gilbert, D. T. (2004). When to fire: Anticipatory versus post-event reconstrual of uncontrollable events. Personality and Social Psychology Bulletin, 30, 340 –351. 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. Zadro, L., Williams, K. D., & Richardson, R. (2004). How low can you go? Ostracism by a computer is sufficient to lower self-reported levels of belonging, control, self-esteem, and meaningful existence. Journal of Experimental Social Psychology, 40, 560 –567.

Received February 24, 2005 Revision received August 15, 2005 Accepted August 25, 2005 䡲

Journal of Personality and Social Psychology 2006, Vol. 91, No. 1, 154 –170

Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.154

Discrepancies Between Explicit and Implicit Self-Concepts: Consequences for Information Processing Pablo Brin˜ol

Richard E. Petty

Universidad Auto´noma de Madrid

Ohio State University

S. Christian Wheeler Stanford University Individuals with discrepancies among their explicit beliefs often engage in greater elaboration of discrepancy-related information in a presumed attempt to reduce the discrepancy. The authors predicted that individuals with discrepancies between explicit and implicit self-conceptions might similarly be motivated to engage in processing of discrepancy-related information, even though they might not be aware of the discrepancy. Four studies were conducted in which various self-dimensions were assessed with explicit and implicit measures. Across several different self-dimensions (e.g., need to evaluate, self-esteem), the authors found that as the discrepancy between the explicit and implicit measure increased (regardless of direction), people engaged in more thinking about information framed as related to the self-dimension on which the discrepancy existed. This research suggests that individuals might be motivated to examine relevant information as a strategy to minimize the implicit doubt that accompanies an inconsistency between explicit and implicit self-conceptions. Keywords: discrepancy, explicit–implicit, self, persuasion, attitude change

subjective discomfort that results from the discrepancy (e.g., Ha¨nze, 2001; Hodson, Maio, & Esses, 2001; Jonas, Diehl, & Bromer, 1997; Katz, Wackenhut, & Hass, 1986). For example, Maio, Bell, and Esses (1996) measured participants’ ambivalence regarding the issue of immigration to Canada (i.e., the extent to which they had both positive and negative reactions to the issue) and then exposed them to a discrepancyrelated message favoring immigration from Hong Kong to Canada that contained either strong or weak arguments. The extent to which participants processed the message information was assessed by examining the extent to which the quality of the arguments affected postmessage immigration attitudes (Petty, Wells, & Brock, 1976). When people are thinking carefully about information, they should be affected by the quality of the arguments a message contains (see Petty & Cacioppo, 1986). As expected, Maio et al. (1996) found that individuals who had ambivalent attitudes toward immigration were more influenced by argument quality than were unambivalent individuals, suggesting that they engaged in enhanced scrutiny of the information. Although research has focused extensively on explicit discrepancies, relatively little work has examined the potential existence of and consequences of discrepancies in which one cognitive element may not be easily reported. Theory suggests that just as people can hold conscious, explicit self-beliefs, they may also hold less conscious (or explicitly denied) automatic self-associations that can conflict with the more consciously endorsed ones (e.g., Greenwald et al., 2002). The present research examines the information-processing consequences of discrepancies between self-conceptions as assessed with explicit and implicit measures of individual differences.

The psychological literature has clearly documented that people can simultaneously hold incompatible explicit beliefs, attitudes, feelings, and behavioral tendencies regarding oneself and others (e.g., Abelson & Rosenberg, 1958; Bem & Allen, 1974; Brehm & Cohen, 1962; Cacioppo, Gardner, & Berntson, 1997; Heider, 1958; Higgins, 1987; Kaplan, 1972; Newcomb, 1968; Norton, 1975; Osgood & Tannenbaum, 1955; Priester & Petty, 1996). Virtually every relevant theory holds that such internal discrepancies tend to be unpleasant and can result in psychologically undesirable outcomes (e.g., Campbell, 1990; Carver & Scheier, 1990; Greenier, Kernis, & Waschull, 1995; Higgins, 1987; Kernis & Waschull, 1995). Because of this, people often attempt to resolve these internal discrepancies. Perhaps the most common approach to addressing discrepancy is enhanced thinking or information processing (e.g., Abelson et al., 1968; Aronson, 1969; Festinger, 1957; Heider, 1958; Hass, Katz, Rizzo, Bailey, & Moore, 1992). By considering additional information, individuals may hope to gain enough information for one or the other side of the discrepancy to resolve or minimize the inconsistency, or at least the

Pablo Brin˜ol, Department of Social Psychology and Methodology, Universidad Auto´noma de Madrid, Madrid, Spain; Richard E. Petty, Department of Psychology, Ohio State University; S. Christian Wheeler, Graduate School of Business, Stanford University. We thank all of the members of the 2000 –2004 Groups for Attitudes and Persuasion at Ohio State University for providing feedback on the experiments reported in this article. We also thank Javier Horcajo and Isamel Gallardo for their help with data collection. Correspondence concerning this article should be addressed to Richard E. Petty, Department of Psychology, Ohio State University, 1835 Neil Avenue Mall, Columbus, OH 43210-1222. E-mail: [email protected] 154

EXPLICIT–IMPLICIT DISCREPANCY

155

The Relationship Between the Implicit and the Explicit Self-Concept

Simultaneous Activity of Explicit–Implicit Self-Discrepant Dimensions

Compared with an explicit or conscious self-conception, an implicit self-view is generally defined as one that is held outside of conscious awareness, or at least is an association that is not endorsed consciously (Fazio & Olson, 2003; Petty, Wheeler, & Tormala, 2003). Such self-conceptions can influence judgment and action automatically without the need for reflection and may only be apparent on disguised or implicit measures. Considerable research has examined the predictive utility of such measures in the domains of motivation (e.g., McClelland, 1985), memory (e.g., Lindsay & Johnson, 1989), personality (e.g., Bornstein, 1995), self-esteem (e.g., Greenwald & Banaji, 1995; Hetts, Sakuma, & Pelham, 1999), attitudes (e.g., Greenwald, McGhee, & Schwartz, 1998; Fazio, Jackson, Dunton, & Williams, 1995), and stereotypes (e.g., Blair & Banaji, 1996; Dovidio, Kawakami, & Beach, 2001). Currently the most popular implicit measures of self-conception are based on reaction times, for example, the Implicit Association Test (IAT; Greenwald et al., 1998) and the Automatic Evaluation Task (Fazio et al., 1995). These measures capitalize on automatic associations between the self and other constructs—reactions that may occur too quickly to come under deliberative control (see Petty, Fazio, & Brin˜ol, in press, for reviews).1 Sometimes, what is assessed with implicit and explicit measures is the same, suggesting that there is just one underlying motive, attitude, personality, or self-conception that is open to conscious awareness. However, explicit and implicit assessments are not always congruent. Although there are various explanations for this incongruency, some researchers have developed theoretical frameworks that account for the divergence by proposing that explicit and implicit constructs sometimes stem from two independent systems that operate in different contexts and influence different types of behavior. For example, Dovidio et al. (2001) have argued that self-report (explicit) and response latency (implicit) measures of attitudes can tap into different evaluations that predict behaviors in different situations (spontaneous vs. deliberative; see also Carlston & Skowronski, 1986; Gawronski & Bodenhausen, in press; Wilson, Lindsey, & Schooler, 2000). A similar distinction has been made in the domain of motivation (see, e.g., McClelland, 1985; Patten & White, 1977). Indeed, Carver (2005) documented that psychologists have promulgated a wide array of dual system approaches (e.g., Lieberman, 2000; Metcalfe & Mischel, 1999; Shastri & Ajjanagadde, 1993; Sloman, 1996; Smith & DeCoster, 2000; Smolensky, 1988). Although the various frameworks differ in some details, each has tended to emphasize the fact that some behavior is guided in a relatively automatic and unconscious way, whereas other behavior is guided in a more deliberative and conscious way. For example, cognitive– experiential self-theory (Epstein, 1973, 2003) argues that humans are characterized by two independent, interactive systems of thinking that jointly determine behavior: a preconscious experiential system and a conscious rational system. Similarly, the reflective–impulsive model (Strack & Deutsch, 2004) postulates two systems— one thoughtful and one more automatic—that operate in parallel and interact with one another.

If discrepant explicit and implicit constructs always operated in different situations (e.g., spontaneous vs. deliberative), then they should lead to little conflict in any given situation. But if the two self-conceptions were ever jointly activated or operative, there would be the possibility of some consequences that mirror incongruency between discrepant explicit constructs (Petty et al., 2006). In fact, there is some evidence suggesting that possessing discrepant implicit and explicit self-concepts can be consequential. For example, Shedler, Mayman, and Manis (1993) studied individuals who reported minimal emotional disturbance on Eysenck’s Neuroticism scale (explicit measure) but who were simultaneously judged on the basis of their early memories (implicit measure) to be relatively disturbed. In comparison with participants in the genuine self-esteem group (who scored as healthy on both measures), those with explicit–implicit discrepancies were significantly more reactive on a combined index of heart rate and blood pressure and scored higher on behavioral indices of anxiety (see also Weinberger & Haradaway, 1990). In conceptually similar research, Zelenski and Larsen (2003) found that having incongruent explicit (e.g., self-ratings) and implicit (measured by the Thematic Apperception Test; Proshansky, 1943) motive profiles was associated with reduced emotional well-being. Other recent research has demonstrated that people who scored relatively high on an explicit measure of self-esteem, but relatively low on an implicit measure (the IAT), exhibited the most self-aggrandizement across different indices (Bosson, Brown, Zeigler-Hill, & Swann, 2003), which is the main characteristic of a narcissistic personality (e.g., Wing & Gough, 1990). Additionally, individuals with the combination of relatively high scores on explicit measures of self-esteem and relatively low scores on implicit measures have been shown to be particularly defensive (Jordan, Spencer, Zanna, Hoshino-Browne, & Correll, 2003). Finally, we have found that discrepancy between explicit and implicit self-esteem scores is associated with implicit but not explicit self-doubt (Brin˜ol, Petty, & Wheeler, 2003). Specifically, as explicit–implicit self-esteem discrepancy (as assessed using the absolute value of the difference between participants’ standardized explicit and implicit self-esteem scores) increased, the strength with which participants automatically associated doubt words with self-words on an IAT also increased. However, increased discrepancy was not associated with explicit reports of self-uncertainty. Similarly, recent research has shown that when people’s attitudes change on explicit measures, they show more doubt on an implicit but not explicit measure of confidence regarding the attitude object compared with people whose attitudes have not changed (Petty et al., 2006). Because recent research also shows that when attitudes 1

When not compatible with the endorsed (explicit) self-conception, the self-associations measured with contemporary implicit measures can represent many things (e.g., repressed evaluations of oneself, prior evaluations of oneself, others’ evaluations of oneself, hopes for oneself, strong societal prescriptions for self-conduct, and so forth). We postulate that regardless of the particular source of the association, inconsistency with one’s consciously endorsed self-view can lead to psychological conflict (see Petty, in press; Petty, Tormala, Brin˜ol, & Jarvis, 2006; Priester & Petty, 2001).

156

˜ OL, PETTY, AND WHEELER BRIN

change, a residue of the prior attitude may still be available on an implicit (automatic) measure (e.g., Gregg, Seibt, & Banaji, 2006), the enhanced implicit doubt that accompanies attitude change is plausibly due to a discrepancy between old (implicit/automatic) and new (explicit/deliberative) evaluations. Together, these studies suggest that having discrepant explicit and implicit self-dimensions is associated with numerous consequences that often appear to be negative, unpleasant, or dysfunctional. Because of this, people with such discrepancies should be motivated to engage in some discrepancy resolution. Specifically, we postulate that like possession of an explicit discrepancy (e.g., attitudinal ambivalence), possession of a discrepancy between implicit and explicit self-conceptions will be associated with attempts at discrepancy reduction, and that this would occur regardless of the direction of the discrepancy. As described earlier, by thinking about information presumed relevant to the issue on which there is a discrepancy, individuals with explicit–implicit discrepancies can possibly restore internal consistency. Thus, the purpose of the present research was to examine the influence of explicit–implicit discrepancies on information processing. We examined conditions under which the discrepancy results both from one’s explicit self-conception being relatively higher (more positive) than one’s implicit self-conception within the sample distribution as well as from one’s explicit self-conception being relatively lower (less positive) than one’s implicit self-conception.

Overview of the Present Research We hypothesized that to reduce a discrepancy between explicit and implicit self-conceptions, individuals with such discrepancies would engage in more effortful elaboration of information presumed relevant to the self-dimension on which the discrepancy exists. To test this hypothesis, we conducted four studies in which explicit and implicit assessments of the self along four different dimensions were collected. In each study we used a standard self-report as the explicit measure and an IAT as the implicit measure. The IAT was selected because (a) it taps into automatic associations that are less subject to conscious control and (b) previously validated procedures were available for some of the dimensions of interest. We then examined the impact of explicit– implicit divergence on the extent of processing of information framed as discrepancy related. The logic was similar across the four studies. In each study, an index of the relative explicit–implicit discrepancy was formed for each participant following procedures used in prior research (see Kehr, 2004, for a discussion). Then, all participants were exposed to a persuasive message. The extent to which participants processed the information in the message was assessed by using an information quality manipulation. As noted earlier, the impact that the quality of information in a message has on resulting attitudes is a widely used indicator of the extent to which individuals carefully attend to and think about the information to which they are exposed (Petty & Cacioppo, 1986). Finally, across studies we varied whether the information contained in the message was actually relevant (Study 1) or simply framed as relevant (Studies 2, 3, and 4) to the self-dimension on which discrepancy existed. In two studies (Studies 3 and 4), we manipulated the extent to which participants expected the message to be relevant to the issue on which discrepancy existed, even

though the message actually contained discrepancy-unrelated information. This manipulation was expected to moderate the impact of discrepancy on information processing. If thinking is aimed at reducing the discrepancy, then explicit–implicit incongruence should be associated with enhanced thinking only when participants expect the message to be related to the self-dimension on which there is a discrepancy (i.e., when it presumably could help reduce the discrepancy). When the same information is framed as unrelated, it should be useless in resolving the discrepancy. Thus, in studies in which the information was presumed relevant to the discrepancy (Studies 1 and 2), we predicted a two-way interaction on attitudes between size of the explicit–implicit discrepancy and information quality. Information quality should have a greater impact on attitudes as discrepancy increases. In studies in which we manipulated the presumed relevance of the information to the discrepancy, a three-way interaction on attitudes was expected. Specifically, discrepancy should affect information processing primarily when the information is framed as relevant. We did not expect these results to depend on the direction of the discrepancy (i.e., whether implicit scores were relatively higher or lower than explicit scores). To the extent these hypotheses are supported, it would suggest that the presence of discrepant explicit and implicit self-dimensions, regardless of direction, is associated with enhanced thinking. The enhanced thinking presumably reflects an attempt at discrepancy reduction.

Experiment 1 Experiment 1 was conducted to provide an initial test of the notion that divergence between explicit and implicit self-views can influence the extent of processing of information relevant to those self-views. In our first experiment, we used the trait of shyness as a specific dimension of personality. There were several reasons for this selection. First, shyness is a personality trait that is well represented in common language and lay psychology, is easily judged by oneself, and is readily observable by others (Asendorpf, 1987, 1989). Second, because of its observability and ubiquity in lay psychology, the self-concept dimension of shyness is relatively accessible (Asendorpf, 1990). Third, it is easy to select shynessdescriptive adjectives for both explicit self-ratings and an IAT for shyness, because previous research has pretested instruments for its assessment (Asendorpf, 1987, 1989). Last, research conducted by Asendorpf, Banse, and Mu¨cke (2002) has demonstrated that the implicit measure of shyness uniquely predicted spontaneous (but not controlled) shy behavior, whereas the explicit ratings uniquely predicted controlled (but not spontaneous) shy behavior. That is, Asendorpf et al. validated the implicit and explicit shyness measures used in Study 1 as useful independent indicators of behavior. The explicit shyness measure consisted of a series of selfreported responses to shyness-related adjectives, whereas the implicit measure was a shyness IAT (see Asendorpf et al., 2002). In this study, all of the participants were exposed to a persuasive message containing either strong or weak arguments directly related to shyness. After reading the message, participants were asked to report their attitudes toward the proposal in the message. We expected participants with a large explicit–implicit discrepancy to be more attentive to the persuasive message than those with a small discrepancy. This enhanced thinking would be evi-

EXPLICIT–IMPLICIT DISCREPANCY

denced in greater attitudinal responsiveness to the argument quality manipulation.

Method Participants and Design Eighty-one undergraduate psychology students at the Universidad Auto´noma de Madrid in Madrid, Spain, participated in partial fulfillment of a course requirement. The students were randomly assigned to the argument quality conditions (strong or weak). Additionally, explicit and implicit shyness were assessed for all the participants to form an index of explicit– implicit discrepancy. The independent variables thereby constituted an Argument Quality (strong vs. weak) ⫻ Explicit–Implicit Discrepancy (continuously scored) ⫻ Direction of Discrepancy (higher on explicit or implicit measure) design.

Procedure Upon arrival, participants were seated at individual computer stations and completed the IAT for shyness. After the IAT task, all of the participants reported their explicit shyness and completed several other ancillary questions. Participants were then told that because of extra time remaining in the session, they would also be participating in another experiment designed to examine personality characteristics that were good for being a psychologist. Participants received a message arguing that being shy was a positive trait for a psychologist. The message contained either strong or weak arguments. All of the participants were told in advance that the message they were about to read had to do with shyness. After reading the message, participants were told that it was important to know what their personal views were on the benefits of shyness. Thus, they completed a measure of their attitudes toward shyness as a trait. Finally, the participants were thanked, asked for permission for their responses to be analyzed (all gave permission), and given an appointment for a meeting to provide them with feedback about their results.

Independent Variables Argument quality. Participants were exposed to a message containing information directly relevant to shyness and thus to any explicit–implicit discrepancy on this trait. The shyness-related message participants received contained either strong or weak arguments. This manipulation was designed to assess the extent to which people were attentive to the content of the message (Petty & Cacioppo, 1986). The arguments selected were pretested and were shown to produce the appropriate pattern of cognitive responding. That is, the strong arguments elicited mostly favorable thoughts and the weak arguments elicited mostly unfavorable thoughts when people were instructed to think carefully about them. In brief, 58 students were asked to list their thoughts for each version of the message. Analysis of the thoughts listed revealed that, on average, the strong message elicited more favorable (M ⫽ 1.86, SD ⫽ 1.80) than unfavorable (M ⫽ 0.65, SD ⫽ 1.07) thoughts, t(28) ⫽ 5.54, p ⬍ .001. For the weak version of the message, participants generated more unfavorable (M ⫽ 2.58, SD ⫽ 1.93) than favorable (M ⫽ 0.86, SD ⫽ 1.57) thoughts, t(29) ⫽ 7.19, p ⬍ .001. The gist of one strong argument in favor of shyness was that shy people have more introspective ability, a quality that was highly valuable in the workplace. The gist of another strong argument was that shy people have been rated as better friends and partners, because they tend to have interpersonal relationships that are more sincere, committed, stable, and satisfactory. In contrast, the gist of one weak argument in favor of shyness was that shy people tend to talk less than extraverted interviewers, making other shy people feel more comfortable. The gist of another weak argument was that some students’ parents prefer that their sons and

157

daughters choose shy people as friends and roommates in college because they tend to be more self-controlled. It is important to note that both the strong and weak arguments argued in favor of shyness, but the strong arguments provided more compelling reasons than did the weak arguments. This manipulation can be distinguished from other forms of message variations, such as arguing either in favor of or against the proposal. Because the argument manipulation is used to assess how much thinking people are doing about the message, all arguments need to argue for the same position— but only with high or low convincingness. Because both sets of arguments are ostensibly in favor of the issue, they may be equally persuasive if people do not think about their implications. Individuals not thinking about the message carefully may respond simply to the number of arguments presented or their initial gut reaction to the proposal (e.g., Petty & Cacioppo, 1984; see also Petty & Wegener, 1998). The more attention paid to the information provided, however, the greater the difference in subsequent attitudes to strong versus weak arguments. Explicit measure of shyness. Shyness was assessed by asking participants to rate the extent to which 10 adjectives described them on 4-point scales anchored by extremely uncharacteristic of me (1) and extremely characteristic of me (4). The adjectives were presented in random order and included words such as inhibited, insecure, timid, reticent, reserved, daring, candid, open, secure, and assertive. These words were taken from Asendorpf et al. (2002) and were also the items used for the implicit measure in the present research. Ratings on the scale items were highly consistent with each other (␣ ⫽ .86) and were averaged (reverse scoring where appropriate) to form a single index of explicit shyness for each participant. Implicit measure of shyness. In the IAT measure of shyness, participants classified target concepts (represented by me or other) and attributes (represented by shy or nonshy categories of words) using two designated keys on a computer. The me category was represented by the words I, self, my, me, and own, whereas the other category was represented by the words they, them, your, you, and other. Attributes related to shyness were selected from the items on the explicit measure and included the words inhibited, insecure, timid, reticent, and reserved. In contrast, nonshy attributes included the words daring, candid, open, secure, and assertive. The test was similar to the original IATs used by Greenwald et al. (1998) and paralleled the one used by Asendorpf et al. (2002). There were seven blocks of trials. Blocks 1, 2, and 5 were practice blocks for which participants made single categorizations (me vs. other, and shy vs. nonshy). In the remaining blocks, participants discriminated shy versus nonshy words and me versus other words on separate trials within the same block. In Block 4, participants used one response key to indicate if a word belonged to the nonshy or other categories and the other key if the word belonged to the shy or me categories. In Block 7, participants used one response key to indicate if a word belonged to the nonshy or me categories and the other key if the word belonged to the shy or other categories. Blocks 3 and 6 were combined blocks that served as practice for Blocks 4 and 7. Only data from Blocks 4 and 7 were used to compute IAT scores. The main dependent variable (IAT score) was computed by subtracting participants’ average response latencies during Block 4 from their average response latencies during Block 7. Positive differences in this index indicated faster automatic associations between me and shy than between others and shy. It is evident from the above description that the IAT is a relative measure. For example, in the present studies, the category me is contrasted with the category others. Indeed, opposing me with others might make the IAT scores difficult to interpret in certain cases because they can reflect associations with the self, with the others, or a combination of the two (e.g., Karpinski, 2004). Particularly relevant to the present research, however, the explicit self-report scale also likely involved a similar relative judgment compared with others. Most subjective judgments about the self or others require reference to some standard—typically provided by others. For example, as Mussweiler (2003) noted, “To characterize oneself as athletic

˜ OL, PETTY, AND WHEELER BRIN

158

. . . implies that one is more athletic than others and is thus, in essence, a comparative statement” (p. 472). In fact, it would be hard for people to develop their self-conceptions without comparing with others (Festinger, 1954). Thus, it is plausible to argue that both the implicit and the explicit measures of shyness might include a comparative component. Following typical procedures, stimuli in the IAT task appeared within a centered white window. Reminder labels were positioned on top of the stimuli on the left and right side. These reminders read me and other for single target-classification blocks, shy and nonshy for single attributeclassification blocks. Mixed target ⫹ attribute blocks were also accompanied by appropriate labels (e.g., shy or other; nonshy or me). Incorrect classifications were followed by error feedback (i.e., the word ERROR). Summary feedback was provided at the end of each practice block informing participants about their average response latency and percentage of errors for that block. All practice trials in the IAT were administered in five blocks. Data-collection trials, consisting of combined target ⫹ attribute classifications, were collected in four blocks. Within each block, stimuli were randomly selected without replacement, and no more than two consecutively presented stimuli belonged to the same category. To correct for anticipatory responses and momentary inattention, we recorded latencies shorter than 300 ms and longer than 3,000 ms as 300 and 3,000 ms, respectively. Response latencies were then log transformed to normalize the distribution. Further details about the IAT procedure are provided by Greenwald et al. (1998). Explicit–implicit discrepancy. The explicit and implicit measures of shyness were significantly correlated (r ⫽ .40, p ⬍ .001). This correlation is similar to those obtained by Asendorpf et al. (2002) using similar explicit and implicit measures in a different sample.2 An index of explicit–implicit discrepancy was formed as the absolute value of the difference between the standardized explicit and implicit measures of shyness. The discrepancy index considers where people fall within the distribution of participants in the study on the implicit versus explicit measures. A zero on the index indicates that the person’s place in the distribution is exactly the same on the implicit and explicit measures (e.g., high in the distribution on both, low in the distribution on both, middling on both, and so forth). Discrepancies can be in either direction. That is, people can be higher in the sample distribution on the explicit measure than the implicit measure (a positive discrepancy) or they can be lower in the distribution on the explicit measure than the implicit measure (a negative discrepancy). In this study, 38 participants had positive discrepancies and 39 participants had negative discrepancies. In addition to including size of the discrepancy, analyses also included a factor for whether the discrepancy was positive or negative.

Dependent Measure: Attitudes After reading the message relevant to shyness, participants’ attitudes toward shyness were assessed using a single-item 9-point semantic differential scale ranging from bad (1) to good (9) on which they rated how favorably they viewed shyness.

Results Attitudes toward shyness were submitted to a hierarchical regression analysis, with extent of discrepancy (continuous variable), manipulated argument quality (strong–weak; dummy coded), and direction of discrepancy (explicit ⬎ implicit vs. explicit ⬍ implicit; dummy coded) as the independent variables. Scores on the discrepancy index were centered by subtracting the mean from each person’s score (Aiken & West, 1991). Main effects were interpreted in the first step of the regression, the two-way interactions in the second step, and the three-way interaction in the final step (Cohen & Cohen, 1983). Responses to the attitude scales were scored so that higher values represented more favorable opinions toward shyness.

Results of this analysis revealed a significant main effect for extent of discrepancy, ␤ ⫽ ⫺.25, t(73) ⫽ ⫺2.19, p ⫽ .03, showing that participants’ attitudes were less favorable toward shyness as explicit–implicit discrepancy increased (all ␤s reported are standardized coefficients). In addition, a significant interaction between argument quality and direction of the discrepancy emerged, ␤ ⫽ .43, t(70) ⫽ 2.42, p ⫽ .02, revealing that argument quality had a larger effect on attitudes for the explicit ⬎ implicit direction than the implicit ⬎ explicit direction. This effect was not obtained in any of the other studies and is not discussed further. More critical to our hypothesis, a significant interaction between argument quality and extent of discrepancy was evident, ␤ ⫽ .46, t(70) ⫽ 3.17, p ⫽ .002, revealing that as discrepancy increased, argument quality had a larger effect on attitudes. Specifically, decomposition of this interaction by recentering discrepancy at one standard deviation (SD) above and below the mean (Aiken & West, 1991) indicated that there was a significant effect of argument quality among participants with high discrepancy, ␤ ⫽ .36, t(69) ⫽ 2.37, p ⫽ .02, but not among those with relatively low discrepancy, ␤ ⫽ ⫺.25, t(69) ⫽ ⫺1.69, p ⫽ .10. The three-way interaction between argument quality, extent of discrepancy, and direction of the discrepancy was not significant, ␤ ⫽ .26, t(69) ⫽ 1.14, p ⫽ .26, indicating that the effects of explicit–implicit discrepancy on elaboration were not restricted to any particular direction of the discrepancy.

Discussion Experiment 1 demonstrated that as the discrepancy between implicit and explicit measures of shyness increases, people are more likely to think carefully about a shyness-related persuasive message. This conclusion was supported by the finding that the attitudes of relatively discrepant individuals were more reflective of the quality of the persuasive message that they received about shyness than were the attitudes of individuals who were low in discrepancy. These findings suggest that participants with a high explicit–implicit discrepancy paid more careful attention to the message than those with a low discrepancy, presumably in an attempt to resolve the discrepancy. Experiment 1 demonstrated that explicit–implicit discrepancies in shyness can lead to greater thinking about information directly relevant to the dimension on which the discrepancy exists. As noted earlier, shyness is a general and somewhat broad personality self-dimension. To extend the utility and generality of our findings, we sought to test whether the same effects would emerge when the explicit–implicit discrepancy concerned a more specific dimension of the self: a person’s need to evaluate (Jarvis & Petty, 1996).

Experiment 2 Our second study was designed to provide a conceptual replication and extension of the first. In Study 2, we used the same paradigm to assess the extent to which participants engaged in 2

Asendorpf et al. (2002) reported correlations between explicit and implicit shyness that ranged from .20 to .44 depending on the format of the explicit measure. For example, the correlation between the IAT and the explicit measure based on the shyness adjectives scale was .40.

EXPLICIT–IMPLICIT DISCREPANCY

effortful elaboration (i.e., an argument quality manipulation). However, several changes were introduced. First, instead of assessing explicit and implicit shyness as in Study 1, we focused on a motivational self-concept dimension, need to evaluate (NE; Jarvis & Petty, 1996; Petty & Jarvis, 1996), which refers to individual differences in people’s tendencies to engage in evaluative thought. People who are high in NE tend to spontaneously assess whether things are good or bad (e.g., Tormala & Petty, 2001; see also the “need to assess”; Kruglanski et al., 2000). Knowing whether things in the world are good or bad helps people to understand the environment. Probably because of this and other functions (e.g., Maio & Olson, 2000), people tend to form attitudes about nearly everything (e.g., Bargh, Chaiken, Govender, & Pratto, 1992; Roskos-Ewoldsen & Fazio, 1992). Nevertheless, some people are more chronic and spontaneous than others in their tendency to evaluate, and the NE scale assesses this. In contrast to the explicit self-report measure typically used to assess NE, we developed an implicit NE measure using an IAT. Similar to the previous study, an index of self-discrepancy was formed as the absolute value of the difference between the standardized values of the explicit and implicit measures. Second, to generalize our results across topics, we used a different persuasive issue. Instead of presenting information directly related to the discrepancy dimension as in Study 1, we used a message containing substantive information that actually was not relevant to the issue of the discrepancy. However, for all participants, the message was framed to appear relevant to NE. That is, participants were told before reading the message that the research concerned their evaluations, and so all participants had an expectation that the message would contain information directly relevant to the domain of the discrepancy. The topic of the message the participants were to evaluate was increasing the amount of vegetables in the diet. Comparing the persuasive effect of strong and weak arguments tested the effects of explicit–implicit self-discrepancy on information processing. We expected participants with a relatively high discrepancy between their explicit and implicit NE to think about information expected to be discrepancy related to a greater extent than individuals with a relatively low discrepancy. That is, we expected argument quality to have a larger impact on attitudes for participants with a high discrepancy between explicit and implicit NE compared with participants with a low discrepancy. Thus, as in Study 1, we expected to find a Discrepancy ⫻ Argument Quality interaction on the measure of attitudes that was unmoderated by the direction of the discrepancy.

Method Participants and Design Ninety-nine undergraduates in psychology courses at the Universidad Auto´noma de Madrid participated in partial fulfillment of a course requirement. We randomly assigned the students to the argument quality conditions (strong or weak) and assessed their explicit and implicit (IAT) NE to form an index of explicit–implicit self-discrepancy. Direction of discrepancy (higher on explicit or implicit measure) was also coded.

Procedure The procedure was similar to Study 1. Participants were seated at individual computer stations and were told that they were going to partic-

159

ipate in two different research projects. First, participants completed the IAT for the NE measure, ostensibly as part of a research project in the cognitive psychology program. After the IAT task, participants were told that because of extra time remaining in the session, they would also be participating in another experiment designed to assess their attitudes toward a particular issue. To create a context of evaluation, all of the participants were explicitly told that the goal of the second research project was to measure their opinions and evaluations of a current commercial campaign. Participants received a persuasive message in favor of the consumption of vegetables containing either strong or weak arguments. Then, participants were told that it was important to know more specifically what their opinions about the consumption of vegetables were. After reporting their attitudes toward the proposal, participants completed the explicit NE scale and several ancillary measures.

Independent Variables Argument quality. The message about vegetable consumption contained either strong or weak arguments in favor of this topic. This manipulation was designed to influence the favorability of participants’ cognitive responses if participants were thinking about the message (Petty & Cacioppo, 1986). The gist of one of the strong arguments in favor of vegetable consumption was that vegetables have more vitamins than the majority of vitamin supplements on the market, making them especially appropriate during exams and workout periods. The gist of one of the weak arguments was that vegetables are becoming more popular for wedding celebrations because they are colorful and look beautiful on plates. The argument quality of the two messages was pretested to ensure that the strong version of the message produced mostly favorable thoughts, whereas the weak one produced mostly negative thoughts when people were instructed to think carefully about them (e.g., Brin˜ol, Horcajo, Becerra, Falces, & Sierra, 2002). Explicit measure of need to evaluate. Participants completed the 16item NE scale (Jarvis & Petty, 1996). This scale assesses the chronic tendency to engage in evaluative responding. Jarvis and Petty (1996) demonstrated that, compared with people low in NE, those high in the NE are more likely to form attitudes toward a variety of social and political issues (see also Bizer et al., 2004), and the attitudes of high NE individuals tend to be more accessible (Hermans, DeHouwer, & Eelen, 2001; for a review, see Brin˜ol & Petty, 2005). The NE scale contains statements such as “I enjoy strongly liking and disliking new things” and “I am pretty much indifferent to many important issues” (reverse scored). Participants responded to each statement on a 5-point scale anchored by extremely uncharacteristic of me (1) and extremely characteristic of me (5). The items on this self-concept dimension were intercorrelated (␣ ⫽ .79), so responses to each item were summed to form a composite score of NE. Participants’ NE scores were not affected by the argument quality manipulation, F(1, 98) ⫽ 0.44, p ⫽ .5. Implicit measure of need to evaluate. As in Study 1, we used an IAT procedure as the implicit assessment of this self-conception (Greenwald et al., 1998). The NE IAT was administered at the beginning of the experimental session and was presented as part of a research project designed to study how taxonomies are represented in people’s minds. In this IAT, participants classified target concepts (represented by me or other) and attributes (represented by neutral or extreme) using two designated keys. The words extreme and neutral were pretested as representative of high and low evaluation categories, respectively. Although the NE scale was designed to measure the tendency to engage in evaluation per se rather than the tendency to engage in extreme evaluation, a number of items on the NE scale clearly refer to extremity. According to Jarvis and Petty (1996, p. 190), the reason those items were included was to maximize the variance in participants’ scores on the NE scale. Also for that reason, we selected the words that were related to extremity to increase the variability in the IAT. The me category was represented by the words I, me, mine, my, and self, whereas the other category was represented by the words they, others,

˜ OL, PETTY, AND WHEELER BRIN

160

them, theirs, and their. The extreme category included words such as extremity, limit, radical, total, and extreme. In contrast, the neutral category included words as moderate, caution, prudent, temperate, and neutral. These words were equally familiar to participants and were selected based on some of the items of the NE scale (e.g., “It bothers me to remain neutral,” “I prefer to avoid taking extreme positions” [reversed], and “I often prefer to remain neutral about complex issues” [reversed]). The difference in response latencies for (me ⫹ extreme and other ⫹ neutral) versus (other ⫹ extreme and me ⫹ neutral) provided our implicit measure of NE. Regarding the combination of blocks, random assignment of stimuli, incorrect classifications, practice trials, anticipatory responses, momentary inattention, and data transformation, we followed the standard IAT procedures described in our first study (see also Greenwald et al., 1998). Explicit–implicit discrepancy. Explicit and implicit NE were not correlated (r ⫽ ⫺.13, p ⫽ .17). An index of discrepancy was formed as the absolute value of the difference between the standardized explicit and standardized implicit measures. Higher scores on that variable reflected greater differences between the explicit and the implicit measures, and thus higher explicit–implicit discrepancy in NE. As in the previous study, direction of discrepancy was symmetrical, with 50 participants showing a positive discrepancy (i.e., higher in the sample distribution on the explicit measure than the implicit measure) and 48 participants showing a negative discrepancy (i.e., lower in the distribution on the explicit than the implicit measure).

Dependent Measure: Attitudes Similar to Study 1, participants’ postmessage attitudes toward vegetable consumption were assessed using a single item 9-point semantic differential scale anchored at bad (1) and good (9).

Results Attitudes toward vegetables were submitted to the same hierarchical regression analysis used in Study 1. Thus, the independent variables included extent of discrepancy (continuous variable), manipulated argument quality (strong–weak; dummy coded), and direction of discrepancy (explicit ⬎ implicit vs. explicit ⬍ implicit; dummy coded). Scores on the discrepancy index were centered by subtracting the mean from each person’s score (Aiken & West, 1991). Main effects were interpreted in the first step of the regression, the two-way interactions in the second step, and the three-way interaction in the final step (Cohen & Cohen, 1983). Responses to the attitude scales were scored so that higher values represented more favorable opinions toward the proposal. Participants’ attitudes were more favorable toward consuming vegetables after receiving the strong (M ⫽ 8.29, SD ⫽ 0.71) than the weak (M ⫽ 7.98, SD ⫽ 0.79) message, ␤ ⫽ .20, t(95) ⫽ 1.96, p ⫽ .05. No main effect for extent of discrepancy emerged ( p ⫽ .66). More important and consistent with expectations, the main effect of argument quality was qualified by a marginally significant interaction between argument quality and extent of discrepancy, ␤ ⫽ .28, t(92) ⫽ 1.83, p ⫽ .07. This interaction revealed that as discrepancy increased, argument quality had a larger effect on attitudes. That is, there was a significant effect of argument quality among participants with relatively high discrepancy (recentered at ⫹1 SD), ␤ ⫽ .38, t(91) ⫽ 2.64, p ⫽ .01, but not among those with relatively low discrepancy (recentered at –1 SD), ␤ ⫽ .005, t(91) ⫽ 0.03, p ⫽ .97. It is important to note that the three-way interaction between argument quality, extent of discrepancy, and direction of the discrepancy was not significant, ␤ ⫽ ⫺.28, t(91) ⫽ ⫺1.42, p ⫽ .16, revealing that the effects of explicit–implicit

discrepancy on elaboration were not restricted to any particular direction of the discrepancy.3

Discussion Experiment 2 conceptually replicated the first experiment by showing that participants with relatively high explicit–implicit discrepancy in their NE processed the message more carefully than participants with relatively low discrepancy. That is, compared with those with a small discrepancy between their explicit and implicit NE, individuals with a high discrepancy were more influenced by the quality of the arguments in the message than were those with a low discrepancy. Although the message in the present study did not actually contain information directly related to the issue on which the explicit–implicit discrepancies existed (i.e., a person’s own NE), the whole context of the study was framed as dealing with the participants’ opinions. We argue that emphasizing that the task was related to evaluation was sufficient for participants with high explicit–implicit discrepancies in NE to engage in more extensive thinking. From our first two studies, it appears that information can be directly related to the issue of discrepancy (Study 1) or simply framed as related to the dimension in which the discrepancy exists (Study 2). We argue that the mere existence of a discrepancy does not result in the indiscriminate processing of any information present in the situation. Enhanced thinking is expected only if the dimension on which the discrepancy exists is activated by leading people to believe that the message is going to pertain or be relevant to that dimension. Without that, there should be no differential processing. To address this issue, our third study included a manipulation of the message frame designed to induce participants to expect the message to be related or unrelated to the issue on which there was an explicit–implicit discrepancy. If having an explicit–implicit discrepancy enhances information processing in general, then high discrepancy individuals should be equally likely to process messages framed as relevant or irrelevant to the issue on which the discrepancy exists. If the information processing is in service of discrepancy reduction, however, individuals with high discrepan3 Because of possible concerns about the use of difference scores in our analyses (i.e., the discrepancy index), we also conducted an alternative analysis treating implicit and explicit measures separately. For maximum power, data from both Studies 1 and 2 were combined and submitted to another hierarchical regression analysis, with the implicit and explicit measures (continuous variables) and manipulated argument quality (dummy coded) as independent variables. Study was also included as a factor in this analysis so that we could examine whether the results generalized across the study differences. As expected, this analysis revealed a significant main effect for argument quality, ␤ ⫽ .14, t(175) ⫽ 1.97, p ⫽ .05, which was qualified by a three-way interaction between argument quality, explicit self-concept, and implicit self-concept, ␤ ⫽ ⫺.16, t(175) ⫽ ⫺2.08, p ⫽ .03. Also importantly, this significant threeway interaction was not moderated by the study independent variable, as is evident in the absence of a four-way interaction ( p ⫽ .64). To facilitate ease of presentation (i.e., interpreting two-way rather than three-way interactions) and matching more closely our conceptual variable (i.e., psychological discrepancy), we present the discrepancy analysis in the text and figures.

EXPLICIT–IMPLICIT DISCREPANCY

cies should be more likely to process the messages framed as relevant than irrelevant.

161

crepancy related. As in previous studies, this enhanced thinking would be evidenced by greater attitudinal responsiveness to the manipulation of argument quality.

Experiment 3 Studies 1 and 2 provided initial evidence supporting the notion that divergence between explicit and implicit self-views can influence the extent of information processing of ostensibly discrepancy-related information. Experiment 3 was conducted to replicate and extend the findings from these studies. Thus, several changes were introduced. First, instead of examining a general dimension of personality (shyness) or motivation (NE), we focused on a more specific dimension of the self-concept: one’s beliefs concerning one’s own resistance to persuasion. Beliefs regarding one’s resistance to change play a central role in people’s values and identities. For example, Schwartz’s (1992) theory about universal human values is organized by two motivational dimensions: the self-transcendence/self-enhancement dimension and the openness to change/conservation dimension. This work implies that almost everyone might have beliefs about their own resistance to change and that such beliefs might be an integral part of the self-concept. In fact, resistance to change constitutes one of the most basic dimensions of personality according to the Big Five framework (e.g., McCrae & Costa, 1985; Wiggins & Trapnell, 1997). Although the construct of resistance to change can be conceptualized quite differently in terms of personality, cognitive ability, psychic structure, openness to experience, or openness as culture (McCrae & Costa, 1997), we focus specifically on personal perceptions of resistance to persuasion. In this experiment, the explicit self-dimension of resistance to persuasion was assessed using the Resistance to Persuasion Scale (Brin˜ol, Rucker, Tormala, & Petty, 2004), and implicit resistance was assessed using an IAT developed for this study. As in our previous studies, an index of explicit–implicit discrepancy was created as the absolute value of the difference between the standardized explicit and implicit scores. Second, to further generalize our results across topics, we used a different persuasive issue. Instead of presenting information about a relatively proattitudinal topic (i.e., increasing the amount of vegetables in the diet; Study 2) or a rather neutral one (i.e., the benefits of shyness; Study 1), we moved to a relatively counterattitudinal topic (i.e., institution of required exams for college students). Finally, and most importantly, we manipulated the presumed discrepancy-relevance of the message information. Because the explicit–implicit discrepancy was related to individual differences in susceptibility to change in attitudes and opinions, half of the participants were told that the study had to do with their attitudes and opinions (similar to the frame in Study 2). This condition was compared with a discrepancy-unrelated frame in which participants were told that the experiment was part of a text comprehension study. Thus, all of the participants were exposed to a persuasive message containing either strong or weak arguments that were framed as either related or unrelated to the dimension on which the discrepancy existed. After reading the message, participants were asked to report their attitudes toward the proposal. We expected participants with a large explicit–implicit discrepancy to think more about the persuasive message than those with a small discrepancy, but only when the message was framed to appear dis-

Method Participants and Design One hundred seventy-three introductory psychology students at Ohio State University participated in partial fulfillment of a course requirement. The students were randomly assigned to the argument quality conditions (strong or weak) and the message context conditions (discrepancy related or unrelated), which were manipulated orthogonally. Additionally, we measured participants’ explicit and implicit resistance to persuasion to form an index of explicit–implicit discrepancy. The independent variables thereby constituted an Argument Quality (strong vs. weak) ⫻ Message Frame (discrepancy-related vs. discrepancy-unrelated) ⫻ Extent of Explicit–Implicit Discrepancy (continuously scored) ⫻ Direction of Discrepancy (higher on implicit or explicit measure) design.

Procedure Upon arrival, participants were seated at individual computer stations and were presented with all of the materials on the computer using MediaLab software (Jarvis, 2000). All of the participants were told that they were going to participate in two different research projects. First, participants completed the IAT for resistance to persuasion, ostensibly as part of a research project on semantic recognition and categorization conducted by the cognitive psychology program. After the IAT task, participants were told that because of extra time remaining in the session, they would also be participating in another experiment designed to assess possible changes in university policies. They read about a new school policy and were told that students’ opinions about this policy were of importance to the university. Participants received a message in favor of instituting senior comprehensive exams that contained either strong or weak arguments. After reading the message, participants were told that it was important for the Board of Trustees to know what their opinions on the topic were. Thus, they completed measures of their attitudes toward the comprehensive exam policy. Finally, participants completed the Resistance to Persuasion Scale (Brin˜ol et al., 2004) and several ancillary questions.

Independent Variables Argument quality. The comprehensive exam message participants received contained either strong or weak arguments. The arguments selected were adopted from previous research and have been shown many times to produce the appropriate pattern of cognitive responding (see Petty & Cacioppo, 1986). That is, the strong arguments elicited mostly favorable thoughts and the weak arguments elicited mostly unfavorable thoughts when people were instructed to think carefully about them. The gist of some strong arguments in favor of the exam policy were that students’ grades would improve if the exams were adopted and that the average starting salary of graduates would increase. The gist of some weak arguments in favor of the exam policy were that implementing the exams would allow the university to take part in a national trend and that the exams would give students the opportunity to compare their scores with those of students at other universities. Message frame. The frame of the message was manipulated to appear related or unrelated to participants’ opinions. That is, because the explicit– implicit discrepancy was related to persuasion (openness vs. resistance to changing one’s opinions), we framed the study as related (or not) to opinions and attitudes. Thus, the message was introduced as part of an opinion-related study (discrepancy-related frame) or as part of a text

˜ OL, PETTY, AND WHEELER BRIN

162

comprehension study (unrelated frame). Participants in the discrepancyrelated frame condition read the following: OPINION SURVEY This part of the study consists of a survey designed to obtain your opinions and thoughts about a campus issue. The message we are going to ask you to read is based on the transcript of an editorial from a college radio station. The editorial was introduced on the radio in order to familiarize students with this important issue. Please pay attention because your opinions will be measured. In the comprehension frame, participants were told that the message was designed to measure their appreciation of new information. Comprehending a text would presumably not trigger the idea that the study involved any explicit or blatant persuasive attempt. In this condition, participants read the following: TEXT COMPREHENSION This part of the study has been designed to measure your appreciation of new information. The information we are going to ask you to read and to examine is based on a transcript of a class project broadcast on a college radio station. We want you to examine the content of the information presented carefully because the extent to which you understand the text will be measured. It is important to note that although attitude change was not salient for the text comprehension frame, opinions toward the proposal were still measured following the message. At the time opinions were measured, the idea of persuasibility may have become apparent to these participants, but by this time the extent of processing of the message was already determined. That is, in the discrepancy-relevant case, participants already expected to give their opinion while they were reading the message because the study was about opinions. In the discrepancy-irrelevant case, they thought that the study was on text comprehension while reading and processing the message, so forming opinions was not salient. We pretested the effectiveness of the induction by randomly assigning 50 students to one of the two frames and asking them what kind of information they expected to receive after the frame. Specifically, we asked participants to respond on a 5-point scale to the question: “To what extent do you expect to receive any information that might be related to your resistance to persuasion?” As expected, participants who received the discrepancy-related frame reported significantly higher expectations to receive persuasion-related information (M ⫽ 3.14, SD ⫽ 1.00) than those who received the discrepancy-unrelated frame (M ⫽ 2.68, SD ⫽ 1.21), t(48) ⫽ 10.39, p ⬍ .0001. It is also important to note that although the message was framed to be related or not to the basis of the self-discrepancy, the actual message participants received was identical. Explicit measure of resistance. The Resistance to Persuasion Scale was used to assess participants’ explicit perceptions of their resistance to persuasion. This scale measures individuals’ perceptions and beliefs about their own vulnerability to persuasion, willingness to change, and motivation and ability to resist persuasion. In prior research validating the scale, it has predicted the number of counterarguments people generate to a message and how resistant they were to influence (Brin˜ol et al., 2004). The scale contains 11 statements such as “My attitudes are open to change” and “It is hard for me to change my ideas.” Participants responded to each statement on a 5-point scale anchored by extremely uncharacteristic of me (1) and extremely characteristic of me (5). Ratings on the scale items were highly consistent with each other (␣ ⫽ .86) and were averaged to form a single index of resistance to persuasion for each participant. Resistance to persuasion scores were not affected by the argument quality manipulation, F(1, 173) ⫽ 0.26, p ⫽ .61; the frame manipulation, F(1, 173) ⫽ 0.68, p ⫽ .41; or the interaction of the two, F(1, 173) ⫽ 0.02, p ⫽ .87. Implicit measure of resistance. In the IAT measure of resistance, participants classified target concepts (represented by me or other) and

attributes (represented by easy to be persuaded or hard to be persuaded categories of words) using two designated keys. The me and other categories included the same words used in Study 2. Attributes related to persuasibility were selected from the items on the explicit scale and included the words easy, flexible, open, variable, and changeable. In contrast, resistant attributes included the words resistant, stable, hard, consistent, and committed. The difference in response latencies for me– easy (me ⫹ easy and other ⫹ hard) versus me– hard (other ⫹ easy and me ⫹ hard) responses provided a measure of relative self-association with resistance (i.e., the IAT effect). Regarding the combination of blocks, random assignment of stimuli, incorrect classifications, practice trials, anticipatory responses, momentary inattention, and data transformation, we followed the standard IAT procedures described in our previous studies (see also Greenwald et al., 1998). Explicit–implicit discrepancy. The explicit and implicit measures of personal resistance were uncorrelated (r ⫽ ⫺.10, p ⫽ .16). An index of explicit–implicit discrepancy was formed as the absolute value of the difference between the standardized explicit and the implicit measures of resistance. Higher scores on that variable reflected greater differences between the explicit and the implicit measures (i.e., higher discrepancy). The distributions of scores using this index revealed roughly equal numbers of people on each side of the discrepancy, with 87 participants showing positive discrepancy (i.e., higher in the sample distribution on the explicit measure than the implicit measure) and 86 showing negative discrepancy (i.e., lower in the distribution on the explicit than the implicit measure). Perhaps more relevant for the present research, the number of participants in each direction was also equivalent in the conditions for which the information was presented as discrepancy related, with 42 participants revealing positive discrepancies and also 42 participants showing negative discrepancies.

Dependent Measure: Attitudes Participants’ attitudes toward the proposal were assessed using a series of five 9-point semantic differential scales ranging from 1 to 9 (i.e., bad– good, unfavorable–favorable, pro–against, foolish–wise, harmful– beneficial) on which they rated the comprehensive exam policy. Ratings on these items were highly intercorrelated (␣ ⫽ .84), so they were averaged to form one overall attitude index.

Results Attitudes were submitted to a hierarchical regression analysis, with extent of discrepancy (continuous variable), manipulated argument quality (strong–weak; dummy coded), message frame (relevant–irrelevant; dummy coded), and direction of discrepancy (explicit ⬎ implicit vs. explicit ⬍ implicit; dummy coded) as the independent variables. Scores on the discrepancy index were centered by subtracting the mean from each person’s score (Aiken & West, 1991). Main effects were interpreted in the first step of the regression, two-way interactions in the second step, three-way interactions in the third step, and the four-way interaction in the final step (Cohen & Cohen, 1983). Responses to the attitude scales were scored so that higher values represented more favorable opinions toward the proposal. Participants’ attitudes were more favorable toward the proposal after receiving the strong (M ⫽ 4.46, SD ⫽ 2.34) than the weak (M ⫽ 3.68, SD ⫽ 2.09) message, ␤ ⫽ .18, t(168) ⫽ 2.35, p ⫽ .02. A significant interaction between argument quality and extent of discrepancy also emerged, ␤ ⫽ .33, t(162) ⫽ 3.31, p ⫽ .001, revealing that as discrepancy increased, argument quality had a larger effect on attitudes. An interaction between frame and extent

EXPLICIT–IMPLICIT DISCREPANCY

of discrepancy also emerged, ␤ ⫽ .23, t(162) ⫽ 1.97, p ⫽ .05. This interaction suggested that attitudes tended to become more positive as discrepancy increased for the relevant frame but more negative as discrepancy increased for the irrelevant frame. This interaction was only obtained in this study and is not discussed further. More critical to our primary concerns, the predicted three-way interaction between argument quality, extent of discrepancy, and message frame was significant, ␤ ⫽ .42, t(158) ⫽ 2.84, p ⫽ .005. As depicted in Figure 1, this three-way interaction indicated that the two-way interaction between discrepancy and argument quality was only significant for the discrepancy-related frame, ␤ ⫽ .56, t(157) ⫽ 4.26, p ⬍ .0001, but not for the discrepancy-unrelated frame, ␤ ⫽ ⫺.02, t(157) ⫽ ⫺.14, p ⫽ .89. For the discrepancyrelated frame, there was a significant effect of argument quality among participants with relatively high discrepancy (analyzed at ⫹1 SD), ␤ ⫽ .72, t(157) ⫽ 4.86, p ⬍ .0001, but not among those with relatively low discrepancy (analyzed at –1 SD), ␤ ⫽ ⫺.13, t(157) ⫽ ⫺.93, p ⫽ .36. The four-way interaction was not significant, ␤ ⫽ ⫺.12, t(157) ⫽ ⫺.51, p ⫽ .61, revealing that the effects of explicit–implicit discrepancy on elaboration were symmetrical and not restricted to any particular direction of the discrepancy.

Discussion Experiment 3 conceptually replicated our previous findings by showing that people who have a large discrepancy between their

163

implicit and explicit views of their resistance to persuasion are more likely to think carefully about a discrepancy-related persuasive message than are people who have a small discrepancy between their implicit and explicit self-conceptions. As in Studies 1 and 2, this conclusion was supported by the finding that the attitudes of relatively discrepant individuals were more reflective of the quality of the discrepancy-related persuasive message than were the attitudes of less discrepant individuals. The fact that the enhanced information processing only occurred for individuals with large discrepancies when the message was framed so as to seem related to the basis of the discrepancy (i.e., an opinion context) rather than unrelated to the discrepancy (i.e., a comprehension context) is consistent with the idea that the purpose of the processing was to resolve the discrepancy. Experiment 3 demonstrated that explicit–implicit discrepancies in resistance to persuasion can lead to greater thinking about information framed as related to the discrepancy. However, resistance to change is a very specific and descriptive self-dimension. Thus, Studies 1, 2, and 3 all focused on descriptive (shyness, NE, and persuasibility, respectively) rather than evaluative selfdimensions. To extend the utility and generality of our findings, in a final study we sought to test whether the same effects can emerge when the explicit–implicit discrepancy concerns a broad, global evaluation of the self. Thus, in our fourth study we used a similar paradigm to test whether the information-processing effects can be found for explicit–implicit divergences on a general evaluative self-dimension: a person’s self-esteem.

Experiment 4

Figure 1. Attitudes as a function of argument quality, message frame, and extent of explicit–implicit (e-i) discrepancy graphed at –1 standard deviation (low discrepancy) and ⫹1 standard deviation (high discrepancy) in Experiment 3.

Our fourth study was designed to extend the previous findings to the domain of self-evaluation. That is, in this study we focused on an evaluative dimension of the self-concept, self-esteem, and assessed it with both explicit and implicit measures. Implicit selfesteem typically has been defined as an evaluation of the self that occurs automatically and unintentionally and can differ from one’s more controlled and deliberative self-assessments (e.g., Farnham, Greenwald, & Banaji, 1999; Greenwald & Banaji, 1995; Greenwald & Farnham, 2000; Hetts & Pelham, 2001; Koole, Dijksterhuis, & van Knippenberg, 2001). In Study 4, we used the same paradigm to assess the extent to which participants engaged in effortful thinking (i.e., an argument quality manipulation). Similar to Study 3, we also manipulated the ostensible discrepancy relatedness of the message information by framing it as related or unrelated to the basis of the discrepancy. Finally, in this study we used the relatively proattitudinal topic used in Study 2 (i.e., increasing the amount of vegetables in the diet). We expected participants with a relatively large discrepancy between their explicit and implicit self-esteem to elaborate the information more than individuals with a relatively small discrepancy, but only when the information was framed to seem discrepancy related. That is, we expected argument quality to have a larger impact on attitudes for participants with a large discrepancy between explicit and implicit self-esteem compared with participants with a small discrepancy, but only when the message was framed to seem related to the discrepancy (i.e., when the message was framed to seem related to their self-concept). More specifically, as in Study 3 we expected to find a Discrepancy ⫻ Argument Quality ⫻ Message Frame interaction on attitudes toward the message proposal.

˜ OL, PETTY, AND WHEELER BRIN

164 Method Participants and Design

One hundred seventeen undergraduate psychology students at the Universidad Auto´noma de Madrid were randomly assigned to the argument quality (strong or weak) and the message frame (discrepancy-related or discrepancy-unrelated) conditions, which were manipulated orthogonally. Additionally, we measured participants’ explicit and implicit self-esteem to form an index of explicit–implicit discrepancy and coded for the direction of discrepancy. The independent variables thereby constituted an Argument Quality (strong vs. weak) ⫻ Message Frame (discrepancy-related vs. discrepancy-unrelated) ⫻ Extent of Explicit–Implicit Discrepancy in SelfEsteem (continuously scored) ⫻ Direction of Discrepancy (relatively higher in explicit or implicit self-esteem) design.

Procedure Upon arrival, participants were seated at individual computer stations and were told that they were going to participate in two different research projects. As in previous studies, participants first completed an IAT, ostensibly as part of a research project on categorization conducted by the cognitive psychology program. After the IAT task, participants were told that because of extra time remaining in the session, they would also be participating in another experiment designed to assess their attitudes toward an opinion topic. Half of the participants were told that the message they were about to read had to do with plants and vegetables and their qualities and properties (discrepancy-unrelated frame). The rest of the participants were told that the message concerned their personal habits and the way they interact with the world (discrepancy-related frame). Then, participants received a message containing strong or weak arguments in favor of the consumption of vegetables. After reading the message, all of the participants were told that it was important to know more specifically what their opinions about the consumption of vegetables were. Finally, after reporting their attitudes toward the proposal, participants completed the Rosenberg Self-Esteem Scale (Rosenberg, 1965).

Independent Variables Argument quality. The message about vegetable consumption contained either strong or weak arguments in favor of this topic. This manipulation was identical to that used in Study 2. Message frame. The introduction to the message was framed to seem as though the message would contain information unrelated or related to the discrepancy domain. In the discrepancy-unrelated frame condition, participants were told, “You are about to read an article about the characteristics and properties of different plants and vegetables.” In the discrepancy-related frame condition, they were told, “You are about to read an article about your self-concept, your personal diet habits, and the way you see your world.” To further enhance the discrepancy relevance, in the discrepancy-unrelated frame, the title of the message was “Research About Vegetables” and in the discrepancy-related frame, the title of the message was “Research About the Self-Concept.” It is important to note that, as in Experiments 2 and 3, we deliberately did not attempt to provide any real information about students’ selfconceptions. Rather, the manipulation was oriented to influence what participants perceived was going to be the content of the message (not the actual content itself). Thus, there would be no differences in ability to process the message across conditions or other irrelevant confounds. We pretested the effectiveness of the induction by randomly assigning 58 students to one of the two frames and asked them what kind of information they expected to receive after that. Specifically, we asked participants to respond on a 5-point scale to the question, “To what extent do you expect to receive any information that might be relevant to your self-concept and self-esteem?” As expected, participants who received the discrepancy-

related frame reported significantly higher expectations to receive selfrelated information (M ⫽ 3.70, SD ⫽ 0.91) than those who received the discrepancy-unrelated frame (M ⫽ 1.82, SD ⫽ 0.94), t(56) ⫽ 7.68, p ⬍ .0001. Explicit self-esteem. We used the Rosenberg Self-Esteem Scale as our explicit measure (Rosenberg, 1965). Ratings on the items were highly consistent with each other (␣ ⫽ .76) and were averaged to form a single index of self-esteem for each participant. Self-esteem scores were not affected by the argument quality manipulation, F(1, 117) ⫽ 0.08, p ⫽ .77; the frame manipulation, F(1, 117) ⫽ 1.41, p ⫽ .24; or the interaction of the two, F(1, 117) ⫽ 0.11, p ⫽ .73. Implicit self-esteem. As in the prior studies, we used an IAT procedure to assess participants’ implicit self-evaluations (Greenwald et al., 1998). The self-esteem IAT was administered at the beginning of the experimental session and was presented as part of a research project designed to study how taxonomies are represented in people’s minds. In the IAT, participants classified target concepts (represented by me or other) and attributes (represented by good and bad) using two designated keys. The me and other categories included the same words used in Studies 2 and 3. The good category words included freedom, peace, love, cheer, and paradise, and the bad category words included poison, cancer, death, vomit, and disaster (selected from Greenwald et al., 1998). The difference in response latencies for me ⫹ bad and other ⫹ good trials versus other ⫹ bad and me ⫹ good trials provided a measure of relative automatic self-esteem (i.e., the IAT effect). Comparable implicit measures have been used in prior research and have been shown to be effective in predicting a variety of thoughts and behaviors relevant to self-evaluations, especially under low thinking (spontaneous) circumstances (e.g., Greenwald & Farnham, 2000; Hetts & Pelham, 2001; Jones, Pelham, Mirenberg, & Hetts, 2002; Koole et al., 2001; see also Bosson, Swann, & Pennebaker, 2000). Explicit–implicit discrepancy. Explicit and implicit self-esteem showed a small negative correlation (r ⫽ ⫺.19, p ⫽ .04). Previous research has also shown negative correlations between explicit and implicit selfesteem (e.g., Bosson et al., 2000; Hetts et al., 1999; Karpinski, 2004; Kitayama & Uchida, 2003). An index of discrepancy was formed as the absolute value of the difference between the standardized explicit and implicit measures. Higher scores on that variable reflected greater differences between the explicit and the implicit measures, and thus higher explicit–implicit self-esteem discrepancy. As in all previous studies, the distributions of scores revealed equivalent numbers of participants on each side of the discrepancy, with 54 participants showing positive discrepancies and 63 showing negative discrepancies. This was also true for the conditions in which the information was presented as discrepancy related, with 26 participants revealing positive discrepancies and 31 showing negative discrepancies.

Dependent Measure: Attitudes As in the previous studies, participants were instructed to report their attitudes following the message. Participants’ attitudes toward the proposal (i.e., the increased consumption of vegetables) were assessed using a series of five 9-point semantic differential scales anchored by 1 and 9 (i.e., bad– good, unfavorable–favorable, pro–against, foolish–wise, harmful– beneficial). Ratings on these items were highly intercorrelated (␣ ⫽ .84) and were averaged to form one overall attitude index. Responses to the attitude scales were scored so that higher values represented more favorable opinions toward increasing consumption of vegetables.

Results Attitudes were submitted to a hierarchical regression analysis, with extent of discrepancy (continuous variable), manipulated argument quality, message frame, and direction of discrepancy as the independent variables. Scores on the discrepancy index were

EXPLICIT–IMPLICIT DISCREPANCY

centered by subtracting the mean from each person’s score (Aiken & West, 1991). Main effects were interpreted in the first step of the regression, two-way interactions in the second step, three-way interactions in the third step, and the four-way interaction in the final step (Cohen & Cohen, 1983). The regression procedure described above revealed a significant main effect for direction of discrepancy, ␤ ⫽ .21, t(110) ⫽ 2.23, p ⫽ .03, showing that participants’ attitudes were more favorable toward the proposal in one direction of discrepancy (explicit ⬎ implicit) than the other (implicit ⬎ explicit). This effect was not obtained in any of the other studies and is not discussed further. In addition, a significant interaction between argument quality and extent of discrepancy was obtained, ␤ ⫽ .31, t(104) ⫽ 2.54, p ⫽ .01, showing that argument quality had a greater impact on attitudes as discrepancy increased. Of most interest, the predicted three-way interaction between argument quality, extent of discrepancy, and frame was significant, ␤ ⫽ .37, t(100) ⫽ 2.94, p ⫽ .004. To examine the basis of this interaction, we decomposed the interaction by frame conditions. As depicted in Figure 2, the two-way interaction between discrepancy and argument quality was only significant for the discrepancy-related frame, ␤ ⫽ .86, t(99) ⫽ 3.76, p ⫽ .0002, but not for the discrepancy-unrelated frame, ␤ ⫽ .11, t(99) ⫽ 0.64, p ⫽ .52. More specifically, for the discrepancy-related frame, there

165

was a significant effect of argument quality among participants with relatively high discrepancy (analyzed at ⫹1 SD), ␤ ⫽ .82, t(99) ⫽ 4.37, p ⬍ .0001. In contrast, those with relatively low discrepancy (analyzed at –1 SD) showed no effect of argument quality, ␤ ⫽ ⫺.26, t(99) ⫽ ⫺1.34, p ⫽ .18. The effects of explicit–implicit discrepancy on information processing were not restricted to any particular direction of the discrepancy, as indicated by a nonsignificant four-way interaction, ␤ ⫽ .06, t(99) ⫽ .29, p ⫽ .77.4

Discussion Experiment 4 conceptually replicated the earlier experiments by showing that participants with relatively high explicit–implicit discrepancy in their self-esteem processed more carefully the information presumably related to the discrepancy dimension than participants with relatively low discrepancy. That is, individuals with a large discrepancy found strong arguments to be more persuasive than weak ones whereas those with a small discrepancy did not, but only when the messages were potentially relevant to the discrepancy. Those with a high discrepancy presumably devoted more cognitive resources to new information framed as discrepancy related to reduce the uncertainty or subjective discomfort that might result from holding simultaneously favorable and unfavorable implicit and explicit self-evaluations.

General Discussion Previous research has shown that there are many possible sources of internal discrepancies (e.g., attitudes, motives, selfconceptions), and these discrepancies are often associated with negative affect and undesirable psychological outcomes. Although most of the discrepancy research has relied on explicit and deliberative self-reports, previous studies identifying people who possess specific explicit and implicit self-discrepant dimensions have also found that discrepancy is associated with some notable difficulties in functioning. The present research extends previous literature by showing that explicit–implicit discrepancy is also associated with enhanced 4

Figure 2. Attitudes as a function of argument quality, message frame, and extent of explicit–implicit (e-i) discrepancy graphed at –1 standard deviation (low discrepancy) and ⫹1 standard deviation (high discrepancy) in Experiment 4.

Similar to Studies 1 and 2, alternative analyses were conducted in which implicit and explicit measures were treated separately. For maximum power, data from both Studies 3 and 4 were combined and submitted to another hierarchical regression analysis, with the implicit and explicit measures (continuous variables) and manipulated argument quality and message frame (dummy coded) as independent variables. Study was also included as a factor in this analysis so that we could examine whether the results generalized across the study differences. As expected, this analysis revealed a significant main effect for argument quality, ␤ ⫽ .18, t(250) ⫽ 3.61, p ⫽ .001, which was qualified by a three-way interaction between argument quality, explicit self-concept, and implicit self-concept, ␤ ⫽ ⫺.16, t(250) ⫽ ⫺2.31, p ⫽ .02. Most important for our concerns, a significant four-way interaction also emerged, ␤ ⫽ ⫺.26, t(250) ⫽ 3.27, p ⫽ .001. As predicted, this four-way interaction indicates that the threeway interaction between argument quality, explicit, and implicit selfconcept was only significant for the discrepancy-related frame, ␤ ⫽ ⫺.38, t(127) ⫽ ⫺4.0, p ⬍ .0001, but not for the discrepancy-unrelated frame, ␤ ⫽ ⫺.06, t(122) ⫽ ⫺0.58, p ⫽ .56. Also importantly, this significant four-way interaction was not moderated by the study independent variable, as is evident in the absence of a five-way interaction ( p ⫽ .44).

166

˜ OL, PETTY, AND WHEELER BRIN

processing of ostensibly discrepancy-related information, presumably in an attempt at inconsistency reduction. The present studies also demonstrate that the tendency to process discrepancy-related information applies to both narrow (Study 3) and relatively broad (Studies 1, 2, and 4), as well as to descriptive (Studies 1, 2, and 3) and evaluative (Study 4), dimensions of the self. Across different types of self-concept elements, we have demonstrated that individuals for whom explicit and implicit self-dimensions are incongruent process information that either is (Study 1) or is framed as (Studies 2, 3, 4) discrepancy related more carefully than individuals with relatively congruent explicit–implicit self-dimensions. By thinking about information that is presumably related to the explicit–implicit dimension of conflict, people may be attempting to reduce the discrepancy between the implicit and explicit dimensions, perhaps unconsciously.

Symmetric Versus Asymmetric Discrepancy In prior research, most researchers interested in explicit–implicit discrepancies have concentrated their studies in just one of the possible directions of discrepancy: people with more positive explicit than implicit self-views. For example, studies on narcissism (e.g., Bosson et al., 2003) and defensive self-esteem (Jordan et al., 2003; Shedler et al., 1993) have particularly focused on cases in which explicit self-esteem is higher than implicit selfesteem and compared this particular discrepancy with cases in which both implicit and explicit self-esteem are relatively high. When looking at these asymmetric discrepancies, prior research has usually divided participants on the basis of relative scores within the sample distribution, as we did in the present research (see Weinberger & Hardaway, 1990, for an exception). Several different procedures have been used for dividing people into discrepancy groups, such as by splitting people at the median, tertiles, or quartiles of the sample distribution (Barger, Kircher, & Croyle, 1997; Emmons & Colby, 1995; Mendolia, Moore, & Tesser, 1996; Newman, Duff, & Baumeister, 1997; Weinberger, 1990) and by recentering at one SD above and below the mean of the distribution (e.g., Bosson et al., 2003; Jordan et al., 2003; McGregor & Marigold, 2003). Other research (e.g., Kehr, 2004), like ours, used a procedure that created a continuous score by taking the difference between standardized explicit and implicit measures of the self-construct. It is important to note that we had comparable numbers of individuals on each side of the discrepancy (implicit ⬎ explicit and explicit ⬎ implicit), and we were able to show that direction of discrepancy did not moderate the effects of amount of discrepancy on information processing. For exploration, we tried an alternative but conceptually similar analysis in which we used the raw discrepancy scores rather than the absolute values and tested for curvilinear argument quality effects across levels of raw discrepancy. In this alternative analysis, both high-positive and high-negative discrepancy individuals should process the relevant frame message more than people who have little discrepancy. That is, because the direction of discrepancy does not matter according to our conceptualization, both high-positive and high-negative discrepancies should look the same. When analyzed in this manner, the results from our experiments look exactly as predicted. Figure 3 shows the significant interaction between discrepancy (quadratic term) and argument quality, ␤ ⫽ .70, t(315) ⫽ 6.87, p ⬍ .0001, collapsing across

Figure 3. Attitudes (standardized scores) as a function of argument quality and raw discrepancy scores across studies (only relevant-frame conditions are included for Studies 3 and 4). Actual discrepancy scores in the studies range from ⫺5.01 to ⫹ 4.52. E-i ⫽ explicit–implicit.

studies and controlling for study as a factor (which did not moderate the critical interaction). Although our discrepancy index does a good job of capturing people’s relative standing in the distribution (as in most prior research), our index can only deal with the range of scores observed in the sample. Thus, by an absolute criterion, it could be the case that all of the participants are relatively high in both explicit and implicit self-esteem, high in explicit self-esteem and low in implicit self-esteem, and so forth. If this were true, it would still be the case that some people were higher than others on each measure and that the relative discrepancies our index captures is successful in predicting the extent of information processing. Nevertheless, the generality of our conclusions would be limited if all of our studies had the same profile of responses on the explicit and implicit measures (e.g., all absolutely high on both, all high on one and low on the other, etc.). To explore this issue further, we created an “absolute” discrepancy index in each study by using the middle point of the explicit scale (e.g., 3 in a 5-point scale) and the zero point in the IAT to approximate absolute differences between participants high and low in each of the self-concept dimensions. This categorization revealed that only Studies 1 and 3 had reasonable numbers of individuals with discrepancies in both directions by this criterion (e.g., implicit measure suggests shyness but explicit measure suggests not shy; implicit measure suggests not shy but explicit measure suggests shy). Notably, these studies produced the same pattern of results as the studies in which the discrepancies were present mostly on a relative basis. Our speculation is that the size of the discrepancy matters more than whether the discrepancy is relative or absolute. For example, an individual with trivially positive self-esteem on an explicit measure and trivially negative self-esteem on an implicit measure

EXPLICIT–IMPLICIT DISCREPANCY

would be categorized as discrepant in absolute terms, whereas an individual with trivially positive self-esteem on an explicit measure and extremely positive self-esteem on an implicit measure would not, even though the difference between implicit and explicit scores of the latter individual would be much larger and more consequential in our view. Nevertheless, future research might profitably recruit larger samples and use alternative measures so that the issue of absolute versus relative discrepancies can be examined more systematically.

Motivation to Reduce Self-Inconsistency As noted earlier, explicit discrepancies are often associated with doubt. Campbell (1990) found that self-concept instability correlated with lower subjective confidence in one’s own trait ratings and also with slower “me/not me” responses to trait adjectives (see Wright, 2001, for a review). Previous literature has also related explicit evaluative inconsistency (attitudinal ambivalence) to lack of confidence. For example, Jonas et al. (1997) provided empirical evidence that evaluative inconsistency evokes elaboration of related information to achieve a sufficient level of confidence with respect to the overall evaluation of the object. Bargh et al. (1992) suggested that evaluative inconsistency might be related to doubt, because response latencies (i.e., attitude accessibility) were found to be slower for ambivalent participants (see also Costelo, Rice, & Schoenfeld, 1974; Gilmore, 1982). Indeed, one function particular to ambivalent attitudes—and, perhaps, also to explicit–implicit discrepant selves—seems to be reducing action readiness and promoting further and elaborated thinking about related information to amplify confidence and knowledge about the target (e.g., Ha¨nze, 2001; Hodson et al., 2001; Jonas et al., 1997). On the basis of this research, it seems reasonable that explicit– implicit self-discrepancies might also be associated with uncertainty or doubt (Brin˜ol et al., 2003). If explicit–implicit selfdiscrepancies are associated with doubt, the enhanced informationprocessing effects we observed might be due in part to this uncertainty. Individuals who are induced to doubt before receiving a message have been shown to engage in greater thinking (Tiedens & Linton, 2001; Weary & Jacobson, 1997). For example, in one study Tiedens and Linton (2001) had participants write about a sad experience in which they felt uncertain about what was happening or a sad experience in which they felt certain. Following this doubt induction, participants received a message containing strong or weak message arguments. The primary result was that uncertain participants engaged in greater information processing (i.e., greater attitudinal differentiation between strong and weak arguments) than certain participants. We speculate that just as explicit uncertainty can guide information processing, so too might implicit uncertainty (see also Petty et al., 2006).

Future Directions Future research might examine other consequences of explicit– implicit discrepancies apart from enhanced elaboration. In particular, research might explore other ways in which people could reduce these discrepancies. These could include changing selfdiscrepant elements (e.g., Festinger, 1957; Harmon-Jones & Mills, 1999), minimizing the salience or perceived (explicit or implicit) importance of the dimension on which the inconsistency exists

167

(e.g., Steele, Southwick, & Critchlow, 1981), or affirming oneself by expressing important values (e.g., Steele & Liu, 1983). Additionally, ignoring or defensively avoiding discrepancy-related information might constitute another strategy when exposure does not automatically occur, as in the present experiments. Different mechanisms of reducing self-inconsistency might be substitutable for each other (e.g., Tesser, 2001). If such mechanisms were interchangeable, then future research might profitably explore the potential differential impact of each of the discrepancy reduction strategies. A final avenue for further examination concerns the possibility of actual resolution of the explicit–implicit discrepancy. For example, in our Study 1 participants received information directly relevant to the dimension (shyness) on which the discrepancy existed. Thus, this information might have influenced their explicit or implicit (or both) self-dimensions, thereby affecting the subsequent explicit–implicit discrepancy. Because follow-up measures related to the self-dimension were not included in that study after message processing, such an influence on the resulting discrepancy could not be assessed. In the other studies of the present research, it was less likely that message processing would help participants to resolve their internal conflict toward their implicit or their explicit self-view because the information provided to them was not actually relevant to resolving the discrepancy. The experiments in this article examine whether people are motivated to process information when they have an explicit–implicit discrepancy, rather than examining the direction or means of discrepancy resolution. This issue constitutes an intriguing question for future consideration.

References Abelson, R. P., Aronson, E., McGuire, W. J., Newcomb, T. M., Rosenberg, M. J., & Tannenbaum, P. H. (1968). Theories of cognitive consistency: A sourcebook. Chicago: Rand McNally. Abelson, R. P., & Rosenberg, M. J. (1958). Symbolic psycho-logic: A model of attitudinal cognition. Behavioral Science, 3, 1–13. Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting interactions. Thousand Oaks, CA: Sage. Aronson, E. (1969). Cognitive dissonance: A current perspective. In L. Bercowitz (Ed.), Advances in experimental social psychology (Vol. 4, pp. 1–34). New York: Academic Press. Asendorpf, J. B. (1987). Videotape reconstruction of emotions and cognitions related to shyness. Journal of Personality and Social Psychology, 53, 542–549. Asendorpf, J. B. (1989). Shyness as a final common pathway for two different kinds of inhibition. Journal of Personality and Social Psychology, 57, 481– 492. Asendorpf, J. B. (1990). The expression of shyness and embarrassment. In W. R. Crozier (Ed.), Shyness and embarrassment: Perspectives from social psychology (pp. 87–118). New York: Cambridge University Press. Asendorpf, J. B., Banse, R., & Mu¨cke, D. (2002). Double dissociation between implicit and explicit personality self-concept: The case of shy behavior. Journal of Personality and Social Psychology, 83, 380 –393. Barger, S. D., Kircher, J. C., & Croyle, R. T. (1997). The effects of social context and defensiveness on the physiological responses of repressive copers. Journal of Personality and Social Psychology, 73, 1118 –1128. Bargh, J. A., Chaiken, S., Govender, R., & Pratto, F. (1992). The generality of the automatic attitude activation effect. Journal of Personality and Social Psychology, 62, 893–912. Bem, D. J., & Allen, A. (1974). On predicting some of the people some of

168

˜ OL, PETTY, AND WHEELER BRIN

the time: The search for cross-situational consistencies in behavior. Psychological Review, 81, 506 –520. Bizer, G. Y., Krosnick, J. A., Holbrook, A. L., Petty, R. E., Wheeler, S. C., & Rucker, D. D. (2004). The impact of personality on cognitive, behavioral, and affective political processes: The effects of need to evaluate. Journal of Personality, 72, 995–1027. Blair, I. V., & Banaji, M. R. (1996). Automatic and controlled processes in stereotype priming. Journal of Personality and Social Psychology, 70, 1142–1163. Bornstein, R. F. (1995). Sex differences in objective and projective dependency tests: A meta-analytic review. Assessment, 2, 319 –331. Bosson, J. K., Brown, R. P., Zeigler-Hill, V., & Swann, W. B., Jr. (2003). Self-enhancement tendencies among people with high self-esteem: The moderating role of implicit self-esteem. Self and Identity, 2, 169 –187. Bosson, J. K., Swann, W. B., Jr., & Pennebaker, J. W. (2000). Stalking the perfect measure of implicit self-esteem: The blind men and the elephant revisited? Journal of Personality and Social Psychology, 79, 631– 643. Brehm, J. W., & Cohen, A. R. (1962). Explorations in cognitive dissonance. New York: Wiley. Brin˜ol, P., Horcajo, J., Becerra, A., Falces, C., & Sierra, B. (2002). Cambio de actitudes implı´citas [Implicit attitude change]. Psicothema, 14, 771– 775. Brin˜ol, P., & Petty, R. E. (2005). Individual differences in persuasion. In D. Albarracı´n, B. T. Johnson, & M. P. Zanna (Eds.), The handbook of attitudes and attitude change (pp. 575– 615). Hillsdale, NJ: Erlbaum. Brin˜ol, P., Petty, R. E., & Wheeler, S. C. (2003, May). Implicit ambivalence: Implications for attitude change. Paper presented at the Attitudinal Incongruence and Information Processing Symposium, University of Amsterdam, Amsterdam, the Netherlands. Brin˜ol, P., Rucker, D., Tormala, Z. L., & Petty, R. E. (2004). Individual differences in resistance to persuasion: The role of beliefs and metabeliefs. In E. S. Knowles & J. A. Linn (Eds.), Resistance and persuasion (pp. 83–104). Mahwah, NJ: Erlbaum. Cacioppo, J. T., Gardner, W. L., & Berntson, G. G. (1997). Beyond bipolar conceptualizations and measures: The case of attitudes and evaluative space. Personality and Social Psychology Review, 1, 3–25. Campbell, J. D. (1990). Self-esteem and clarity of the self-concept. Journal of Personality and Social Psychology, 59, 538 –549. Carlston, D. E., & Skowronski, J. J. (1986). Trait memory and behavior memory: The effects of alternative pathways on impression judgment response times. Journal of Personality and Social Psychology, 50, 5–13. Carver, C. S. (2005). Impulse and constraint: Some perspectives from personality psychology, convergence with theory in other areas, and potential for integration. Personality and Social Psychology Review, 9, 312–333. Carver, C. S., & Scheier, M. F. (1990). Origins and functions of positive and negative affect: A control-process view. Psychological Review, 97, 19 –35. Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Erlbaum. Costelo, R. M., Rice, D. P., & Schoenfeld, L. S. (1974). Attitudinal ambivalence with alcoholic respondents. Journal of Consulting and Clinical Psychology, 42, 303–304. Dovidio, J. F., Kawakami, K., & Beach, K. (2001). Implicit and explicit attitudes: Examination of the relationship between measures of intergroup bias. In R. Brown & S. L. Gaertner (Eds.), Blackwell handbook of social psychology: Vol. 4. Intergroup relations (pp. 175–197). Oxford, England: Blackwell. Emmons, R. A., & Colby, P. M. (1995). Emotional conflict and well-being: Relation to perceived availability, daily utilization, and observer reports of social support. Journal of Personality and Social Psychology, 68, 947–959. Epstein, S. (1973). The self-concept revisited: Or a theory of a theory. American Psychologist, 28, 404 – 416.

Epstein, S. (2003). Cognitive-Experiential Self-Theory of personality. In T. Millon & M. J. Lerner (Eds.), Handbook of psychology (Vol. 5, pp. 159 –184). Hoboken, NJ: Wiley. Farnham, S. D., Greenwald, A. G., & Banaji, M. R. (1999). Implicit self-esteem. In D. Abrams & M. A. Hogg (Eds.), Social identity and social cognition (pp. 230 –248). Oxford, England: Blackwell. Fazio, R. H., Jackson, J. R., Dunton, B. C., & Williams, C. J. (1995). Variability in automatic activation as an unobtrusive measure of racial attitudes: A bona fide pipeline? Journal of Personality and Social Psychology, 69, 1013–1027. Fazio, R. H., & Olson, M. (2003). Implicit measures in social cognition research: Their meaning and uses. Annual Review of Psychology, 54, 297–327. Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7, 117–140. Festinger, L. (1957). A theory of cognitive dissonance. Stanford, CA: Stanford University Press. Gawronski, B., & Bodenhausen, G. V. (in press). Associative and propositional processes in evaluation: An integrative review of implicit and explicit attitude change. Psychological Bulletin. Gilmore, R. F. (1982). Conflict or dissonance: A reinterpretation of counterattitudinal essay writing. Dissertation Abstracts International, 43, 1659. Greenier, K. D., Kernis, M. H., & Waschull, S. B. (1995). Not all high (or low) self-esteem people are the same: Theory and research on stability of self-esteem. In M. H. Kernis (Ed.), Efficacy, agency, and self-esteem (pp. 51–71). New York: Plenum Press. Greenwald, A. G., & Banaji, M. (1995). Implicit social cognition: Attitudes, self-esteem, and stereotypes. Psychological Review, 102, 4 –27. Greenwald, A. G., Banaji, M. R., Rudman, L. A., Farnham, S. D., Nosek, B. A., & Mellott, D. S. (2002). A unified theory of implicit attitudes, stereotypes, self-esteem, and self-concept. Psychological Review, 109, 3–25. Greenwald, A. G., & Farnham, S. D. (2000). Using the Implicit Association Test to measure self-esteem and self-concept. Journal of Personality and Social Psychology, 79, 1022–1038. Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring individual differences in implicit cognition: The implicit association task. Journal of Personality and Social Psychology, 74, 1464 –1480. Gregg, A. P., Seibt, B., & Banaji, M. R. (2006). Easier done than undone: Asymmetry in the malleability of implicit preferences. Journal of Personality and Social Psychology, 90, 1–20. Ha¨nze, M. (2001). Ambivalence, conflict, and decision making: Attitudes and feelings in Germany towards NATO’s military intervention in the Kosovo war. European Journal of Social Psychology, 31, 693–706. Harmon-Jones, E., & Mills, J. S. (Eds.). (1999). Cognitive dissonance: Progress on a pivotal theory in social psychology. Washington, DC: American Psychological Association. Hass, R. G., Katz, I., Rizzo, N., Bailey, J., & Moore, L. (1992). When racial ambivalence evokes negative affect, using a disguised measure of mood. Personality and Social Psychology Bulletin, 18, 786 –797. Heider, F. (1958). The psychology of interpersonal relations. New York: Wiley. Hermans, D., De Houwer, J., & Eelen, P. (2001). A time course analysis of the affective priming effect. Cognition and Emotion, 15, 143–165. Hetts, J. J., & Pelham, B. W. (2001). A case for the nonconscious selfconcept. In G. B. Moskowitz (Ed.), Cognitive social psychology: The Princeton symposium on the legacy and future of social cognition (pp. 105–123). Washington, DC: American Psychological Association. Hetts, J. J., Sakuma, M., & Pelham, B. W. (1999). Two roads to positive regard: Implicit and explicit self-evaluation and culture. Journal of Experimental Social Psychology, 35, 512–559. Higgins, E. T. (1987). Self-discrepancy: A theory relating self and affect. Psychological Review, 94, 319 –340.

EXPLICIT–IMPLICIT DISCREPANCY Hodson, G., Maio, G. R., & Esses, V. M. (2001). The role of attitudinal ambivalence in susceptibility to consensus informative. Basic and Applied Social Psychology, 23, 197–205. Jarvis, W. B. G. (2000). MediaLab (Version Beta 2000) [Computer software]. Columbus, OH: Empirisoft. Jarvis, W. B. G., & Petty, R. E. (1996). The need to evaluate. Journal of Personality and Social Psychology, 70, 172–194. Jonas, K., Diehl, M., & Bromer, P. (1997). Effects of attitudinal ambivalence on information processing and attitude-intention consistency. Journal of Experimental Social Psychology, 33, 190 –210. Jones, J. T., Pelham, B. W., Mirenberg, M. C., & Hetts, J. J. (2002). Name letter preferences are not merely mere exposure: Implicit egotism as self-regulation. Journal of Experimental Social Psychology, 38, 170 – 177. Jordan, C. H., Spencer, S. J., Zanna, M. P., Hoshino-Browne, E., & Correll, J. (2003). Secure and defensive high self-esteem. Journal of Personality and Social Psychology, 85, 969 –978. 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. Karpinski, A. (2004). Measuring self-esteem using the Implicit Association Test: The role of the other. Personality and Social Psychology Bulletin, 30, 22–34. Katz, I., Wackenhut, J., & Hass, R. G. (1986). Racial ambivalence, value duality, and behavior. In J. F. Dovidio & S. L. Gaertner (Eds.), Prejudice, discrimination, and racism (pp. 35–59). New York: Academic Press. Kehr, H. M. (2004). Implicit/explicit motive discrepancies and volitional depletion among managers. Personality and Social Psychology Bulletin, 30, 315–327. Kernis, M. H., & Waschull, S. B. (1995). The interactive roles of stability and level of self-esteem: Research and theory. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 27, pp. 93–141). San Diego, CA: Academic Press. Kitayama, S., & Uchida, Y. (2003). Explicit self-criticism and implicit self-regard: Evaluating self and friend in two cultures. Journal of Experimental Social Psychology, 39, 476 – 482. Koole, S. L., Dijksterhuis, A., & van Knippenberg, A. (2001). What’s in a name: Implicit self-esteem and the automatic self. Journal of Personality and Social Psychology, 80, 669 – 685. Kruglanski, A. W., Thompson, E. P., Higgins, E. T., Atash, M. N., Pierro, A., Shah, J. Y., & Spiegel, S. (2000). To “do the right thing” or to “just do it”: Locomotion and assessment as distinct self-regulatory imperatives. Journal of Personality and Social Psychology, 79, 793– 815. Lieberman, M. D. (2000). Intuition: A social cognitive neuroscience approach. Psychological Bulletin, 126, 109 –137. Lindsay, D. S., & Johnson, M. K. (1989). The eyewitness suggestibility effect and memory for source. Memory & Cognition, 17, 349 –358. Maio, G. R., Bell, D. E., & Esses, V. M. (1996). Ambivalence and persuasion: The processing of messages about immigrant groups. Journal of Experimental Social Psychology, 32, 513–536. Maio, G. R., & Olson, J. M. (Eds.). (2000). Why we evaluate: Functions of attitudes. Mahwah, NJ: Erlbaum. McClelland, D. C. (1985). How motives, skills, and values determine what people do. American Psychologist, 40, 812– 825. McCrae, R. R., & Costa, P. T., Jr. (1985). Openness to experience. In R. Hogan & W. H. Jones (Eds.), Perspectives in personality (Vol. 1, pp. 145–172). Greenwich, CT: JAI Press. McCrae, R. R., & Costa, P. T., Jr. (1997). Conceptions and correlates of openness to experience. In R. Hogan, J. Johnson, & S. Briggs (Eds.), Handbook of personality psychology (Vol. 1, pp. 825– 847). New York: Academic Press. McGregor, I., & Marigold, D. C. (2003). Defensive zeal and the uncertain

169

self: What makes you so sure? Journal of Personality and Social Psychology, 85, 838 – 852. Mendolia, M., Moore, J., & Tesser, A. (1996). Dispositional and situational determinants of repression. Journal of Personality and Social Psychology, 70, 856 – 867. Metcalfe, J., & Mischel, W. (1999). A hot/cool-system analysis of delay of gratification: Dynamics of willpower. Psychological Review, 106, 3–19. Mussweiler, T. (2003). Comparison processes in social judgment: Mechanisms and consequences. Psychological Review, 110, 472– 489. Newcomb, T. M. (1968). Interpersonal balance. In R. Abelson et al. (Eds.), Theories of cognitive consistency: A sourcebook (pp. 28 –51). Chicago: Rand McNally. Newman, L. S., Duff, K. J., & Baumeister, R. F. (1997). A new look at defensive projection: Thought suppression, accessibility, and biased person perception. Journal of Personality and Social Psychology, 72, 980 –1001. Norton, G. R. (1975). Measurement of ambiguity tolerance. Journal of Personality Assessment, 39, 607– 619. Osgood, C. E., & Tannenbaum, P. H. (1955). The principle of congruity in the prediction of attitude change. Psychological Review, 62, 42–55. Patten, R. L., & White, L. A. (1977). Independent effects of achievement behavior. Motivation and Emotion, 1, 39 –59. Petty, R. E. (in press). A meta-cognitive model of attitudes. Journal of Consumer Research. Petty, R. E., & Cacioppo, J. T. (1984). The effects of involvement on responses to argument quantity and quality: Central and peripheral routes to persuasion. Journal of Personality and Social Psychology, 46, 69 – 81. Petty, R. E., & Cacioppo, J. T. (1986). Communication and persuasion: Central and peripheral routes to attitude change. New York: SpringerVerlag. Petty, R. E., Fazio, R. H., & Brin˜ol, P. (Eds.). (in press). Attitudes: Insights from the new implicit measures. Mahwah, NJ: Erlbaum. Petty, R. E., & Jarvis, B. G. (1996). An individual differences perspective on assessing cognitive processes. In N. Schwarz & S. Sudman (Eds.), Answering questions: Methodology for determining cognitive and communicative processes in survey research (pp. 221–257). San Francisco: Jossey-Bass. Petty, R. E., Tormala, Z. L., Brin˜ol, P., & Jarvis, W. B. G. (2006). Implicit ambivalence from attitude change: An exploration of the PAST model. Journal of Personality and Social Psychology, 90, 21– 41. Petty, R. E., & Wegener, D. T. (1998). Attitude change: Multiple roles for persuasion variables. In D. Gilbert, S. Fiske, & G. Lindzey (Eds.). The handbook of social psychology (4th ed., Vol. 1, pp. 323–390). New York: McGraw-Hill. Petty, R. E., Wells, G. L., & Brock, T. C. (1976). Distraction can enhance or reduce yielding to propaganda: Thought disruption versus effort justification. Journal of Personality and Social Psychology, 34, 874 – 884. Petty, R. E., Wheeler, S. C., & Tormala, Z. L. (2003). Persuasion and attitude change. In T. Millon & M. J. Lerner (Eds.), Handbook of psychology: Vol. 5. Personality and social psychology (pp. 353–382). Hoboken, NJ: Wiley. Priester, J. R., & Petty, R. E. (1996). The gradual threshold model of ambivalence: Relating the positive and negative bases of attitudes to subjective ambivalence. Journal of Personality and Social Psychology, 71, 431– 449. Priester, J. R., & Petty, R. E. (2001). Extending the bases of subjective attitudinal ambivalence: Interpersonal and intrapersonal antecedents of evaluative tension. Journal of Personality and Social Psychology, 80, 19 –34. Proshansky, H. M. (1943). A projective method for the study of attitudes. Journal of Abnormal and Social Psychology, 38, 393–395.

170

˜ OL, PETTY, AND WHEELER BRIN

Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University Press. Roskos-Ewoldsen, D. R., & Fazio, R. H. (1992). On the orienting value of attitudes: Attitude accessibility as a determinant of an object’s attraction of visual attention. Journal of Personality and Social Psychology, 63, 198 –211. Schwartz, S. H. (1992). Universals in the content and structure of values: Theoretical advances and empirical test in 20 countries. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 24, pp. 1– 65). San Diego, CA: Academic Press. Shastri, L., & Ajjanagadde, V. (1993). From simple associations to systematic reasoning: A connectionist representation of rules, variables, and dynamic bindings using temporal synchrony. Behavioral and Brain Sciences, 16, 417– 494. Shedler, J., Mayman, M., & Manis, M. (1993). The illusion of mental health. American Psychologist, 48, 1117–1131. Sloman, S. A., (1996). The empirical case for two forms of reasoning. Psychological Bulletin, 119, 3–22. 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. Smolensky, P. (1988). On the proper treatment of connectionism. Behavioral and Brain Sciences, 11, 1–23. Steele, C. M., & Liu, T. J. (1983). Dissonance processes as selfaffirmation. Journal of Personality and Social Psychology, 45, 5–19. Steele, C. M., Southwick, L., & Critchlow, B. (1981). Dissonance and alcohol: Drinking your troubles away. Journal of Personality and Social Psychology, 41, 831– 846. Strack, F., & Deutsch, R. (2004). Reflective and impulsive determinants of social behavior. Personality and Social Psychology Review, 8, 220 –247. Tesser, A. (2001). On the plasticity of self-defense. Current Directions in Psychological Science, 10, 66 – 69. 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.

Tormala, Z. L., & Petty, R. E. (2001). On-line versus memory-based processing: The role of “need to evaluate” in person perception. Personality and Social Psychology Bulletin, 27, 1599 –1612. Weary, G., & Jacobson, J. A. (1997). Causal uncertainty beliefs and diagnostic information seeking. Journal of Personality and Social Psychology, 73, 839 – 848. Weinberger, D. M. (1990). The construct validity of the repressive coping style. In J. L. Singer (Ed.), Repression and dissociation: Implications for personality, psychopathology, and health (pp. 337–386). Chicago: University of Chicago Press. Weinberger, J., & Hardaway, R. (1990). Separating science from myth in subliminal psychodynamic activation. Clinical Psychology Review, 10, 727–756. Wiggins, J. S., & Trapnell, P. D. (1997). Personality structure: The return of the Big Five. In R. Hogan, J. Johnson, & S. Briggs (Eds.), Handbook of personality psychology (Vol. 1, pp. 737–765). San Diego, CA: Academic Press. Wilson, T. D., Lindsey, S., & Schooler, T. Y. (2000). A model of dual attitudes. Psychological Review, 107, 101–126. Wing, P., & Gough, H. G. (1990). New narcissism scales for the California Psychological Inventory and MMPI. Journal of Personality Assessment, 54, 3– 4. Wright, R. (2001). Self-certainty and self-esteem. In T. J. Owens, S. Stryker, & N. Goodman (Eds.), Extending self-esteem theory and research: Sociological and psychological currents (pp. 101–134). Cambridge, England: Cambridge University Press. Zelenski, J. M., & Larsen, R. J. (2003, February). The importance of knowing what we want: How implicit and explicit motives interact to predict well-being. Paper presented at the meeting of the Society for Personality and Social Psychology, Los Angeles, CA.

Received June 2, 2003 Revision received February 9, 2006 Accepted February 17, 2006 䡲

Journal of Personality and Social Psychology 2006, Vol. 91, No. 1, 16 –32

Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.16

The Single Category Implicit Association Test as a Measure of Implicit Social Cognition Andrew Karpinski and Ross B. Steinman Temple University The Single Category Implicit Association Test (SC-IAT) is a modification of the Implicit Association Test that measures the strength of evaluative associations with a single attitude object. Across 3 different attitude domains—soda brand preferences, self-esteem, and racial attitudes—the authors found evidence that the SC-IAT is internally consistent and makes unique contributions in the ability to understand implicit social cognition. In a 4th study, the authors investigated the susceptibility of the SC-IAT to faking or self-presentational concerns. Once participants with high error rates were removed, no significant self-presentation effect was observed. These results provide initial evidence for the reliability and validity of the SC-IAT as an individual difference measure of implicit social cognition. Keywords: implicit social cognition, associative processes, individual differences

Over the past 20 years, there has been an increasing awareness that much social cognition occurs outside of conscious awareness or conscious control (Bargh & Ferguson, 2000; Greenwald, 1992; Kihlstrom, 1990). Implicit social cognition may be inaccessible to conscious introspection, and thus it is necessary to develop measures that do not rely on introspection and self-report in order to understand and measure these processes (Greenwald & Banaji, 1995). One particularly fruitful approach to measuring implicit social cognition has been the development of individual difference measures of associative strength. Although a large number of these association-based measures of implicit social cognition have been developed (for a review, see Fazio & Olson, 2003), the Implicit Association Test (IAT; Greenwald, McGhee, & Schwartz, 1998) has become the most commonly used among the implicit measurement techniques because it is reliable, easy to administer, and robust and produces large effect sizes, particularly in comparison to other implicit measures of social cognition (Greenwald et al., 1998; Greenwald & Nosek, 2001). The IAT is unique among the recently developed association-based measures of implicit social cognition because it cannot reveal the evaluative associations with a single target concept.

young– old, weak–strong, warm– cold, liberal– conservative, aggressive– peaceful, and so forth. (Greenwald & Farnham, 2000, p. 1023)

This property is both a strength and a limitation of the measure. As highlighted by Greenwald and Farnham (2000), many attitude objects have a complementary category, and it makes sense to consider these attitude objects relative to another category. Yet for some research questions, evaluative associations with a single target concept may be of interest. For example, to measure self-esteem by using the IAT, researchers have measured the positive and negative associations a person has with the self in comparison to an unspecified other (or with me in comparison to not me). At the same time, an alternative approach to measuring self-esteem would be to measure only evaluative associations with the self with no complementary category (see Karpinski, 2004). This approach is not possible within the standard IAT paradigm. Furthermore, there are also instances in which the choice of a complement is not obvious. Consider a researcher interested in predicting President Bush’s job approval rating. This researcher may want to obtain a measure of evaluative associations with President Bush, but relative to whom? In such instances, it may be useful to have an IAT-type task that does not require the use of a complementary category (see also Blanton & Jaccard, 2006; Blanton, Jaccard, Gonzales, & Christie, 2006; De Houwer, 2002). In addition, greater information may be obtained by measuring the evaluative associations with two concept domains independently rather than examining only comparative associations. Whereas measures of two concept domains can reveal two dimensions of information, the IAT provides only one. For example, on a Black–White IAT, scores are interpreted as a comparison of one’s positive White associations and/or negative Black associations with one’s negative White associations and/or positive Black associations. A high score could indicate (a) the presence of many positive White associations, (b) the presence of many negative Black associations, (c) the lack of negative White associations, and/or (d) the lack of positive Black associations. From the single IAT score, it is impossible to determine which of these factors, or which combination of these factors, contributes to the overall score

Because it uses complementary pairs of concepts and attributes, the IAT is limited to measuring the relative strengths of pairs of associations rather than absolute strengths of single associations. In practice, however, the IAT can nevertheless be effectively used because many socially significant categories form complementary pairs, such as positive–negative (valence), self– other, male–female, Jewish–Christian,

Andrew Karpinski and Ross B. Steinman, Department of Psychology, Temple University. Ross B. Steinman is now at the Department of Psychology, Widener University. We thank Shelley Keiper and Jennifer Steinberg for their helpful comments. Correspondence concerning this article should be addressed to Andrew Karpinski, Department of Psychology, Temple University, 1701 North 13th Street, Philadelphia, PA 19122-6085. E-mail: [email protected] 16

SINGLE CATEGORY IMPLICIT ASSOCIATION TEST

17

attitude object. The Single Target IAT (ST-IAT; Wigboldus, Holland, & van Knippenberg, 2005) is conceptually identical to our SC-IAT, differing only in minor procedural details. In an initial test of its utility, a Christian ST-IAT was found to correlate with explicit attitude questions about Christianity, and an Islamic STIAT was found to correlate with explicit attitude questions about Islam. Neither of the implicit measures correlated with a comparative Christianity versus Islam explicit attitude measure. Additionally, the difference between the Christian ST-IAT and Islamic ST-IAT correlated weakly and only marginally significantly with a Christian–Islamic IAT (Wigboldus et al., 2005). These findings provide some empirical evidence that a measure of associations with a single category or target may reveal different information than a comparative IAT; however, additional research is needed to determine whether single-category or target IAT measures are reliable, valid across content domains, and relatively impervious to self-presentation.

(Blanton & Jaccard, 2006; Blanton et al., 2006; Nosek, Greenwald, & Banaji, 2005). If a single category IAT-type task were available, then a measure of the evaluative associations with Whites and of the evaluative associations with Blacks could be obtained independently, to eliminate some of the ambiguity in the interpretation of IAT scores. Several implicit social cognition measures have been developed to assess evaluative associations with a single attitude object. Priming-based measures (see Dovidio, Kawakami, Johnson, Johnson, & Howard, 1997; Fazio, Jackson, Dunton, & Williams, 1995; Wittenbrink, Judd, & Park, 1997), the Go/No-Go Association Task (Nosek & Banaji, 2001), and the Extrinsic Affective Simon Task (De Houwer, 2003) can all be used to assess associations with a single attitude object. Each of these measures may be used to assess the evaluative associations with a single target; however, the reliability of these measures has been very low (see Bosson, Swann, & Pennebaker, 2000; De Houwer, 2003; Nosek & Banaji, 200l; Olson & Fazio, 2003; Teige, Schnable, Banse, & Asendorpf, 2004). Thus, there is still the need for a reliable individualdifference measure of the evaluative associations with a single attitude object.

Overview of Studies In the four studies that follow, we examined the reliability, validity, and susceptibility to faking of the SC-IAT in four different concept domains. For all four studies, we also examined the IAT and explicit measures of attitudes. In cases where the concept of interest was a comparative domain, a difference between two SC-IATs (such as a White SC-IAT and a Black SC-IAT) was computed to obtain a comparative SC-IAT score. We expected an IAT and a comparative SC-IAT to reveal similar findings in terms of known groups validity, correlations with explicit measures, and predictive validity. Unlike the IAT, a comparative SC-IAT can be decomposed into its components, and thus, the SC-IATs may allow for specific conclusions in these domains, to supplement the overall findings obtained from the IAT and comparative SC-IAT. In cases where measurement of associations with a single category may be preferable to a comparative measure, compared with the IAT, we expected the SC-IAT to provide unique and meaningful information about the category of interest. We determined a fixed order of measures for each study in order to reduce the large sample size requirements if we were to counterbalance the presentation of all the measures. We presented all implicit measures of attitudes prior to the explicit measures of

The Single Category IAT (SC-IAT) Another possible type of measure to assess the strength of evaluative associations with a single attitude object is a modified version of the IAT that eliminates the need for the second contrast category. We designed the SC-IAT as a two-stage modification of IAT procedure to measure the evaluative associations with a single category or attitude object. Because the SC-IAT is a modification of IAT procedure, it shares many properties with the IAT, including its ease of use and interpretation. In each stage, target words associated with the attitude object and an evaluative dimension are presented in random order. In the first stage, good words and attitude object words are categorized on one response key, and bad words are categorized on a different key. In the second stage, bad words and attitude object words are categorized on one response key, and good words are categorized on a different key (see Table 1 for a comparison of a self– other IAT to a self-SC-IAT). Independently, other researchers have developed similar modifications of the IAT to assess evaluative associations with a single

Table 1 Comparison of the Implicit Association Test (IAT) and Single Category IAT (SC-IAT) IAT

SC-IAT

Block

Trials

Function

Left-key response

1 2 3a

30 30 30

Practice Practice Practice

4a

30

Test

5 6b

30 30

Practice Practice

7b

30

Test

Pleasant words Self words Pleasant words self words Pleasant words self words Other words Pleasant words other words Pleasant words other words

⫹ ⫹ ⫹ ⫹

Right-key response Unpleasant words Other words Unpleasant words other words Unpleasant words other words Self words Unpleasant words self words Unpleasant words self words

Block

Trials

Function

Left-key response

⫹

1c

24

Practice

Bad words

⫹

2c

72

Test

Good words ⫹ self words Good words ⫹ self words

⫹

3d

24

Practice

Good words

Bad words ⫹ self words

⫹

4d

72

Test

Good words

Bad words ⫹ self words

Note. Blocks with a common subscript were experienced as one continuous block.

Right-key response

Bad words

KARPINSKI AND STEINMAN

18

attitudes. Recently, Nosek and colleagues (2005) have argued that the order of presentation of implicit and explicit measures does not alter their psychometric properties or their intercorrelation. Other studies have found evidence that the completion of explicit measures of attitudes affects subsequent responses on implicit measures and artificially increases the observed correlation between implicit and explicit measures of attitudes (Bosson et al., 2000). We know of no studies where completing the implicit measures has significantly affected the responses on subsequent explicit measures. Thus, to be conservative, we always presented the implicit measures prior to the explicit self-report measures.

Study 1 The goal of Study 1 was to examine the SC-IAT in a situation where a comparative measure, such as the IAT, would be useful. We investigated attitudes toward soda brands, with the goal of predicting whether participants preferred Coke products to Pepsi products. Previous studies using the IAT to examine evaluative associations with Coke and Pepsi have measured associations at the product level (such as associations with the beverages CocaCola and Pepsi-Cola; see Maison, Greenwald, & Bruin, 2004). In Study 1, we used the IAT and SC-IAT to measure associations at the brand level. The Coke brand included the beverages Coke and Diet Coke; the Pepsi brand included the beverages Pepsi and Pepsi One. As a consequence, the results of this study were expected to be similar, but not identical to, the results of previously conducted studies. Because this study was the first test of the SC-IAT methodology, we examined whether the SC-IAT revealed known soda attitudes. That is, the SC-IAT should reveal that Coke product drinkers have more favorable associations with Coke products than with Pepsi products and that Pepsi product drinkers have more favorable associations with Pepsi products than with Coke products. When participants choose between a Coke and a Pepsi product, the outcome variable is a dichotomous choice. Because the outcome variable is a comparative, dichotomous choice, a comparative Coke–Pepsi IAT would seem to be an ideal attitude measure. As a consequence, we expected a soda SC-IAT (the difference between a Coke SC-IAT and a Pepsi SC-IAT) to be similar in its effects to a Coke–Pepsi IAT, with regards to known group differences, correlations with explicit attitude measures, and prediction of soda choice.

Method Participants Fifty-six students (41 women, 15 men) enrolled in an introductory psychology course at Temple University participated in this experiment. All participants received course credit for their participation.

Procedure Participants were tested in groups of up to 3 at a time. Each participant was seated at a desk with a Gateway 1.5 Gz Pentium 4 desktop computer using Medialab and Direct RT software. All tasks were presented on the computer, and all participants completed the tasks in the same order: a Coke–Pepsi IAT measure of soda associations, a Coke SC-IAT, a Pepsi SC-IAT, and explicit measures of soda preferences. At the conclusion of the session, the participants were thanked and completely debriefed.

IAT measure of soda brand associations. A Coke–Pepsi IAT procedure followed the standard IAT paradigm (see Greenwald et al., 1998) with minor modifications. Specifically, participants completed seven stages in the same order (see Table 1 for an example). In each stage, participants responded to 30 target presentations, and the target stimuli were selected randomly, without replacement. The evaluative dimension was labeled pleasant and unpleasant, and the object dimension was labeled Coke and Pepsi. Five target words were used for each of the evaluative dimensions (pleasant: brilliant, diamond, joy, truth, and sunrise; unpleasant: awkward, hate, failure, slum, and stink). All target words were presented in lowercase letters. Five target pictures were also selected to be associated with the Coke brand (pictures of two-liter bottles and six-packs of cans of Coke and Diet Coke) and with the Pepsi brand (pictures of two-liter bottles and six-packs of cans of Pepsi and Pepsi One). The procedure of the IAT was similar to the SC-IAT with five exceptions. First, participants responded by using the a key and the 5 key on the number keypad to categorize target words and pictures. Second, the target word remained on the screen until the participants responded. Third, participants were not given feedback regarding the accuracy of their responses. Fourth, the category reminder labels were appropriately positioned in the center of the screen, immediately to the left or right side of the target word. Fifth, the evaluative dimension was labeled pleasant and unpleasant. SC-IAT measure of Coke and Pepsi brand associations. The Coke SC-IAT consisted of two stages, which all participants completed in the same order. Each stage consisted of 24 practice trials immediately followed by 72 test trials (three blocks of 24 trials each). In the first stage (Coke ⫹ good), Coke pictures and good words were categorized on the z key, and bad words were categorized on the 2 key on the numeric keypad. In an attempt to prevent a response bias from developing, Coke pictures, good words, and bad words were not presented at equal frequency, but were presented in a 7:7:10 ratio so that 58% of correct responses were on the z key and 42% of correct responses were on the 2 key. In the second stage (Coke ⫹ bad), good words were categorized on the z key, and Coke pictures and bad words were categorized on the 2 key on the numeric keypad. Coke pictures, good words, and bad words were presented in a 7:10:7 ratio so that 42% of correct responses were on the z key and 58% of correct responses were on the 2 key. The evaluative dimension was labeled good and bad, and the object dimension was labeled Coke. Twenty-one target words were used for each of the evaluative dimensions (see Appendix), and all target words were presented in lowercase letters. Seven target pictures were selected to be associated with Coke (pictures of six-packs and two-liter bottles of Coke and Diet Coke). Within each category, words and pictures were selected randomly without replacement. Each stage was preceded by a set of instructions concerning the dimensions of the categorization task and the appropriate key responses. Each target word or picture appeared centered on the screen. Category reminder labels were appropriately positioned on the bottom fourth of the screen. The target word remained on the screen until the participants responded or for 1,500 ms. If participants failed to respond within 1,500 ms, a reminder to “Please respond more quickly!” appeared for 500 ms. This response window is largely window dressing; pilot testing revealed that the response window truncates less than 1% of all critical responses. Nevertheless, the response window creates a sense of urgency and may decrease the likelihood that participants engage in controlled processing during the task. Following each response, participants were given feedback regarding the accuracy of their response. A green O in the center of the screen for 150 ms followed correct responses; a red X in the center of the screen for 150 ms followed each incorrect response. For the Pepsi SC-IAT, the procedure was repeated with the target category Pepsi and target pictures of Pepsi products (pictures of six-packs and two-liter bottles of Pepsi and Pepsi One). All participants completed the Pepsi ⫹ good task followed by the Pepsi ⫹ bad task.

SINGLE CATEGORY IMPLICIT ASSOCIATION TEST Explicit measures of soda preferences. Next, participants completed semantic differential, feeling thermometer, and rating scale measures regarding their attitudes toward the Coke and Pepsi brands. For the semantic differential, participants rated the Coke and Pepsi brands on five bipolar dimensions: ugly– beautiful, bad– good, unpleasant–pleasant, foolish– wise, and awful–nice. Each dimension was rated on a 7-point scale ranging from ⫺3 (the negative pole) to 3 (the positive pole), and participants were instructed to circle zero if the anchoring adjectives were irrelevant to the concept (Coke, ␣ ⫽ .88; Pepsi, ␣ ⫽ .92). A semantic differential measure of soda brand preference was computed by subtracting semantic differential ratings of the Pepsi brand from semantic differential ratings of the Coke brand. For the feeling thermometer, participants were asked to rate how positive or negative they found the Coke and Pepsi brands on a scale from 0 (extremely negative) to 100 (extremely positive). A feeling thermometer measure of soda brand preference was computed by subtracting feeling thermometer ratings of Pepsi from thermometer ratings of Coke. A rating scale measure of Coke brand enjoyment was obtained by asking participants to indicate their agreement or disagreement on a 6-point scale with the following statements: “I enjoy drinking Coke (and Coke products)” and “Coke products satisfy my thirst.” Higher numbers indicated more agreement with the statement (Coke, ␣ ⫽ .49). These questions were repeated with Pepsi as the target brand (Pepsi, ␣ ⫽ .67). A rating scale measure of soda preference was computed by subtracting ratings of Pepsi from ratings of Coke. All three explicit brand attitude measures correlated strongly with each other (Coke attitudes, ␣ ⫽ .81; Pepsi attitudes, ␣ ⫽ .90; and soda [Coke – Pepsi] attitudes, ␣ ⫽ .92). As a result, the three explicit measures were standardized and averaged to create standardized explicit brand attitude ratings of Coke, Pepsi, and soda (Coke – Pepsi). Finally, participants answered a behavioral intention question. Participants indicated whether they would prefer a free Coke or Pepsi product (indifference and refusal were also response options). On the basis of this question, participants were defined as Coke drinkers (n ⫽ 17), Pepsi drinkers (n ⫽ 30), or neither (n ⫽ 6).

Results

19

more positive than negative associations with Coke and Pepsi. Finally, a soda SC-IAT D score was computed by subtracting the Pepsi SC-IAT D score from the Coke SC-IAT D score.

Reliability of the SC-IAT and IAT To determine the reliability of the SC-IAT, we divided each SC-IAT into thirds (blocks of 24 test trials) and calculated a SC-IAT score separately for each third of the trials without dividing by the standard deviation of correct response times. A measure of internal consistency was obtained by calculating the average intercorrelation among these scores. Dividing the task into thirds (or halves) underestimates the reliability of the entire measure. Fortunately, the Spearman–Brown correction can be applied to compensate for this underestimate of the true internal consistency for the entire measure (designated adjusted r; Nunnally, 1978). All internal consistency correlations reported in this article have been adjusted by using the Spearman–Brown correction. These adjusted reliability coefficients are conceptually equivalent and directly comparable to the Cronbach’s alphas computed for the explicit measures. A reliability analysis on the SC-IAT measures from Study 1 revealed a reasonable level of internal consistency (Coke, adjusted r ⫽ .61; Pepsi, adjusted r ⫽ .69). For the IAT, a reliability correlation was computed by correlating the IAT score computed from the practice trials with an IAT score computed from the test trials (following the procedure outlined by Greenwald et al., 2003; adjusted r ⫽ .82). Overall, the reliability of the SC-IAT is somewhat low compared with the IAT. However, the reliability of the SC-IAT is similar to the reliability typically observed for IAT measures (see Greenwald et al., 2003; Nosek et al., 2005) and higher than the reliability of other implicit measures (see Bosson et al., 2000; Olson & Fazio, 2003).

IAT and SC-IAT Data Reduction Compared with the IAT, error rates were significantly higher on both the Coke SC-IAT, t(55) ⫽ 4.19, p ⬍ .01, and the Pepsi SC-IAT, t(55) ⫽ 3.61, p ⬍ .01. This result is not surprising given that the response window in the SC-IAT procedure was included to facilitate quick responding, and quicker responding is likely to be accompanied by increased error rates. Participants with an error rate greater than 20% on the soda IAT, the Coke SC-IAT, or the Pepsi SC-IAT were excluded from analysis, resulting in the elimination of 3 participants (average error rates: Coke SC-IAT ⫽ 6.60%; Pepsi SC-IAT ⫽ 6.43%; soda IAT ⫽ 3.25%). IAT scores were computed by using the newer D-score algorithm for IAT data (Greenwald, Nosek, & Banaji, 2003). For the resulting IAT scores, higher numbers indicated a bias for Coke compared with Pepsi. For the SC-IAT, a scoring algorithm was modeled on the D-score algorithm used for the IAT data. Because the 24 practice trials in each stage were truly practice, data from the practice blocks were discarded (Blocks 1 and 3). Responses less than 350 ms were eliminated, nonresponses were eliminated, and error responses were replaced with the block mean plus an error penalty of 400 ms. The average response times of Block 2 (e.g., Coke ⫹ good) were subtracted from the average response times of Block 4 (e.g., Coke ⫹ bad). This quantity was divided by the standard deviation of all correct response times within Blocks 2 and 4. Thus, Coke SC-IAT and Pepsi SC-IAT D scores indicate

Implicit and Explicit Measures of Soda Attitudes First, we divided the sample into Coke drinkers and Pepsi drinkers, on the basis of the behavioral choices. The predicted differences emerged on all the comparative soda measures (see top of Table 2). Coke drinkers displayed a greater bias for Coke compared with Pepsi than did Pepsi drinkers on the IAT, SC-IAT, and explicit attitude measures ( ps ⬍ .02, ds ⱖ 0.87). Single category measures also tended to reveal the expected group differences (see bottom of Table 2). Coke drinkers had more favorable explicit attitudes toward Coke than did Pepsi drinkers (d ⫽ 0.69), and Pepsi drinkers had more favorable attitudes toward Pepsi than did Coke drinkers (d ⫽ 2.40). A Coke SC-IAT revealed no significant difference in evaluative Coke associations for Coke and Pepsi drinkers (d ⫽ 0.20), but a Pepsi SC-IAT revealed that Pepsi drinkers had more positive associations with Pepsi than did Coke drinkers (d ⫽ 0.94).

Correlational Analyses Soda IAT scores were significantly correlated with soda SCIAT scores, r(51) ⫽ .29, p ⫽ .04, suggesting that the soda brand associations measured by the SC-IAT were significantly related to the soda brand associations measured by the IAT, as expected. However, soda IAT scores were unrelated to explicit soda ratings,

KARPINSKI AND STEINMAN

20

Table 2 Study 1: Summary Statistics for the IAT and Explicit Attitude Measures by Soda Preference

Attitude measure

Coke drinkers (n ⫽ 17)

Pepsi drinkers (n ⫽ 30)

M

M

SD

Difference

SD

d

t(45)

p

0.44 0.53 0.66

0.87 0.74 2.11

2.91 2.47 7.08

⬍.01 .02 ⬍.01

Comparative measures Soda IAT Soda SC-IAT Explicit soda attitudes

0.35 0.34 0.91

0.21 0.44 0.72

0.02 ⫺0.03 ⫺0.56

Single category measures SC-IAT Coke SC-IAT Pepsi SC-IAT Explicit attitudes Explicit Coke attitude Explicit Pepsi attitude

0.25 ⫺0.09

0.30 0.33

0.18 0.22

0.35 0.31

0.20 0.94

0.68 3.16

.50 ⬍.01

0.42 ⫺0.79

0.83 0.77

⫺0.15 0.60

0.80 0.42

0.69 2.40

2.32 8.08

.03 ⬍.01

Note. For comparative soda measures, positive values indicate a bias or preference for Coke over Pepsi. For single category Coke and Pepsi measures, positive values indicate a bias or preference for Coke or Pepsi, respectively. IAT ⫽ Implicit Association Test; SC-IAT ⫽ Single Category Implicit Association Test.

r(51) ⫽ .18, p ⫽ .20. Follow-up analyses revealed that soda IAT scores were also unrelated to explicit Coke or explicit Pepsi attitudes, r(51) ⫽ .21, p ⫽ .14, and r(51) ⫽ ⫺.05, p ⫽ .74, respectively. As expected, soda SC-IAT scores were positively correlated with explicit soda attitudes, r(51) ⫽ .29, p ⫽ .04. Follow-up analyses revealed that Coke SC-IAT scores were positively correlated with explicit Coke attitudes, r(51) ⫽ .27, p ⫽ .05, and uncorrelated with explicit Pepsi attitudes, r(51) ⫽ ⫺.01, p ⫽ .97. Similarly, Pepsi SC-IAT scores were positively correlated with explicit Pepsi attitudes, r(51) ⫽ .26, p ⫽ .06, and uncorrelated with explicit Coke attitudes, r(51) ⫽ ⫺.07, p ⫽ .61. These results suggest that Coke and Pepsi SC-IAT scores reveal specific information pertaining to Coke and Pepsi preferences, respectively, and not general information about soda preferences. Additionally, attitudes and associations with Coke were unrelated to attitudes and associations with Pepsi on both explicit ratings, r(51) ⫽ ⫺.01, p ⫽ .93, and the SC-IAT measures, r(51) ⫽ ⫺.06, p ⫽ .66.

Prediction of Soda Choice Next, we examined the ability of the IAT, SC-IAT, and explicit measures to predict soda choice. For the soda choice outcome variable, participants who chose a Coke product were given a value of 1, participants who indicated no preference were given a value of 0, and participants who chose a Pepsi product were given a value of ⫺1. First, the Coke SC-IAT, Pepsi SC-IAT, and soda IAT were entered in a simultaneous regression predicting soda choice (see top of Table 3). These results indicated that both the IAT and Pepsi SC-IAT significantly predicted soda choice, but the Coke SC-IAT was unrelated to soda choice. Next, the analysis was repeated with explicit soda attitudes added to the model (see bottom of Table 3). Explicit attitudes were strong predictors of soda choice. However, even once explicit preferences were con-

trolled, the IAT, Coke SC-IAT, and Pepsi SC-IAT significantly predicted soda choice.1 An alternative method of comparing the predictive validity of the SC-IAT and the IAT is to examine the percentage of variance accounted for by SC-IAT and IAT measures in soda choice. A series of regression analyses revealed that the Coke and Pepsi SC-IATs predicted 17% of the variance in soda choice (7% beyond the variance explained by explicit measures). A soda IAT predicted 12% of the variance in soda choice (4.9% beyond the variance explained by explicit measures). Taken together, these results provide strong evidence for the utility of the SC-IAT above and beyond the IAT and explicit attitude measures.

Information Obtained From the SC-IAT and IAT One advantage to using two single attitude measures as opposed to one comparative attitude is that a greater amount of information is obtained from the two single attitude measures (see Figure 1). On the top panel of Figure 1, Coke SC-IAT scores are plotted against Pepsi SC-IAT scores. From this panel, information about Coke, Pepsi, and soda (Coke – Pepsi) associations may be obtained. Some participants had favorable associations with both Coke and Pepsi (the upper right quadrant), some had unfavorable associations with both Coke and Pepsi (the lower left quadrant), and some had favorable associations toward one soda and unfa1 If the 6 individuals who did not indicate a preference for a Coke or Pepsi product are eliminated from the analysis, then these relationships can also be examined by using a logistic regression predicting soda choice. For a logistic regression with implicit measures only, the results closely parallel the standard regression analysis. IAT scores ( p ⫽ .05) and Pepsi SC-IAT scores ( p ⫽ .06) predicted the soda choice, whereas Coke SC-IAT scores did not ( p ⫽ .63). However, when explicit soda attitudes were added to the equation, none of the implicit measures or the explicit measure uniquely predicted soda choice (all ps ⬎ .28).

SINGLE CATEGORY IMPLICIT ASSOCIATION TEST

Table 3 Study 1: Predicting Soda Choice Prediction of soda choice

␤

Predictor

t

p

0.47 2.74 2.10

.64 ⬍.01 .04

Implicit measures only Coke SC-IAT Pepsi SC-IAT IAT

.06 ⫺.35 .28

Implicit and explicit measures Coke SC-IAT Pepsi SC-IAT IAT Explicit soda attitudes

.16 ⫺.22 .23 .67

1.76 2.33 2.43 7.37

.09 .02 .02 ⬍.01

Note. N ⫽ 52. All variables were entered simultaneously. SC-IAT ⫽ Single Category Implicit Association Test; IAT ⫽ Implicit Association Test.

vorable associations toward the other soda (the remaining two quadrants). Information about participants’ relative soda associations can be obtained by observing where the participant falls in reference to the dashed diagonal line. Participants with SC-IAT scores falling above the line had more positive and/or less negative associations with Coke than Pepsi; participants with SC-IAT scores falling below the line had more positive and/or less negative associations with Pepsi than Coke. In the bottom panel of Figure 1, soda IAT scores are plotted. Only information about participants’ relative soda associations is available from this panel. Participants with scores on the left side of the graph (scores less than zero) had more positive and/or less negative associations with Pepsi than Coke, whereas participants with scores on the right side of the graph (scores greater than zero) had more positive and/or less negative associations with Coke than Pepsi.

21

it is somewhat surprising that the soda SC-IAT measure correlated significantly with explicit soda preferences, whereas the IAT failed to correlate with explicit measures of soda attitudes. Overall, these results suggest that the SC-IAT may have some utility above and beyond the IAT even in situations where the outcome variable is comparative. Although the Coke and Pepsi SC-IATs performed well on a number of aspects, there were a couple of curious aspects of these results. First, the reliability of the SC-IAT measures was lower than the reliability observed for the IAT. One possibility is that not having a comparative category for the attitude object may result in extra error variance in responses. If this were true, then SC-IAT measures would inherently have lower reliabilities than IAT measures. A second possible explanation for the low reliability of the SC-IATs is that the 1,500-ms response window not only resulted in a higher error rate, but also increased unreliability for the SC-IAT. A final possibility is that the order of the tasks adversely affected the reliability of the SC-IAT. For the IAT, all responses to the attitude objects of interest are comparative. After completing an IAT, participants may continue to think about the attitude objects in a comparative manner. If a SC-IAT were to follow an IAT, participants’ responses may be influenced by the comparative mind-set induced by the IAT, perhaps resulting in increased error

Discussion These results provide initial support for the SC-IAT as a valid measure of evaluative associations with a single attitude object. The validity of the SC-IAT was established on multiple levels: known groups validity, convergent validity, and predictive validity. First, along with the IAT and explicit soda preferences, the soda SC-IAT discriminated between Coke and Pepsi drinkers. Second, the SC-IAT measure of soda preferences correlated significantly with explicit soda preferences. The Coke and Pepsi SC-IATs were specific in their measurement of associations. The Pepsi SC-IAT revealed that Pepsi drinkers had more positive than negative associations with Pepsi, compared with Coke drinkers. The Pepsi SC-IAT also correlated with explicit Pepsi attitudes, but not with explicit Coke attitudes. Likewise, the Coke SC-IAT correlated with explicit Coke attitudes, but not with explicit Pepsi attitudes. Third, the Pepsi SC-IAT, and to a lesser extent the Coke SC-IAT, reliably predicted intended soda choice, even when controlling for IAT scores and explicit attitude ratings. This study was ideally suited for a comparative IAT style of measurement, and as expected, the IAT discriminated between Coke and Pepsi drinkers and predicted intended soda choice. Thus,

Figure 1. A comparison of information available from independent Coke and Pepsi Single Category Implicit Association Tests (SC-IATs) and a Coke–Pepsi Implicit Association Test (IAT). IAT and SC-IAT scores were calculated by using the D-score algorithm (N ⫽ 53).

KARPINSKI AND STEINMAN

22

variance in the SC-IAT. The issue of the reliability of the SC-IAT is one that will be revisited in subsequent studies. A second curious finding in Study 1 is that all Pepsi measures (SC-IAT and explicit measures) discriminated between Coke and Pepsi drinkers better and predicted soda choice better than did Coke measures. We are puzzled by this finding, but it may be the result of the Temple University environment. Temple University has an exclusive agreement with Pepsi to provide soft drinks on campus. All vending machines and restaurants on campus serve only Pepsi products. Thus, Temple University students may receive more information regarding Pepsi products than Coke products, and this environmental bias may lead to greater predictive validity and known-groups validity for the Pepsi SC-IAT than for the Coke SC-IAT. Regardless, evidence for the validity of the Coke SC-IAT was obtained; the Coke SC-IAT correlated with explicit Coke attitudes and not explicit Pepsi attitudes.

Study 2 A comparative attitude or associative measure is not ideally suited for all contexts. For example, when investigators wish to predict smoking behavior, there is no clear complementary behavior to smoking; it depends on the nature of the research question. Similarly, self-esteem researchers may be interested in assessing evaluative associations with the self and not comparative self– other associations (see Karpinski, 2004). In these situations, a SC-IAT measure of associations may provide a more specific measure of the evaluative associations in question than an IAT. Because self-report measures of self-esteem are not explicitly comparative, we expected to find that a self-SC-IAT would correlate more strongly with explicit measures of self-esteem than would a self– other IAT. Study 2 was designed to investigate the use of a SC-IAT as a measure of self-associations. The procedure of Study 2 closely followed the procedure of Study 1, with one key exception. To examine the possibility that taking a (comparative) IAT prior to a SC-IAT adversely affected the reliability of the SC-IAT, we switched the order of the SC-IAT and the IAT. Thus, participants completed a self-SC-IAT, a self– other IAT, and then explicit measures of self-esteem.

Method Participants Sixty-six students (16 men, 41 women, 9 unknown) enrolled in an introductory psychology course at Temple University participated in this experiment. All participants received course credit for their participation.

Procedure All tasks were presented on the computer, and all participants completed the tasks in the same order: a SC-IAT measure of self-associations, an IAT measure of self– other associations, and explicit measures of self-esteem. At the conclusion of the session, the participants were thanked and completely debriefed. SC-IAT measure of self-associations. The self-SC-IAT was identical to the Coke and Pepsi SC-IATs used in Study 1, with the exception of the target words and category labels. The evaluative dimension remained labeled good and bad, and the object dimension was labeled self. Five target words were selected to be associated with the category self ( partic-

ipant’s first name, participant’s last name, me, I, and myself). The good and bad target words were identical to the words used in Study 1. All target and category words were presented in lowercase letters. Participants first completed the self ⫹ positive blocks, followed by the self ⫹ negative blocks. IAT measure of self– other associations. The IAT procedure was identical to the procedure used in Study 1, with the exception of the target words and category labels. The category labels self and other replaced the labels of Coke and Pepsi, respectively. Five target words were also selected to be associated with each of the attitude objects (self: participant’s first name, participant’s last name, me, I, and myself; other: he, her, his, hers, and person). Explicit measures of self-esteem. Participants next completed three explicit measures of self-esteem: a self semantic differential, a self feeling thermometer, and the Rosenberg Self-Esteem Scale (Rosenberg, 1965). For the semantic differential, participants rated self on five bipolar dimensions: ugly– beautiful, bad– good, unpleasant–pleasant, foolish–wise, and awful– nice. Each dimension was rated on a 7-point scale ranging from ⫺3 (negative pole) to 3 (positive pole). The five items were summed to form a semantic differential measure of self-esteem (␣ ⫽ .55). The self feeling thermometer consisted of a single item, with participants rating themselves on a thermometer ranging from 0 (cold or unfavorable) to 100 (warm or favorable). For presentation purposes, feeling thermometer scores have been rescaled to range from ⫺50 to 50 so that zero indicates neutral self-attitudes. For the Rosenberg scale, participants responded to each item on a 7-point scale ranging from 1 (disagree strongly) to 7 (agree strongly). The 10 items were averaged to compute a measure of self-esteem (␣ ⫽ .86). These three explicit measures of self-esteem were strongly interrelated (␣ ⫽ .80), and thus, the three measures were standardized and averaged to compute a single explicit measure of self-esteem.

Results IAT and SC-IAT Data Reduction and Reliability Analysis Because of a computer error, the explicit attitude data were lost for 9 participants. Replicating the results of Study 1, error rates were significantly higher for the self-SC-IAT than for the self– other IAT, t(65) ⫽ 3.15, p ⬍ .01. Participants with an IAT error rate greater than 20% were excluded from analysis, resulting in the elimination of 5 participants. Participants with a SC-IAT error rate greater than 20% were also excluded, resulting in elimination of 9 additional participants. The resulting error rates were consistent with those observed in Study 1 (self-SC-IAT ⫽ 6.60%, self– other IAT ⫽ 4.53%). IAT and SC-IAT scores were computed by using the scoring algorithms described for Study 1. Self– other IAT scores were computed so that higher numbers indicate more positive associations with the self (and/or negative associations with an other) than negative associations with the self (and/or positive associations with an other), and the self-SC-IAT scores were such that higher scores indicate greater positive than negative associations with the self. To determine the reliability of the SC-IAT, we divided each SC-IAT into thirds and calculated a SC-IAT score separately for each block of 24 test trials. In this study, the SC-IAT displayed a level of reliability that is similar to the level of reliability typically found for the IAT (adjusted r ⫽ .73). For the IAT, a reliability correlation was computed by correlating the IAT score computed from the practice trials with the IAT score computed from the test trials. The observed reliability coefficient was on the lower end of what is typically observed in IAT data (adjusted r ⫽ .58).

SINGLE CATEGORY IMPLICIT ASSOCIATION TEST

Average Levels of Self-Esteem Overall, all measures showed a pattern consistent with an interpretation of high self-esteem (see Table 4). All three explicit measures of self-esteem revealed the presence of positive selfevaluations (all ds ⱖ 1.35). Implicit measures revealed a similar pattern. The IAT revealed that participants had more positive self-associations and/or negative other associations than negative self-associations and/or positive other associations (d ⫽ 1.84). The SC-IAT revealed that participants had more positive selfassociations than negative self-associations (d ⫽ 1.13). No significant gender differences were observed on any of the measures of self-esteem (all ps ⱖ .13).

Relationship Between Implicit and Explicit Measures of Self-Esteem It is interesting to note that the self-SC-IAT and self– other IAT scores were only marginally correlated, r(50) ⫽ .25, p ⫽ .07. In other words, the self-associations measured by the SC-IAT were only weakly related to the self– other associations measured by the self– other IAT. The self– other IAT failed to correlate with the standardized explicit measure of self-esteem, r(42) ⫽ .01, p ⫽ .93. Conversely, a significant positive correlation was observed between the self-SC-IAT and the explicit measure of self-esteem, r(42) ⫽ .38, p ⫽ .01. A regression analysis was conducted to investigate whether the self-SC-IAT predicted unique variance in explicit reports of selfesteem. Consistent with the correlation findings, the regression analysis revealed that self-SC-IAT scores were uniquely predictive of explicit self-esteem (␤ ⫽ .40, p ⫽ .01), whereas IAT scores were not uniquely predictive of explicit self-esteem (␤ ⫽ ⫺.09, p ⫽ .54).

Discussion Study 2 provided additional evidence for the reliability and validity of the SC-IAT as a measure of implicit social cognition. Like the self– other IAT and explicit measures of self-esteem, the

Table 4 Study 2: Descriptive Statistics Difference from midpoint Measure

M

SD

t

p

d

t(51) ⫽ 13.24 t(51) ⫽ 8.13

⬍.01 ⬍.01

1.84 1.13

⬍.01 ⬍.01 ⬍.01

1.35 2.98 1.49

Implicit measures Self–other IAT Self-SC-IAT

0.58 0.45

0.32 0.40

Explicit measures Rosenberg SE Scale Self semantic differential Self feeling thermometer

5.39 8.74 27.70

1.03 2.93 18.54

t(43) ⫽ 8.94 t(43) ⫽ 19.75 t(43) ⫽ 9.91

Note. Midpoint is the middle point of the scale or the point of the scale at which a person has neutral self-associations. IAT ⫽ Implicit Association Test; SC-IAT ⫽ Single Category Implicit Association Test; SE ⫽ SelfEsteem.

23

self-SC-IAT revealed more positive than negative selfassociations, suggesting positive self-esteem in the sample. Unlike the self– other IAT, the self-SC-IAT correlated significantly with explicit measures of self-esteem. These correlations are larger than correlations typically observed between implicit and explicit measures of self-esteem (see Bosson et al., 2000; Greenwald & Farnham, 2000) but are similar in size to correlations found between an affective priming measure of self-esteem and explicit measures of self-esteem (Wentura, Kulfanek, & Greve, 2005). In addition, a small and nonsignificant relationship was observed between self-SC-IAT scores and self– other IAT scores. Karpinski (2004) hypothesized that the evaluative self– other associations measured by the IAT may be qualitatively different from evaluative self-associations, and these results provide support for this claim. The self– other IAT has proven to be a useful measure of self-esteem in many contexts, yet the current findings suggest that a self-SC-IAT may provide additional information about implicit self-esteem that is not captured by the self– other IAT. However, the correlation we observed between explicit measures of self-esteem and the self– other IAT is lower than what is typically reported (e.g., see Bosson et al., 2000). The reasons for this discrepancy are unclear, but the end result is that this study may underestimate the IAT– explicit self-esteem relationship and the SC-IAT/IAT relationship. The results of Study 2 provided stronger evidence for the reliability of the SC-IAT than did Study 1. In this study, the SC-IAT displayed a level of internal consistency similar to the reliabilities that are typically found by using the IAT. One difference between this study and Study 1 is that in the current study, the SC-IAT measure was obtained prior to the IAT measure. This result provides indirect support to the hypothesis that the completion of a (comparative) IAT may interfere with ensuing SC-IAT measures. Thus, for subsequent studies, we will present the SCIAT measures prior to the IAT measures.

Study 3 One of the more interesting applications of the IAT is its use as a measure of implicit racial associations. Studies using a Black– White IAT have typically found that White participants display a large racial bias in favor of Whites and/or against Blacks (Dasgupta, McGhee, Greenwald, & Banaji, 2000; Greenwald et al., 1998; Monteith, Voils, & Ashburn-Nardo, 2001; Nosek, Banaji, & Greenwald, 2002). One possible interpretation of this effect is that “virtually all White Americans may have automatic negative associations to African American names” (Greenwald et al., 1998, p. 1475). Yet, because of the comparative nature of the IAT, there are multiple interpretations of this IAT race bias. For example, a person who has no evaluative associations with Blacks, and mostly positive associations with Whites, would display the typical IAT race bias. Likewise, an IAT racial bias may emerge for participants who have mostly positive associations with Blacks but also have more positive associations with Whites (Blanton & Jaccard, 2006; Blanton et al., 2006; Gehring, Karpinski, & Hilton, 2003). In these alternative interpretations of the IAT race bias, participants would not have automatic negative associations with Blacks. We hypothesized that the SC-IAT could be used to help interpret meaning of the IAT race bias. For Study 3, separate Black and White SC-IATs were used to assess the strength of evaluative

KARPINSKI AND STEINMAN

24

associations with Blacks and Whites independently. We expected a race bias SC-IAT (White SC-IAT–Black SC-IAT) to reveal a similar race bias as the IAT; White SC-IAT scores were hypothesized to be higher than Black SC-IAT scores. If this effect is due to out-group prejudice, then participants would also have negative Black SC-IAT scores. However, if the SC-IAT and IAT race biases are a result of in-group favoritism, then we would expect to observe positive evaluative associations on a White SC-IAT and less positive (but not negative) evaluative associations on a Black SC-IAT. Furthermore, we explored differences in automatic race biases between White and Black participants. Previous studies have found that Black participants tend to show either no IAT race bias or a small IAT race bias in favor of Whites and/or against Blacks (Nosek et al., 2002). In Study 3, participants completed a White SC-IAT, a Black SC-IAT, a Black–White IAT, and explicit measures of racial attitudes. Several modifications were made to both the procedures of the IAT and SC-IAT. First, participants taking the IAT received response feedback, seeing a green O following a correct response and a red X following an incorrect response. This modification is in line with standard versions of the IAT that typically include response feedback. Second, we reduced the number of critical trials in each SC-IAT from 72 to 48 in order to reduce the time required for each SC-IAT. Third, we increased the response window for the SC-IAT from 1,500 ms to 2,000 ms. If the 1,500-ms response window was contributing to high error rates on the SC-IAT, then increasing the response window may alleviate this problem.

Method Participants Eighty-one White students (24 men, 57 women) and 37 Black students (9 men, 28 women) enrolled in an introductory psychology course at Temple University participated in this experiment. All participants received course credit for their participation.

Procedure The procedure mirrored the procedure of Study 1, with Black American and White American as the categories of interest. All tasks were presented on the computer, and all participants completed the tasks in the same order: a White-SC-IAT, a Black-SC-IAT, a Black–White IAT, and explicit measures of racial attitudes. At the conclusion of the session, the participants were thanked and completely debriefed. SC-IAT measure of Black and White associations. The Black SC-IAT and White SC-IAT were similar to the soda SC-IATs used in Study 1, with three exceptions. First, the target words and category labels were changed to be relevant for the categories Black and White. The evaluative dimension remained labeled good and bad; the category dimension was labeled White American for the White SC-IAT and Black American for the Black SC-IAT. Six Black faces and six White faces were selected to be associated with the categories Black and White, respectively. These face stimuli were identical to those used by Nosek et al. (2002). Second, the response window was lengthened from 1,500 ms to 2,000 ms. If participants did not respond within 2,000 ms of the presentation of the target word or face, they were reminded to respond more quickly. Third, the number of critical trials was reduced from 72 trials to 48 trials. IAT measure of Black–White associations. The IAT procedure was identical to the procedure used in Study 1, with two exceptions. First, the target words and pictures and category labels were modified to be appro-

priate for a Black–White IAT. The category labels White American and Black American replaced the labels of Coke and Pepsi, respectively. Six target faces were also selected to be associated with each of the attitude objects (three male and three female for each race, identical to the stimuli used for the SC-IATs). In addition, several of the unpleasant target words were changed to avoid overlap with the Black stereotype (see Appendix). Finally, unlike Studies 1 and 2, participants received response feedback for correct and incorrect responses. Participants viewed a green O for 150 ms following a correct response and a red X for 150 ms following an incorrect response. Explicit measures of race attitudes. Participants next completed two explicit measures of White and Black attitudes: a semantic differential (White, ␣ ⫽ .84; Black, ␣ ⫽ .89) and a feeling thermometer. These measures were identical to those used in the previous studies, but the target category labels White American and Black American were used to assess attitudes toward Whites and Blacks, respectively. These two explicit attitudes measures were strongly related (White, ␣ ⫽ .64; Black, ␣ ⫽ .81), and thus they were standardized and averaged to form composite explicit attitude measures of Whites and Blacks. Explicit measures of comparative racial attitudes were obtained by subtracting the explicit measures of Blacks from the explicit measures of Whites, such that higher numbers indicated more favorable attitudes of Whites compared with Blacks.

Results IAT and SC-IAT Data Reduction and Reliability Analysis Compared with the IAT, error rates were significantly higher on the White SC-IAT, t(117) ⫽ 7.93, p ⬍ .01, and the Black SC-IAT, t(117) ⫽ 7.40, p ⬍ .01. Participants with an error rate greater than 20% on the Black SC-IAT, the White SC-IAT, or race IAT were excluded from analysis, resulting in the elimination of 5 participants. The resulting average error rates were consistent with error rates observed in previous studies (White SC-IAT ⫽ 6.15%; Black SC-IAT ⫽ 5.97%; race IAT ⫽ 3.08%). These findings suggest that increasing the response window did not result in a reduction of the SC-IAT error rate. As in the previous studies, IAT and SC-IAT scores were computed by using the D algorithm. A comparative race SC-IAT score was obtained by subtracting Black SC-IAT scores from White SC-IAT scores. For both the race SC-IAT and IAT, positive scores indicate more positive associations with Whites (and/or negative associations with Blacks) than negative associations with Whites (and/or positive associations with Blacks). To determine the reliability of the SC-IAT, we calculated a SC-IAT score for each of the two test blocks of 24 test trials. A reliability analysis on these scores revealed greater internal consistency for the White SC-IAT (adjusted r ⫽ .70) than for the Black SC-IAT (adjusted r ⫽ .55). For the IAT, a reliability correlation was computed by correlating IAT scores computed from the practice trials with IAT scores computed from the test trials (adjusted r ⫽ .75). Similar to Study 1, the SC-IAT displayed lower levels of internal consistency than did the IAT.

Implicit and Explicit Measures of Racial Attitudes Explicit measures. For the comparative explicit race bias measures, White and Black participants reported an in-group bias (see top of Table 5). On both the feeling thermometer and semantic differential, White participants rated Whites more positively than Blacks ( ps ⱕ .01, ds ⱖ 1.39), and Black participants rated Blacks

SINGLE CATEGORY IMPLICIT ASSOCIATION TEST

25

Table 5 Study 3: Descriptive Statistics

Whites (n ⫽ 79) Attitude measure

M

SD

d

Difference between White and Black participants

Blacks (n ⫽ 34) M

SD

d

t(111)

p

d

16.26 17.36 15.82 0.75 0.82 0.98

⫺0.94 1.09 2.15 ⫺0.62 0.75 1.10

7.33 1.75 5.63 4.45 1.38 3.60

⬍.01 .08 ⬍.01 ⬍.01 .17 ⬍.01

1.39 0.33 1.07 0.85 0.26 0.68

0.42 0.63 0.44 0.43

0.22 ⫺0.10 0.36 0.52

7.68 2.03 0.18 2.80

⬍.01 .04 .86 ⬍.01

1.46 0.39 0.03 0.49

Explicit measures Race feeling thermometer White feeling thermometer Black feeling thermometer Race semantic differential White semantic differential Black semantic differential

9.19 24.35 15.16 0.40 0.85 0.45

16.22 14.19 16.65 1.02 0.84 0.80

0.57 1.72 0.91 0.39 1.01 0.56

⫺15.21 18.91 24.12 ⫺0.47 0.61 1.08

Implicit measures Race IAT Race SC-IAT White SC-IAT Black SC-IAT

0.63 0.16 0.17 0.01

0.30 0.52 0.37 0.33

2.07 0.30 0.46 0.04

0.09 ⫺0.07 0.16 0.22

Note. For comparative race measures, higher numbers indicate more positive (or fewer negative) attitudes or associations with Whites than with Blacks. For single category measures, higher numbers indicate more positive attitudes. IAT ⫽ Implicit Association Test; SC-IAT ⫽ Single Category Implicit Association Test.

more positively than Whites ( ps ⱕ .01, ds ⱖ .62). Examination of the separate ratings of Whites and Blacks revealed that these in-group biases were due to in-group favoritism and not out-group prejudice. On both the semantic differential and feeling thermometer measures, White and Black participants reported favorable attitudes toward both Whites and Blacks (all ps ⱕ .01, ds ⱖ 0.56). Implicit measures. As expected, for White participants, the implicit comparative measures of racial attitudes revealed significant race biases (see bottom of Table 5). The IAT revealed a large race bias (d ⫽ 2.07, p ⬍ .01), whereas the SC-IAT revealed a small- to medium-sized race bias (d ⫽ 0.30, p ⬍ .01). Echoing the findings on the explicit measures, these findings indicate that White participants showed an in-group bias at the implicit level, favoring Whites over Blacks. For Black participants, both implicit measures of race bias revealed no significant racial biases (IAT, d ⫽ 0.22, p ⫽ .22; SC-IAT, d ⫽ ⫺0.10, p ⫽ .57). A test for the difference between these effects revealed that White participants displayed a significantly stronger in-group bias than did Black participants on both the IAT (d ⫽ 1.46, p ⬍ .01) and the SC-IAT (d ⫽ 0.39, p ⫽ .04). Because the IAT cannot be decomposed into component attitudes of Whites and attitudes of Blacks, no further conclusions can be made about the nature of the race bias observed with the IAT. However, the race SC-IAT is composed of a difference between a White SC-IAT and a Black SC-IAT, and each of these components can be analyzed separately. An analysis of the White SC-IAT component revealed that both White and Black participants had more positive than negative associations with Whites (ds ⱖ 0.34, ps ⱕ .04), and there was no significant difference in the valence of White associations held by Whites and Blacks, t(111) ⫽ 0.18, p ⫽ .86, d ⫽ 0.03. But for the Black SC-IAT component, Black participants had significantly more positive associations with Blacks than did White participants, t(111) ⫽ 2.80, p ⬍ .01, d ⫽

0.49. Follow-up analyses revealed that Black participants had positive associations with Blacks (d ⫽ 0.52, p ⬍ .01), and White participants had equally positive and negative associations with Blacks (d ⫽ 0.04, p ⫽ .72). These results suggest that, on average, the SC-IAT racial bias emerged for Whites because of implicit in-group favoritism (favorable associations with Whites) and not because of implicit out-group prejudice (negative associations with Blacks). SC-IAT scores can also be used to help interpret the contribution of evaluative White and Black associations to IAT scores. To investigate whether IAT scores reflect meaningful variations in both evaluative White and evaluative Black associations, a regression analysis was conducted to predict IAT scores from Black SC-IAT and White SC-IAT scores. This analysis revealed that race IAT scores were positively associated with White SC-IAT scores (␤ ⫽ .24, p ⬍ .01) and negatively associated with Black SC-IAT scores (␤ ⫽ ⫺.30, p ⬍ .01). Thus, as expected, a Black–White IAT measured evaluative associations with both Blacks and Whites. Correlations between measures. As expected, race SC-IAT scores were significantly correlated with IAT scores, r(111) ⫽ .40, p ⬍ .01. This relationship remained significant when examining only White participants, r(77) ⫽ .36, p ⬍ .01, and also when examining only Black participants, r(32) ⫽ .40, p ⫽ .02. Furthermore, White SC-IAT scores were uncorrelated with Black SC-IAT scores, r(111) ⫽ ⫺.11, p ⫽ .23, suggesting that evaluative associations with Whites were independent from evaluative associations with Blacks. All of the implicit measures tended to not be correlated with explicit measures of racial bias (see Table 6). For White participants in this sample, both the IAT and a race SC-IAT failed to correlate with explicit measures of racial attitudes (all ps ⬎ .22), whereas for Black participants, both the IAT and a race SC-IAT

KARPINSKI AND STEINMAN

26

Table 6 Study 3: Correlations Between Implicit and Explicit Measures SC-IAT measures Race IAT

Explicit attitude measure

Race SC-IAT

White SC-IAT

Black SC-IAT

.17 .16 .17

⫺.08 ⫺.08 ⫺.02

White participants (n ⫽ 79) Explicit race attitudes Explicit attitudes toward Whites Explicit attitudes toward Blacks

⫺.05 .08 .14

.14 .17 .01

Black participants (n ⫽ 34) Explicit race attitudes Explicit attitudes toward Whites Explicit attitudes toward Blacks

.29† .07 ⫺.23

.29† .01 ⫺.23

.48* .25 ⫺.15

.10 .23 .16

Note. SC-IAT ⫽ Single Category Implicit Association Test; IAT ⫽ Implicit Association Test. † p ⬍ .10. * p ⬍ .01.

correlated marginally with explicit measures of racial attitudes (all ps ⱕ .10). Additionally, little evidence was found for correlations between the White and Black SC-IATs and their respective explicit attitude measures, either for White or Black participants (all ps ⬎ .15).

Discussion Study 3 provided evidence for the validity of the SC-IAT on multiple levels. First, similar mean-level effects were observed on the IAT and the race SC-IAT. On both these measures, White participants displayed a pro-White bias, and Black participants displayed no significant evaluative race bias. For White participants, the race bias observed by using the IAT was substantially larger than the race bias observed by using the SC-IAT. However, because no objective measure of implicit race bias exists, it is unclear whether this difference in effect sizes between the measures is meaningful. Second, the White SC-IAT and Black SC-IAT facilitated the interpretation of IAT and race SC-IAT scores. For White participants, the SC-IAT race bias emerged because of positive evaluative White associations and neutral evaluative Black associations; for Black participants, the lack of a SC-IAT race bias emerged because of positive evaluative Black associations and equally positive evaluative White associations. This pattern of results suggests that the SC-IAT and IAT race biases observed in White participants were, on average, a result of in-group favoritism and not out-group prejudice. At the very least, the current results show that IAT and SC-IAT race biases can be due to in-group favoritism and may occur in the absence of out-group negativity. Third, the race SC-IAT correlated significantly with the race IAT. Thus, there is some evidence that the race SC-IAT and the race IAT are measuring similar aspects of Black–White evaluative associations. These results suggest that a Black–White IAT, a White SC-IAT, and a Black SC-IAT may all provide useful information regarding evaluative associations regarding Blacks and

Whites. However, neither the IAT nor the SC-IATs consistently correlated with the explicit attitudes measures. One area of concern regarding the SC-IAT is the low reliability coefficients of the SC-IAT measures, particularly the Black SCIAT. Consistent with the findings of Study 1, the SC-IATs displayed lower levels of internal consistency than did the IAT. Although the White SC-IAT displayed adequate levels of reliability, the reliability of the Black SC-IAT was poor. In this study, the number of critical trials in the SC-IAT procedure was reduced from 72 to 48. It is possible, and indeed likely, that this modification adversely affected the reliability of the SC-IAT measure. A second modification to the SC-IAT procedure was an increase in the length of the response window from 1,500 ms to 2,000 ms. This change was intended to reduce the error rate, but this change was ineffective. The reliability and validity of the SC-IAT appear to be greater when the SC-IAT procedure includes at least 72 critical trials and a 1,500-ms response window, as was used in Study 2. A more troubling possibility is that the low reliability of the SC-IAT was due to participants attempting to respond to the task in a socially desirable manner. This concern arises from the finding that the Black SC-IAT had poor reliability and from the reduced effect size on the race SC-IAT in White participants, compared with the IAT. The SC-IAT was designed to be relatively impervious to controlled processing in general and especially to social desirability concerns. The moderate-sized correlation between the IAT and a race SC-IAT suggests that social desirability pressures might not have affected SC-IAT responses in a sizable way. Nevertheless, the possibility that the SC-IAT may not be as implicit as previously thought is a disturbing finding, and Study 4 was designed to investigate this issue more thoroughly.

Study 4 One of the advantages of implicit measures of social cognition is that they are impervious to self-presentational motivations and social desirability concerns. Thus, for the SC-IAT to be useful as an implicit measure of social cognition, it must be shown that participants cannot control or fake their responses on the task. There are reasons to suspect that it may be easier for participants to control or fake their responses on the SC-IAT than on the IAT. Undoubtedly, it is easier for participants to reconceptualize the instructions in order to disrupt measurement of the associations of interest on the SC-IAT than on the IAT. For example, participants may recode the SC-IAT instructions as follows: “If it is a ‘bad’ target word, press the right key; for all other target words, press the left key.” Similarly, if the target category is represented by pictures and the evaluative dimension is represented by words (as in Studies 1 and 3), participants may use an “all pictures to the left” rule and avoid encoding the content of the target pictures. Although participants could use these strategies to control or fake their responses on the SC-IAT, it is not clear that they do use these strategies spontaneously. Study 4 was conducted to investigate participants’ ability to control or fake their responses on a SC-IAT measuring evaluative associations with women. Participants were randomly assigned to display positive or negative attitudes toward women and then completed a SC-IAT measuring evaluative associations with

SINGLE CATEGORY IMPLICIT ASSOCIATION TEST

women. Because the modifications of the SC-IAT procedure used in Study 3 (reducing the number of critical trials in each stage to 48 and increasing the response window to 2,000 ms) were ineffective, we used a SC-IAT with 72 critical trials in each stage and with a 1,500-ms response window (as was used in Study 2). For comparison purposes, participants also completed a male–female IAT and explicit attitude measures toward women. We expected that participants would easily be able to present the desired response on the explicit attitude measures. A number of researchers have investigated the susceptibility of the IAT to faking. Steffens (2004) reviewed the literature and concluded that small, nonsignificant effects of faking on the IAT were typically observed. Thus, we anticipated a small effect of instruction on the male–female IAT scores. The primary question of interest was the effect of instruction on the female SC-IAT scores and how this effect compared with the effect observed for the IAT and explicit attitude measures.

27 Results

IAT and SC-IAT Data Reduction and Reliability Analysis As in the previous studies, IAT scores were computed by using the D algorithm, such that higher numbers indicate a profemale (and/or antimale) bias. For the IAT, a reliability coefficient was computed by correlating an IAT score computed from the practice trials with an IAT score computed from the test trials (adjusted r ⫽ .78). SC-IAT scores were computed by using the modified D algorithm, such that higher numbers indicate more positive than negative associations with women. To determine the reliability of the SC-IAT, a SC-IAT score was calculated for each of the three test blocks of 24 test trials (adjusted r ⫽ .85). Similar to Study 2, the SC-IAT displayed slightly higher levels of internal consistency than the IAT.

Effect of the Self-Presentation Instructions Method Participants Eighty-four students (20 men, 64 women) enrolled in an introductory psychology course at Temple University participated in this study. All participants received course credit for their participation.

Procedure Participants were randomly assigned to respond to all questions and tasks as if they had very positive attitudes or very negative attitudes toward women. Participants then completed a SC-IAT measure of evaluative associations with women, a male–female IAT, and self-report measures of attitudes toward women. SC-IAT measure of female associations. The female SC-IAT was identical to the self-SC-IAT procedure used in Study 2, with appropriate modifications to measure associations with women. The evaluative dimension was labeled good and bad, and the concept dimension was labeled female. Five target words were used for the category female (her, woman, girl, she, and lady). IAT measure of male–female associations. The IAT procedure was identical to the procedure used in Study 3 (i.e., unlike Studies 1 and 2, the procedure included response feedback), with appropriate modifications to measure associations with men and women. The evaluative dimension was labeled pleasant and unpleasant, and the attitude object dimension was labeled female and male. Five target words were used for each of the attitude object dimensions (male: him, man, boy, he, and guy; female: her, woman, girl, she, and lady). In Blocks 3 and 4, participants paired female ⫹ pleasant and male ⫹ unpleasant. This pairing was reversed in Blocks 6 and 7: male ⫹ pleasant and female ⫹ unpleasant. Explicit measures of attitudes toward females. Participants next completed three explicit measures of attitudes toward women: a female semantic differential (␣ ⫽ .98), a female feeling thermometer, and the Modern Sexism Scale (Swim, Aikin, Hall, & Hunter, 1995). The semantic differential and feeling thermometer measures were identical to those used in the previous studies, but the target category was changed to be females. Participants responded to the items of the Modern Sexism Scale on a scale from 1 (strongly disagree) to 5 (strongly agree; ␣ ⫽ .88). Modern sexism scores were recoded so that higher numbers indicated less sexism. The three attitude measures were highly interrelated (␣ ⫽ .92) and thus were averaged to compute a composite measure of self-reported attitudes toward women.

As expected, participants instructed to present positive attitudes toward women displayed significantly more positive explicit attitudes toward women than did participants instructed to present negative attitudes toward women (present positive, M ⫽ 0.71; present negative, M ⫽ ⫺0.78), t(82) ⫽ 11.84, p ⬍ .01, d ⫽ 2.61. For the IAT, we first examined whether the error rate differed across conditions. A t test of the IAT error rate revealed no significant difference in the IAT error rate as a function of presentation instructions, t(82) ⫽ 0.40, p ⫽ .69, d ⫽ 0.09. Following the procedure from the previous studies, 1 participant with an error rate larger than 20% was excluded from the remaining IAT analyses, and the resulting error rate was 5.30%. Somewhat surprisingly, a medium instruction effect was observed for the IAT. Participants instructed to present positive attitudes toward women displayed significantly higher IAT scores than did participants instructed to present negative attitudes toward women (present positive, M ⫽ 0.60; present negative, M ⫽ 0.37), t(81) ⫽ 2.03, p ⫽ .05, d ⫽ 0.45.2 For the SC-IAT, we also examined whether the error rate differed across conditions. A t test of the SC-IAT error rate revealed a significant medium-sized difference in the SC-IAT error rate as a function of presentation instructions, t(82) ⫽ 2.15, p ⫽ .04, d ⫽ 0.48. Participants who attempted to present negative attitudes toward women had a significantly higher error rate on the task (M ⫽ 18.04%) than did participants who attempted to present positive attitudes toward women (M ⫽ 12.94%). An analysis of the SC-IAT scores of all participants revealed a large instruction effect (present positive, M ⫽ 0.68; present negative, M ⫽ 0.28), t(82) ⫽ 3.31, p ⬍ .01, d ⫽ 0.73. However, the standard practice with IAT and SC-IAT data is to exclude participants with high error rates from the analyses. Once the 25 participants with a SC-IAT error rate larger than 20% were removed from the data (10 from the positive condition and 15 from the negative condition), the overall SC-IAT error rate no longer differed by condition (M ⫽ 9.42%), 2

This presentation effect on IAT scores was observed only when the D-score algorithm was used to compute IAT scores; when the original log-based IAT scoring algorithm was used, no presentation effect emerged, t(81) ⫽ 0.10, p ⫽ .92, d ⫽ 0.02.

28

KARPINSKI AND STEINMAN

t(57) ⫽ 0.47, p ⫽ .64, d ⫽ 0.12. Furthermore, a t test revealed that SC-IAT scores did not differ significantly as a function of condition, although a small presentation effect in the anticipated direction was observed (present positive, M ⫽ 0.60; present negative, M ⫽ 0.50), t(57) ⫽ 0.92, p ⫽ .36, d ⫽ 0.24. These analyses suggest that participants can fake a SC-IAT score, but to do so they also increase their error rate. An alternative interpretation of these results is that the fakers (those who had large SC-IAT error rates) are different from nonfakers, confounding the interpretation of the results. Although we cannot entirely rule out this possibility, we think it is unlikely. First, the fakers were not significantly different from the nonfakers on any demographic variable we had collected (gender, ethnicity, and native language; all ps ⬎ .25). Second, it is possible that the nonfakers were insensitive to the presentation manipulation and/or that the fakers tried harder than the nonfakers. However, the nonfakers did show the predicted self-presentation effect on the explicit attitude measures, t(57) ⫽ 7.73, p ⬍ .01, d ⫽ 2.05, and the size of this self-presentation effect did not differ between fakers and nonfakers, t(80) ⫽ 1.45, p ⫽ .15, d ⫽ 0.32. Furthermore, if the nonfakers did not take the SC-IAT task seriously, then we would expect their responses to be mostly error variance. On the contrary, these participants displayed a strong bias in favor of women, t(58) ⫽ 12.60, p ⬍ .01, d ⫽ 1.64, a result that is consistent with other female SC-IAT findings (Karpinski & Lytle, 2005). Finally, the fact that the nonfakers had low error rates also suggests that they approached the task with sincerity.

A Closer Analysis of SC-IAT Error Rates The high error rates observed in this study appear to be specifically associated with the instruction manipulation. The SC-IAT error rate was higher in Study 4 than in Studies 1–3, and other studies have found normal error rates on the female SC-IAT when no presentation instructions were provided (see Karpinski & Lytle, 2005). We conducted several follow-up analyses of the SC-IAT error rates to better understand why the instruction manipulation resulted in such high error rates on the SC-IAT. First, we investigated whether the increased error rate was due to an increase in true errors (pressing the wrong response key) or due to an increase in the failure to respond within the 1,500-ms response window. An inspection of the error rates revealed that the overall error rate (15.37%) was due to a high rate of incorrect responses (13.69%) and a low rate of failure to respond within the response window (1.68%). Next, we computed overall error rates separately for each class of target word (female vs. good vs. bad) within each block of the SC-IAT (Block 1: female ⫹ good vs. Block 2: female ⫹ bad). A Target Word ⫻ Block ⫻ Instruction Condition mixed ANOVA on these error rates revealed main effects for each of the factors, but these main effects were qualified by two significant higher order effects: a Block ⫻ Condition interaction, F(1, 164) ⫽ 12.00, p ⬍ .01, and a Target Word ⫻ Block interaction, F(2, 164) ⫽ 6.70, p ⬍ .01 (see Figure 2). The Block ⫻ Condition interaction indicates that participants who were instructed to present positive attitudes toward women had higher error rates when pairing female ⫹ bad than when

Figure 2. Single Category Implicit Association Test error rates by target, stage, and instruction condition (N ⫽ 84).

pairing female ⫹ good (d ⫽ 0.92), whereas participants who were instructed to present negative attitudes did not significantly differ in their error rates across the blocks (d ⫽ 0.05). On average, participants in both instruction conditions showed evidence of more positive than negative associations with women, suggesting that in the absence of faking, most participants have more positive than negative associations with women (see also Karpinski & Lytle, 2005). Thus, participants instructed to present positive attitudes may not have had to adjust their responses for the female ⫹ good pairings as a result of the instructions— only when they reached Block 2 (pairing female ⫹ bad) did they have to start monitoring their responses, resulting in higher error rates. Conversely, participants instructed to present negative attitudes had to adjust their responses for the female ⫹ good pairings in response to the instructions. As a consequence, they may have adjusted their responses from the start of the task, resulting in high error rates in both blocks. Therefore, this Block ⫻ Condition interaction is consistent with the claim that participants make more errors when they are attempting to fake a SC-IAT score. Consistent with this interpretation, when participants with high error rates were excluded from the analysis, the Block ⫻ Condition interaction dropped to nonsignificance, F(1, 114) ⫽ 1.52, p ⫽ .22. The Target Word ⫻ Block interaction indicates that participants’ error rates for the types of target words differed across blocks. For Block 1 (female ⫹ good), participants had a lower error rate for female target words than for good target words (d ⫽ 0.44) or for bad target words (d ⫽ 0.61; and error rates for good and bad target words did not significantly differ, d ⫽ .17). For Block 2 (female ⫹ bad), participants had higher error rates for good target words than for female or bad target words (ds ⱖ 0.43), and error rates for female and bad words did not significantly differ (d ⫽ 0.01). The interpretation of this interaction is unclear, but it does not appear to be a result of the presentation instructions. Supporting this conclusion, the Target Word ⫻ Block interaction remained significant once individuals with high error rates were excluded from the analysis, F(2, 114) ⫽ 7.87, p ⬍ .01. Finally, we examined whether the D-score algorithm might exaggerate the effect of a high error rate on SC-IAT scores. The

SINGLE CATEGORY IMPLICIT ASSOCIATION TEST

D-score algorithm replaces errors with the block mean plus a 400-ms penalty. To determine whether this error penalty affected the results, we computed an alternative SC-IAT score by using the D-score algorithm, but we eliminated all error responses from the computation. An analysis on these adjusted SC-IAT scores paralleled the previous analysis. When all participants were analyzed, an instruction effect was observed, t(82) ⫽ 2.53, p ⫽ .01, d ⫽ 0.55, and this effect disappeared when participants with high error rates were excluded, t(57) ⫽ 0.54, p ⫽ .59, d ⫽ 0.14. Thus, the observed effects are not dependent on the inclusion or exclusion of incorrect response times in the scoring algorithm.

Discussion The results of Study 4 suggest that participants can fake a SC-IAT score, but to do so they also significantly increase their error rate. Once participants with high error rates were excluded from the analysis, only a small, nonsignificant presentation effect remained. The bad news about these findings is that when participants wish to fake a SC-IAT score, a significant number of them spontaneously develop strategies that enable them to present the desired attitude. Participants in this study were not provided with strategies to fake the SC-IAT; they discovered strategies on their own. The good news about this finding is that when participants attempt to fake a SC-IAT score or to present a certain attitude on a SC-IAT, they are likely to make many errors, and they can be identified and excluded from subsequent analyses. Somewhat surprisingly, participants were also able to fake IAT scores in this study. Participants who were able to present a particular attitude on the IAT did so without significantly increasing their error rate, making it impossible to identify them and remove them from the sample. We are somewhat skeptical of this result, as no faking effect was found when IAT scores were computed by using the older log-based scoring algorithm. Nevertheless, the concerns about the faking of SC-IAT scores appear to be no worse than concerns about faking of IAT scores. The results of Study 4 suggest that several cautions regarding the use of the SC-IAT are in order. First, the ability of participants to fake a SC-IAT score may decrease power (due to the discarding of those with high error rates). Second, it may be the case that participants who are able to fake a SC-IAT score are different from those who were unable to fake a SC-IAT, resulting in a biased sample, although there is no evidence for this claim in the current study. Third, as previously highlighted, it may be easier for participants to fake a SC-IAT score if category targets are pictures and evaluative targets are words. For potentially sensitive domains, it may be prudent to present category targets and evaluative targets by using the same form of presentation (but there was no evidence of high error rates or faking of SC-IAT scores in Study 3 when Black and White faces were used as the target stimuli). Finally, although we did not find presentation effects on the SC-IAT once participants with high error rates were removed, it does not mean that participants cannot devise other strategies to fake a SC-IAT. Although participants did not spontaneously devise effective strategies to fake a SC-IAT score without increasing their error rate, if we had provided participants with a strategy, it is likely that they could have carried out the strategy and we would

29

have found a presentation effect on the SC-IAT. Participants may be more likely to spontaneously discover these alternative strategies if they complete multiple SC-IATs (although in another study, participants had to complete as many as four SC-IATs in a study session and showed no evidence of faking or presentation effects; see Karpinski & Lytle, 2005).

General Discussion We developed the SC-IAT as a single-category measure of social cognition to complement the IAT. Whereas the IAT measures comparative associations between two attitude objects, the SC-IAT can assess the evaluative associations with a single attitude object. Across three different attitude domains—soda brand preferences, self-esteem, and racial attitudes—we found evidence that the SC-IAT makes unique contributions in the ability to measure and understand implicit social cognition. In the soda brand domain, a Coke SC-IAT and a Pepsi SC-IAT predicted behavioral intentions above and beyond what was predicted by the IAT and explicit measures of soda brand attitudes. As a measure of self-esteem, the SC-IAT demonstrated a significant medium- to large-sized correlation with explicit measures of self-esteem. In addition, the self-SC-IAT and the self– other IAT score were only marginally correlated, supporting the claim that the SC-IAT’s measure of self-associations is theoretically distinct from the IAT’s measure of self– other associations. As a measure of racial attitudes, similar mean-level effects were observed on the race IAT and the race SC-IAT. However, the use of a White SC-IAT and a Black SC-IAT allowed for a more detailed interpretation of the implicit race bias. We also extensively examined the reliability of the SC-IAT. In general, the reliability of implicit measures of social cognition has been relatively poor, with the exception of the IAT (Bosson et al., 2000; Olson & Fazio, 2003). Across the four studies and six different SC-IAT measures, the internal consistency of the SC-IAT was reasonable (average r ⫽ .69; ranging from r ⫽ .55 to r ⫽ .85). These reliability coefficients are similar to the internal consistency observed for the IATs used in these studies (average r ⫽ .73; ranging from r ⫽ .58 to r ⫽ .82) and similar to the internal consistency observed for the IAT in previous research (Greenwald et al., 2003; Nosek et al., 2005). Thus, the SC-IAT appears to have a sufficient level of reliability to be of use as an individual difference measure of implicit social cognition. For the SC-IAT to be useful as a measure of implicit social cognition, the SC-IAT must be relatively unsusceptible to faking or self-presentational concerns. We found that when participants attempted to self-present an attitude on the SC-IAT, many of them had high error rates. Once participants with large error rates were removed from the sample, there was only a small, nonsignificant effect of self-presentation (d ⫽ .24). This result is consistent with previous research on the IAT, suggesting that there may be a small effect of faking or self-presentation on the IAT (Steffens, 2004). Together, these findings provide strong evidence for the reliability and validity of the SC-IAT as a measure of implicit social cognition.

30

KARPINSKI AND STEINMAN

Interpreting the SC-IAT as a Measure of Evaluative Associations There has been some controversy surrounding the interpretation of IAT scores (for overviews, see Arkes & Tetlock, 2004; Fazio & Olson, 2003). Because the methodology of the SC-IAT is a modification of the IAT, it is possible that the various proposed limitations and interpretations of the IAT also apply to the SCIAT. A common theme of these alternative interpretations is that IAT effects may be due, at least in part, to factors other than affective valence. For example, IAT effects may be observed because of the cost of task switching (Mierke & Klauer, 2001), a criterion shift across the different blocks of the task (Brendl, Markman, & Messner, 2001), and the salience of the attitude objects (Rothermund & Wentura, 2004). Additionally, the IAT has been described as a measure of environmental associations, or extrapersonal associations, rather than as a measure of one’s personal attitudes (Arkes & Tetlock, 2004; Karpinski & Hilton, 2001; Olson & Fazio, 2004). Some of these alternative interpretations are likely to apply to the SC-IAT as well. For example, the SC-IAT may reveal more about one’s environmental associations than one’s personal evaluative beliefs. Additionally, we have been careful to describe the SC-IAT as a measure of associations with a single attitude object and not as an absolute measure of attitudes toward an attitude object. It is likely that no attitude can be measured in absolute terms and that all attitudes require some type of comparative judgment (Festinger, 1950). For example, to determine that one likes Pepsi, a person must know how much he or she likes other beverages or how much other people like Pepsi. Although a contrast category is not specified in the SC-IAT and the SC-IAT is less comparative than the IAT (which explicitly requires a comparison category), the SCIAT also may not be an absolute measure of associations, in the purest sense. Furthermore, although the SC-IAT avoids using a contrast category for the target category of interest, it still measures the evaluation dimension comparatively (see Blanton et al., 2006). The presence of these possible alternative explanations for SCIAT effects and methodological issues regarding the SC-IAT do not suggest that the SC-IAT should not be used as a measure of associations, only that SC-IAT scores should be interpreted cautiously and with these potential limitations in mind.

Recommendations for Using the SC-IAT On the basis of the results of the studies reported here and other studies conducted in our lab, we have several recommendations for researchers interested in using the SC-IAT as a measure of implicit social cognition. First, as previously mentioned, the SC-IAT is conceptually identical to the ST-IAT and has only minor procedural differences from the ST-IAT (Wigboldus et al., 2005). Compared with the SC-IAT, the ST-IAT includes an initial practice stage with only good and bad target words, has fewer target words in each stage, and does not use a response window. Furthermore, the internal consistency of the SC-IAT tends to be higher than the internal consistency for the ST-IAT (Christian ST-IAT, adjusted r ⫽ .39; Islamic ST-IAT, adjusted r ⫽ .68). Until studies are conducted to investigate the effects of these procedural differences

on the reliability of single category or target measures, we recommend following the SC-IAT procedure to keep the reliability of the measure high. Second, we recommend that the SC-IAT be used with at least 24 practice trials and 72 critical trials in each critical block. Unlike the IAT, the SC-IAT does not have separate practice stages; within each block, it is necessary to include practice trials and to exclude those trials from the final calculation of SC-IAT scores. SC-IATs with only 48 trials have been found to have lower internal consistencies than SC-IATs with 72 trials. Likewise, the internal consistency of the ST-IAT, with only 20 trials per block, has also been found to be lower than the internal consistency of the SC-IAT (Wigboldus et al., 2005). Thus, we suggest that 72 trials be the lower bound for the number of trials that should be used for each stage in a SC-IAT. Third, we recommend that a 1,500-ms response window be included in the SC-IAT procedure. The response window may decrease the likelihood that participants engage in controlled processing during the task. Longer response windows do not appear to increase the reliability or to decrease the error rate, but it may be possible to decrease the response window without adversely affecting the error rate or the reliability of the measure. Fourth, we recommend using the D-score algorithm to compute SC-IAT scores. For the IAT, the D-score algorithm has been shown to increase reliability, increase the correlations between the IAT and explicit measures, reduce the correlation between the IAT measure and speed of responding, and reduce the effect of procedural variables (Greenwald et al., 2003). Although not presented here, we also examined SC-IAT scores by using a log-based scoring algorithm, modeled on the original IAT scoring algorithm (Greenwald et al., 1998). We observed a very small trend for larger effect sizes and larger SC-IAT-explicit correlations when the D-score algorithm is used compared with the log-based algorithm. We did not have the large samples required to detect small differences between the scoring algorithms by using tests of statistical significance, but because of the similar methodology shared by the SC-IAT and IAT, it is likely that many of the advantages of the D-score algorithm will carry over to the SC-IAT. Fifth, when the SC-IAT and the IAT are used in the same experimental session, we recommend that the SC-IAT measures precede any IAT measure. The IAT encourages a complementary, dichotomous mind-set toward the categories of interest. When an IAT is completed prior to a SC-IAT, the complementary mind-set may carry over to the SC-IAT, possibly resulting in a task that is not a measure of the associations with a single target category and in lower reliability for the task. However, it is also possible that completing the SC-IAT may adversely affect the IAT, and future studies are needed to examine the interplay between these two measures. Finally, we have presented the SC-IAT as a measure of the evaluative associations with a single target category. However, like the IAT, the SC-IAT is a flexible measure that can be modified to assess other aspects of implicit social cognition. For example, the IAT has been used as a measure of gender identity by assessing the strength of associations between the self and the male–female dimension, relative to the strength of associations between others and the male–female dimension (Aidman & Carroll, 2003; Greenwald & Farnham, 2000; Rudman & Goodwin, 2004). The SC-IAT can be easily modified to assess the strength of associations

SINGLE CATEGORY IMPLICIT ASSOCIATION TEST

between the self and the male–female dimension, without also assessing other associations. In this case, the gender dimension (male–female) would still be assessed as a comparative dimension. The SC-IAT can be used to eliminate one comparative dimension but not both of the comparative dimensions. Likewise, the SC-IAT can be modified to provide measures of implicit stereotypes and other aspects of implicit social cognition (see Karpinski & Lytle, 2005). Additionally, for investigators who wish to reduce the contaminating effects of environmental, or extrapersonal, associations (see Olson & Fazio, 2004), a personalized SC-IAT can be constructed by replacing the category labels good and bad with I like and I don’t like.

Conclusion The IAT has been a breakthrough in the ability to measure and understand implicit processes. The SC-IAT is a measure that can be used alone or in combination with the IAT to further illuminate these implicit processes. By continually developing and refining individual difference measures of implicit social cognition, social psychologists can improve their understanding of implicit and explicit social cognition.

References Aidman, E. V., & Carroll, S. M. (2003). Implicit individual differences: Relationships between implicit self-esteem, gender identity, and gender attitudes. European Journal of Personality, 17, 19 –36. Arkes, H. R., & Tetlock, P. E. (2004). Attributions of implicit prejudice, or “Would Jesse Jackson ‘Fail’ the Implicit Association Test?” Psychological Inquiry, 15, 257–278. Bargh, J. A., & Ferguson, M. L. (2000). Beyond behaviorism: On the automaticity of higher mental processes. Psychological Bulletin, 126, 925–945. Blanton, H., & Jaccard, J. (2006). Arbitrary metrics in psychology. American Psychologist, 61, 27– 41. Blanton, H., Jaccard, J., Gonzales, P. M., & Christie, C. (2006). Decoding the Implicit Association Test: Perspectives on criterion prediction. Journal of Experimental Social Psychology, 42, 192–212. Bosson, J. K., Swann, W. B., & Pennebaker, J. W. (2000). Stalking the perfect measure of implicit self-esteem: The blind men and the elephant revisited? Journal of Personality and Social Psychology, 79, 631– 643. Brendl, C. M., Markman, A. B., & Messner, C. (2001). How do indirect measures of evaluation work? Evaluating the inference of prejudice in the Implicit Association Test. Journal of Personality and Social Psychology, 81, 760 –773. Dasgupta, N., McGhee, D. E., Greenwald, A. G., & Banaji, M. R. (2000). Automatic preference for White Americans: Eliminating the familiarity explanation. Journal of Experimental Social Psychology, 33, 316 –328. De Houwer, J. (2002). The Implicit Association Test as a tool for studying dysfunctional associations in psychopathology: Strengths and limitations. Journal of Behavior Therapy and Experimental Psychiatry, 33, 115–133. De Houwer, J. (2003). The Extrinsic Affective Simon Task. Experimental Psychology, 50, 77– 85. Dovidio, J. F., Kawakami, K., Johnson, C., Johnson, B., & Howard, A. (1997). On the nature of prejudice: Automatic and controlled processes. Journal of Experimental Social Psychology, 33, 510 –540. Fazio, R. H., Jackson, J. R., Dunton, B. C., & Williams, C. J. (1995). Variability in automatic activation as an unobtrusive measure of racial attitudes: A bona fide pipeline? Journal of Personality and Social Psychology, 69, 1013–1027.

31

Fazio, R. H., & Olson, M. A. (2003). Implicit measures in social cognition research: Their meaning and use. Annual Review of Psychology, 54, 297–327. Festinger, L. (1950). A theory of social comparison. Human Relations, 7, 117–140. Gehring, W. J., Karpinski, A., & Hilton, J. L. (2003). Cognition during interracial interactions. Nature Neuroscience, 6, 1241–1243. Greenwald, A. G. (1992). New Look 3: Unconscious cognition reclaimed. American Psychologist, 47, 766 –779. Greenwald, A. G., & Banaji, M. R. (1995). Implicit social cognition: Attitudes, self-esteem, and stereotypes. Psychological Review, 102, 4 –27. Greenwald, A. G., & Farnham, S. D. (2000). Using the Implicit Association Test to measure self-esteem and self-concept. Journal of Personality and Social Psychology, 79, 1022–1038. Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring individual differences in implicit cognition: The Implicit Association Test. Journal of Personality and Social Psychology, 74, 1464 –1480. Greenwald, A. G., & Nosek, B. A. (2001). Health of the Implicit Association Test at age 3. Zeitschrift fu¨r Experimentelle Psychologie, 48, 85–93. Greenwald, A. G., Nosek, B. A., & Banaji, M. R. (2003). Understanding and using the Implicit Association Test: I. An improved scoring algorithm. Journal of Personality and Social Psychology, 85, 197–216. Karpinski, A. (2004). Measuring self-esteem using the Implicit Association Test: The role of the other. Personality and Social Psychology Bulletin, 30, 22–34. Karpinski, A., & Hilton, J. L. (2001). Attitudes and the Implicit Association Test. Journal of Personality and Social Psychology, 81, 774 –788. Karpinski, A., & Lytle, J. (2005). Measuring implicit gender attitudes, gender identity, and self-esteem using the Single Category Implicit Association Test. Unpublished manuscript. Kihlstrom, J. (1990). The psychological unconscious. In L. A. Pervin (Ed.), Handbook of personality: Theory and research (pp. 445– 464). New York: Guilford Press. Maison, D., Greenwald, A. G., & Bruin, R. H. (2004). Predictive validity of the Implicit Association Test in studies of brands, consumer attitudes, and behavior. Journal of Consumer Psychology, 14, 405– 415. Mierke, J., & Klauer, K. C. (2001). Implicit association measurement with the IAT: Evidence for effects of executive control processes. Zeitschrift fu¨r Experimentelle Psychologie, 48, 107–122. Monteith, M. J., Voils, C. I., & Ashburn-Nardo, L. (2001). Taking a look underground: Detecting, interpreting, and reacting to implicit racial biases. Social Cognition, 19, 395– 417. Nosek, B. A., & Banaji, M. R. (2001). The go/no-go association task. Social Cognition, 19, 625– 664. Nosek, B. A., Banaji, M. R., & Greenwald, A. G. (2002). Harvesting implicit group attitudes and beliefs from a demonstration Web site. Group Dynamics: Theory, Research, and Practice, 6, 101–115. Nosek, B. A., Greenwald, A. G., & Banaji, M. R. (2005). Understanding and using the Implicit Association Test: II. Method variables and construct validity. Personality and Social Psychology Bulletin, 31, 166 – 180. Nunnally, J. (1978). Psychometric theory. New York: McGraw-Hill. Olson, M. A., & Fazio, R. H. (2003). Relations between implicit measures of prejudice: What are we measuring? Psychological Science, 14, 636 – 639. Olson, M. A., & Fazio, R. H. (2004). Reducing the influence of extrapersonal associations on the Implicit Association Test: Personalizing the IAT. Journal of Personality and Social Psychology, 86, 653– 657. Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University Press. Rothermund, K., & Wentura, D. (2004). Underlying processes in the

KARPINSKI AND STEINMAN

32

Implicit Association Test (IAT): Dissociating salience from associations. Journal of Experimental Psychology: General, 133, 139 –165. Rudman, L. A., & Goodwin, S. A. (2004). Gender differences in automatic in-group bias: Why do women like women more than men like men? Journal of Personality and Social Psychology, 87, 494 –509. Steffens, M. C. (2004). Is the Implicit Association Test immune to faking? Experimental Psychology, 51, 165–179. Swim, J. K., Aikin, K. J., Hall, W. S., & Hunter, B. A. (1995). Sexism and racism: Old-fashioned and modern prejudices. Journal of Personality and Social Psychology, 68, 199 –214. Teige, S., Schnable, K., Banse, R., & Asendorpf, J. B. (2004). Assessment of multiple implicit self-concept dimensions using the Extrinsic

Affective Simon Task (EAST). European Journal of Personality, 18, 495–520. Wentura, D., Kulfanek, M., & Greve, W. (2005). Masked affective priming by name letters: Evidence for a correspondence of explicit and implicit self-esteem. Journal of Experimental Social Psychology, 41, 657– 663. Wigboldus, D. H. J., Holland, R. W., & van Knippenberg, A. (2005). Single target implicit associations. Unpublished manuscript. Wittenbrink, B., Judd, C. M., & Park, B. (1997). Evidence for racial prejudice at the implicit level and its relationship with questionnaire measures. Journal of Personality and Social Psychology, 72, 262– 274.

Appendix Target Words Used in the SC-IAT and IAT IAT target words

SC-IAT target words

Pleasant

Unpleasant

Good

Bad

brilliant diamond joy truth sunrise

awkward deatha hateb failureb filtha slumb stink uglya

beautiful celebrating cheerful excellent excitement fabulous friendly glad glee happy laughing likable loving marvelous pleasure smiling splendid superb paradise triumph wonderful

angry brutal destroy dirty disaster disgusting dislike evil gross horrible humiliate nasty noxious painful revolting sickening terrible tragic ugly unpleasant yucky

Note. SC-IAT ⫽ Single Category Implicit Association Test; IAT ⫽ Implicit Association Test. a Unpleasant words used in IAT for Study 3 only. b Unpleasant words used in IAT for Studies 1, 2, and 4 only.

Received May 20, 2005 Revision received August 18, 2005 Accepted August 19, 2005 䡲

Journal of Personality and Social Psychology 2006, Vol. 91, No. 1, 171–187

Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.171

Investigating the Dopaminergic Basis of Extraversion in Humans: A Multilevel Approach Jan Wacker, Mira-Lynn Chavanon, and Gerhard Stemmler Philipps-Universita¨t Marburg A recent theory suggests that the agency facet of Extraversion (E) is based on brain dopamine (DA). The paucity of human data relevant to this model is probably due to the lack of widely accessible noninvasive psychophysiological indices and well-established behavioral measures sensitive to both E and manipulations of DA activity. Aiming to identify such measures, the authors assessed the electroencephalogram and n-back task performance in groups of introverts and extraverts after administration of either placebo or a selective DA D2 receptor antagonist. As predicted, the antagonist’s effects on n-back reaction time measures and frontal versus parietal electroencephalogram theta activity were strongly and specifically modulated by E. New research avenues and theoretical extensions suggested by these results are discussed. Keywords: agentic Extraversion, dopamine D2 receptor antagonist, n-back task, electroencephalogram theta activity, frontal electroencephalogram asymmetry

Luciana, Arbisi, Collins, & Leon, 1994) or extremely expensive neuroimaging techniques (e.g., Farde, Gustavsson, & Jonsson, 1997). These methodological requirements considerably impede empirical evaluation of the model because the necessary laboratory equipment is typically either not available to personality psychologists or too expensive to be used with the large sample sizes necessary for the investigation of individual differences. Therefore, it is not surprising that the paucity of relevant data noted by Depue and Collins (1999) 7 years ago is still largely uncompensated. Noninvasive and more accessible psychophysiological indices of central DA functioning are needed for progress in this exciting field of research. Furthermore, even if the hypothesized link between E and brain DA had already been firmly established empirically, the important question of how individual differences in DA functioning are translated into extraverted versus introverted patterns of behavior would remain largely unanswered. To begin to address this problem at the human level, behavioral paradigms that are sensitive to both manipulations of functional brain DA activity and individual differences in E need to be established. The aims of the present study were: (a) to introduce an easily obtainable electroencephalogram (EEG) index that can be used to measure those individual differences in functional brain DA activity that are relevant to E and (b) to identify a behavioral paradigm that is sensitive to both manipulations of functional brain DA activity and individual differences in E. These psychophysiological and behavioral measures could then be used as a basis to further extend the theory proposed by Depue and Collins (1999) to the human level.

Thirty-five years ago, Jeffrey A. Gray (1970) proposed that Extraversion (E) and other personality traits reflect individual differences in motivational systems that evolved to deal with certain classes of stimuli associated with positive and negative reinforcement. Depue and Collins (1999) have integrated this general idea with the extensive database generated by behavioral neuroscience during the last decades into a novel psychobiological theory of E. In brief, their model holds that E is based on a broad motivational system that “facilitates” behavior motivated by positive incentives and that is neurobiologically based on the mesocorticolimbic dopamine (DA) system. Although Depue and Collins (1999) firmly rooted their theory of E in basic neuroscience insights gleaned from animal research, they have thus far not been particularly interested in using their model to derive new hypotheses and viable empirical approaches for personality research with humans. As also noted by others, “the model in its present form specifies, at the human level, little more than an expected pattern of correlations between trait measures and indices of dopaminergic neurotransmission” (commentary by Pickering in Depue & Collins, 1999, p. 534). However, currently, such indices of dopaminergic neurotransmission either require invasive measurements of blood hormone levels (e.g., Depue,

Jan Wacker, Mira-Lynn Chavanon, and Gerhard Stemmler, Faculty of Psychology, Philipps-Universita¨t Marburg, Germany. This article is based on the doctoral dissertation of Jan Wacker under the supervision of Gerhard Stemmler. The reported research was conducted with the help of Deutsche Forschungsgemeinschaft Grant Ste 405/9-1. We thank Thomas Scherer and Erwin Hennighausen for hardware and software support and Christof Kemper for his assistance in conducting the experiment. Correspondence concerning this article should be addressed to Jan Wacker, Faculty of Psychology, Philipps-Universita¨t Marburg, Gutenbergstrasse 18, D-35032, Marburg, Germany. E-mail: [email protected]

E and DA Although in several influential psychobiological theories (e.g., Cloninger, Svrakic, & Przybeck, 1993; Pickering & Gray, 1999; Zuckerman, 1991) the dopaminergic system is regarded as (part of) the basis of one of several highly correlated traits from the 171

172

WACKER, CHAVANON, AND STEMMLER

impulsivity-sensation seeking spectrum (novelty seeking, impulsivity, impulsive antisocial sensation seeking, or impulsive sensation seeking), based on psychometric analyses and theoretical considerations Depue and Collins (1999) argue that impulsivitysensation seeking is rather heterogeneous in terms of both the underlying motivations and brain systems and that therefore consistent and replicable associations with markers of DA neurotransmission are not to be expected. Depue and Collins (1999) further note that E can be subdivided into sociability on the one hand and agency on the other. It is the agency facet of E (i.e., a motivational disposition that comprises social dominance, enthusiasm, energy, assertiveness, ambitiousness, and achievement striving), which they regard as the central characteristic of E and for which they propose a dopaminergic basis in their theory. Depue and Collins (1999) suggest that agentic E is based on a broad motivational system, which they term the behavioral facilitation system (BFS). The BFS modulates or “facilitates” behavior motivated by positive incentives through increasing their motivational salience. Agentic extraverts possess a more reactive BFS and therefore are generally more enduring, motivated, and vigorous in their pursuits of positive incentive goals. In the neurobiological part of their theory, Depue and Collins (1999) propose that the mesocorticolimbic DA system (i.e., the dopaminergic projections of the ventral tegmental area in the midbrain to the nucleus accumbens, medial orbital prefrontal cortex (PFC), and other cortical and subcortical structures) form the basis of the BFS and that individual differences in functional properties of this neurobiological system are the basis of agentic E. To be more specific, they argue in their “psychobiological threshold model” that positive incentive stimuli induce a certain state level of postsynaptic DA receptor activation in terminal structures of the mesocorticolimbic DA system. This state level adds to the tonic trait level of DA activation, which is higher in individuals high versus low in agentic E. A certain level of trait plus state DA activity is necessary to induce behavioral facilitation. As a consequence, agentic extraverts reach this critical threshold more easily. Therefore, for these individuals, relatively weak incentive stimuli are sufficient to induce behavioral facilitation; in accordance, they experience and display the affective and behavioral concomitants of positive incentive motivation more frequently. To date, the strongest empirical support for Depue and Collins’ (1999) central theoretical proposals comes from studies in which a selective DA receptor agonist was administered while blood samples were taken to measure plasma levels of the hormone prolactin. It is known that the secretion of prolactin is modulated by central DA activity. Thus, changes in prolactin blood levels in response to challenges with dopaminergic drugs can be used as a peripheral indicator of central DA activity. In their landmark study, Depue et al. (1994) reported strong and specific correlations between agentic E as measured by Tellegen’s Multidimensional Personality Questionnaire (MPQ; scale Positive Emotionality; Tellegen & Waller, in press) and several characteristics of the DA agonistinduced inhibition of prolactin secretion in a sample of N ⫽ 11 female volunteers. These findings were replicated using the same methods in a larger sample, and no significant correlations emerged for the MPQ scales Negative Emotionality and Constraint (Depue, 1995, 1996). Although these observations strongly support the theory, both a replication by other research groups and a direct statistical comparison of the validities of agentic E and

impulsivity sensation seeking with respect to DA indicators remain to be conducted. The development of dopaminergic tracers that can be used in neuroimaging studies to more directly measure certain aspects of central DA activity recently has opened up new possibilities to investigate the hypothesized association between personality traits and DA. However, the few studies that have already employed this novel technique do not yet provide a conclusive picture (Breier et al., 1998; Farde et al., 1997; Gray, Pickering, & Gray, 1994; Kestler, Malhotra, Finch, Adler, & Breier, 2000; Laakso et al., 2000, 2003). In summarizing the evidence, it can still be concluded with Pickering and Gray (2001, p. 124) that “it seems premature to conclude that measures of E are more strongly and/or more consistently associated with DA functioning than measures of ImpASS [impulsive antisocial sensation seeking].” We would add to this conclusion that fast and significant progress on this question is unlikely as long as the established approaches to measurement of central DA activity in humans [i.e., invasive measurement of blood hormone levels and positron emission tomography (PET) scans with dopaminergic tracers that cost several thousand dollars per participant] are largely inaccessible to personality psychologists. Eysenck’s (1967) famous arousal theory of E inspired more than 1,000 empirical studies and stimulated some of the most creative research in personality psychology for the last 4 decades. It is conceivable that this enormous success of Eysenck’s (1967) theoretical proposals depended significantly on easily accessible noninvasive psychophysiological methods (e.g., EEG indices) to measure the central construct of the theory (i.e., general cortical arousal) in humans. Thus, easily obtainable noninvasive psychophysiological indices of individual differences in DA functioning that are relevant to E could prove very useful for progress in this field of research. In the following section, we selectively review the current literature in search of such indices in the EEG.

E, DA, and Frontal EEG Activity The EEG signal measured on the scalp is generated largely in cortical areas (Davidson, Jackson, & Larson, 2000). Therefore, the EEG cannot be used to directly measure activity in those subcortical elements of the mesocorticolimbic system that constitute the neurobiological core of the BFS. However, the PFC receives strong dopaminergic inputs from mesolimbic structures, and it is well known that DA plays an important modulatory role in PFC functioning (e.g., Seamans & Yang, 2004). As a consequence, EEG indices of PFC activity are a good place to start searching for noninvasive psychophysiological measures of central DA functioning.

Asymmetrical Frontal EEG Activity During the past 25 years, research on the relationship between frontal cortical asymmetry and emotion, motivation, and personality has attracted considerable attention (for recent review, see Coan & Allen, 2004a). Besides many other interesting associations, significant correlations between relative left frontal cortical activity, inferred from greater right than left frontal EEG activity in the alpha band (8 –13 Hz), and several personality traits from the E spectrum have been reported repeatedly (e.g., Coan & Allen,

EXTRAVERSION AND DOPAMINE

2003; Harmon-Jones & Allen, 1997; Schmidt, 1999; Sutton & Davidson, 1997). Findings like these typically have been interpreted within the theoretical framework proposed by Davidson and colleagues (e.g., Davidson, 1998), according to which the left and the right anterior region of the brain are part of two separate neural systems underlying approach and withdrawal motivation, respectively. The approach system is activated by the perception of goals and is thought to underlie initiation of goal-directed approach behavior toward these goals as well as elicitation of approachrelated motivational and emotional states (i.e., desire, wanting, enthusiasm, pregoal attainment positive affect, but also anger, see Harmon-Jones, 2004). On the neuroanatomical level, the approach system encompasses the left dorsolateral and medial prefrontal cortices and the dopaminergic circuitry of the basal ganglia. Several investigators (see, e.g., the commentary by Kline in Depue & Collins, 1999; Harmon-Jones & Allen, 1997; Sutton & Davidson, 1997; Wacker, Heldmann, & Stemmler, 2003) have noted a considerable conceptual overlap between Davidson’s (1998) approach system and Depue and Collins’ (1999) BFS or Gray’s behavioral activation system (BAS; e.g., Pickering & Gray, 1999). However, it is still a matter of debate whether and how Davidson’s description of a left frontal approach system can be integrated with the latter two models that focus on dopaminergic neurotransmission in largely subcortical brain structures and do not make any assumptions about asymmetric cortical activation patterns (Coan & Allen, 2004b; Hewig, Hagemann, Seifert, Naumann, & Bartussek, 2004; Wacker et al.). An empirical demonstration that frontal cortical asymmetry is sensitive to changes in brain DA functioning (or vice versa) would, thus, constitute an important step toward conceptual integration. In summary, frontal EEG asymmetry in the alpha band has not only been associated with traits from the E spectrum in several empirical studies, but there are also good theoretical reasons to expect that this psychophysiological index is related to certain aspects of central DA activity.

Symmetrical Frontal EEG Activity Several researchers recently have emphasized the importance of bilateral (or symmetrical) resting frontal EEG activity in the context of personality traits (Davidson, 2004; Hewig et al., 2004; Hewig, Hagemann, Seifert, Naumann, & Bartussek, 2006) and affective disorders (Nitschke, Heller, Etienne, & Miller, 2004). For example, Hewig et al. proposed that low resting bilateral frontal EEG activity in the alpha band (i.e., presumably high frontal cortical activity) is associated with high habitual activity in Gray’s BAS. The BAS is a DA-based motivational system closely related to Depue and Collins’ (1999) BFS, which responds to stimuli of reward and nonpunishment, elicits positive emotions, and motivates goal-directed behavior (e.g., Pickering & Gray, 1999). Hewig et al. (2004, 2006) repeatedly observed significant negative correlations between scores in Carver and White’s (1994) BAS scale and resting EEG alpha power at frontal scalp sites. Furthermore, several recent functional magnetic resonance imaging studies have found substantial correlations between traits from the E spectrum and either increased or reduced activity in the anterior cingulate cortex (ACC) during a number of cognitive and emotional tasks (Canli, Amin, Haas, Omura, & Constable, 2004; Canli et al., 2001; Gray & Braver, 2002; Kumari, ffytche, Williams, &

173

Gray, 2004). The ACC is an area of the PFC that receives strong dopaminergic inputs (Holroyd & Coles, 2002; Seamans & Yang, 2004) and whose activity is influenced by genes controlling PFC DA neurotransmission (Blasi et al., 2005). It is interesting that there is now substantial evidence (e.g., Pizzagalli, Oakes, & Davidson, 2003) that activity in this region is reflected in frontal midline EEG activity in the theta band (4 – 8 Hz). Taken together, the recent literature suggests that both increased activity in the frontal cortex (as measured by reduced EEG alpha activity at frontal sites) and either increased or reduced activity in the ACC (as measured by increased or reduced frontal midline EEG theta activity) are not only related to E and/or related traits but are also modulated by brain DA. To measure individual differences in resting frontal EEG activity, two issues need to be considered. First, irrelevant individual differences in skull thickness affect the amplitude of the EEG signal measured on the scalp. Second, due to volume conduction, electrical activity generated in other areas of the cortex is also measurable at the sites of interest and vice versa. In prior research, two approaches have been used to control for these effects. The first approach employs a regression procedure that corrects the EEG power at the target site in the frequency band of interest for global EEG power in that frequency band computed as the average power across all electrodes. However, to use this method, the EEG needs to be recorded from a relatively large number of electrodes to derive valid estimates of global EEG power (a minimum of 20 electrodes is recommended by Davidson et al., 2000). The second approach is to simply compute difference values between EEG power measured at the site of interest and EEG power measured at a separate site, thought to be influenced by different brain processes. This approach has been employed with great success in research on frontal EEG asymmetry by calculating an asymmetry index as the difference between homologous right and left electrodes. Although this method only yields measures of relative activity at the two sites used to compute the difference values, it bears the advantage that as few as two active electrodes are sufficient to obtain the EEG index of interest. Application of this approach, however, requires that the second electrode site is not chosen randomly but based on relevant theoretical and empirical information. Difference values should only be computed when the relative activity at the two sites arguably captures a (theoretical) meaningful aspect of EEG activity (e.g., EEG lateralization in research on frontal asymmetry). It is interesting that Hewig et al. (2004, 2006) did not only repeatedly observe negative correlations between BAS sensitivity and EEG alpha power at frontal scalp sites, as already noted above, but also an (unexpected) positive correlation of similar magnitude between BAS sensitivity and EEG alpha power at parietal and parietooccipital sites. As a consequence, in these two studies, a difference value computed as the difference between EEG alpha activity at frontal sites minus activity at parietal sites would have been maximally suited to capture the aspect of brain electrical activity related to habitual BAS activation. Also, it has long been known that two different manifestations of EEG theta rhythm can be distinguished (Pizzagalli et al., 2003; Schacter, 1977). In addition to the already described frontal midline variant, which has been linked to alert states characterized by focused attention and mental effort, a second variant exists, which exhibits a more posterior and widespread distribution and has been associated with

174

WACKER, CHAVANON, AND STEMMLER

decreased prestimulus alertness and less focused, more freely floating attention (e.g., during hypnagogic states). Because the mental processes associated with the two types of theta rhythm are at least in part mutually exclusive, it would seem likely that the activity in brain sources of frontal midline and posterior theta are also reciprocally related. If this were in fact the case, the difference between frontal midline theta activity and theta activity at posterior (e.g., parietal) sites would be well suited to capture a meaningful aspect of EEG theta activity in a single measure. In summary, it can be concluded that based on prior findings, symmetrical frontal EEG activity both in the alpha and theta bands can be hypothesized to be associated with both E-related traits and brain DA functioning and that difference values of EEG band power at frontal sites minus band power at parietal sites are well-suited to capture these aspects of brain electrical activity in two simple measures.

E, DA, and Behavior: The N-Back Task There is ample evidence for an involvement of DA in working memory (WM) functions from a variety of different methodological approaches (for recent reviews, see, e.g., Barch, 2004; Seamans & Yang, 2004). It is interesting that researchers have repeatedly reported that DA agonists (e.g., amphetamine or the D2 receptor agonist bromocriptine) improve performance in WM tasks (and other tasks known to strongly recruit the PFC, such as the Wisconsin Card Sorting Task) only in individuals with poor baseline performance but impair performance in individuals with good baseline performance (e.g., Kimberg, D’Esposito, & Farah, 1997; Mattay et al., 2000, 2003). Using the well-known n-back WM task with increasing levels of task difficulty (i.e., WM loads of 1-, 2-, and 3-back), Mattay et al. (2000, 2003) recently demonstrated that there are genetically determined individual differences in DA neurotransmission that are associated with both performance in the most difficult version of the n-back task (3-back) under normal conditions and changes in performance and PFC activity induced by dopaminergic drugs. However, are these individual differences in DA neurotransmission also associated with E? Several recent observations indicate that this may indeed be the case. It has been demonstrated that both E and BAS sensitivity predict the magnitude of increases in brain activity in dorsolateral PFC and the ACC produced by the difficult 3-back task (Gray & Braver, 2002; Kumari et al., 2004). In addition, Lieberman and Rosenthal (2001) have shown that reaction times (RTs) in the difficult (2- and 3-back) but not in the easy (0- and 1-back) versions of the n-back task are negatively correlated with E, that is, extraverts reacted faster than introverts specifically in those versions of the task that impose strong demands on WM. However, until now, the negative correlation between E and n-back RTs has not been successfully replicated, and it is unknown whether it is indeed (partly) based on individual differences in DA functioning as proposed by Lieberman and Rosenthal (2001). If a dopaminergic basis could be demonstrated for this E effect, this would not only support Depue and Collins’ (1999) claim of a link between DA and E but also suggest that the n-back task could be a useful behavioral paradigm to investigate the question of how individual differences in DA functioning are translated into extraverted versus introverted patterns of behavior.

The Present Study Based on prior findings, agentic E is hypothesized to be associated with: (a) asymmetrical frontal EEG activity and in the alpha band and symmetrical frontal EEG activity in the alpha and theta bands and (b) performance in the n-back task. One aim of the present study was to demonstrate these associations empirically. It is more important that the present work aimed at examining whether these associations can be attributed to a common dopaminergic basis. To this end, we measured the EEG and n-back performance in extreme groups of participants either high or low in agentic E after administration of either placebo or a selective dopaminergic drug. Because dopaminergic drugs have been shown to invert preexisting group differences in measures of brain activity and task performance (e.g., Corr & Kumari, 1997; Kimberg et al., 1997; Mattay et al., 2000, 2003; Takeshita & Ogura, 1994), we predicted that the differences between introverts and extraverts1 expected for the placebo condition would be inverted in the drug condition, resulting in a significant interaction between E (introverted vs. extraverted) and pharmacological substance (placebo vs. dopaminergic drug). If such a significant interaction could be demonstrated for our behavioral and/or EEG parameters, this would not only add to the still very limited database currently supporting the suggested link between agentic E and DA (Depue & Collins, 1999) but also open up new avenues for experimental personality research on the human level by providing more accessible indicators of central DA functioning.

Method Participants A sample of 46 participants who were either high or low in agentic E was selected from a pool of N ⫽ 507 male student volunteers (see below). We only used male participants because in females, the possibility of an undetected pregnancy constitutes an additional risk in pharmacological studies and because the menstrual cycle is known to introduce error variance into physiological recordings. Further inclusion criteria were German native language, age 18 to 40 years, right-handedness, and no history of mental disorders as assessed with a standardized clinical interview (Margraf, 1997). Exclusion criteria were treatment with prescription drugs during the last 3 months, acute diseases, history of abnormal conditions of the liver or the kidney, habitual smoking of more than 10 cigarettes per day, or use of illegal drugs. Six participants were excluded because of strong artifacts in the EEG recordings (excessive blinking and/or muscle artifacts). The average age of the N ⫽ 40 participants in the final sample was M ⫽ 24.4 years (range, 18 –34 years), and all were university or high school students. Participants were told that they were selected based on self-ratings of certain aspects of their personality and that the aim of the study was to further delineate the brain processes and neurotransmitters underlying a RT task and to find out how these relate to personality. They were paid 45 EUR (55 USD) for approximately 5 hours of involvement in the study.

1 To facilitate reading, we will occasionally use the terms E or extraverted/introverted instead of agentic E and high/low in agentic E when referring to the extreme groups of participants examined in the present study. However, we do not wish to imply that agentic E and E are one and the same: In addition to the agency facet, the super trait E also includes sociability and, in some conceptions, impulsivity (see Depue & Collins, 1999).

EXTRAVERSION AND DOPAMINE

Experimental Design and Extreme Group Selection Participants were assigned to one of four experimental groups formed by the combination of E (extraverted, introverted) and Substance (placebo, D2 antagonist).2 For the analyses of n-back performance and posttask waiting period EEG, this 2 ⫻ 2 design was supplemented by the repeated factor Task (0- to 3-back task with the possibility to obtain reward and 2-back task without reward; see below) with the effect of presentation order controlled via cyclic permutation in a Latin Square. Depue et al. (1994) recommended the MPQ scale Positive Emotionality (Tellegen & Waller, in press) for measuring agentic E. Therefore, we developed a German short version of this scale to use it as an economical screening instrument for extreme group selection. This brief questionnaire contains three positively correlated 10-item scales that correspond to the three MPQ primary scales most relevant to Positive Emotionality: Wellbeing, Achievement, and Social Potency (see Tellegen & Waller, in press). In prior unpublished work from our laboratory, the three newly developed scales correlated highly with the respective MPQ primary scales (r ⬎ .85, N ⫽ 140). The sum of the three scales constituted our aggregate measure of agentic E. A total of N ⫽ 507 male study volunteers were contacted on several locations across the university campus and filled out the newly developed 30-item personality questionnaire. Internal reliabilities calculated in this pool of volunteers were satisfactory for both the three 10-item primary scales (Cronbach’s ␣ ⱖ .81) and the 30-item aggregate scale (Cronbach’s ␣ ⫽ .87). Participants scoring above the median in all three primary scales constituted the extraverted group, whereas participants scoring not higher than the median on any of the three scales constituted the introverted group. The 40 participants in the sample of the present study scored either above the 67th percentile (20 participants in the extraverted extreme group) or below the 33rd percentile (20 participants in the introverted extreme group) on the aggregate scale of agentic E. Ten participants of each extreme group were randomly assigned to either the placebo or the D2 antagonist group.

Sulpiride: A Selective DA D2 Receptor Antagonist We used the selective D2 antagonist sulpiride (200 mg) to challenge the DA system. The substituted benzamide sulpiride exerts its antagonistic effects predominantly at D2 receptor sites to which it binds with high affinity (Missale, Nash, Robinson, Jaber, & Caron, 1998). D1 receptors, GABA receptors, and adrenergic, cholinergic, histaminergic, and serotonergic receptors are largely unaffected by sulpiride (Caley & Weber, 1995; Mauri, Bravin, Bitetto, Rudelli, & Invernizzi, 1996). The drug accumulates more strongly in the mesolimbic than in the nigrostriatal DA system, and it completely blocks stimulatory effects induced by the selective D2 agonist bromocriptine (Jackson & Hashizume, 1987). Sulpiride reaches its maximum plasma concentration 1 to 6 h after oral administration (Caley & Weber, 1995; Mauri et al., 1996). A single dose of sulpiride is well tolerated by healthy volunteers, who are typically not able to tell whether they received placebo or sulpiride (e.g., McClelland, Cooper, & Pilgrim, 1990; Meyer-Lindenberg, Rammsayer, Ulferts, & Gallhofer, 1997).

175

In an adjacent room were placed a 32-channel SynAmps 5083 amplifier (NeuroScan), a Macintosh Power Mac G4/450 (Apple Computer, Cupertino, CA) with a PCI 6503 SCSI card (National Instruments, Austin, TX) that performed recording and storage of the digitized EEG data under Labview 5.0 (National Instruments), and a personal computer that performed experimental control under Presentation 0.5 (Neurobehavioral Systems, Albany, CA).

Procedure Informed consent was obtained at the beginning of the session. The experimenter then conducted a standardized clinical interview (Margraf, 1997) to check for lifetime absence of Diagnostic and Statistical Manual of Mental Disorders (American Psychiatric Association, 1994) Axis I psychiatric disorders. If the participant met this final inclusion criterion, the experimenter administered identical capsules of either sulpiride (200 mg) or placebo under double-blind conditions at 9:30 a.m. ⫾ 75 minutes. Participants fasted and abstained from nicotine and caffeine from 10 p.m. of the previous night until completion of the experiment with the exception of a light breakfast (two rolls with butter, cheese, and/or jam) and an 8-ounce glass of 1.5% milk taken at the time of medication (or placebo) to prevent nausea. Directly after completion of the breakfast, a brief (20 –25 minute) intelligence test was conducted; then, participants completed several personality questionnaires (see below). The experimenter, aided by an assistant, then positioned electrodes and transducers and explained the n-back task. Afterward, 40 practice trials of the 2-back task were presented. Before leaving the room, the experimenter reminded the participant to sit quietly to help prevent artifacts in the physiological recordings and also told him that further instructions and task feedback were prerecorded and would be presented to him over the loudspeakers. On average, the experiment was started 186 minutes (range, 150 –205 minutes) after administration of the medication (or placebo) capsule. First, participants were instructed to relax during a 10-min rest period (with three embedded 1-min EEG recording periods) while keeping their eyes open. Next, the n-back tasks were presented in five consecutive blocks with the effect of block position controlled via cyclic permutation in a Latin Square. Each block was followed by a 1-min period, during which participants were instructed to simply wait for a moment, before either the next block would begin (2-back task without possibility to obtain reward) or the experimenter would briefly enter to give performance feedback and, if the participant passed the criterion, to hand out the promised reward (0-, 1-, 2-, and 3-back task with the possibility to obtain reward). During each of the five 1-min posttask (prereward) waiting periods, the EEG was recorded. After completion of the final task block, the participant filled out a task questionnaire presented on his flat screen display and answered a brief postexperimental interview. The experiment ended approximately 4 h and 20 min after administration of the medication (or placebo) capsule.

N-Back Task and Incentive Motivational Context The version of the n-back task employed was similar to the one used by Braver et al. (1997). Single white letters (Times New Roman, 60 point)

Setting and Apparatus The experimental room (4 ⫻ 3.4 m) was sound-attenuated and airconditioned and had a largely nontechnical appearance. Participants sat comfortably in a reclined position. Electrodes were connected to a customized headbox (NeuroScan, Sterling, VA), where signals were preamplified with a gain of 30 (input impedance, 10 M⍀). N-back stimuli were presented on a 15-inch flat screen display placed in front of the participant (distance to the eyes, approximately 60 –70 cm). Participants reacted by pressing buttons on a response box (XQMS, Frankfurt, Germany). Prerecorded instructions and verbal task feedback were presented via two loudspeakers located on the walls of the experimental room.

2

Initially, the factor Substance included a third group of participants who received a low dose (1.25 mg) of the selective D2 agonist bromocriptine. However, unexpectedly, several participants in this drug condition experienced severe side effects (nausea, dizziness). Because blindness to the experimental condition thus could not be secured for the bromocriptine group, rendering the obtained results difficult to interpret, we decided not to report the bromocriptine data. The few statistically reliable effects that were observed for bromocriptine despite the error variance that was probably introduced by the aversive side effects were similar to those reported for sulpiride, although somewhat smaller in magnitude.

176

WACKER, CHAVANON, AND STEMMLER

were presented successively on a black computer screen. For each of the four levels of difficulty (0- to 3-back), there was a different decision criterion for making the target response of either yes or no (see Braver et al., 1997) with the index or middle fingers of the right hand, respectively. For each version of the task, a sequence of 60 practice trials (20 targets) followed by 120 evaluated trials (40 targets) was presented. The last 50 practice trials of each task version were used to compute the 90th percentile of the individual RT distribution for correct reactions (only RTs that were not slower than the individual RT mean plus 3 times the individual RT SD were considered). This served as an individual RT criterion for the following evaluated task sequence. Each letter was presented for 500 ms followed by a 1,650-ms interval, during which the screen was black. The participant was expected to react during this 2,150-ms interval. Each trial ended with the presentation of a verbal feedback (350 ms) that indicated whether the participant’s preceding reaction was “correct”, “wrong” (i.e., either a wrong reaction or no reaction), or “slow” (i.e., correct, but slower than the individual RT criterion). Directly after the feedback had ended, the next letter was presented. As explained in more detail elsewhere (Pauls, Wacker, & Crost, 2005; Stemmler, 1997), we conceptualize E and other personality traits as dispositions that are activated only in relevant situational contexts (for discussion of related theoretical perspectives, see Mischel & Shoda, 1998). Even though the current study was not designed to provide evidence for this traits-as-dispositions approach, this general concept nonetheless significantly influenced certain aspects of the present experimental setup. When traits are only activated in relevant situational contexts, their predictive power concerning behavior and the associated brain states will also be limited to such relevant contexts. For agentic E, as defined by Depue and Collins (1999), relevant contexts are characterized by the presence of positive incentives. As a consequence, we took special precautions to create an experimental setting that is strongly characterized by positive incentives. Right at the beginning of the laboratory session, the participant was told that in each version of the task, he had the chance to obtain a reward for good performance. The potential rewards consisted of a variety of popular goodies each worth about 1 EUR (1.3 USD). Goodies were displayed throughout the whole experimental session in a large cardboard box placed across the room from the sitting participant. After completion of each task version, the participant was allowed to pick one of the goodies if he reached the following relatively low performance criterion not known to him: He had to react correctly and fast enough to more than 50% of the targets and to more than 50% of the nontargets. Because, by definition, the first three letters of the evaluated sequence could not be targets in the 3-back version of the task, only the last 117 trials were used to check the performance criterion and to compute performance measures for statistical analyses. Participants were told to react as quickly and as accurately as possible. To check whether the possibility to obtain a reward moderated E effects, the 2-back version of the task was presented twice (using different sequences of letters): once with and once without the possibility to obtain reward. In contrast to the task presentation with possible reward already described above, in the presentation without possible reward, the participant was told that this time, he could not obtain a reward and that the experimenter would not enter the room after the task to give performance feedback. However, the participant was explicitly asked to engage in the task even though a reward could not be obtained. For each of the five task presentations, the following three performance measures were calculated: (a) the average RT for correct reactions to targets, (b) the percentage of correct reactions to targets, and (c) the SD of RTs for all correct reactions. These choices were based on the following considerations: (a) Gray, Chabris, and Braver (2003) have convincingly argued that the performance in target trials is a more valid indicator of frontal cortical functions than the average RT of all correct reactions as used by Lieberman and Rosenthal (2001). (b) It has long been known that when given the choice, introverts compared with extraverts tend to work more slowly but also more accurately on a variety of tasks (e.g., Wilson,

1978). As a consequence, in studies like the present one, it is mandatory to consider the possibility that potential E effects are based on a significant speed–accuracy trade-off. To simplify the presentation of the results and to ensure that measures of speed and accuracy were based on the same trials, we decided to quantify accuracy as the percentage of correct target reactions (see also Gray et al.). An alternative analysis using signal detection measures (d⬘ and C) did not yield any additional information. (c) Individual distributions of RTs are typically skewed leading to a positive correlation between average RTs and individual SD values of RTs: Individuals whose RTs vary more strongly on a trial to trial basis (e.g., between targets and nontargets) also tend to have longer average RTs. Therefore, it is possible that the RT differences between introverts and extraverts observed by Lieberman and Rosenthal (2001) would have been even more pronounced had these authors used measures of RT variability instead of RT means. To prevent distortion of the results by invalid reactions, all performance measures were based only on trials in which participants pressed either the yes button or the no button in a 1,950-ms time window starting 200 ms after the beginning of the presentation of the letter and ending with the beginning of the verbal feedback. The mean percentage of valid trials was at least 96.7% in all possible combinations of tasks and experimental groups. For the calculation of all statistical tests, the two RT measures were square root transformed to normalize distributions. Reliability estimates for the three performance parameters are shown in Table 1.

EEG Recording and Analysis In keeping with the 10-20 International System (Jasper, 1958) and using Easy Cap electrode caps (Falk Minow Services, Herrsching-Breitbrunn, Germany), recordings were made from midfrontal (F3/F4), frontal midline (Fz), and parietal midline (Pz) regions of the scalp. EEG data were also obtained from central (C3/C4/Cz), anterior temporal (T3/T4), and parietal (P3/P4) sites, but these recordings are not analyzed here. All sites were referenced to resistor-linked mastoids (M1–M2; 5-K⍀ resistors). To record eye blinks and vertical eye movements, electrodes were placed midline above and below the right eye. Electrodes on the outer canthi of both eyes were applied to record horizontal eye movements. Electrode impedances were kept under 5 K⍀ for the EEG electrodes and under 1 K⍀ for the ground electrode by cleaning the skin with alcohol and treating it with a mild abrasive. For all EEG sites, InVivo-Metrics (Healdsburg, CA) Ag-

Table 1 Grand Means (SDs) and Reliabilities of the Three Performance Measures in the Five Versions of the N-Back Task Target RT Task 0-Back 1-Back 2-Back 2-Back (no reward) 3-Back

M (SD) 452 (57) 490 (91) 600 (149) 595 (147) 672 (214)

Rel .91 .96 .94 .95 .95

RT Variability M (SD) 87 (27) 116 (38) 157 (53) 165 (63) 177 (72)

Rel .79 .86 .89 .90 .92

Correct target reactions (%) M (SD) 94 (05) 87 (10) 72 (17) 72 (16) 55 (14)

Rel .66 .62 .82 .85 .58

Note. N ⫽ 40. Rel ⫽ reliability estimated by applying the SpearmanBrown prophecy formula to the correlations between individual values calculated for the first and second halves of the trials. RT values were square root transformed before calculation of correlations. For each task, both halves contained 20 target trials each.

EXTRAVERSION AND DOPAMINE AgCl electrodes (8 mm) were used. For the ground electrode and the electrooculogram (EOG) sites, disposable VivoMed (Servoprax, Wesel, Germany) Ag-AgCl electrodes (10 mm) were employed. EEG and EOG were amplified (EEG, gain ⫽ 500; EOG, gain ⫽ 100), filtered (bandpass set to 1–50 Hz for EEG, lowpass set to 1,000 Hz for EOG, 50-Hz notch filter enabled), digitized at 2,000 Hz, and stored. In a second step, the signal was down-sampled to 250 Hz and converted to physical units and visually scored for artifact. Data were excluded if muscle, eye movement, or other artifacts were present. Spectral power was then computed for 1-s artifact-free epochs overlapping by 50% using a fast Fourier transform with a Hamming window. The Fourier transform was performed on 4-s segments each containing 1-s epochs of windowed data padded with zeros on both sides. The resulting estimates of power density (microvolts squared per hertz) for each artifact-free 0.5-s interval were clustered into two bands, theta (4.00 –7.75 Hz) and low alpha (8.00 –10.25 Hz). We decided to focus on low instead of broad alpha (8 –13 Hz) because low alpha power has been shown to be more strongly associated with PET measures of cortical activity (Oakes et al., 2004) and because we have previously found frontal asymmetry effects to be more pronounced in the low alpha band (Wacker et al., 2003). Power values for the two bands were transformed to natural logarithms to normalize the distributions of scores to be used in statistical analyses (see, e.g., Davidson et al., 2000). Because estimates of spectral power based on short time intervals are unreliable (Davidson et al.), means of ln-transformed power estimates per 1-min recording phase were obtained only if more than 20% of the 0.5-s data (i.e., at least 12 s of EEG data) were nonmissing, otherwise the phase was set missing data for a given regionband combination. We then computed the EEG indices of interest separately for each recording phase as follows: (a) frontal EEG asymmetry ⫽ ln-transformed power at F4 minus ln-transformed power at F3 and (b) frontal versus parietal EEG activity ⫽ ln-transformed power at Fz minus ln-transformed power at Pz. An analysis of variance (ANOVA) with E (extraverted, introverted) and Substance (placebo, sulpiride) as between factors did not reveal any significant group differences in the mean number of artifact-free epochs used to compute within-subject averages of frontal EEG asymmetry and symmetrical frontal versus parietal EEG activity for each of the eight data recordings, F(1, 36) ⱕ 0.34, p ⬎ .57. On average, the resting EEG indices for each 1-min recording were based on M ⫽ 62.0 (SD ⫽ 20.5) artifact-free epochs. Finally, we averaged the three recordings of the rest period.

Intelligence Test and Personality Questionnaires We assessed general fluid intelligence using the German short version of Cattell’s Culture Fair Test (Scale 3, Cattell & Weiß, 1971). Besides our short version of the MPQ scale Positive Emotionality (see above) used for selection of extreme groups according to agentic E, we also used German versions of the revised Eysenck Personality Questionnaire (EPQ-R; Ruch, 1999) and the Zuckerman-Kuhlman Personality Questionnaire (ZKPQ; Zuckerman, 2002). The EPQ-R measures Eysenck’s personality traits of E (mainly the sociability component), Neuroticism, and Psychoticism. The ZKPQ measures Zuckerman’s “Alternative Big Five” AggressionHostility, Neuroticism-Anxiety, Sociability, Activity, and Impulsive Sensation Seeking. The ZKPQ was employed for two reasons: (a) It nicely separates the sociability component from the agency component of E by providing separate scales for Sociability and Activity. Based on Depue and Collins’ (1999) theory, strong relationships with indicators of brain DA are expected only for Activity. (b) It also measures Impulsive Sensation Seeking, that is, the trait considered by some to be more closely related to brain DA than agentic E (see, e.g., Pickering & Gray, 1999).

Statistical Data Analysis N-back performance measures and EEG parameters from the posttask waiting periods were analyzed with three-way analysis of covariance

177

(ANCOVA) with Substance and E as between-subjects factors, Task (5; 0to 3-back with reward and 2-back without reward) as repeated factor, and Block Number (coded from ⫺2 to 2) as a covariate. Making full use of the Latin Square design employed to control for possible effects of task presentation order, the covariate Block Number was included to reduce the error term by controlling irrelevant variance attributable to linear changes across task blocks. Using PROC MIXED of SAS/STAT (SAS 6.12 for Macintosh, SAS Institute Inc., Cary, NC), the error variance-covariance matrix for the repeated factor was specified to be completely general (TYPE ⫽ UN option). EEG parameters from the initial rest period were analyzed with two-way ANOVAs with Substance (2; placebo, sulpiride) and E (2; introverted, extraverted) as between-subjects factors. ANOVAs and ANCOVAs were followed by a priori-specified contrasts tested with an ␣ level of .05. Contrasts for the EEG parameters from the initial rest period were defined on the level of the Substance ⫻ E interaction and depicted E effects within substance groups and Substance effects within E groups. The E effects for the two EEG alpha measures (frontal asymmetry and frontal vs. parietal activity) under placebo were tested one-tailed, and the remaining contrasts for EEG variables were tested two-tailed. Contrasts for the behavioral variables and the EEG parameters from the posttask waiting periods were defined on the level of the Substance ⫻ E ⫻ Task interaction and were designed to test for: (a) differences between introverts and extraverts within substance groups for each of the five n-back tasks; (b) differences of linear changes in performance with increasing task difficulty (0- to 3-back) between introverts and extraverts within substance groups; (c) differences of these E effects (in each of the five tasks and in the linear performance changes) between substance groups or, put differently, differences of the substance effects (sulpirideplacebo) between introverts and extraverts; and (d) differences between the 2-back tasks with versus without possible reward within each of the four experimental groups. For the behavioral variables with the exception of the contrasts subsumed under (d) directed hypotheses could be derived from prior findings (see the Results section). Thus, for the behavioral variables, one-tailed tests were calculated for contrasts (a), (b), and (c), whereas contrasts (d) were tested two-tailed. For the two EEG alpha measures (frontal asymmetry and frontal vs. parietal activity) during the waiting periods, the differences between introverts and extraverts in the placebo group for each of the five tasks were tested one-tailed (compared with introverts, extraverts were expected to display greater relative right vs. left frontal EEG alpha activity and lower frontal vs. parietal EEG alpha activity). The remaining EEG contrasts were tested two-tailed.

Results Check for Side Effects and Blindness to Substance Group Participants did not report any adverse side effects. The ratings of nausea and dizziness averaged across experimental phases were very low (⬍1) in all four experimental groups (0 ⫽ not at all applicable and 1 ⫽ not applicable). Also, the percentage of participants, who guessed (in a forced choice question posed after completion of the experiment) that they had received a pharmacologically active substance, did not differ between the four experimental groups (extraverts-sulpiride, 2/10; introverts-sulpiride, 5/10; extraverts-placebo, 5/10; introverts-placebo, 3/10). When asked to evaluate the confidence in their own guess concerning the substance received in a probability rating, none of the participants reported to be 100% sure (M ⫽ 63.8%, SD ⫽ 18.7). Thus, it can be concluded that participants were indeed blind to the experimental condition as intended.

178

WACKER, CHAVANON, AND STEMMLER

Check for Preexisting Differences Between Experimental Groups A series of ANOVAs with Substance and E as between-subjects factors and age, general fluid intelligence, and all ZKPQ and EPQ-R scales as dependent variables revealed no significant main effects of Substance and no significant E ⫻ Substance interactions. Several significant main effects of E were observed; supporting convergent validity of our brief MPQ scale of agentic E extraverts compared with introverts scored higher in the alternative measure of agentic E; ZKPQ Activity, F(1, 36) ⫽ 36.82, p ⬍ .0001, ␩2 ⫽ .48; as well as in ZKPQ Sociability, F(1, 36) ⫽ 16.25, p ⬍ .001, ␩2 ⫽ .30, and in broader E as assessed by EPQ-R E, F(1, 36) ⫽ 39.01, p ⬍ .0001, ␩2 ⫽ .50. However, extraverts compared with introverts also scored somewhat lower on EPQ-R Neuroticism, F(1, 36) ⫽ 8.80, p ⬍ .01, ␩2 ⫽ .19, and ZKPQ NeuroticismAnxiety, F(1, 36) ⫽ 8.35, p ⬍ .01, ␩2 ⫽ .19, possibly as a result of the moderate negative correlation between E and neuroticism factors frequently described in the literature (e.g., John & Srivastava, 1999). Finally, extraverts compared with introverts unexpectedly also scored lower in general fluid intelligence, F(1, 36) ⫽ 8.63, p ⬍ .01, ␩2 ⫽ .18. No further main effects of E were observed.

N-Back Task Performance The main effect of Task was significant for all performance measures: target RT, F(4, 36) ⫽ 27.18, p ⬍ .0001; RT variability, F(4, 36) ⫽ 47.56, p ⬍ .0001; and percent correct target reactions, F(4, 36) ⫽ 111.27, p ⬍ .0001. As expected, both target RT and RT variability increased with increasing task difficulty, whereas the percentage of correct target reactions decreased (see Ms and SDs in Table 1). For each of the three performance measures, a significant linear effect of task block was observed: target RT, F(1, 36) ⫽ 74.08, p ⬍ .0001; RT variability, F(1, 36) ⫽ 54.33, p ⬍ .0001; and percent correct target reactions, F(1, 36) ⫽ 6.01, p ⬍ .05. The more task blocks participants had already completed, the shorter were the observed RTs in the current task block and the higher were the observed percentages of correct target reactions. These differences between task blocks are probably due to practice effects that are not relevant to our hypotheses. For the percentage of correct target reactions, no main effects or interactions besides an unpredicted interaction of Substance and Task were significant, F(4, 36) ⫽ 2.93, p ⬍ .05; all remaining p ⱖ .09.3 However, for both RT measures, a significant two-way interaction, E ⫻ Substance was observed: target RT, F(1, 36) ⫽ 4.26, p ⬍ .05; RT variability, F(1, 36) ⫽ 8.85, p ⬍ .01; and for RT variability, but not for target RT, the predicted three-way interaction E ⫻ Substance ⫻ Task was significant: target RT, F(4, 36) ⫽ 2.08, p ⫽ .10; RT variability, F(4, 36) ⫽ 3.27, p ⬍ .05. Besides a two-way interaction, E ⫻ Task for RT variability, F(4, 36) ⫽ 2.39, p ⫽ .07, no further main effects or interactions approached significance (all p ⱖ .28).

Tests of a Priori Contrasts For none of the performance measures, a significant difference between the 2-back tasks with versus without possible reward was

observed in any of the four experimental groups (all p ⱖ .07, twotailed). The tests of the central a priori contrasts are documented in Table 2. The corresponding means are shown in Figure 1. Based on the findings by Lieberman and Rosenthal (2001), we predicted performance differences between introverts and extraverts under placebo specifically in the more difficult n-back tasks and a reversal of these differences under sulpiride, as suggested by prior reports of reversals of preexisting performance differences after application of dopaminergic agents (e.g., Corr & Kumari, 1997; Kimberg et al., 1997). Indeed, the pattern of results observed for the target RTs in the placebo group closely replicates the findings reported by Lieberman and Rosenthal (2001): Introverts had longer target RTs than extraverts in all tasks, and this difference was statistically reliable only for the 3-back task. Also, the linear increase in target RTs from 0- to 3-back tasks under placebo was more pronounced in introverts than in extraverts. In addition, the effects of sulpiride on target RTs differed significantly between introverts and extraverts specifically in the 2and 3-back tasks (significant tetrad differences in the sulpirideplacebo column of Table 2): Sulpiride tended to slow target RTs in extraverts and to speed up target RTs in introverts.4 Put differently, under sulpiride, the pattern of results observed under placebo tended to be reversed as predicted; compared with introverts, extraverts had numerically slower target RTs in the 2- and 3-back tasks and significantly larger linear increases in RTs from 0- to 3-back tasks. It is interesting that both the E effects observed under placebo and the modulation of the sulpiride effects by E could not only be documented for target RTs but also (and even more strongly) for RT variability (see Table 2 and Figure 1). In addition, under placebo, introverts compared with extraverts did not only react more slowly but also more accurately in target trials of the two difficult n-back tasks, indicating a significant speed–accuracy trade-off, correlation between target RT and percentage of correct target reactions for the 3-back task, r(40) ⫽ .30, p ⬍ .05, onetailed.5 Also, the linear decrease in accuracy from 0- to 3-back was significantly less pronounced for introverts than for extraverts. However, apart from a weak tendency in the expected direction observed for the rewarded 2-back task, the predicted reversal of the E effects under sulpiride could not be shown for the percentage of correct target reactions (see Table 2).

3

This effect resulted from the fact that compared with placebo the percentage of correct target responses was significantly higher under sulpiride only in the 2-back task without reward, t(36) ⫽ 3.19, p ⫽ .003, but not in the other tasks, p ⬎ .40. Because all other observed sulpiride effects on performance measures (see Table 2) were similar in the 2-back task with and without reward, this unpredicted observation is difficult to interpret. 4 For the 3-back task this tetrad difference (i.e., difference of the effects of sulpiride-placebo in introverts vs. extraverts) was also significant when an analogous ANCOVA model was calculated for mean RT of all correct answers, t(36) ⫽ ⫺2.18, p ⬍ .05, one-tailed, or for mean RT of incorrect answers, t(36) ⫽ ⫺2.30, p ⬍ .05, one-tailed. 5 The correlations between target RT and percentage of correct target reactions for the 3-back task within the two substance groups were of similar magnitude, although not significant due to the smaller sample size: placebo, r(20) ⫽ .27, p ⫽ .13, one-tailed; sulpiride, r(20) ⫽ .33, p ⫽ .08, one-tailed.

EXTRAVERSION AND DOPAMINE

179

Table 2 Differences in N-Back Performance Between Introverts and Extraverts Within the Two Substance Groups and in Substance Effects (Sulp-Plac): t-Values of Contrasts Target RT Extraversion contrast (I-E) 0-back 1-back 2-back 2-back (no reward) 3-back Linear trend 0- to 3-back

RT Variability (SD)

% correct target reactions

Plac

Sulp

Sulp-Plac

Plac

Sulp

Sulp-Plac

Plac

Sulp

Sulp-Plac

1.86 1.15 1.38 1.40 2.54†† 1.96†

0.60 ⫺0.00 ⫺1.10 ⫺1.54 ⫺1.26 ⫺1.78†

⫺0.89 ⫺0.81 ⫺1.76† ⫺2.08† ⫺2.68†† ⫺2.64††

1.78 2.07* 1.96† 2.69†† 3.41†† 2.60††

0.29 ⫺0.32 ⫺1.83† ⫺2.08† ⫺1.33 ⫺2.06†

⫺1.06 ⫺1.69 ⫺2.68†† ⫺3.38†† ⫺3.36†† ⫺3.29††

1.66 0.27 2.09† 0.93 1.69† 1.73†

0.47 ⫺0.61 ⫺0.59 0.45 0.63 0.38

⫺0.84 ⫺0.62 ⫺1.90† ⫺0.34 ⫺0.75 ⫺0.96

Note. RT parameters were square root transformed before calculation of statistical analyses. df ⫽ 36. Plac ⫽ placebo; Sulp ⫽ sulpiride. Contrasts for the 0-back task and for the 1-back task were tested two-tailed; the remaining contrasts were tested one-tailed. * p ⬍ .05, two-tailed. † p ⬍ .05, one-tailed. †† p ⬍ .01, one-tailed.

EEG Indices: Posttask Waiting Periods Frontal EEG Asymmetry Three-way ANCOVAs with E and Substance as between-groups factors, Task as within-group factor, and Block Number as covariate did not reveal any significant main effects or interactions for frontal EEG asymmetry in the theta or low alpha bands, for all effects F(1, 36) ⱕ 1.65, p ⱖ .21 or F(4, 36) ⱕ 2.25, p ⱖ .08.

Frontal Versus Parietal EEG Activity Three-way ANCOVAs computed for frontal versus parietal EEG activity in the theta and low alpha bands revealed the following statistically reliable effects: a main effect for Task in the theta band, F(4, 36) ⫽ 3.21, p ⬍ .05, resulting from a linear decrease from 0- to 3-back waiting period, t(36) ⫽ ⫺3.02, p ⬍ .01, two-tailed, a two-way interaction E ⫻ Substance for the theta band, F(1, 36) ⫽ 19.77, p ⬍ .0001, and, less pronounced, for the low alpha band, F(1, 36) ⫽ 4.40, p ⬍ .05, and finally a three-way interaction, E ⫻ Substance ⫻ Task, for both the theta band, F(4, 36) ⫽ 4.99, p ⬍ .01, and the low alpha band, F(4, 36) ⫽ 4.67, p ⬍ .01.

Tests of a Priori Contrasts Neither for the theta band nor for the low alpha band a significant difference between the 2-back tasks with versus without

possible reward was observed in any of the four experimental groups (all p ⱖ .12, two-tailed). The tests of the central a priori contrasts are documented in Table 3. For the theta band, the corresponding means are shown in Figure 2. Under placebo, compared with extraverts, introverts had greater relative frontal versus parietal EEG theta activity during the waiting periods after all tasks except the 0-back task. Under sulpiride, an opposite difference between introverts and extraverts was observed after all tasks except the 0-back task. Furthermore, under placebo, extraverts, but not introverts, showed a linear decrease in frontal versus parietal theta activity from 0- to 3-back task, t(36) ⫽ ⫺2.02, p ⫽ .05, two-tailed, whereas under sulpiride only, introverts, but not extraverts, showed this decrease, t(36) ⫽ ⫺2.62, p ⫽ .01, two-tailed. Thus, sulpiride did not only completely reverse the overall differences in frontal versus parietal theta activity observed under placebo, it also reversed the differences in linear decreases from 0- to 3-back: Under placebo, extraverts showed a (numerically) larger decrease than introverts, whereas under sulpiride, introverts showed a significantly larger decrease than extraverts. As shown in Table 3, the pattern of results observed for relative frontal versus parietal EEG activity in the low alpha band was similar to the one just described for relative frontal versus parietal EEG activity in the theta band. However, the effects were generally less pronounced with the differences between introverts and extraverts within substance groups only significant for the 1-back task.

Figure 1. Target RT and RT variability in the four rewarded n-back tasks (M and SEM values of untransformed variables).

WACKER, CHAVANON, AND STEMMLER

180

Table 3 Differences in Frontal vs. Parietal EEG Activity (Post Task Waiting Periods) Between Introverts and Extraverts Within the Two Substance Groups and in Substance Effects (Sulp-Plac): t-Values of Contrasts Frontal vs. Parietal Theta Extraversion Contrast (I-E) 0-back 1-back 2-back 2-back (no reward) 3-back Linear trend 0- to 3-back

Frontal vs. Parietal Low Alpha

Plac

Sulp

Sulp-Plac

Plac

Sulp

Sulp-Plac

1.55 3.17** 3.12** 2.49* 2.42* 1.29

⫺1.11 ⫺4.08** ⫺2.47* ⫺2.44* ⫺3.67** ⫺3.51**

⫺1.89 ⫺5.12** ⫺3.95** ⫺3.49** ⫺4.29** ⫺3.35**

0.16 2.02† 0.27 0.73 1.21 0.64

⫺1.01 ⫺2.91** ⫺1.15 ⫺1.26 ⫺1.72 ⫺0.46

⫺0.82 ⫺3.48** ⫺1.01 ⫺1.40 ⫺2.07* ⫺0.78

Note. df ⫽ 36. Plac ⫽ placebo; Sulp ⫽ sulpiride. Contrasts for low alpha in the placebo group were tested two-tailed except for the linear trend contrast; the latter as well as the remaining contrasts were tested one-tailed. * p ⬍ .05, two-tailed. ** p ⬍ .01, two-tailed. † p ⬍ .05, one-tailed.

EEG Indices: Initial Rest Period As explained in the Method section, our conceptualization of traits as dispositions that are activated only in relevant situational context led us to expect strong correlations between agentic E and the EEG, particularly when the recording situation is characterized by the presence of positive incentives. In the present study, positive incentives were arguably most salient in the waiting periods after each of the four potentially rewarded n-back tasks, during which participants were waiting to be rewarded for their performance. As documented above, we found some strong relationships between agentic E and EEG parameters obtained for these prereward recordings. However, the pattern of results, although somewhat weaker for the 0-back task, was quite similar across the four

task versions and, most notably, did not differ between the rewarded and the nonrewarded version of the 2-back task. In addition, prior research on the relationship between personality traits and asymmetrical and symmetrical frontal EEG activity focused almost exclusively on recordings made under resting conditions without any explicit reward (or punishment) signals (Coan & Allen, 2004a; Hewig et al., 2004, 2006). Therefore, we also checked whether the effects reported for the posttask waiting periods can be documented for the initial rest period.

Frontal EEG Asymmetry Paralleling the observations for the posttask waiting periods, two-way ANOVAs with the factors E and Substance did not reveal significant main effects or interactions for frontal EEG asymmetry in the theta and low alpha bands, for all effects F(1, 36) ⱕ 1.42, p ⱖ .24, ␩2 ⱕ .038. Also, a priori contrasts did not reveal any significant differences between introverts and extraverts within either the placebo group or the sulpiride group (all p ⱖ .38, two-tailed).

Frontal Versus Parietal EEG Activity

Figure 2. Frontal versus parietal EEG theta activity during the waiting periods after the four rewarded n-back tasks (M and SEM values).

Two-way ANOVAs with the factors E and Substance did not show significant main effects for frontal versus parietal EEG activity in the theta or low alpha bands, for all main effects F(1, 36) ⱕ 1.88, p ⱖ .18, ␩2 ⱕ .045. However, paralleling the observations obtained for the posttask waiting periods, the two-way interaction E ⫻ Substance was highly significant for the theta band, F(1, 36) ⫽ 18.62, p ⬍ .0001, ␩2 ⫽ .332, and just failed to reach formal significance for the low alpha band, F(1, 36) ⫽ 3.21, p ⫽ .08, ␩2 ⫽ .076. A priori contrasts revealed that introverts had greater frontal versus parietal EEG theta activity than extraverts under placebo, t(36) ⫽ 2.17, p ⬍ .05, two-tailed, whereas the opposite was true under sulpiride, t(36) ⫽ ⫺3.39, p ⬍ .01, twotailed (M ⫽ 0.40, 0.66, 0.77, and 0.44, SD ⫽ 0.18, 0.22, 0.22, and 0.23, for Extraverts-Placebo, Introverts-Placebo, ExtravertsSulpiride, and Introverts-Sulpiride, respectively). The difference between sulpiride and placebo was significant for both introverts and extraverts, t(36) ⫽ ⫺2.25, p ⬍ .05, two-tailed, and t(36) ⫽ 3.86, p ⬍ .001, two-tailed, respectively.

EXTRAVERSION AND DOPAMINE

Relationship Between Resting EEG and 3-Back Performance As shown above, the pattern of results observed for frontal versus parietal EEG activity in the theta band is not only highly consistent across the different recording periods (initial rest period and posttask waiting periods) but also highly similar to the pattern observed for RT measures in the 3-back task. Thus, the question of whether 3-back RT measures can be predicted from resting frontal versus parietal EEG activity in the theta band arises. Correlation analyses revealed that 3-back target RT and RT variability were not only highly interrelated, r(40) ⫽ .88, p ⬍ .0001, one-tailed, but also could be predicted from frontal versus parietal EEG theta activity measured during the initial rest period, target RT, r(40) ⫽ .26, p ⫽ .05, one-tailed; RT variability, r(40) ⫽ .29, p ⬍ .05, one-tailed.

Are the Observed Effects Specific to Agentic E? To check whether the effects of sulpiride on target RT, RT variability, and frontal versus parietal EEG theta activity during the initial rest period were indeed only modulated by agentic E and not by other (correlated) personality traits, age, or general fluid intelligence, we calculated a series of ANCOVAs for each of the three dependent variables using the statistical models described above, but now entering an additional covariate (either an EPQ-R scale, a ZKPQ scale, age, or intelligence scores), its two-way interaction with Substance, and, for the RT data, its two-way interaction with Task and its three-way interaction with Substance and Task (see Table 4 for the intercorrelations of the questionnaire scales, age, and intelligence scores). If the effects reported above are indeed specific to agentic E, they should disappear when the alternative measure of agentic E (ZKPQ Activity) is used as a covariate but remain significant when either one of the other personality scales, age, or general fluid intelligence is used as a covariate. In brief, this is exactly what we found. After controlling for ZKPQ Activity, the E ⫻ Substance interaction was no longer statistically reliable for frontal versus parietal EEG theta activity, F(1, 34) ⫽ 3.18, p ⬎ .05, Cohen’s d ⫽ 0.61. However, the E ⫻ Substance interaction was still highly significant when either one

181

of the other personality scales, age or general fluid intelligence, was controlled for, F(1, 34) ⱖ 12.93, p ⱕ .001, Cohen’s d ⱖ 1.23. Likewise, substance no longer modulated the E effects (cf. sulpiride-placebo column in Table 2) for 3-back target RT, t(34) ⫽ ⫺0.55, p ⫽ .29, one-tailed, and 3-back RT variability, t(34) ⫽ ⫺1.38, p ⫽ .08, one-tailed, after controlling for ZKPQ Activity (Cohen’s d ⱕ 0.47). However, these contrasts remained significant for 3-back target RT, t(34) ⱕ ⫺2.15, p ⱕ .02, one-tailed, and 3-back RT variability, t(34) ⱕ ⫺2.72, p ⱕ .005, when either one of the other personality scales, age or general fluid intelligence, was controlled for (Cohen’s d ⱖ 0.74). It is notable that controlling for general fluid intelligence even strengthened the E ⫻ Substance interaction (i.e., the tetrad difference contrast) for both 3-back target RT, t(34) ⫽ ⫺3.63, p ⫽ .0013, and 3-back RT variability, t(34) ⫽ ⫺4.05, p ⫽ .0003, Cohen’s d ⱖ 1.25. In summary, these results show that (a) the effects are indeed specific for agentic E (⫽discriminant validity) and (b) not specific to the measure of agentic E used to select groups of introverts and extraverts but also observable with an alternative questionnaire scale (⫽convergent validity; see Figure 3 for a scatterplot of the simple correlations between ZKPQ Activity and frontal vs. parietal EEG theta activity within the two substance groups).

Discussion With the present study, we aimed at identifying an easily obtainable EEG index and a behavioral paradigm that can be used to investigate Depue and Collins’ (1999) suggestion of a dopaminergic basis for the personality trait agentic E in humans. Based on recent findings, we argued that performance in the difficult versions of the n-back task and both asymmetrical frontal EEG activity in the low alpha band and symmetrical frontal versus parietal EEG activity in both theta and (low) alpha bands can be hypothesized to be related not only to agentic E but also to certain aspects of central DA functioning. To put these hypotheses to an initial test, we assessed performance in the n-back task and the EEG variables of interest in groups of healthy young men high versus low in agentic E after administration of either placebo or the selective DA D2 antagonist sulpiride. We predicted that under

Table 4 Intercorrelations of Personality Scales, General Fluid Intelligence, and Age in the Whole Sample

ZKPQ Sy ZKPQ N-Anx ZKPQ ImpSS ZKPQ Agg-H EPQ-R E EPQ-R N EPQ-R P EPQ-R L CFT 3 (gf) Age

ZKPQ Act

ZKPQ Sy

ZKPQ N-Anx

ZKPQ ImpSS

ZKPQ Agg-H

EPQ-R E

EPQ-R N

EPQ-R P

EPQ-R L

CFT 3 (gf)

.60** ⫺.23 .21 .22 .60** ⫺.23 ⫺.26 .08 ⫺.37* ⫺.26

⫺.31 .38* .34* .81** ⫺.24 ⫺.10 ⫺.19 ⫺.29 ⫺.31

⫺.23 .21 ⫺.53** .77** .00 .02 .04 .11

.11 .58** ⫺.11 .30 ⫺.19 ⫺.06 ⫺.22

.22 .13 .30 ⫺.17 ⫺.26 .02

⫺.31 ⫺.10 ⫺.05 ⫺.44** ⫺.16

.06 ⫺.07 .20 .09

⫺.35* .20 ⫺.09

⫺.32* .28

⫺.08

Note. N ⫽ 40. Act ⫽ activity; Agg-H ⫽ aggression-hostility; CFT 3 (gf) ⫽ general fluid intelligence as measured by the short version of the Culture Fair Test (Scale 3); E ⫽ extraversion; ImpSS ⫽ impulsive sensation seeking; L ⫽ lie scale; N ⫽ neuroticism; N-Anx ⫽ neuroticism-anxiety; P ⫽ psychoticism; Sy ⫽ sociability. * p ⬍ .05, two-tailed. ** p ⬍ .01, two-tailed.

182

WACKER, CHAVANON, AND STEMMLER

Figure 3. Scatterplot of the correlation between agentic E (ZKPQ Activity) and frontal versus parietal EEG theta activity during the initial rest period within sulpiride (closed triangles; r ⫽ .64, p ⬍ .01, two-tailed) and placebo (open circles; r ⫽ ⫺.50, p ⬍ .05, two-tailed). The correlations differed significantly between sulpiride and placebo, z ⫽ 3.81, p ⬍ .0002, two-tailed. Significant correlations within substance groups were not observed for age, general fluid intelligence, the EPQ-R scales, or the remaining ZKPQ scales.

placebo, participants high versus low in agentic E would show (a) shorter target RTs and lower RT variability in the difficult n-back tasks, (b) greater relative left frontal cortical activity (i.e., greater relative right frontal EEG activity in the low alpha band; see Oakes et al., 2004), and (c) lower EEG activity in the low alpha band and either lower or higher EEG theta activity at frontal versus parietal electrode locations. Furthermore, we predicted that all of these differences between participants high versus low in agentic E would be reversed under sulpiride, resulting in significant two-way interactions of the two factors E (high vs. low) and Substance (placebo vs. sulpiride) for each of the dependent variables. These predictions were strongly supported by the present data for n-back RT measures and for relative frontal versus parietal resting EEG theta activity but not for frontal EEG asymmetry. In addition, supplementary analyses showed that the observed effects were indeed specific to agentic E and not due to different personality traits (e.g., Sociability, Neuroticism, or Impulsive Sensation Seeking), age, or general fluid intelligence. We will now discuss each of these findings in more detail before sketching some implications for further empirical and theoretical work on the psychobiological basis of agentic E.

E, DA, and N-Back Performance The differences in n-back performance observed between participants high versus low in agentic E in the placebo group closely replicate the observations of Lieberman and Rosenthal (2001): Exactly as reported by these authors, (agentic) introverts had longer RTs than extraverts, and this difference was maximal (and statistically significant) in the 3-back task. Also, in our study, just like in theirs, (agentic) introverts showed a significantly larger

linear increase in RTs from the 0- to the 3-back task. In addition, under sulpiride, the RT differences between introverts and extraverts tended to be completely reversed as predicted. Two further results of the present study indicate that the introverts’ slower responding in the difficult n-back tasks does not represent a simple performance deficit, as suggested by Lieberman and Rosenthal (2001), but may be more accurately interpreted as a different way of performing the task: First, the pattern of differences between introverts and extraverts under placebo and the partial reversal of these differences under sulpiride documented for mean target RT were also observable (even somewhat more pronouncedly) for the variability of RTs across trials. Thus, it is unclear whether the observed RT effects represent differences in the overall speed level (i.e., mean RTs), in the variability of this level across trials, or both. In future studies, an in-depth examination of differences in patterns of RT trial-to-trial variation between introverts and extraverts could provide some interesting insights concerning this issue (e.g., introverts compared with extraverts may show more pronounced post-error slowing, resulting in increased RT variability). Second, several observations indicate the presence of a significant speed–accuracy trade-off at least for the more difficult versions of the n-back task: Under placebo and relative to extraverts, introverts reacted not only slower on target trials, but they also made fewer mistakes. Furthermore, speed and accuracy in the target trials of the 3-back task were negatively correlated: Participants who reacted relatively slowly on target trials (e.g., introverts) tended to make relatively few mistakes. The fact that the predicted reversal of the E effect under sulpiride could be observed more clearly for the RT measures than for the percentage of correct target trials is difficult to interpret because it may simply be a consequence of the lower reliability of the latter parameter (see Table 1). Even though the n-back performance differences between introverts and extraverts cannot be represented on a continuum ranging from good to poor performance and, thus, seem to be more complex than suggested by Lieberman and Rosenthal (2001), the present results provide initial support for these authors’ proposal that they are related to brain DA. Because the RT differences between introverts and extraverts (in particular, the differences in the linear RT increases from easy to difficult tasks) were almost completely reversed by application of the selective D2 antagonist sulpiride, it seems prudent to conclude that the observed individual differences in behavior are at least partly based on individual differences in functional brain DA activity at D2 receptor sites. These conclusions certainly are not only limited by several caveats that we will discuss below, but they also do not speak to the important question of which presumably dopaminergically modulated cognitive processes underlie the observed differences in behavior. This question cannot be answered with the present data because it is well known that the difficult n-back tasks involve a rather large variety of cognitive processes (e.g., Gray et al., 2003), and it is currently unclear which of these processes are tapped by the available performance measures. Because there is at least some consensus in the literature that the difficult n-back tasks tap certain aspects of cognitive control, our findings conform with the conclusion recently drawn by Gray and Burgess (2004, p. 36) that “some component of the cognitive control network depends on BAS [or agentic E] in a true functional sense, albeit the relation is complex.” Recent models of DA functioning may serve as a starting point to tackle this complexity in future studies: Seamans and Yang (2004) suggested that DA acts to contract or expand the breadth of

EXTRAVERSION AND DOPAMINE

information held in WM buffers in PFC networks. In a similar manner, Dreisbach and Goschke (2004) proposed that DA adjusts the balance of goal maintenance versus flexible switching between multiple goals. It is possible that under normal conditions, extraverts compared with introverts tend more toward flexible switching and to holding a greater breadth of information in WM (see Avila, Barros, Ortet, Parcet, & Ibanez, 2003). Of course, this interpretation fits nicely with Lieberman and Rosenthal’s (2001) suggestion that the E-related differences in n-back RTs result from superior multitasking ability in extraverts. However, the notion of a trade-off between flexible switching and stable maintenance of goals also implies that life may not be quite as unfair for introverts as surmised by these authors: The extraverts’ greater multitasking or switching ability may come at the price of increased distractibility.

E, DA, and Frontal EEG Activity Asymmetrical frontal EEG activity. In the present study, under placebo, agentic E was not significantly related to resting frontal EEG asymmetry in either the theta or low alpha bands. With regard to the low alpha band, these observations contrast with several studies (e.g., Coan & Allen, 2003; Harmon-Jones & Allen, 1997; Schmidt, 1999; Sutton & Davidson, 1997), in which E and related traits have been linked to relative left frontal resting cortical activity (i.e., relative right frontal EEG activity in the alpha band). However, Hagemann, Naumann, Thayer, and Bartussek (2002) have shown that this measure is strongly influenced by unknown occasion-specific state factors, which account for almost as much variance as a temporally stable latent trait factor. Thus, it is possible that in the present study, these unknown occasion-specific influences on frontal EEG asymmetry completely masked an existing association of the latent trait component with agentic E. An alternative explanation for the unexpected absence of correlations between E and frontal asymmetry is suggested by our conceptualization of E and other personality traits as dispositions that predict behavior and the associated brain states only in relevant situational contexts (for details, see Stemmler, 1997). In the present study, participants knew that they were about to engage in a challenging task, and the goodies that could be obtained for good performance were placed directly in front of them during the whole experiment, including the initial rest period. However, because we observed virtually no differences between the potentially rewarded and the nonrewarded 2-back task, we simply do not know whether these efforts to create an experimental setting that is strongly characterized by positive incentives and therefore suitable to activate agentic E were actually successful. It is possible that different features of the EEG recording situation were even more salient for the participants. For instance, recent findings suggest that for male participants female experimenters represent a situational context that activates an affiliative form of approach motivation that is not mediated by brain DA but nonetheless strongly related to frontal EEG asymmetry (Kline, Blackhart, & Joiner, 2002; Pauls et al., 2005). In the present study, both men and women served as experimenters. Thus, besides the possibility that in the predicted correlations between agentic E and EEG asymmetry were masked by a sizable contribution of irrelevant occasion-specific variance to asymmetry scores, it could also be that we did not succeed in providing a sufficiently “pure” positive incentive context to specifically produce variance in EEG asym-

183

metry relevant to agentic E. Direct comparisons between these two alternatives are needed to reach reliable conclusions concerning the reasons for the inconsistent trait-asymmetry correlations observed here and elsewhere (for review, see Hagemann et al., 2002). Likewise, it would certainly be premature to reject the idea of a partly dopaminergic basis of frontal EEG asymmetry based on the complete absence of reliable effects of the selective DA D2 receptor antagonist sulpiride observed in this study. As detailed in the Introduction, there are well-founded reasons to expect an association between resting frontal brain asymmetry and DA, and initial indirect evidence that such an association indeed exists comes from studies in which significant associations between resting frontal EEG asymmetry in the alpha band and spontaneous eye blink rates (i.e., a peripheral indicator of DA activity, see Karson, 1983) were observed (e.g., Pauls et al., 2005). It is interesting that very recent work with monkeys and humans suggests that spontaneous eye blink rate may depend more strongly on DA D1 than DA D2 receptor signaling (Jutkiewicz & Bergman, 2004; van der Post, de Waal, de Kam, Cohen, & van Gerven, 2004).6 If this relative D1-specificity of dopaminergic effects also holds for frontal EEG asymmetry, this would offer a straightforward explanation for the absence of reliable D2 antagonist effects in the present study. Research with different dopaminergic agents that also tap the D1 receptor system is needed to directly test this hypothesis. Symmetrical frontal versus parietal EEG activity. As predicted, we observed a significant difference in EEG theta activity at frontal relative to parietal electrode locations between participants low versus high in agentic E under placebo: High agentic E was associated with stronger parietal (vs. frontal) theta activity. This finding is consistent with earlier work by Stenberg (1992), who reported an association of a BAS-E factor with posterior theta activity and an association of a behavioral inhibition system (BIS)Anxiety factor with frontal theta activity. The finding that E is related to posterior versus frontal brain activity is also echoed by observations from a PET study of cerebral blood flow correlates of E. In summarizing their findings, Johnson et al. (1999) note that on the whole, introversion was associated with activity in frontal areas, whereas E was associated with activity in more posterior areas. In addition, as already noted above, a number of recent functional magnetic resonance imaging studies reported significant associations between E-related traits and activity in the ACC (Canli et al., 2001, 2004; Gray & Braver, 2002; Kumari et al., 2004). Given the evidence that frontal midline EEG theta activity reflects activity in the ACC (Pizzagalli et al., 2003), it seems prudent to assume that the effects observed in the present study for frontal versus parietal theta activity partly result from E-related activity differences in the ACC. If this hypothesis holds up in future studies that combine EEG source localization algorithms and brain imaging technology or at least employ somewhat more densely spaced EEG derivations that allow to tease apart the individual contributions of frontal midline and parietal 6 Also, note that Depue et al. (1994) did not observe a statistically reliable main effect of the D2 agonist bromocriptine on (maximum) eye blink rate, nor a modulation of such an effect by E. Only the time to blink maximum after administration of bromocriptine was found to correlate with E. However, in contrast to the link between E and the bromocriptineinduced prolactin response, Depue and colleagues did not report a replication of this correlation in a larger sample (Depue, 1995, 1996).

184

WACKER, CHAVANON, AND STEMMLER

theta by providing a means to control for volume conduction and individual differences in skull thickness (see Introduction section), then the current knowledge on the functions of the ACC could be used as a basis to develop hypotheses concerning the functional significance of frontal versus parietal theta activity; currently, the ACC is hypothesized to function as a discrepancy detector that signals the organism when things are going worse than expected (e.g., Holroyd & Coles, 2002) and that serves to recruit lateral prefrontal cortices to exert top-down control to resolve the detected discrepancies and/or bias responding in a goal-congruent manner (e.g., MacDonald, Cohen, Stenger, & Carter, 2000). Thus, greater frontal versus parietal theta may indicate the initiation of conflict- or discrepancy-triggered goal shielding or, in terms of Seamans and Yang’s (2004) model of DA functioning in the PFC, a contraction (vs. expansion) of the breadth of information held in WM. Although the present data thus do not allow definite conclusions concerning both the underlying neural generators and the functional significance of frontal versus parietal theta activity, they do provide some interesting insights with respect to the implicated neurochemistry. For this EEG measure, we observed a strong interaction of the two factors E (high vs. low) and Substance (placebo vs. sulpiride); under sulpiride, E effects were completely reversed, with high agentic E associated with stronger frontal versus parietal theta activity. The selective DA D2 receptor antagonist sulpiride resulted in a shift toward more frontal versus parietal EEG theta activity in participants high in agentic E and in a shift toward more parietal versus frontal EEG theta activity in participants low in agentic E leading to no discernible effects for the group as a whole. These observations suggest that frontal versus parietal EEG theta activity contains a considerable amount of trait variance associated both with agentic E and DA functioning. Therefore, the present study provides initial evidence that frontal versus parietal EEG theta activity qualifies as an easily obtainable EEG index that can be used as an efficient tool to further probe the DA basis of E. For frontal versus parietal EEG activity in the low alpha band, we observed only weak and (with the exception of the waiting period after the 1-back task) statistically unreliable effects that mirrored the strong effects observed for the theta band. Why were we not able to demonstrate a more consistent relationship between frontal versus parietal EEG activity and E in the placebo group as the results reported by Hewig et al. (2004, 2006) would seem to suggest? A number of possible explanations are offered by several significant methodological differences between the Hewig et al. studies and the present work (e.g., sample size, number and length of EEG recordings, parametrization, and EEG reference scheme). Anyway, the similarity of the patterns we observed in the low alpha and theta bands points to the intriguing possibility that both the Hewig et al. findings and the present results are based on the same electrocortical phenomenon: Boundaries between traditional EEG frequency bands are somewhat arbitrary. Maybe the presumably DA-related aspects of oscillatory brain activity that we observed most clearly in the theta range can sometimes (or for some individuals) also be detected in the alpha band.

(Agentic) E and the BIS? In contrast to the theoretical perspective adopted here (see Depue & Collins, 1999), other personality theorists (e.g., Gray,

1970; Patterson & Newman, 1993) maintain that E is not only related to increased sensitivity or activity of a BFS-like system that facilitates behavior in response to incentives (i.e., in Gray’s terminology, the BAS) but also to reduced sensitivity or activity of a second broad motivational system, the BIS, that serves to inhibit behavior in response to events that are potentially incompatible with the organism’s current goals (e.g., signals for punishment and nonreward). Thus, for these authors, E represents the habitual balance between behavioral activation/facilitation on the one hand and behavioral inhibition on the other. It is interesting that several of the present findings may also be viewed from this theoretical perspective. According to Gray and McNaughton (2000), the core function of the BIS lies in the detection of conflict among concurrently activated goals. As explained above, the detection of such goal conflicts is thought to be associated with activity in the ACC and frontal midline EEG theta. Also, some authors (e.g., Stenberg, 1992) have reported a positive correlation between traits from the BIS spectrum and frontal EEG theta. Thus, the greater frontal versus parietal theta we observed in introverts compared with extraverts under placebo may indicate stronger BIS sensitivity in addition to (or even rather than) weaker BAS/BFS sensitivity in the former. In a similar manner, the introverts’ tendency to perform more slowly but somewhat more accurately than extraverts in the difficult n-back tasks under placebo may indicate not only lower BAS/BFS sensitivity, but also greater BIS sensitivity, resulting in a greater responsiveness to the negative feedback associated with mistakes (see also Patterson & Newman, 1993). In brief, our suggestion that both the electrocortical and the behavioral differences between introverts and extraverts reported here indicate individual differences in the balance of the contraction versus expansion of the breadth of information held in WM could potentially be mapped onto earlier suggestions that E results from individual differences in the balance between two neurobehavioral systems, the BIS and the BAS/BFS. This would of course provide a straightforward explanation for the fact that some of the associations we observed for agentic E are about as suggestive of individual differences in BIS activity as of individual differences in BAS/BFS activity. However, in the present study, we did not observe any significant effects for traits more closely aligned with habitual BIS sensitivity (i.e., EPQ-R Neuroticism and ZKPQ Neuroticism-Anxiety), and the current psychobiological personality literature provides no theoretical basis to predict an association between habitual BIS sensitivity and brain DA. Therefore, it may be more appropriate to stick to Depue and Collins’ (1999) suggestion that agentic E maps only onto a single, affectively unipolar motivational system, which is neurobiologically based on brain DA (i.e., the BFS) and to conceptualize the involvement of the BIS as somewhat more indirect. For example, one could simply add the idea that high state activity in the BFS/BAS is incompatible with high state activity in the BIS, which places partly antagonistic constraints on behavior and the associated brain states.

Underlying Mechanisms: The Return of Eysenck’s (1967) Inverted U? The present findings do not speak to the important question of why the D2 antagonist sulpiride had opposite psychophysiological and behavioral effects in individuals high versus low in agentic E. Opposing effects of dopaminergic drugs in different groups of

EXTRAVERSION AND DOPAMINE

185

participants have been described frequently before (e.g., Corr & Kumari, 1997; Kimberg et al., 1997; Mattay et al., 2000, 2003; Takeshita & Ogura, 1994). It is typical that such effects are explained by assuming that the different groups occupy separate locations on a continuum of baseline dopaminergic functioning and that the relationship between the dependent variable, in which effects of the dopaminergic drug were observed, and baseline DA follows an inverted U-shaped function (e.g., Depue & Collins, 1999; Mattay et al., 2003). Invoking such an inverted U-shaped function to explain differences between introverts and extraverts will surely remind many readers of Eysenck’s (1967) famous arousal theory of E in which he postulated that by virtue of hardwired individual differences in the brain’s ascending reticular activating system, introverts are typically more aroused/arousable than extraverts except for particularly arousing conditions in which the protective mechanism of transmarginal inhibition leads to paradoxical arousal decreases in introverts, resulting in a reversal of the typical arousal differences. Indeed, given that the ascending DA fibers are an important (although not exclusive) component of the ascending reticular activating system (Robbins, 1997), the present findings could be interpreted as evidence that, on the whole, Eysenck (1967) was right all along (see also Corr & Kumari, 1997). However, what exactly are the hypothesized differences in baseline DA functioning between extraverts and introverts, and where do they come from? Novel and more specific hypotheses concerning this issue may eventually come from molecular genetic studies that have recently begun to identify functionally relevant gene polymorphisms that are linked to both E and functional differences in DA neurotransmission (see, e.g., Mattay et al., 2003; Reuter & Hennig, 2005). Combining such molecular genetic assessments with the methods employed in the present study could thus be a promising approach for further research aimed at going beyond the inverted U to specify the precise mechanisms underlying the differences between introverts and extraverts reported here.

specific to DA or rather attributable to unknown unspecific side effects of sulpiride. In addition, in this initial study, we only employed one single dosage of sulpiride and made no attempt to investigate dose-response relationships. Future studies will have to use a greater variety of pharmacological agents and dosages to provide converging evidence for the interpretation in terms of a DA mechanism suggested here. Also, it should be noted that because at present no pharmacological agents exist that exclusively influence the mesocorticolimbic DA system rather than the other DA systems of the brain (nigrostriatal, tuberoinfundibular, and incertohypothalamic), the strong Depue and Collins (1999) hypothesis of a specific role of the mesocorticolimbic DA system cannot be validated with the methods employed here. Furthermore, because we only investigated preselected groups of male participants, it remains an open question whether the present findings generalize to randomly drawn female samples. Finally, because we conceptualize traits as dispositions that are only operative in relevant situational contexts (see Stemmler, 1997), the present study was conducted in an incentive motivational setting specifically designed to activate agentic E. However, because we made only a relatively weak and absolutely unsuccessful attempt to manipulate the situational context (i.e., announcing for one of the more challenging tasks that no reward and no performance feedback would be given afterward), we cannot tell whether our observations are indeed specific to trait-relevant positive incentive situations. If such a situational specificity of the observed associations could be demonstrated in future studies, this would not only support the general concept of traits as dispositions (Stemmler, 1997), but also Depue and Collins’ (1999) claim that agentic E is based on positive incentive motivation, thereby forging further links between these authors’ theoretical model and the extensions proposed here.

Limitations and Future Directions

In the present study, we demonstrated that the effects of the selective DA D2 antagonist sulpiride on both RT measures in the difficult n-back tasks and frontal versus parietal EEG theta activity are strongly and specifically modulated by agentic E. These observations not only offer novel support for Depue and Collins’ (1999) DA theory of agentic E, they also provide new opportunities to further extend this theory to the human level by identifying behavioral indicators as well as an easily accessible, noninvasive EEG measure that are sensitive to both E and manipulations of functional brain DA activity and thus open up a range of new research avenues for further work on the psychobiological basis of this important dimension of human personality.

Even though we were able to demonstrate that the observed effects were indeed due to agentic E rather than several other correlated personality traits, age, or general fluid intelligence, of course the possibility still remains that some other third variable correlated with E actually underlies our findings. Such potential alternative variables range from other unmeasured personality dimensions to established correlates of E like the responsiveness to pharmacological manipulations of general arousal (e.g., with caffeine) and the sensitivity of task performance to time of day (Revelle, Humphreys, Simon, & Gilliand, 1980), just to name a few. As one of the reviewers noted, such potential third variable explanations represent a general problem for this kind of research as long as we have not found a way to experimentally vary E itself, which would probably be akin to changing the engrained hardwiring of the brain. For the time being, all we can do is include multiple methods of assessment for the targeted construct to foster integration of different lines of investigation of E and measure as many potentially relevant third variables as possible to rule out at least the more obvious alternative explanations.7 Because we only used one (albeit highly selective) dopaminergic drug, it remains unclear whether the observed effects are in fact

Conclusions

7 At least a third variable explanation based on the Revelle et al. (1980) findings concerning E, time of day, and caffeine effects would not seem particularly convincing because these authors observed consistent findings only for impulsivity, which is a facet of Eysenck’s early E concept quite distinct from agentic E as defined and measured in the present article. There were no associations between ZKPQ Impulsive Sensation Seeking and either one of our two alternative measures of agentic E. Also, ZKPQ Impulsive Sensation Seeking (even though strongly correlated to EPQ-R E) did not produce any significant effects in the present study.

WACKER, CHAVANON, AND STEMMLER

186 References

American Psychiatric Association. (1994). Diagnostic and statistical manual of disorders (4th ed.). Washington, DC: Author. Avila, C., Barros, A., Ortet, G., Parcet, M. A., & Ibanez, M. I. (2003). Set-shifting and sensitivity to reward: A possible dopamine mechanism for explaining disinhibitory disorders. Cognition and Emotion, 17, 951–959. Barch, D. M. (2004). Pharmacological manipulation of human working memory. Psychopharmacology (Berlin), 174, 126 –135. Blasi, G., Mattay, V. S., Bertolino, A., Elvevag, B., Callicott, J. H., Das, S., et al. (2005). Effect of catechol-O-methyltransferase val158met genotype on attentional control. Journal of Neuroscience, 25, 5038 –5045. Braver, T. S., Cohen, J. D., Nystrom, L. E., Jonides, J., Smith, E. E., & Noll, D. C. (1997). A parametric study of prefrontal cortex involvement in human working memory. Neuroimage, 5, 49 – 62. Breier, A., Kestler, L., Adler, C., Elman, I., Wiesenfeld, N., Malhotra, A., et al. (1998). Dopamine D2 receptor density and personal detachment in healthy subjects. American Journal of Psychiatry, 155, 1440 –1442. Caley, C. F., & Weber, S. S. (1995). Sulpiride: An antipsychotic with selective dopaminergic antagonist properties. Annals of Pharmacotherapy, 29, 152–160. Canli, T., Amin, Z., Haas, B., Omura, K., & Constable, R. T. (2004). A double dissociation between mood states and personality traits in the anterior cingulate. Behavioral Neuroscience, 118, 897–904. Canli, T., Zhao, Z., Desmond, J. E., Kang, E., Gross, J., & Gabrieli, J. D. (2001). An fMRI study of personality influences on brain reactivity to emotional stimuli. Behavioral Neuroscience, 115, 33– 42. Carver, C. S., & White, T. L. (1994). Behavioral inhibition, behavioral activation, and affective responses to impending reward and punishment: The BIS/BAS scales. Journal of Personality and Social Psychology, 67, 319–333. Cattell, R. B., & Weiß, R. H. (1971). Grundintelligenztest Skala 3 (CFT 3). Go¨ttingen, Germany: Hogrefe. Cloninger, C. R., Svrakic, D. M., & Przybeck, T. R. (1993). A psychobiological model of temperament and character. Archives of General Psychiatry, 50, 975–990. Coan, J. A., & Allen, J. J. B. (2003). Frontal EEG asymmetry and the behavioral activation and inhibition systems. Psychophysiology, 40, 106–114. Coan, J. A., & Allen, J. J. B. (2004a). Frontal EEG asymmetry as a moderator and mediator of emotion. Biological Psychology, 67,7– 49. Coan, J. A., & Allen, J. J. B. (2004b). The state and trait nature of frontal asymmetry in emotion. In K. Hugdahl & R. J. Davidson (Eds.), The asymmetrical brain (pp. 565– 615). Cambridge, MA: MIT Press. Corr, P. J., & Kumari, V. (1997). Sociability/impulsivity and attenuated dopaminergic arousal: Critical flicker/fusion frequency and procedural learning. Personality and Individual Differences, 22, 805– 815. Davidson, R. J. (1998). Affective style and affective disorders: Perspectives from affective neuroscience. Cognition and Emotion, 12, 307–330. Davidson, R. J. (2004). What does the prefrontal cortex “do” in affect: Perspectives on frontal EEG asymmetry research. Biological Psychology, 67, 219–233. Davidson, R. J., Jackson, D. C., & Larson, C. L. (2000). Human electroencephalography. In J. T. Cacioppo, L. G. Tassinary & G. G. Berntson (Eds.), Handbook of psychophysiology (2nd ed., pp. 27–52). New York: Cambridge University Press. Depue, R. A. (1995). Neurobiological factors in personality and depression. European Journal of Personality, 9, 413– 439. Depue, R. A. (1996). Neurobiology and the structure of personality: Implications for the personality disorders. In J. Clarkin & M. F. Lenzenweger (Eds.), Major theories of personality disorders. New York: Guilford Press. Depue, R. A., & Collins, P. F. (1999). Neurobiology of the structure of personality: Dopamine, facilitation of incentive motivation, and extraversion. Behavioral and Brain Sciences, 22, 491–569. Depue, R. A., Luciana, M., Arbisi, P., Collins, P., & Leon, A. (1994). Dopamine and the structure of personality: Relation of agonist-induced

dopamine activity to positive emotionality. Journal of Personality and Social Psychology, 67, 485– 498. Dreisbach, G., & Goschke, T. (2004). How positive affect modulates cognitive control: Reduced perseveration at the cost of increased distractibility. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30, 343–353. Eysenck, H. J. (1967). The biological basis of personality. Springfield, IL: Charles C Thomas. Farde, L., Gustavsson, J. P., & Jonsson, E. (1997). D2 dopamine receptors and personality traits. Nature, 385, 590. Gray, J. A. (1970). The psychophysiological basis of introversionextraversion. Behaviour Research and Therapy, 8, 249 –266. Gray, J. A., & McNaughton, N. (2000). The neuropsychology of anxiety (2nd ed.). New York: Oxford University Press. Gray, J. R., & Braver, T. S. (2002). Personality predicts working-memoryrelated activation in the caudal anterior cingulate cortex. Cognitive, Affective and Behavioral Neuroscience, 2, 64 –75. Gray, J. R., & Burgess, G. C. (2004). Personality differences in cognitive control? BAS, processing efficiency, and the prefrontal cortex. Journal of Research in Personality, 38, 35–36. Gray, J. R., Chabris, C. F., & Braver, T. S. (2003). Neural mechanisms of general fluid intelligence. Nature Neuroscience, 6, 316 –322. Gray, N. S., Pickering, A. D., & Gray, J. A. (1994). Psychoticism and dopamine D2 binding in the basal ganglia using single photon emission tomography. Personality and Individual Differences, 17, 431– 434. Hagemann, D., Naumann, E., Thayer, J. F., & Bartussek, D. (2002). Does resting electroencephalograph asymmetry reflect a trait? An application of latent statetrait theory. Journal of Personality and Social Psychology, 82, 619–641. Harmon-Jones, E. (2004). Contributions from research on anger and cognitive dissonance to understanding the motivational functions of asymmetrical frontal brain activity. Biological Psychology, 67, 51–76. Harmon-Jones, E., & Allen, J. J. B. (1997). Behavioral activation sensitivity and resting frontal EEG asymmetry: Covariation of putative indicators related to risk for mood disorders. Journal of Abnormal Psychology, 106, 159 –163. Hewig, J., Hagemann, D., Seifert, J., Naumann, E., & Bartussek, D. (2004). On the Selective Relation of Frontal Cortical Asymmetry and Anger-Out Versus Anger-Control. Journal of Personality and Social Psychology, 87, 926–939. Hewig, J., Hagemann, D., Seifert, J., Naumann, E., & Bartussek, D. (2006). The relation of cortical activity and BIS/BAS on the trait level. Biological Psychology, 71, 42–53. Holroyd, C. B., & Coles, M. G. (2002). The neural basis of human error processing: Reinforcement learning, dopamine, and the error-related negativity. Psychological Review, 109, 679 –709. Jackson, D. M., & Hashizume, M. (1987). Bromocriptine-induced locomotor stimulation in mice is modulated by dopamine D-1 receptors. Journal of Neural Transmission, 69, 131–145. Jasper, H. H. (1958). The ten-twenty electrode system of the International Federation. Electroencephalography and Clinical Neurophysiology, 10, 371–375. 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: theory and research (2nd ed., pp. 102–138). New York: Guilford Press. Johnson, D. L., Wiebe, J. S., Gold, S. M., Andreasen, N. C., Hichwa, R. D., Watkins, G. L., et al. (1999). Cerebral blood flow and personality: A positron emission tomography study. American Journal of Psychiatry, 156, 252–257. Jutkiewicz, E. M., & Bergman, J. (2004). Effects of dopamine D1 ligands on eye blinking in monkeys: Efficacy, antagonism, and D1/D2 interactions. Journal of Pharmacology and Experimental Therapeutics, 311, 1008 –1015. Karson, C. N. (1983). Spontaneous eye-blink rates and dopaminergic systems. Brain, 106, 643– 653. Kestler, L. P., Malhotra, A. K., Finch, C., Adler, C., & Breier, A. (2000). The relation between dopamine D2 receptor density and personality:

EXTRAVERSION AND DOPAMINE Preliminary evidence from the NEO personality inventory-revised. Neuropsychiatry, Neuropsychology, and Behavioral Neurology, 13, 48 –52. Kimberg, D. Y., D’Esposito, M., & Farah, M. J. (1997). Effects of bromocriptine on human subjects depend on working memory capacity. Neuroreport, 8, 3581–3585. Kline, J. P., Blackhart, G. C., & Joiner, T. E. (2002). Sex, lie scales, and electrode caps: An interpersonal context for defensiveness and anterior electroencephalographic asymmetry. Personality and Individual Differences, 33, 459 – 478. Kumari, V., ffytche, D. H., Williams, S. C., & Gray, J. A. (2004). Personality predicts brain responses to cognitive demands. Journal of Neuroscience, 24, 10636 –10641. Laakso, A., Vilkman, H., Kajander, J., Bergman, J., Paranta, M., Solin, O., et al. (2000). Prediction of detached personality in healthy subjects by low dopamine transporter binding. American Journal of Psychiatry, 157, 290–292. Laakso, A., Wallius, E., Kajander, J., Bergman, J., Eskola, O., Solin, O., et al. (2003). Personality traits and striatal dopamine synthesis capacity in healthy subjects. American Journal of Psychiatry, 160, 904 –910. Lieberman, M. D., & Rosenthal, R. (2001). Why introverts can’t always tell who likes them: Multitasking and nonverbal decoding. Journal of Personality and Social Psychology, 80, 294 –310. MacDonald, A. W., Cohen, J. D., Stenger, V. A., & Carter, C. S. (2000). Dissociating the role of the dorsolateral prefrontal and anterior cingulate cortex in cognitive control. Science, 288, 1835–1838. Margraf, J. (1997). Mini-DIPS (Kurz-)Interview bei psychischen Sto¨rungen. Berlin: Springer. Mattay, V. S., Callicott, J. H., Bertolino, A., Heaton, I., Frank, J. A., Coppola, R., et al. (2000). Effects of dextroamphetamine on cognitive performance and cortical activation. Neuroimage, 12, 268 –275. Mattay, V. S., Goldberg, T. E., Fera, F., Hariri, A. R., Tessitore, A., Egan, M. F., et al. (2003). Catechol O-methyltransferase val158-met genotype and individual variation in the brain response to amphetamine. Proceedings of the National Academy of Sciences, USA, 100, 6186 – 6191. Mauri, M. C., Bravin, S., Bitetto, A., Rudelli, R., & Invernizzi, G. (1996). A risk-benefit assessment of sulpiride in the treatment of schizophrenia. Drug Safety, 14, 288 –298. McClelland, G. R., Cooper, S. M., & Pilgrim, A. J. (1990). A comparison of the central nervous system effects of haloperidol, chlorpromazine and sulpiride in normal volunteers. British Journal of Clinical Pharmacology, 30, 795– 803. Meyer-Lindenberg, A., Rammsayer, T., Ulferts, J., & Gallhofer, B. (1997). The effects of sulpiride on psychomotor performance and subjective tolerance. European Neuropsychopharmacology, 7, 219 –223. Mischel, W., & Shoda, Y. (1998). Reconciling processing dynamics and personality dispositions. Annual Review of Psychology, 49, 229 –258. Missale, C., Nash, S. R., Robinson, S. W., Jaber, M., & Caron, M. G. (1998). Dopamine receptors: From structure to function. Physiological Review, 78, 189 –225. Nitschke, J. B., Heller, W., Etienne, M. A., & Miller, G. A. (2004). Prefrontal cortex activity differentiates processes affecting memory in depression. Biological Psychology, 67, 125–143. Oakes, T. R., Pizzagalli, D. A., Hendrick, A. M., Horras, K. A., Larson, C. L., Abercrombie, H. C., et al. (2004). Functional coupling of simultaneous electrical and metabolic activity in the human brain. Human Brain Mapping, 21, 257–270. Patterson, C. M., & Newman, J. P. (1993). Reflectivity and learning from aversive events: Toward a psychological mechanism for the syndromes of disinhibition. Psychological Review, 100, 716 –736. Pauls, C. A., Wacker, J., & Crost, N. W. (2005). The two components of social desirability and their relations to resting frontal brain asymmetry. Journal of Individual Differences, 26, 29 – 42. Pickering, A. D., & Gray, J. A. (1999). The neuroscience of personality. In

187

L. A. Pervin & O. P. John (Eds.), Handbook of personality: Theory and research (2nd ed., pp. 277–299). New York: Guilford Press. Pickering, A. D., & Gray, J. A. (2001). Dopamine, appetitive reinforcement, and the neuropsychology of human learning: An individual differences approach. In A. Eliasz & A. Angleitner (Eds.), Advances in research on temperament (pp. 113–149). Lengerich, Germany: Pabst. Pizzagalli, D. A., Oakes, T. R., & Davidson, R. J. (2003). Coupling of theta activity and glucose metabolism in the human rostral anterior cingulate cortex: An EEG/PET study of normal and depressed subjects. Psychophysiology, 40, 939 –949. Reuter, M., & Hennig, J. (2005). Association of the functional catecholO-methyltransferase VAL158MET polymorphism with the personality trait of extraversion. Neuroreport, 16, 1135–1138. Revelle, W., Humphreys, M. S., Simon, L., & Gilliand, K. (1980). The interactive effect of personality, time of day, and caffeine: A test of the arousal model. Journal of Experimental Psychology General, 109, 1–31. Robbins, T. W. (1997). Arousal systems and attentional processes. Biological Psychology, 45, 57–71. Ruch, W. (1999). Die revidierte Fassung des Eysenck Personality Questionnaire und die Konstruktion des deutschen EPQ-R bzw. EPQ-RK. The Eysenck Personality Questionnaire-Revised and the construction of the German standard and short versions (EPQ-R and EPQ-RK). Zeitschrift fu¨r Differentielle und Diagnostische Psychologie, 20, 1–24. Schacter, D. L. (1977). EEG theta waves and psychological phenomena: A review and analysis. Biological Psychology, 5, 47– 82. Schmidt, L. A. (1999). Frontal brain electrical activity in shyness and sociability. Psychological Science, 10, 316 –320. Seamans, J. K., & Yang, C. R. (2004). The principal features and mechanisms of dopamine modulation in the prefrontal cortex. Progress in Neurobiology, 74, 1–58. Stemmler, G. (1997). Selective activation of traits: Boundary conditions for the activation of anger. Personality and Individual Differences, 22, 213–233. Stenberg, G. (1992). Personality and the EEG: Arousal and emotional arousability. Personality and Individual Differences, 13, 1097–1113. Sutton, S. K., & Davidson, R. J. (1997). Prefrontal brain asymmetry: A biological substrate of the behavioral approach and inhibition systems. Psychological Science, 8, 204 –210. Takeshita, S., & Ogura, C. (1994). Effect of the dopamine D2 antagonist sulpiride on event-related potentials and its relation to the law of initial value. International Journal of Psychophysiology, 16, 99 –106. Tellegen, A., & Waller, N. G. (in press). Exploring personality through test construction: Development of the Multidimensional Personality Questionnaire (MPQ). Minneapolis: University of Minnesota Press. van der Post, J., de Waal, P. P., de Kam, M. L., Cohen, A. F., & van Gerven, J. M. (2004). No evidence of the usefulness of eye blinking as a marker for central dopaminergic activity. Journal of Psychopharmacology, 18, 109 –114. Wacker, J., Heldmann, M., & Stemmler, G. (2003). Separating emotion and motivational direction in fear and anger: Effects on frontal asymmetry. Emotion, 3, 167–193. Wilson, G. (1978). Introversion/extroversion. In H. London & J. E. Exner (Eds.), Dimensions of personality (pp. 217–262). New York: Wiley. Zuckerman, M. (1991). The psychobiology of personality. New York: Cambridge University Press. Zuckerman, M. (2002). Zuckerman-Kuhlman personality questionnaire (ZKPQ): An alternative five-factorial model. In B. d. Raad & M. Perugini (Eds.), Big Five assessment (pp. 377–396). Go¨ttingen, Germany: Hogrefe & Huber.

Received August 12, 2005 Revision received January 25, 2006 Accepted February 3, 2006 䡲

Journal of Personality and Social Psychology 2006, Vol. 91, No. 1, 188 –204

Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.188

Possible Selves and Academic Outcomes: How and When Possible Selves Impel Action Daphna Oyserman, Deborah Bybee, and Kathy Terry The University of Michigan Puzzled by the gap between academic attainment and academic possible selves (APSs) among lowincome and minority teens, the authors hypothesized that APSs alone are not enough unless linked with plausible strategies, made to feel like “true” selves and connected with social identity. A brief intervention to link APSs with strategies, create a context in which social and personal identities felt congruent, and change the meaning associated with difficulty in pursuing APSs (n ⫽ 141 experimental, n ⫽ 123 control low-income 8th graders) increased success in moving toward APS goals: academic initiative, standardized test scores, and grades improved; and depression, absences, and in-school misbehavior declined. Effects were sustained over a 2-year follow-up and were mediated by change in possible selves. Keywords: possible selves, African American, Hispanic, prevention, self-regulation

of self-concept. Another possibility is that PSs fail to sustain self-regulatory action because sustaining self-regulatory effort over time is difficult, and youth may misinterpret difficulty as evidence that the PS is not a reasonable goal and should be abandoned. To predict when a particular PS is likely to motivate action, is it necessary to have a model that predicts which PSs are likely to be online in working memory and, of these PSs, which will be invested in over time. In the current paper, we focus on school-related PSs and outline and test a predictive model. We propose that youth have difficulty creating and sustaining school-focused PSs when they perceive these PSs to be incongruent with important social identities (e.g., racial-ethnic identities), misinterpret difficulties in working on these PSs as evidence that academic goals are unrealistic PSs, and live in social contexts that fail to cue strategies for attaining their PS goal. Youth will commit sustained self-regulatory effort to a PS when the PS itself is effective and contains behavioral strategies and social context supports working on the PS, when the PS feels congruent with important social identities, and when difficulty working on the PS is construed as normative. In the following sections, we outline each of these links and the intervention we developed based on this model with a goal of enhancing the self-regulatory impact of PS.

From early adolescence, the future is an important component of self-concept (McGuire & Padawe-Singer, 1976), and doing well in school is a common element of youths’ future-oriented selves (Oyserman, Johnson, & Bybee, 2006). These future or possible selves (PSs) are positive and negative images of the self already in a future state—the “clever” self who passed the algebra test, the “fat” self who failed to lose weight, the “fast” self who fell in with the “wrong” crowd (Oyserman & Markus, 1990). A number of self-regulatory models have posited positive and negative consequences of PS–PS motivate current action (Baumeister, 1998; Carver & Scheier, 1990; Eccles & Wigfield, 2002; Higgins, 1996a; Oyserman & Markus, 1990; Weinstein, 1993). Failure to attain PSs (Oyserman & Fryberg, 2006), hoped for selves, or ideal selves (Strauman, 2002; Strauman & Higgins, 1988) may increase risk of depression. There is some evidence that imagining successful PSs improves well-being (King, 2001) and performance (Ruvolo & Markus, 1992). However, PSs do not always sustain selfregulatory action. Youth fail algebra, fall off weight maintenance, and engage in risky behavior, perhaps also increasing their risk of feeling depressed at their inability to move toward their self-goals. Why might PSs fail to sustain self-regulatory action? Given that the self is multidimensional and includes multiple potentially competing goals (e.g., Abrams, 1994; Burke, 2003; King & Smith, 2004; Oyserman, 2001; Settles, 2004), a particular PS may fail to sustain self-regulatory action because it conflicts with other parts

PSs and Self-Regulatory Behavior Effective PSs

Daphna Oyserman, Deborah Bybee, and Kathy Terry, The Institute for Social Research, The University of Michigan. This work was supported by National Institute of Mental Health Grant R01 MH 58299. We thank the students, parents, trainers, teachers, school staff, and community members involved in the intervention, data collection, and tracking; Jim Klein for help in obtaining school data; and our coordinating and database managers, Carol Carlin, Johnessa Dimicks, Tami Hart-Johnson, and Angelique Lange. Correspondence concerning this article should be addressed to Daphna Oyserman, The Institute for Social Research, The University of Michigan, Ann Arbor, MI 48106-1248. E-mail: [email protected]

A number of studies suggest that PSs differ in self-regulatory effectiveness. Self-regulatory effort improves when youth have both positive PSs (goals) and negative PSs (fears) in the same domain (“balanced” PSs) (Oyserman & Markus, 1990) and when youth have incorporated detailed strategies into their PSs (“plausible” PSs) (Oyserman, Bybee, Terry, & Hart-Johnson, 2004). When PSs are balanced, individuals select strategies that both increase the likelihood of becoming like the positive PSs and decrease the likelihood of becoming like the negative PSs, thereby focusing self-regulation and broadening effort (Oyserman & 188

POSSIBLE SELVES AND ACADEMIC OUTCOMES

Markus, 1990). Plausibility provides automatic cuing of predeveloped strategies (e.g., “set my alarm,” “go to class even if my friends skip”); indeed, youth with plausible academic PSs (APSs) are able to improve their grades over the course of the school year, whereas those who simply had APSs are not (Oyserman et al., 2004). Despite the fact that most low-income youth have at least one PS focused on school, few of these PSs include strategies (Oyserman et al., 2006). Many youths fail to attain even the basic academic PS goal of graduating high school. Thus, the national average for on-time graduation is 75% overall, 50% for African Americans, and 53% for Hispanics) (Orfield, Losen, Wald, & Swanson, 2004). In urban centers like Detroit, graduation estimates are even lower (between 40% and 44% on-time graduation) (Detroit News, 2005).

Contextual Cuing of PSs Not only is the potential self-regulatory impact of having a PS goal undermined when PSs are not balanced or plausible, selfregulatory effectiveness is also undermined when social contexts do not cue the PS. Because information that is cued (chronically or situationally made salient) is likely to be used in judgments and decision-making (Higgins, 1996b), contextually cued PSs should influence self-regulatory behavior more than those that are not cued. Moreover, to sustain ongoing engagement in school, PSs must be linked with behavioral strategies; positive expected APSs need to be linked with strategies to attain them, and feared APSs need to be linked with strategies to avoid them. Although resource-rich contexts such as a middle-class neighborhood and school provide models of success and a developed structure to guide the process of attaining APSs, this is unlikely to be the case in resource-limited contexts. In middle-class contexts, strategies may be automatically cued; parents, teachers, and parents of friends all converge to emphasize homework, persistence in the face of difficulty, tutoring, or staying after school if needed. In low-socioeconomic status (SES) contexts, strategies are unlikely to be automatically cued because these contexts are less likely to present easily accessible models or to guide success (e.g., Roderick, 2003). In low-SES contexts, youth encounter adults who are likely to be unemployed, have low academic attainment, and hold nonprofessional jobs. Given lack of easily accessible models or automatically cued strategies, youth may maintain an abstract commitment to education without connecting these PSs to everyday behavior, expressing high aspirations even as their behavior reflects avoidance or even flight from school (for qualitative description, see Roderick, 2003).

APSs and Social Identities Self-concepts include both personal and social identities. Social identities are aspects of self-concept based not in individual traits and goals but on group-based traits and goals (Oyserman, in press). It seems reasonable that social identities will incorporate community expectations about the occupations and academic attainment of in-group members. Minority youth living in low-SES contexts are exposed to images of the in-group as low achieving (Thomas, Townsend, & Belgrave, 2003), raising the possibility that the selves possible for in-group members may not feel congruent with APSs.

189

The negative link youths perceive between minority status and academic attainment and eventual occupation can be seen both in ethnographic descriptions and in experimental paradigms. Ethnographic research suggests that high school students perceive Latinos as more likely to become manual laborers, Asians to do well in school, and African Americans to do poorly in school (Kao, 2000). The same results emerge from scenario-based experimental research, whether focused on the link between minority status and low academic attainment or on the link between low social class and low academic attainment. Thus, when a failing student is described, Latino and African American students are more likely to predict that the target is Latino or African American than White (Graham, 2001). When asked to predict academic performance of a target student, low-income students infer worse performance from low (vs. middle) social class peers (Re´gner, Huguet, & Monteil, 2002; Weinger, 2000). When imagining what future is possible for one’s self, such negative preformed group images are likely to be highly accessible, making social group membership feel like it conflicts with APSs. Working toward one’s APS is likely to feel harder in the presence of accessible images of in-group members engaging in behaviors that undermine chances of attaining APSs and failing to engage in behaviors that would help attain APSs. Although there is debate as to the degree of evidence that school success is viewed as a White middle-class goal (Cook & Ludwig, 1997; Ferguson, 1998), there is consistent evidence of the stereotyped link between minority status and low achievement (e.g., Steele, 1997). The stereotype threat literature documents that simply bringing to mind category membership as minority or working class dampens academic performance (e.g., Croizet & Claire, 1998; Steele, 1997). Individuation or separation of self from in-group alleviates this effect (Ambady, Paik, Steele, Owen-Smith, & Mitchell, 2004), but the idea of intervening to disconnect youth from their racialethnic in-group (e.g., creating “racelessness”; see Fordham & Ogbu, 1986) is unappealing and likely to have other negative consequences (Arroyo & Zigler, 1995). Thus, rather than attempting to dampen the centrality of important social identities as a way to improve success in attaining APSs, increasing felt congruence between APSs and social identity, is a more reasonable strategy.

APSs and Inoculation From Overinterpretation of Difficulty In addition to the linkage of the APS to self-regulatory behavior and integration with important social identities, taking into account research on how salient information influences judgment and behavior may close what appears to be a puzzling gap between the high value placed on education and the actual attainment of lowincome and minority youth. Working on one’s APS is likely to be difficult. To judge what this difficulty means, whether APSs are “true” PSs or contradict other important social identities, to judge whether attaining these PSs is plausible or not worth the effort, teens must answer the implied questions “Why is engaging in this APS so hard for me; is this really the true me? Do we have PSs like this?” Likewise, they must judge whether particular behavioral patterns (e.g., asking for help) are likely to work and if they contradict in-group identity (e.g., “Will asking the teacher for help actually help me succeed in school or is it just a ‘White’ thing to do?”).

OYSERMAN, BYBEE, AND TERRY

190

To better understand how the feelings of difficulty that are experienced as adolescents imagine and pursue their PS influence commitment to that PS, it is fruitful to reconsider James’ (1890) original formulation of the self. James viewed the self as comprised both of content—what one thinks about when one thinks about one’s self and the accompanying metacognitive process— the feeling of thinking about one’s self. This implies that selfjudgments about who one is or may become are based on both content (what comes to mind) and process. Considering the content of thoughts about the self separately from the feelings associated with these thoughts parallels work in social cognition (Schwarz, 1998, 2002). This work proposes that human reasoning is accompanied by metacognitive experiences of relative ease (difficulty) and fluency (disfluency) (Schwarz, 1998, 2002). Following from this research in social cognition, when imagining a PS is accompanied by a metacognitive experience of difficulty, the feeling of difficulty is interpreted with a naı¨ve theory— things that are hard to think of are less likely to be true (Higgins, 1998; Schwarz & Bless, 1992; Schwarz & Clore, 1996). The experience of ease or difficulty when bringing to mind a PS can provide the basis for inferring whether a PS is a “true” self that is worth pursuing and investing effort in or a “false” self, conflicting with social identities. Metacognitive experience of ease also provides feedback as to whether the gap between the current and PS is manageable or unmanageable and therefore whether the PS should be expanded or abandoned. Although the experience of metacognitive difficulty is generally interpreted as meaning “not true for me,” a number of studies have documented that other interpretations are possible (Rothman & Schwarz, 1998). Sports stories abound with reinterpretation of the meaning of experienced difficulty (e.g., “no pain, no gain”) and the need to keep trying (e.g., “you miss 100% of the shots you don’t take”). In the case of attempting to attain APS, although the metacognitive experience of difficulty is generally interpreted as “not the true me,” the experience of difficulty could be reinterpreted to mean other things. Difficulty can be viewed as a norma-

tive part of the process (e.g., “success is 1% inspiration and 99% perspiration”). Difficulty can also provide evidence of progress (e.g., “the important things in life are the ones you really have to work for”); if difficulty and failures along the way are viewed as critical to eventual success, then difficulty is evidence of striving. Successful movement toward positive APSs and away from feared APSs requires ongoing behavior; it is not enough to complete one homework assignment or stay after class one day. If one’s metacognitive experience is that working on a PS is difficult and if this difficulty is interpreted with a naı¨ve theory that ease is associated with truth, then difficulties associated with working toward the PS will undermine it. As we have outlined, low-income and minority youth are likely to experience at least three sources of difficulty— difficulty bringing to mind APSs and linking them to strategies, difficulty sustaining the behavioral self-regulation PS strategies entail, and difficulty integrating APSs and social identities. Youth growing up in low-SES contexts have multiple models of adults who failed to attain their PSs, making it unlikely that they will recognize the normativeness of difficulties and instead likely that they will misinterpret feelings of difficulty as a sign of inevitable failure. This misinterpretation is crucial because it is likely to undermine behavioral persistence in pursuit of PS goals. Thus, rather than assuming that youth are able to make sense of difficulty as normative, low-SES youth are likely to need specific inoculation from overinterpreting current difficulty and failure as predictive of future possibilities. Taken together, their metacognitive experience of difficulty is likely to provide feedback that APSs are false rather than true selves, cuing disengagement from these PSs and the goal pursuit they imply.

PSs: A Process Model and Proposed Intervention Figure 1 presents our process model of the connections among PSs, self-regulatory behaviors, academic outcomes, and mental health. The process model links social identities and metacognitive

Social Identity

STJ Intervention

Meta-cognitive Experience

Academic Outcomes

Possible Selves

Self-Regulatory Behaviors

Well-being Mental Health

Figure 1. Process model. Theorized effects of the STJ intervention on PSs and of PSs on academic engagement, academic outcomes, and depression.

POSSIBLE SELVES AND ACADEMIC OUTCOMES

experience to PSs and links PSs to persistent engagement in self-regulatory behavior. To test this model, we developed an intervention (labeled STJ in Figure 1). The goals of the intervention were to evoke PSs and strategies to attain them, forge links between PSs and strategies that are not otherwise automatic, inoculate youth from misinterpreting failure and setbacks in attaining these PSs, and create a link between social identity and PSs. Ease is typically the basis for metacognitive judgment, so we took care to ensure that sessions felt easy, so that students would not immediately develop a metacognitive sense that thinking of the future is hard and therefore “not for me.” We based the intervention in school so that strategies articulated in the intervention would be cued in school. We structured intervention activities to make APSs salient, create linking connections between APSs and strategies, make salient naı¨ve theories of the meaning of difficulty that link difficulty with progress toward meaningful goals, and create a space in which APSs and social identity are congruent. Together, these activities were designed to “inoculate” youth from withdrawing effort to attain positive APSs and avoid feared APSs. In the long term, this change in PSs was expected to evoke persistent change in self-regulatory behavior. That is, we expected that behavior relevant to APSs (e.g., doing homework) would increase and that behavior undermining APSs (e.g., misbehaving in class) would decrease. Over time, sustained self-regulatory behavior should produce better academic outcomes, that is, grade point average (GPA). Moreover, because sustained self-regulatory behavior should reduce discrepancy between current and future selves, intervention youth should have the added benefit of reduced risk of depression (Higgins, 1997) and other mental health problems. Session 1 focused on making APSs salient and relevant as well as relevant to social identity: Each group member introduced a partner in terms of the skills or ability the partner possessed that would help him or her complete the school year successfully (e.g., “well-organized,” “positive attitude”). This provided an initial example of APSs and social identity as congruent; because all youth engaged in the task, the metamessage was “we all care about school.” Session 2 focused on adult PSs: Youth picked photographs that fit their adult “visions” (PSs). Because most adult PSs youth describe are images of material success, the metamessage was “we all want a good future.” Because all youth choose photographs of adult PSs, successful PSs and social identity are presented as congruent. In Session 3, students drew role models and negative forces— people or things that provide energy to work toward PSs and those that are draining or nay-saying. The metamessage was “everyone faces obstacles and difficulties; this does not make the PSs less part of the ‘true’ self.” In Session 4, students drew timelines into the future, including forks in the road and obstacles. The metamessage was “everyone has difficulties, and failures and setbacks are a normal part of timelines and do not mean that APSs are not true selves.” Session 5 introduced action goals, allowing students to practice articulating specific strategies to attain their APSs, further highlighting the normativeness of difficulty in attaining APSs. Sessions 6 and 7 focused on PSs and strategies to attain them, building on previous sessions, with a different concrete medium, poster board, stickers, and markers. Students chose next year feared and to-beexpected PSs and linked them with current and possible strategies.

191

Sessions 8 to 10 focused on decoupling difficulty and genuineness explicitly through work in smaller groups on everyday problems, social problems, academic problems, and the process of getting to high school graduation, and the metatheme was that all students care about these issues, that difficulties are normative and not self-defining. Session 11 cemented the new metacognitive interpretations by having participants review and critique the sessions. Two follow-up sessions included parent/guardians and community members, with the goal of helping youth broker their APSs in the community outside of school in ways that link APSs with social identities.

Hypotheses Two key hypotheses follow from our model. (a) The intervention will directly influence PSs, self-regulatory behaviors, academic outcomes, and risk of depression. (b) Effects of the intervention on self-regulatory behaviors, academic outcomes, and depression will be mediated by intervention effects on PSs. That is, shift in PSs will improve engagement in self-regulatory behaviors, and engagement in self-regulatory behaviors will increase academic success and reduce feelings of depression. We also hypothesized (c) that the intervention would influence the relationship between APSs and social identity, creating conditions for APSs and social identity to feel congruent. To test these hypotheses, we examined the direct effects of the intervention on change in PSs and the longitudinal effects of the intervention, both direct and as mediated by PSs. Finally, we examined the long-term effects of the intervention on the relationship between APSs and social identity as operationalized by racial-ethnic identity.

Method Sample Data were collected at three Detroit middle schools; 71.6% of students were African American, 17.4% were Latino, and 11.0% were White; total N for analyses ⫽ 264 as detailed below. Students were low-income, as evidenced by school lunch program (two-thirds received free or reduced lunch), neighborhood (in the students’ census tracts, n ⫽ 105, 54.1% of households were below the poverty line and only 43.4% of adults were employed; U.S. Bureau of Census, 2000), housing stability (45.2% moved at least once over the 2 years we tracked them; those who moved averaged 1.54 moves), and school stability (20% changed schools during the 1st year).

Procedure Obtaining permission. School district and school principals approved random assignment and provision of the intervention during the elective hour. School-based data collection in low-income and minority populations typically is hampered by low consent rates (Ellickson & Hawes, 1989). Parents who are lower income and have more academically at-risk students are less likely to return consent forms, reducing generalizability of findings (Kearney, Hopkins, Mauss, & Weisheit, 1983; Pokorny, Jason, Schoeny, Townsend, & Curie, 2001). Because not returning a consent form mailed home (the standard protocol) is not the same as refusing participation when asked directly, these authors argue for passive rather than signed consent to protect generalizability of findings. We also obtained the typical low response rate with the standard mailing procedure (15% return rate). Rather than limit generalizability, we followed up the 85% nonresponders with personal contact (by phone if they had one or by going to the home) to

192

OYSERMAN, BYBEE, AND TERRY

explain the study and what we would do with the data and to ask for permission or refusal. Together, the initial mailing and direct communication were well received, resulting in a total signed consent rate of 94.3% and no differences by race/ethnicity or condition. Data collection. Core subject teachers (blind to experimental assignment) assessed student in-class behavior. Test scores, GPA, and attendance were obtained from school records. Student surveys were administered in-class. Teacher and student baseline data collection occurred prior to the intervention. Teacher and student postintervention data were collected at three additional times, at the end of the 8th grade school year and again in the fall and spring of the following year. School records were obtained each time report cards were issued over the 2-year data collection period (1st quarter of 8th grade through the end of 9th grade). Because the intervention occurred prior to 1st quarter 8th grade grades, the youth baseline survey included self-report of grades to provide a prior academic attainment control variable. Teachers were not present for student surveys to assure confidentiality. Instructions were read aloud, and research staff circulated to assist students. In-home interviews were completed in year 1 for the 56 students suspended (n ⫽ 11), expelled (n ⫽ 9), transferred (n ⫽ 10), or otherwise not in school (n ⫽ 26). By year 2, students were enrolled in 80 schools; the at-home interview procedure was followed when in-school survey completion was not feasible. Teachers were reimbursed $5 for each assessment ($10 in year 2); students were reimbursed $5 (year 2 only). Missing data. Extensive tracking efforts minimized data loss. We obtained data from at least one source across all four semesters of data collection for a full 96% of our sample, 89.8% to 97.3% of student, 92.4% to 95.8% of school records, and 83.3% to 96.6% of teacher ratings. All but 4 youth had data from at least one source at semester 4. To preserve generalizability, expectation maximization methods were used to estimate the approximately 7.9% of the data matrix that were missing due to skipped items or unavailable information. Although missingness was not completely random, Little’s missing completely at random ␹2(df ⫽ 16,115) ⫽ 30,685.46, p ⬍ .001, there was no evidence that it was not ignorable (Little & Rubin, 2002). All results reported use estimated data. Experimental manipulation. In each homeroom, approximately onehalf of students were randomly assigned to attend the regular elective period (control group), and one-half were assigned to attend the intervention as the elective period (experimental group). We called the intervention School-to-Jobs (STJ) and the overall project Pathways for Youth to prevent any potential stigmatization. Students who enrolled in school late (n ⫽ 3) or had an elective period broken by lunch (n ⫽ 34) were not randomized, resulting in an effective cohort of 280. Analyses involve the 264 students with signed parental consent (n ⫽ 141 experimental, n ⫽ 123 control; n ⫽ 140 female; n ⫽ 191 African American, n ⫽ 45 Latino, n ⫽ 28 White). The STJ intervention thumbnail sketched in the introduction is fully manualized and includes a detailed fidelity assessment protocol (both available from the first author). STJ was provided twice weekly over a 7-week period (due to elective periods missed for school half days) with the two supplemental parent-youth STJ sessions provided the following evenings or weekend. Groups averaged 12 participants. Of youth in school, attendance was 80% to 90% for the youth in-school sessions, and 40% attended at least one parent-youth supplemental session. Two trainers ran each group. Fidelity to protocol was maintained via in vivo rating and weekly staff meetings. Trainers and observers were female Detroiters with previous experience who received intensive structured training in the protocol (for trainers, 39.5 hours; for observers, 69.5 hours). Verification of random assignment. Randomization was successful. Comparison of the experimental and control groups on the 13 baseline measures showed no significant differences on the multivariate, F(13, 250) ⫽ 1.07, p ⫽ .39, or on any of the univariate Fs.

PSs. PSs were assessed in 8th grade and spring of 9th grade using the standard open-ended format (available at http://www.sitemaker.umich.edu/ culture.self/files/possible_selves_measure.doc, 07–22-2003, Oyserman & Saltz, 1993; Oyserman, Terry, & Bybee, 2002). Youth generated expected PSs, noted strategies to attain each PS, and repeated this process for feared PSs. PSs were content coded following Oyserman and Markus (1990) by five coders with access only to PS responses. All PSs were double coded, and interrater reliability was 94% (disagreements discussed to agreement). Analyses focused on the two most commonly generated PSs, academic PSs and feared off-track PSs. At baseline, 96.6% of youth generated an expected APS, 63.6% generated a feared APS, and 51.1% generated a feared off-track PSs. Feared off-track PSs. These PSs focused on involvement in gangs or violence (30% of responses), drugs (30%), delinquency or involvement with the police (30%), and becoming pregnant and other “status” offenses (10%), fall 8th grade, M ⫽ 0.74, SD ⫽ .90; spring 8th grade, M ⫽ .85, SD ⫽ .83; spring 9th grade, M ⫽ 1.41, SD ⫽ 1.04. Raw counts were skewed so they were transformed into change scores (spring-fall), improving distribution and focusing on the impact of the intervention on change in PSs during 8th grade; change scores are appropriate when groups are randomly assigned and variances similar across observations (Cribbie & Jamieson, 2000; Maris, 1998). Change in feared off-track PSs ranged from ⫺3 to ⫹ 3, M ⫽ 0.11, SD ⫽ 1.04. APS balance. Balance (the number of pairs of expected and feared APSs, e.g., expecting to “pass the 8th grade” while wanting to avoid “failing and having to be an 8th grader again”) was coded following Oyserman and Markus (1990). About one-third of youth had at least one balanced APS pair, fall 8th grade, M ⫽ 0.39, SD ⫽ .56; spring 8th grade, M ⫽ .40, SD ⫽ .57; spring 9th grade, M ⫽ .30, SD ⫽ .59. Raw counts were skewed so they were transformed into change scores (spring-fall), improving distribution, and focusing on the impact of the intervention on change in PSs during 8th grade; change in APS balance ranged from ⫺2 to ⫹ 3, M ⫽ 0.01, SD ⫽ 0.69. For moderated regression analysis (Hypothesis 3), 9th grade APS balance scores were log-transformed (raw spring 9th grade scores had skew of 2.5 and kurtosis of 8.47), reducing skew and kurtosis to ⬍1.5. APS plausibility. The set of expected and feared APSs and strategies was coded using a 6-point scale (0 ⫽ no APSs to 5 ⫽ at least 4 APSs and APS strategies) (Oyserman et al., 2004). Youth averaged about two APSs and strategies, fall 8th grade, M ⫽ 2.26, SD ⫽ 1.30; spring 8th grade, M ⫽ 2.08, SD ⫽ 1.25; spring 9th grade, M ⫽ 2.18, SD ⫽ 1.26. APS plausibility was not skewed, so log-transformation was not necessary, and intervention effects were assessed by examining plausibility in spring 8th grade. The coding manual and detailed instructions are found on http://sitemaker .umich.edu/daphna.oyserman/files/plausibility_instructions_only012903.doc.

Social Identity Social identity was operationalized with a 4-item, 5-point Likert response (1 ⫽ strongly disagree, 5 ⫽ strongly agree) in-group connectedness scale, spring 9th grade, M ⫽ 4.18, SD ⫽ .59, ␣ ⫽ .76. Students filled in their main racial-ethnic group and then responded. For the group “African American,” example items are “I feel close to African Americans,” “It is important to me to think of myself as an African American,” and “I feel a part of the African American community.” Analysis of connectedness to racial-ethnic in-group excluded the n ⫽ 28 White students because “White identity” is unlikely to have the same meaning as minority racial-ethnic identity.

Self-Regulatory Behavior Measures Demographics. Schools provided gender and birthdates; youths provided their race/ethnic description.

Time spent doing homework. In 8th grade, we asked “How many hours a week do you usually spend doing homework?” Open ended responses ranged from 0 to 20 hours and were log-transformed to reduce skew. On an

POSSIBLE SELVES AND ACADEMIC OUTCOMES 8-point closed ended follow-up question (0 ⫽ no time to 7 ⫽ more than 10 hours), fall M ⫽ 3.89 (SD ⫽ 1.77) and spring M ⫽ 3.89 (SD ⫽ 1.95), mean responses reflected about 2 hours a week. In 9th grade, we used a 1-week event history calendar (Belli, 1998); again, students reported about 2 hours of weekly homework, fall M ⫽ 160.20 minutes, SD ⫽ 151.89; spring M ⫽ 112.15 minutes, SD ⫽ 111.28. Homework time was low; as a benchmark, national daily diary report data for 12 to 14 year olds are 4.40 hours weekly (Juster, Ono, & Stafford, 2004). Disruptive behavior. Eighth grade youth reported “How often does the teacher make you leave the classroom because of your behavior?” on a 7-point scale (1 ⫽ never, 2 ⫽ once or twice a year, 3 ⫽ less than once a month, 4 ⫽ once a month or so, 5 ⫽ once every few weeks, 6 ⫽ once a week or so, 7 ⫽ more than once a week), fall M ⫽ 2.18, SD ⫽ 1.55; spring M ⫽ 2.64, SD ⫽ 1.94. Teachers reported on youth behavior with a 4-item 5-point response (1 ⫽ never, 5 ⫽ always) (Finn Disruptive Behavior Scale, Finn, Pannozzo, & Voelkl, 1995; 8th grade revision, J. Finn, personal communication, October 14, 1998) scale each semester, ␣ ⫽ 79, .81, .78, and .80, respectively. Beginning with the stem “This student . . . ,” items were “annoys peers or interferes with peers’ work,” “is critical of peers who do well in school,” “needs to be reprimanded or sent to the office,” and “is verbally or physically abusive to the teacher.” Initiative-taking behavior. Teachers reported on youth behavior with a 4-item 5-point response (1 ⫽ never, 5 ⫽ always) (Finn Initiative Scale, Finn et al., 1995; 8th grade revision, J. Finn, personal communication, October 14, 1998) scale each semester, ␣ ⫽ .75, .82, .82, and .85, respectively. Beginning with the stem “How often does this student . . . ,” items were “do more than the work assigned?,” “persist when confronted with difficult problems?,” “actively participate in class discussions?,” and “engage me in conversation about subject matter before or after school or outside of class?” Absences. Eighth grade youth reported “How often are you absent from school or miss a class during the day?” on a 7-point scale (1 ⫽ never, 2 ⫽ once or twice a year, 3 ⫽ less than once a month, 4 ⫽ once a month or so, 5 ⫽ once every few weeks, 6 ⫽ once a week or so, 7 ⫽ more than once a week), fall M ⫽ 3.17, SD ⫽ 1.67; spring M ⫽ 3.82, SD ⫽ 1.79. As a second measure of absences, a count of unexcused absences was obtained from school records in 8th and 9th grades (log transformed to reduce skew). Each day not enrolled in any school was counted as an absence.

Academic Outcomes GPA. School record of core (math, history, science, English) grades (0.0 ⫽ F to 4.0 ⫽ A) was obtained; 0.0 was assigned as GPA for students not enrolled in any school during the semester. As a baseline measure of grades prior to the intervention, we obtained youth report of prior grades. Three additional academic outcome measures were obtained at the end of 8th grade: (a) proportion of subject tests passed on the 8th grade Essential Skills Attainment Test (based on the Michigan Educational Assessment Program (MEAP) test), M ⫽ .80, SD ⫽ .25, range ⫽ 0.0 to 1.0), (b) referral to remedial summer school as a prerequisite to high school promotion (47% of youth), and (c) retained in 8th grade according to school records by the end of the summer (9.5% of students). These measures were not available in 9th grade (standardized tests are not administered each year); therefore, 9th grade academic outcomes analyses focus on GPA only.

Depression Youth report of depression, the standard 20-item, 4-point Center for Epidemiological Studies Depression Scale (Radloff, 1977), anchored at 0 ⫽ not at all or less than one day in the past week and 3 ⫽ 5 to 7 days in the past week, was obtained spring of 9th grade (M ⫽ 11.35, SD ⫽ 8.33). Items include affective (e.g., “I felt depressed”) and somatic (e.g., “I did not feel like eating; my appetite was poor”) aspects of depression. Propor-

193

tion of youth meeting the clinical threshold for depression (sum scores of at least 16) (Radloff & Locke, 2000) was 22.7%, comparable with the proportion of youth reaching the threshold for depression in national surveys (Costello, Mustillo, & Erkanli, 2003; Roberts, Attkisson, & Rosenblatt, 1998).

Results We first examined the direct effect of the STJ intervention on PSs, self-regulatory behaviors, academic outcomes, and depression. Then, we examined PSs as mediators of intervention effects on self-regulatory behaviors and, via their impact on selfregulatory behavior, as mediators of intervention effects on academic outcomes and depression. We also examined whether the STJ intervention influenced the association between APSs and social identity. Prior to describing these analyses and results, we outline our decision to present analyses of both the “intention-totreat” (ITT) and compliant sample analyses.

ITT and Compliant Sample Analyses Following standard procedure, direct effects analyses (Hypothesis 1) were conducted twice, once with the ITT sample and a second time with the compliant sample. Analyses of meditating effects of PSs (Hypothesis 2) and of the impact of the intervention on the relationship between APSs and social identity (Hypothesis 3) utilized the compliant sample. The ITT sample included all 264 youth randomized into experimental or control conditions, whether they were in school to receive the intervention or not. Although analysis based on an ITT sample preserves random assignment, it is not a good test of the intervention effects to the extent that some people who did not receive the intervention are included in the treatment group for analyses. The alternative is to compare those who received a meaningful dose of the intervention with those who did not. This is termed compliance analyses. Use of compliance analyses to estimate effects for individuals who received a meaningful dose of an intervention is appropriate if assessed variables provide a useful and valid estimate of those likely to participate and if nonparticipators among those assigned to the experimental condition do not differ in outcomes from control condition youth (Jo, 2002a, 2002b). In the current study, both requirements were met, as outlined in more detail at http:// www.sitemaker.umich.edu/culture.self/files/ appendix_a_on_web_ site.doc, 26.0 kb, the compliant sample (n ⫽ 228) included the “participating” youth (n ⫽ 116) and a comparable sample of control youth (n ⫽ 112). Participating youth were assigned to the experimental condition and received a reasonable dose of the intervention (i.e., attended at least five sessions). The comparable sample of control youth (n ⫽ 112) were youth who had a similar unobserved likelihood of “participating” had they been randomized to intervention according to the expectation maximization compliance model incorporating the full set of preintervention variables (Little & Yau, 1998). Most of the difference in sample size between the ITT and compliant analyses was due to youth being suspended or expelled within the 1st month of school; some youth simply were not in school to receive the intervention. Effects presented are from the compliant (participating) sample unless otherwise specified.

194

OYSERMAN, BYBEE, AND TERRY

Hypothesis 1: STJ Intervention Has Direct Effects on PSs, Self-Regulatory Behavior, and Outcomes Analysis Plan We used multilevel modeling (MLM; Raudenbush & Bryk, 2002; Snijders & Bosker, 1999) to appropriately analyze variance at three levels—time nested within students and students nested within homerooms. We used standard methods for MLM, first building baseline models containing random effects for homeroom and time and then adding condition assignment and baseline control variables to optimize statistical power by accounting for preintervention variability in sex, race, age, baseline GPA (student report), baseline behavioral problems (teacher report), and baseline assessment of each dependent measure. For ease of interpretation, control variables (other than baseline assessment of the dependent variable) were constant across analyses. We show direct effects at the end of 8th grade and over time (through the end of 9th grade). For the longitudinal models predicting change in self-regulatory behavior and academic outcomes, we set the intercept for time at spring of 9th grade to obtain estimates of between-condition effects at the final measurement point. We used Raudenbush and Liu’s (2000, 2001) methods to compute standardized effect sizes; these effects are calibrated to the residual level 1 variance of the unconditional model (i.e., the model containing no predictors other than homeroom) for 8th Grade MLM analyses and to the random

slope and intercept variances of the unconditional model (i.e., the model with no predictors other than time and homeroom) for the longitudinal MLM. Standardized effect sizes are interpreted like Cohen’s d.

STJ Effects by End of 8th Grade Effects of STJ for the total ITT sample and the compliant (participating) sample are presented in Table 1. For each dependent variable, the raw coefficient is interpreted as a regression weight and indicates, in the metric of the dependent variable, the average within-homeroom difference between students assigned to STJ rather than the control group. The standardized effect size is interpreted as Cohen’s d and expresses the same difference in SD units. To further facilitate interpretation of intervention effects, estimated means for spring 8th grade are in the first columns of Table 2. We summarize findings by dependent variable in the sections below. PSs. Following Hypothesis 1, both ITT and compliant sample analyses showed significant intervention effects on PSs in the spring of 8th grade. Experimental youth generated more balanced APSs, more plausible APSs, and more feared off-track PSs, setting the stage for self-regulation. Self-regulatory behavior. As can be seen in the negative coefficients connoting fewer absences, participation in the STJ intervention significantly reduced absences, whether assessed via

Table 1 Multilevel Models: Effects of the STJ Intervention in Spring 8th Grade Participating samplea (N ⫽ 228)

Total ITT sample (N ⫽ 264) Raw coefficients

Raw coefficients

Dependent variables

B

p

B

p

Change in APS balance Change in feared off-track PS APS plausibility Absences (student report) Unexcused absences (school records)c Homework time (hrs/week; open ended)c Homework time (hrs/week; closed ended) Disruptive behavior (student report) Disruptive behavior (teacher report) In-class initiative (teacher report) Core academic GPA (school records) Standardized tests (school records; proportion passed) Referral to remedial summer schoold Retention in 8th graded

.157 .286 .340 ⫺.417 ⫺.183 .081 .263 ⫺.307 ⫺.099 .179 .229 .068 ⫺.194 ⫺.598

.023 .004 .015 .043 .049 .090 .229 .165 .261 .043 .021 .017 .457 .164

.186 .308 .400 ⫺.457 ⫺.202 .099 .406 ⫺.380 ⫺.139 .142 .218 .086 ⫺.419 ⫺.979

.014 .003 .008 .036 .022 .135 .079 .094 .149 .142 .039 .006 .140 .076

Standardized effect sizeb .269 .302 .329 ⫺.267 ⫺.726 .155 .240 ⫺.209 ⫺.172 .170 .252 .360

Note. The cross-sectional multilevel models analyzed variance at two levels: students (Level 1), nested within homerooms (Level 2). All Level 1 intercepts were random, and all modeled fixed effects were at Level 1 (student). Models controlled for the following Level 1 (student) covariates: race, sex, baseline age, preintervention GPA (student-report), preintervention behavioral problems (teacher report), and preintervention assessment of the dependent variable, when available. a The participating sample were 116 students who attended at least five in-school STJ sessions plus 112 students in the control group who had similar likelihood of attending, had they been randomly assigned to the intervention, according to maximum likelihood (EM) modeling of intervention compliance. b Interpretation is similar to d for STJ intervention effects on individuals within homerooms. ES ⫽ B/V⫺1 p , where B ⫽ raw coefficient and Vp ⫽ residual variance in the unconditional intercept model, following Raudenbush & Liu c (2001). These positively skewed dependent variables were log-transformed to reduce the influence of extreme scores. d Multilevel models for these dichotomous dependent variables used a logit link function. Standardized effect sizes could not be computed for logit models.

POSSIBLE SELVES AND ACADEMIC OUTCOMES

195

Table 2 Multilevel Models: Estimated Means in Experimental and Control Conditions: Spring 8th and 9th Grades Estimated condition means/percentages Spring of 8th grade Dependent variables Change in APS balance Change in feared off-track PS APS plausibility Absences (student report) Unexcused absences (school records)b Homework time (hours/week; open ended)b Homework time (hours/week; closed ended) Disruptive behavior (student report) Disruptive behavior (teacher report) In-class initiative (teacher report) Core academic GPA (school records) Standardized tests (school records; proportion passed) Referral to remedial summer school Retention in 8th grade Depression (CESD—spring of 9th grade)c

STJ (n ⫽ 116)

Control (n ⫽ 112)

0.07 0.31 2.36 3.49 12.28 3.49 4.15 2.37 1.74 2.51 1.98 0.83 36.2% 4.3%

⫺0.12 0.00 1.96 3.95 14.53 3.28 3.74 2.63 1.83 2.44 1.83 0.77 48.2% 10.7%

Spring of 9th grade STJ (n ⫽ 116)

Control (n ⫽ 112)

22.52 2.51

24.77 1.57

1.55 2.48 1.64

1.73 2.25 1.36

10.35

12.29

Note. N ⫽ 228 youth in the participating sample. For variables with estimated means for both 8th and 9th grades, estimates were from the intercept term of a three-level longitudinal multilevel model (time, students, homerooms); for variables with estimated means for only one grade, estimates were from cross-sectional two level multilevel models (students in homerooms). b These positively skewed variables were log transformed for analysis; however, to facilitate interpretation, means are presented here in the original metric (i.e., in exponentiated form). c Depression was measured only in the spring of 9th grade; means were estimated by a two-level cross-sectional multilevel model. a

school or youth report, in both ITT and compliant samples. For school-reported absences, the effect of the intervention is in Cohen’s large range, with d near .80. Each semester intervention youth attended more school, over 2 additional days, than control youth. With regard to time spent in homework and behavior in class, results are suggestive of STJ impact (for homework, student and teacher rated behavior, all p ⬍ .15). Academic outcomes. Intervention youth showed significantly better GPA and better standardized test scores using both ITT and compliance analyses. With regard to 8th grade retention, more than twice as many control youths (10.7%) as intervention youths (4.3%) were retained, p ⬍ .10).

STJ Effects by End of 9th Grade Estimated means for the spring of 9th grade are in Table 2. Table 3 presents the direct effects of the STJ intervention on the trajectory of change over time and on the difference between intervention and control groups in the spring of 9th grade. For each dependent variable, the raw slope coefficient indicates the average within-homeroom difference in linear trajectory between students assigned to STJ and the control group. The raw intercept coefficient indicates the average spring 9th grade within-homeroom difference between students assigned to STJ and control group. Standardized effect sizes are interpreted as Cohen’s d and express the same differences in SD units. The trajectories of change for each dependent variable are presented graphically in Figures 2 and 3. Self-regulatory behavior. STJ youth spent significantly more time doing homework, were more likely to take initiative in class,

less disruptive, and less likely to skip class than control group youth. As displayed in Figure 2a, time spent on homework declined for all youth, but the decline was significantly less for STJ youth. By spring of 9th grade, STJ youth were spending on average 2.51 hours a week on homework, nearly an hour more per week than control youth who averaged 1.57 hours per week. Both ITT and compliant sample analyses showed a large and significant effect of STJ on both level and trajectory of change over time in homework. Trajectories of in-class behavior are displayed in Figure 2b (initiative taking) and 2c (disruptive behavior). As can be seen, initiative-taking declined for control youth, not for STJ youth; disruptive behavior declined for all youth, but the decline was significantly steeper for STJ youth. Size of intervention effects is moderate for the trajectory of change in initiative-taking and large for the trajectory of change in disruption over time. With regard to absences (skipping), as displayed in Figure 3a, the large and significant effect of the intervention at the end of 8th grade remained stable and significant through 9th grade. Each semester of 9th grade, STJ youth averaged 2.25 more days in school than control youth. Academic outcomes. STJ and control youth differed significantly in GPA by the end of 9th grade, with the size of gap between groups increasing over time, as displayed graphically in Figure 3b; intervention effect sizes for both end of 9th grade level and trajectory over time were “small-to-moderate” using Cohen’s (1988) rules of thumb. Within-time models comparing the STJ and control groups at each grading period showed that a significant

OYSERMAN, BYBEE, AND TERRY

196

Table 3 Longitudinal Multilevel Models: Effects of STJ Intervention on Change (Slope) and Level (Spring of 9th Grade) of Self-Regulatory Behavior, Academic Outcomes, and Depression Participating samplea (N ⫽ 228)

Total ITT sample (N ⫽ 264) Raw coefficients Dependent variables Homework time (open ended)c Spring 9th grade (intercept) Change trajectory (linear slope) Teacher-report of in-class initiative Spring 9th grade (intercept) Change trajectory (linear slope) Teacher-report disruptive behavior Spring 9th grade (intercept) Change trajectory (linear slope) School record of unexcused absencesc Spring 9th grade (intercept) Change trajectory (linear slope) School record of core academic GPA Spring 9th grade (intercept) Change trajectory (linear slope) Depression (CESD)d Spring 9th grade (intercept)

B

Raw coefficients Standardized effect sizeb

p

B

p

0.166 0.011

.043 .050

0.183 0.012

.035 .041

0.742 1.044

0.166 0.012

.115 .073

0.234 0.015

.037 .036

0.326 0.430

0.085 ⫺0.005

.283 .366

⫺0.178 ⫺0.010

.030 .040

⫺0.326 ⫺0.776

⫺0.144 0.000

.051 .994

⫺0.175 0.001

.027 .847

⫺0.296 0.000

0.211 0.164

.074 .266

0.277 0.030

.031 .055

0.300 0.354

⫺1.355

.187

⫺1.943

.049

⫺0.257

Note. The longitudinal multilevel models analyzed variance at three levels: time (Level 1) nested within students (Level 2), nested within homerooms (Level 3). All Level 2 intercepts and slopes were random. Models controlled for the following Level 2 (student) covariates: race, sex, baseline age, and preintervention GPA (student-report). a The participating sample was 116 students who attended at least five in-school STJ sessions plus 112 students in the control group who had similar likelihood of attending, had they been randomly assigned to the intervention, according to maximum likelihood (EM) modeling of intervention compliance. b Interpretation is similar to d. For STJ intervention effects on the intercept, ES ⫽ B/T⫺1 00 , where B ⫽ raw coefficient and T00 ⫽ intercept variance in the unconditional change model. For STJ effects on change in linear slopes, ES ⫽ B/T⫺1 11 , where B ⫽ raw coefficient and T11 ⫽ variance of the individual linear change trajectories in the unconditional change model. For STJ effects on individuals within homerooms (for depression in spring of 9th grade), ES ⫽ B/V⫺1 p , where B ⫽ raw coefficient and Vp ⫽ residual variance in the unconditional intercept model (Raudenbush & Liu, 2001). c These positively skewed dependent variables were log-transformed to reduce the influence of extreme scores. d Depression was measured only in spring of 9th grade; coefficients are from a cross-sectional, two-level MLM.

difference in GPA emerged at the 3rd quarter, 2 quarters after the intervention (B ⫽ .22, p ⬍ .02). GPA between groups continued to diverge over time. By the end of the 9th grade, estimated average GPA was 1.64 for STJ youth, compared with 1.36 for control group youth. Depression. Intervention effects on depression were significant; estimated mean Center for Epidemiological Studies Depression Scale scores were nearly 2 points lower for STJ (M ⫽ 10.35) compared with control youth (M ⫽ 12.29) at the end of 9th grade.

Hypothesis 2: PSs Mediate STJ Effects on Self-Regulatory Behavior, Academic Outcomes, and Depression Analysis Plan Having found support for hypothesis 1 (significant direct effects of STJ on PSs, self-regulatory behavior, academic outcomes, and depression) allowed for examination of hypothesis 2 (mediating effects of PSs). To test for mediation, we used latent structural equation modeling, employing standard methods to develop the measurement model defining the latent constructs, to assess overall

model fit, and to calculate and test indirect effects. All SEM analyses used AMOS 5 software (Arbuckle & Wothke, 2003). Figure 4 displays the structural model of the hypothesized direct and indirect effects including four observed constructs (8th grade APS balance, APS plausibility, feared off-track PSs, and 9th Grade GPA) and four latent constructs (attendance, homework, classroom behavioral problems, and depression). The measurement model was developed to test associations between latent constructs and the observed indicators that measure them. As presented in Table 4, all loadings were above .46 (all p ⬍ .01). For the multiindicator latent constructs (all but depression), measurement error was reflected in the loadings linking observed indicators to latent constructs. For the single-indicator latent construct (the multiitem depression scale), measurement error was reflected by weighting the scale error variance by 1 ⫺ Cronbach’s alpha for the depression scale (1 ⫺ .84 ⫽ .16). Following standard practice, we allowed correlated errors among those indicator variables that shared measurement variance due to common source and item format. These were closed-ended student-report of attendance, homework, and classroom behavioral problems (error correlations

POSSIBLE SELVES AND ACADEMIC OUTCOMES

197

Figure 2. Effect of the STJ intervention over time on hours spent on homework each week (a), teacher-rated initiative (b), and teacher-rated disruptive behavior (c).

ranged from r ⫽ .12 to r ⫽ .26). Correlations among measurement errors of other indicators were fixed at zero. The structural model of hypothesized direct and indirect effects was applied to the latent constructs defined in the measurement model. Single-headed arrows in Figure 4 indicate the directional conceptualization of the structural model: The influence of the intervention on PSs, self-regulatory behavior, academic outcomes, and depression. Associations among 8th grade measures of selfregulatory behavior (homework effort, classroom behavior problems, and absences) were specified as nondirectional correlations, indicated by curved, double-headed arrows. Multivariate normality of the full model was acceptable: Mardia’s (1985) test of multivariate kurtosis ⫽ 10.57, with univariate skew and kurtosis less than 1.2 for all endogenous variables. Fit of the full structural model to the data was assessed by ␹2, supplemented by tests and critical thresholds recommended by Hu and Bentler (1999): standardized root mean residual ⱕ .08, root mean square error of approximation (RMSEA) ⱕ .06, and Bollen’s incremental fit

index ⱖ .95. With the sample size of 228 and df exceeding 100, power exceeds .80 for a test of overall model fit using RMSEA (MacCallum, Browne, & Sugawara, 1996). By current criteria, our sample size of 228 is adequate for stable and unbiased estimation of indirect effects and SEs (e.g., MacKinnon, Warsi, & Dwyer, 1995). Bootstrapped, bias-corrected SEs were used to test the significance of indirect effects (MacKinnon, Lockwood, Hoffman, West, & Sheets, 2002; MacKinnon, Lockwood, & Williams, 2004). Power to test the significance of mediated effects generally exceeded .80 given sample size, measure reliability, extent of collinearity between independent and mediating variables, and use of latent constructs (see Hoyle & Kenny, 1999).

Testing the Full Process Model The final structural model, shown in Figure 4, was an excellent fit to the data: ␹2(df ⫽ 146) ⫽ 156.36, p ⫽ .26; standardized root mean residual ⫽ .05, RMSEA ⫽ .02, incremental fit index ⫽ .99.

OYSERMAN, BYBEE, AND TERRY

198

Figure 3.

Effect of the STJ intervention on days absent per semester (a) and core academic grades (b).

The intervention had direct positive effects on PSs as well as a direct negative effect on school absences (reducing absences). Each PS variable had significant effects on 8th and 9th grade outcomes; all mediation paths were significant at p ⬍ .05. Nondirectional correlations among constructs were moderate and anticipated: 8th grade classroom behavioral problems were correlated positively with absences (r ⫽ .50) and negatively with homework (r ⫽ ⫺.33). Table 5 lists the total and specific mediation paths linking the intervention with 8th and 9th grade effects.

Figure 4.

Interpreting the Mediating Effects of PSs APSs. The STJ intervention had positive direct effects on APSs (APS balance B ⫽ .14 and APS plausibility, B ⫽ .12), as well as an indirect effect on APS plausibility through APS balance (indirect B ⫽ .02). APS balance and APS plausibility were complete mediators of intervention effects on two of the self-regulatory behaviors in the 8th grade— behavioral problems (indirect B ⫽ ⫺.02) and time spent in homework (indirect B ⫽ .03). On 9th

Effects of the STJ intervention over time as mediated by PSs.

POSSIBLE SELVES AND ACADEMIC OUTCOMES

Table 4 Measurement Coefficients for Latent Structural Equation Model Latent constructs and observed indicators Classroom behavior problems (8th grade) Teacher report of disruptive behaviora Student report of disruptive behavior Teacher report of initiative Homework (8th grade) Closed-ended rating of time spent on homework each weeka Open-ended report of time spent on homework each week Absences (8th grade) School record of unexcused absencesa Student report of absences Depression (9th grade)b CESD

Standardized coefficient 0.817 0.465 ⫺0.468 0.888 0.764 0.736 0.568 0.910

Note. N ⫽ 228 youth in the participating sample. All coefficients significant at p ⬍ .001. a These indicators were set to 1 to define the scale of the construct. b Measurement error in this latent construct was modeled by weighting the error variance of the single indicator (the CESD scale score) by 1-Cronbach’s alpha (⫺.84 ⫽ .16).

grade academic outcomes, APS balance and APS plausibility partially mediated intervention effects on GPA (indirect B ⫽ .03). APS balance and plausibility also partially mediated intervention effects on 9th grade depression (indirect B ⫽ ⫺.01). Feared off-track PSs. The intervention had positive direct effects on feared off-track PSs (B ⫽ .13). Off-track PS was a partial mediator of intervention effects on 8th grade school ab-

199

sences (indirect B ⫽ ⫺.02). For 9th grade outcomes, off-track PS partially mediated intervention effects on academic outcomes ⫺ GPA (indirect B ⫽ .01), as well as 9th grade depression (indirect B ⫽ ⫺.01).

Hypothesis 3: STJ Influences the Relationship Between APS and Social Identity Analysis Plan We used hierarchical multiple regression to examine the moderating effect of the STJ intervention on the Spring 9th Grade APS-social identity relationship. Fall 8th Grade APS was entered as a control variable at Block 1, experimental assignment at Block 2, Social Identity Connectedness at Block 3, and the Connectedness ⫻ Intervention interaction at Block 4. Methods recommended by Aiken and West (1991) were used to probe interaction effects and test significance of simple slopes.

Effects on APS APS plausibility. The full model containing main effects and the STJ ⫻ Social Identity Connectedness interaction was significant, F(4, 223) ⫽ 5.02, p ⬍ .001, with added variance explained by the STJ ⫻ Social Identity Connectedness interaction, B ⫽ .57, ⌬F(1, 223) ⫽ 3.78, p ⫽ .05. For intervention youth, Connectedness and APS plausibility were positively associated, simple slope B ⫽ .80, p ⬍ .001, but for control youth, there was no significant relationship, simple slope B ⫽ .22, p ⫽ .31. APS balance. The full model containing main effects and the STJ ⫻ Social Identity Connectedness interaction was marginally significant, F(4, 223) ⫽ 1 .95, p ⫽ .10, and the STJ ⫻ Social

Table 5 Mediational Paths Linking the STJ Intervention to Spring 8th Grade and Spring 9th Grade Effects General and specific mediational paths STJ effects on spring 8th grade outcomes Total STJ effect on classroom behavior problems (8th grade) STJ 3 APS balance 3 APS plausibility 3 classroom behavior problems STJ 3 APS plausibility 3 classroom behavior problems Total STJ effect on homework (8th grade) STJ 3 APS balance 3 APS plausibility 3 homework STJ 3 APS plausibility 3 homework Total STJ effect on absences (8th grade) STJ 3 feared off-track PS 3 absences STJ 3 absences STJ effects on spring 9th grade outcomes Total STJ effect on GPA (spring 9th grade) STJ 3 APS balance 3 GPA STJ 3 APS balance 3 APS plausibility 3 classroom behavior 3 GPA STJ 3 APS plausibility 3 classroom behavior problems 3 GPA STJ 3 absences 3 GPA STJ 3 feared off-track PS 3 absences 3 GPA Total STJ effect on depression (spring 9th grade) STJ 3 APS balance 3 APS plausibility 3 homework 3 depression STJ 3 APS plausibility 3 homework 3 depression STJ 3 feared off-track PS 3 absences 3 depression STJ 3 absences 3 depression

Standardized effect

p

⫺0.022 ⫺0.003 ⫺0.019 0.026 0.003 0.023 ⫺0.208 ⫺0.021 ⫺0.186

0.023 0.034 0.044 0.015 0.036 0.025 0.002 0.027 0.003

0.090 0.019 0.002 0.009 0.054 0.006 ⫺0.091 ⫺0.001 ⫺0.005 ⫺0.009 ⫺0.076

0.002 0.018 0.028 0.024 0.006 0.024 0.001 0.023 0.014 0.017 0.008

Note. N ⫽ 228 youth in the participating sample. Italicized entries are for general indirect effects; nonitalicized entries are for specific indirect effect paths.

OYSERMAN, BYBEE, AND TERRY

200

Identity Connectedness interaction, B ⫽ .21, ⌬F(1, 223) ⫽ 6.39, p ⫽ .01, added significantly to variance explained. For intervention youth, Connectedness and APS balance were positively associated, simple slope B ⫽ .13, p ⬍ .03, but for control youth, there was no significant relationship, simple slope B ⫽ ⫺.08, p ⫽ .20.

Discussion Although PSs focused on school success are common, so is failure in school. We asked why PSs fail to produce sustained self-regulation, outlining a process model in which self-regulation and, therefore behavior change, is likely when PSs are linked with contextually salient strategies, when they are congruent with social identity, and when difficulty is understood as a normative part of the process of attaining PSs. Focusing explicitly on low-income and minority youth, we developed an intervention to link PSs and strategies, to incorporate difficulty as a normative part of pursuing PSs, and to facilitate a positive link between APSs and important social identities. Our basic premises were that self-concept is not monolithic, that PSs are differentially accessible, and that they are likely to influence behaviors only when linked to strategies, when experienced as compatible with social identity, and when difficulty working toward the PS is viewed as normative. To ensure that the intervention effects would be sustained over time, the intervention occurred in school, with peers, during the school day and targeted each aspect of our process model. Structured group activities evoked academically focused PSs, made clear that APSs were held by peers (and therefore something that “we” aspire to), and highlighted the normativeness of difficulties and failures along the way to attaining PS goals. Thus, the intervention operationalized our theory of how PSs might influence self-regulatory behaviors (and through these behaviors, academic outcomes and well-being). Although attaining and sustaining an intervention effect is notoriously difficult, we documented effects that were stable and even increasing over time. This sustained effect over 2 years is particularly impressive, given that the high-poverty neighborhoods the youth were embedded in and the difficulty of improving academic outcomes when prior academic attainment accounts for much of the variance to be explained. We were able to directly assess each aspect of the process model except the positive selfregulatory consequences of changing metacognitive experience. Moreover, we documented effects using two very different analytic strategies (structural equation modeling and longitudinal MLM) with different strengths and assumptions. Structural equation modeling provided explicit test of hypothesized mediation paths via PSs to self-regulatory behavior to behavioral outcomes, using multiple sources of information to account for measurement error. MLM, by allowing us to look directly at intervention effects on trajectories of individual change as well as at specific points, provided an explicit test of the hypothesized effects over time. MLM documented that intervention effects on self-regulatory behavior (attendance, homework time, and behavior problems) and academic outcomes were stable or increased over our 2-year follow-up, while accounting for the nesting of students within homerooms. Our focus on PSs of low-income and minority youths and their fit with social identity as working class, African American, or Latino was intentional. We chose PSs and social identity because

these social identities are often assumed to be at odds with academic self-goals (Steinitz & Solomon, 1986), and school failure has important consequences, increasing risk of depression and delinquency (Joseph, 1996; Kasen, Cohen, & Brook, 1998), future unemployment, problems in parenting, and mental health problems (e.g., Stoep, Weiss, Kuo, Cheney, & Cohen, 2003) and making successful completion of other developmental tasks less likely (Sandler & Chassin, 2002). Low SES is a clear risk factor for school failure (Blair, Blair, & Madamba, 1999; Orfield et al., 2004); the combination of minority race/ethnicity and low social class together account for about 19% of variance in academic attainment (McDermott, 1995). Improving outcomes in this highrisk group is both critical and difficult given that poverty and minority status are “fixed-risk” factors, unlikely to change dramatically over the student’s middle and high school years. As a test of our model, we chose to intervene in the fall of the final year of middle school. We chose this point both because the transition from middle to high school is associated with increased salience of important social class and racial ethnic social identities and because school failure rates begin to rise dramatically from the 1st year of high school (e.g., Seidman, LaRue, Aber, Lawrence, & Feinman, 1994). The transition to high school is stressing, as can be seen in the outcomes for control group youth; without the intervention, self-regulatory behavior declines, academic outcomes erode, and risk of depression is higher. An intervention that bolsters APSs and behavioral engagement will improve outcomes in the last year of middle school, making the transition to high school less risky. In addition, middle schools are on average smaller and more homogeneous than high schools, making intervention focused on linking PSs and social identity simpler. Youth who are not equipped with specific PSs that make engagement in school (attending, putting effort into homework, engaging teachers) a necessity are more likely to fall behind in high school simply because high schools are more impersonal and less likely to routinely fulfill their educational needs. Indeed, the transition to high school has been described as a consciousnessraising experience due to the influx of students of diverse racialethnic and economic backgrounds (Seidman, Aber, Allen, & French, 1996). Ethnographic accounts suggest that low-SES teens do not realize that they are poor in homogeneously low-SES schools and neighborhoods where they appear average (Roderick, 2003; Steinitz & Solomon, 1986). Low-SES students report learning that they are part of a low-SES group when they enter high school., making the transition to high school very stressing for these youth. Not surprisingly, racial-ethnic identity also becomes more salient in the transition (Altschul, Oyserman, & Bybee, in press). Not only does the transition to high school raise consciousness of social identities and their relative standing, it also highlights ambiguities about the one’s probable future. Schools no longer provide vocational or technical training, so youths not in advanced classes are in “general studies,” implying that the point of school is either college or marking time. For all of these reasons, transition to high school is characterized by dropping grades and school involvement for low-SES and minority children (Seidman et al., 1994). We documented that our intervention does change PSs, increasing both feared off-track and academically focused PSs and strategies to attain them. As hypothesized, intervention youth both had more of these PSs and were better able to use them to improve

POSSIBLE SELVES AND ACADEMIC OUTCOMES

behavioral self-regulation. The intervention produced measurable change in PSs, and change in PSs predicted change in behavioral self-regulation— going to school rather than skipping, behaving and participating in class, and spending time on homework; selfregulation not only improved academic outcomes but, equally importantly, reduced risk of depression. The impact of academically focused PSs on self-regulatory behaviors was distinct from the impact of feared off-track PSs on self-regulatory behaviors. Youth with balanced and plausible academically focused PSs spent more time doing homework, were less disruptive, and more behaviorally engaged in class-room activities. Youth with feared off-track PSs attended school more (had fewer school absences). The distinct role of feared off-track PSs is congruent with a number of self-regulatory models: Carver (2004) describes self-regulation to avoid feared PSs or antigoals as discrepancy-enlarging self-regulation. Larsen (2004) describes the self-regulatory system as vigilant to environmental dangers; when danger is cued, individuals are more cautious about engaging in behaviors that may increase risk. Higgins (Higgins & Spiegel, 2004) describes prevention-focused self-regulation as risk averse. Following these perspectives, youth with feared off-track PSs can be expected to be cautious about risk-increasing behaviors and likely to engage in action (e.g., attending school) they perceive as antithetical to their off-track selves of becoming pregnant, involved in drugs, crime, or the police. Indeed, increased feared off-track PSs reduced risk of school absences. Although vigilant focus on antigoals or prevention is likely to reduce risk of harm, active engagement in goal attainment (discrepancy-reducing selfregulation) is likely to increase chances of success. Although self-regulatory systems theories posit roles for both discrepancy-reducing (promotion) and discrepancy-increasing (prevention) systems, prevention focus, engaging in selfregulatory behavior to avoid feared PSs, is not particularly prominent in the academic goal literature. Perhaps this is because in middle-class contexts, pursuit of academic goals may more commonly involve the discrepancy reducing feedback system— engaging in self-regulatory behavior to attain positive expected PSs. There is evidence that college students are more likely to use promotion- than prevention-focused self-regulation (e.g., Lockwood, Sadler, Fyman, & Tuck, 2004). However, prevention focused self-regulation may become more salient in a number of circumstances. First, prevention focused self-regulation may be more likely when social contextual risk is high, such as in circumstances of poverty. For example, first generation college students are more likely to engaging in strategies to avoid feared PSs than strategies to attain positive PSs (Oyserman, Gant, & Ager, 1995, Study 1). Ethnographic evidence from low-income high school students also highlights the salience of feared off-track PSs— becoming unemployed, homeless, and destitute (Steinitz & Solomon, 1986; see also Kaiser Foundation, 2002). Second, it is possible that culture influences choice of self-regulatory system; Euro-Canadian college students find promotion-focused strategies compelling, whereas Asian Canadian college students find prevention-focused strategies compelling (Lockwood, Marshall, & Sadler, 2005). Similarly, Hong Kong Chinese college students found prevention-focused reasons for action more convincing than did American college students (Lee, Aaker, & Gardner, 2000, Study 4).

201

Whether focused on positive expected or feared PSs, selfregulation fails when individuals do not realize that a particular action is antithetical to goal achievement, when PSs do not provide clear standards of what to attain or avoid, or when PSs are not linked with self-regulatory behaviors (e.g., Carver, 2004; Higgins, 1987). For low-income teens, lapses in self-regulation may be difficult to repair—when risk of failure is high, any misstep can spell disaster. This contrasts with the situation of middle-class teens whose self-regulatory lapses can be compensated for by contextual regulation set in place by neighborhood, school, and parents. Middle-class students are more likely to be provided mentoring, tutoring, monitoring, and enrichment activities whether they seek them out or not (e.g., Sampson, Morenoff, & Earls, 1999). Low-income students are more likely to live in contexts lacking such collective efficacy resources (Sampson et al., 1999). Thus, for low-income youths, self-initiated engagement in selfregulation that focuses on attaining positive PSs and avoiding negative PSs is likely necessary. Our emphasis on the situation of low-income and minority youth is congruent with research using the stereotype threat model (Steele, 1997) that demonstrates the undermining effect on academic outcomes of making stigmatized in-group social identity salient. To date, stereotype threat research has not documented the mechanism by which stereotype threat produces negative consequences over time. Our results suggest that stereotype threat may undermine academic attainment by making APSs less salient and school-focused strategies less accessible as ways of avoiding offtrack PSs. This process model is congruent with a number of studies documenting that when stereotype threat is activated, prevention (discrepancy enhancing) focus increases (Oyserman, Uskul, Yoder, Nesse, & Williams, 2005; Seibt & Fo¨rster, 2004), as do negative thoughts about one’s math capacity (Cadinu, Maass, Rosabianca, & Kiesner, 2005). Our results demonstrate the real-world power of a social psychological conceptualization of the self as a motivational resource. By integrating PSs, social identity and metacognitive perspectives, we developed a process model that, when operationalized, produced lasting change on PSs, self-regulation, academic outcomes, and depression. These results are particularly promising given the difficulty in producing sustained improvement in the educational outcomes of low-income youths and the great need to develop integrative social science models that can be applied to ameliorating this large-scale social problem.

References Abrams, D. (1994). Social self-regulation. Personality and Social Psychology Bulletin, 20, 473– 483. Aiken, L., & West, S. (1991). Multiple regression: Testing and interpreting interactions. Newbury Park, CA: Sage. Altschul, I., Oyserman, D., & Bybee, D. (in press). Racial-ethnic identity in mid-adolescence: Content and change as predictors of academic attainment. Child Development. Ambady, N., Paik, S., Steele, J., Owen-Smith, A., & Mitchell, J. (2004). Deflecting negative self-relevant stereotype activation: The effects of individuation. Journal of Experimental Social Psychology, 40, 401– 408. Arbuckle, J., & Wothke, W. (2003). AMOS 5.0 user’s guide. Chicago: SmallWaters. Arroyo, C., & Zigler, E. (1995). Racial identity, academic achievement, and the psychological well-being of economically disadvantaged adolescents. Journal of Personality and Social Psychology, 69, 903–914.

202

OYSERMAN, BYBEE, AND TERRY

Baumeister, R. (1998). The self. In D. Gilbert, S. Fiske, & G. Lindzey (Eds.), Handbook of social psychology (4th ed., pp. 680 –740). New York: McGraw-Hill. Belli, R. (1998). The structure of autobiographical memory and the event history calendar: Potential improvements in the quality of retrospective reports in surveys, Memory, 6, 383– 406. Blair, S., Blair, M., & Madamba, A. (1999). Racial/ethnic differences in high school students’ academic performance: Understanding the interweave of social class and ethnicity in the family context. Journal of Comparative Family Studies, 30, 539 –555. Burke, P. (2003). Relationships among multiple identities. In P. Burke & T. Owens (Eds.), Advances in identity theory and research (pp. 195– 214). New York: Kluwer Academic/Plenum Press. Cadinu, M., Maass, A., Rosabianca, A., & Kiesner, J. (2005). Why do women underperform under stereotype threat? Evidence for the role of negative thinking. Psychological Science, 16, 572–578. Carver, C. (2004). Self-regulation of action and affect. In R. Baumeister & K. Vohs (Eds.), Handbook of self-regulation: Research, theory and applications (pp. 13–39). New York: Guilford Press. Carver, C., & Scheier, M. (1990). Principles of self-regulation: Action and emotion. In T. Higgins & R. Sorrentinoeds (Eds.), Handbook of motivation and cognition (Vol. 2, pp. 3–52). New York: Guilford Press. Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Erlbaum. Cook, P., & Ludwig, J. (1997). Weighing the “burden of acting White”: Are there race differences in attitudes toward education. Journal of Policy Analysis and Management, 16, 256 –278. Costello, E., Mustillo, S., & Erkanli, A. (2003). Prevalence and development of psychiatric disorders in childhood and adolescence. Archives of General Psychiatry, 60, 837– 844. Cribbie, R., & Jamieson, J. (2000). Structural equation models and the regression bias for measuring correlates of change. Educational and Psychological Measurement, 60, 893–907. Croizet, J., & Claire, T. (1998). Extending the concept of stereotype and threat to social class: The intellectual underperformance of students from low socioeconomic backgrounds. Personality and Social Psychology Bulletin, 24, 588 –594. Detroit News (2005, May 29). Graduation rates. Retrieved September 15, 2005, from http://www.detnews.com/2005/specialreport/0505/30/ A01-196697.htm. Eccles, J., & Wigfield, A. (2002). Motivational beliefs, values, and goals. Annual Review of Psychology, 53, 109 –132. Ellickson, P., & Hawes, J. (1989). An assessment of active versus passive methods for obtaining parental consent. Evaluation Review, 13, 45–55. Ferguson, R. (1998). Comment on the burden of acting White: Do Black adolescents disparage academic achievement? by P. J. Cook and J. Ludwig. In C. Jencks & M. Phillips (Eds.), The Black-White test score gap (pp. 394 –397). Washington, DC: Brookings Institution. Finn, J., Pannozzo, G., & Voelkl, K. (1995). Disruptive and inattentivewithdrawn behavior and achievement among fourth graders. The Elementary School Journal, 95, 421– 434. Fordham, S., & Ogbu, J. (1986). Black students’ school success: Coping with the “burden of acting white.” Urban Review, 18, 176 –206. Graham, S. (2001). Inferences about responsibility and values: Implication for academic motivation. In F. Salili, & C. Chiu (Eds.), Student motivation: The culture and context of learning (pp. 31–59). Dordrecht, Netherlands: Kluwer Academic. Higgins, T. (1987). Self-discrepancy: A theory relating self and affect. Psychological Review, 94, 319 –340. Higgins, T. (1996a). The “self digest”: Self-knowledge serving selfregulatory functions. Journal of Personality and Social Psychology, 71, 1062–1083. Higgins, T. (1996b). Knowledge activation: Accessibility, applicability,

and salience. In T. Higgins & A. Kruglanski (Eds.), Social psychology, handbook of basic principles (pp. 133–168). New York: Guilford Press. Higgins, T. (1997). Beyond pleasure and pain. American Psychologist, 52, 1280 –1300. Higgins, T. (1998). The aboutness principle: A pervasive influence on human inference. Social Cognition, 16, 173–198. Higgins, T., & Spiegel, S. (2004). Promotion and prevention strategies for self-regulation: A motivated cognition perspective. In R. Baumeister & K. Vohs (Eds.), Handbook of self-regulation: Research, theory and applications (pp. 171–187). New York: Guilford Press. Hoyle, R., & Kenny, D. (1999). Sample size, reliability, and tests of statistical mediation. In R. Hoyle (Ed.), Statistical strategies for small sample research (pp. 195–222). Thousand Oaks, CA: Sage. Hu, L., & Bentler, P. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6, 1–55. James, W. (1890). Principles of psychology. New York: Holt. Jo, B. (2002a). Estimation of intervention effects with noncompliance: Alternative model specifications. Journal of Educational and Behavioral Statistics, 27, 385– 409. Jo, B. (2002b). Model misspecification sensitivity analysis in estimating causal effects of interventions with noncompliance. Statistics in Medicine, 25, 305–327. Joseph, J. (1996). School factors and delinquency: A study of African American youths. Journal of Black Studies, 26, 340 –355. Juster, T., Ono, H., & Stafford, F. (2004). Changing times of American youth, 1981–2003. Retrieved September 14, 2005, from the University of Michigan Institute for Social Research Child Development Supplement to the Panel Study for Income Dynamics: http://www.umich.edu/ news/Releases/2004/Nov04/teen_time_report.pdf Kaiser Foundation (2002). National Survey of Latinos. Summary of findings. Retrieved September 14, 2005, from http://www.kff.org/ kaiserpolls/20021217a-index.cfm Kao, G. (2000). Group images and possible selves among adolescents: Linking stereotypes to expectations by race and ethnicity. Sociological Forum, 15, 407– 430. Kasen, S., Cohen, P., & Brook, J. (1998). Adolescent school experiences and dropout, adolescent pregnancy, and young adult deviant behavior. Journal of Adolescent Research, 13, 49 –72. Kearney, K., Hopkins, R., Mauss, A., & Weisheit, R. (1983). Sample bias resulting from a requirement for written parental consent. Public Opinion Quarterly, 47, 96 –102. King, L. (2001). The health benefits of writing about life goals. Personality and Social Psychology Bulletin, 27, 798 – 807. King, L., & Smith, N. (2004). Gay and straight possible selves: Goals, identity, subjective well-being, and personality development. Journal of Personality, 72, 967–994. Larsen, R. (2004, November). Emotion and cognition: The case of automatic vigilance. APA online. Retrieved September 14, 2005, from http:// www.apa.org/science/psa/sb-larsen.html Lee, A., Aaker, J., & Gardner, W. (2000). The pleasures and pains of distinct self-construals: The role of interdependence in regulatory focus. Journal of Personality and Social Psychology, 78, 1122–1134. Little, R., & Rubin, D. (2002). Statistical analysis with missing data (2nd ed). New York: Wiley. Little, R., & Yau, L. (1998). Statistical techniques for analyzing data from prevention trials: Treatment of no-shows using Rubin’s causal model. Psychological Methods, 3, 147–159. Lockwood, P., Marshall, T., & Sadler, P. (2005). Promoting success or preventing failure: Cultural differences in motivation by positive and negative role models. Personality and Social Psychology Bulletin 31, 379 –392. Lockwood, P., Sadler, P., Fyman, K., & Tuck, S. (2004). To do or not to

POSSIBLE SELVES AND ACADEMIC OUTCOMES do: Using positive and negative role models to harness motivation. Social Cognition, 22, 422– 450. MacCallum, R., Browne, M., & Sugawara, H. (1996). Power analysis and determination of sample size for covariance structure modeling. Psychological Methods, 1, 130 –149. MacKinnon, D., Lockwood, C., Hoffman, J., West, S., & Sheets, V. (2002). A comparison of methods to test mediation and other intervening variable effects. Psychological Methods, 7, 83–104. MacKinnon, D., Lockwood, C., & Williams, J. (2004). Confidence limits for the indirect effect: Distribution of the product and resampling methods. Multivariate Behavioral Research, 39, 99 –128. MacKinnon, D., Warsi, G., & Dwyer, J. (1995). A simulation study of mediated effect measures. Multivariate Behavioral Research, 30, 41– 62. Mardia, K. (1985). Mardia’s test of multinormality. In S. Kotz & N. Johnson (Eds.), Encyclopedia of statistical sciences (Vol. 5, pp. 217– 221). New York: Wiley. Maris, E. (1998). Covariance adjustment versus gain scores: revisited. Psychological Methods, 3, 309 –327. McDermott, P. (1995). Sex, race, class, and other demographics as explanations for children’s ability and adjustment: A national appraisal. Journal of School Psychology, 33, 75–91. McGuire, W., & Padawe-Singer, A. (1976). Trait salience in the spontaneous self-concept. Journal of Personality and Social Psychology, 33, 743–754. Orfield, G., Losen, D., Wald, J., & Swanson, C. (2004). Losing our future: How minority youth are being left behind by the graduation rate crisis. Cambridge, MA: The Civil Rights Project at Harvard University. Retrieved September 14, 2005, from http://www.urban.org/UploadedPDF/ 410936_LosingOurFuture.pdf Oyserman, D. (2001). Self-concept and identity. In A. Tesser & N. Schwarz (Eds.), Blackwell handbook of social psychology (pp. 499 – 517). Malden, MA: Blackwell Press. Oyserman, D. (in press.) Social identity and self-regulation. In A. Kruglanski & T. Higgins, Social psychology: Handbook of basic principles. New York: Guilford Press. Oyserman, D., Bybee, D., Terry, K., & Hart-Johnson, T. (2004). Possible selves as roadmaps. Journal of Research in Personality, 38, 130 –149. Oyserman, D., & Fryberg, S. A. (2006). The possible selves of diverse adolescents: Content and function across gender, race, and national origin. In C. Dunkel & J. Kerpelman (Eds.), Possible selves: Theory, research, and application (pp. 17–39). Huntington, NY: Nova. Oyserman, D., Gant, L., & Ager, J. (1995). A socially contextualized model of African American identity: Possible selves and school persistence. Journal of Personality and Social Psychology, 69, 1216 –1232. Oyserman, D., Johnson, E., & Bybee, D. (2006). Effects of low socioeconomic status on early adolescent possible selves. Manuscript under editorial review. Oyserman, D., & Markus, H. (1990). Possible selves and delinquency. Journal of Personality and Social Psychology, 59, 112–125. Oyserman, D., & Saltz, E. (1993). Competence, delinquency, and attempts to attain possible selves. Journal of Personality and Social Psychology, 65, 360 –374. Oyserman, D., Terry, K., & Bybee, D. (2002). A possible selves intervention to enhance school involvement. Journal of Adolescence, 24, 313– 326. Oyserman, D., Uskul, A., Yoder, N., Nesse, R., & Williams, D. (2005). Unfair treatment and self-regulatory focus. Manuscript submitted for editorial review. Pokorny, S., Jason, L., Schoeny, M., Townsend, S., & Curie, C. (2001). Do participation rates change when active consent procedures replace passive consent. Evaluation Review, 25, 567–580. Radloff, L. (1977). The CES-D scale: A self-report depression scale for research in the general population. Applied Psychological Measurement, 1, 385– 401.

203

Radloff, L., & Locke, B. (2000). Center for Epidemiologic Studies Depression Scale (CESD). In Taskforce for the Handbook of Psychiatric Measures (Eds.), Handbook of psychiatric measures: American Psychiatric Association (pp. 523–526). Washington, DC: American Psychiatric Association. Raudenbush, S., & Bryk, A. (2002). Hierarchical linear models: Applications and data analysis methods (2nd ed.). Thousand Oaks, CA: Sage. Raudenbush, S., & Liu, X. (2000). Statistical power and optimal design for multisite randomized trials. Psychological Methods, 5, 199 –213. Raudenbush, S., & Liu, X. (2001). Effects of study duration, frequency of observation, and sample size on power in studies of group differences in polynomial change. Psychological Methods, 6, 387– 401. Re´gner, I., Huguet, P., & Monteil, J. (2002). Effects of socioeconomic status (SES) information on cognitive ability inferences: When low-SES students make use of a self-threatening stereotype. Social Psychology of Education, 5, 253–269. Roberts, R., Attkisson, C., & Rosenblatt, A. (1998). Prevalence of psychopathology among children and adolescents. American Journal of Psychiatry, 155, 715–725. Roderick, M. (2003). What’s happening to the boys? Early high school experiences and school outcomes among African American male adolescents in Chicago. Urban Education, 38, 538 – 607. Rothman, A., & Schwarz, N. (1998). Constructing perceptions of vulnerability: Personal relevance and the use of experiential information in health judgments. Personality and Social Psychology Bulletin, 24, 1053–1064. Ruvolo, A., & Markus, H. (1992). Possible selves and performance: The power of self-relevant imagery. Social Cognition, 10, 95–124. Sampson, R., Morenoff, J., & Earls, F. (1999). Beyond social capital: Spatial dynamics of collective efficacy for children. American Sociological Review, 64, 633– 660. Sandler, I., & Chassin, L. (2002). Training of prevention researchers: Perspectives from the Arizona State University Prevention Research Training Program. Prevention & Treatment, 5, Item 5. Retrieved April 10, 2006, from http://journals.apa.org/prevention/volume5/pre0050006a.html 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. Schwarz, N. (2002). Situated cognition and the wisdom of feelings: Cognitive tuning. In L. Feldman Barrett & P. Salovey (Eds.), The wisdom in feelings (pp. 144 –166). New York: Guilford Press. Schwarz, N., & Bless, H. (1992). Constructing reality and its alternatives: An inclusion/exclusion model of assimilation and contrast effects in social judgment. In: L. Martin & A. Tesser (Eds.), The construction of social judgments (pp. 217–245). Hillsdale, NJ: Erlbaum. Schwarz, N., & Clore, G. (1996). Feelings and phenomenal experiences. In: T. Higgins & A. Kruglanski (Eds.), Social psychology: A handbook of basic principles. New York: Guilford Press. Seibt, B., & Fo¨rster, J. (2004). Stereotype threat and performance: How self-stereotypes influence processing by inducing regulatory foci. Journal of Personality and Social Psychology, 87, 38 –56. Seidman, E., Aber, J., Allen, L., & French, S. (1996). The impact of the transition to high school on the self-esteem and perceived social context of poor urban youth. American Journal of Community Psychology, 24, 489 –515. Seidman, E., Allen, L., Aber, J. L., Mitchell, C., & Feinman, J. (1994). The impact of school transitions in early adolescence on the self-system and perceived social context of poor urban youth. Child Development, 65, 507–522. Settles, I. H. (2004). When multiple identities interfere: The role of identity centrality. Personality and Social Psychology Bulletin, 30, 487–500. Snijders, T., & Bosker, R. (1999). Multilevel analysis: An introduction to basic and advanced multilevel modeling. London: Sage.

204

OYSERMAN, BYBEE, AND TERRY

Steele, C. (1997). A threat in the air: How stereotypes shape intellectual identity and performance. American Psychologist, 52, 613– 629. Steinitz, V., & Solomon, E. (1986). Starting out: Class and community in the lives of working-class youth. Philadelphia: Temple University Press. Stoep, A., Weiss, N., Kuo, E., Cheney, D., & Cohen, P. (2003). What proportion of failure to complete secondary school in the U.S. population is attributable to adolescent psychiatric disorder? Journal of Behavioral Health Services and Research, 30, 119 –124. Strauman, T. (2002). Self-regulation and depression. Self & Identity, 1, 151–157. Strauman, T., & Higgins, T. (1988). Self-discrepancies as predictors of vulnerability to distinct syndromes of chronic emotional distress. Journal of Personality, 56, 685–707.

Thomas, D., Townsend, T., & Belgrave, F. (2003). The influence of cultural and racial identification on the psychosocial adjustment of inner-city African American children in school. American Journal of Community Psychology, 32, 217–228. U.S. Bureau of the Census (2000). Census 2000. Retrieved April 10, 2006, from http://www.census.gov Weinger, S. (2000). Opportunities for career success: Views of poor and middle-class children. Children & Youth Services Review, 22, 13–35. Weinstein, N. (1993). Testing four competing theories of health-protective behavior. Health Psychology, 12, 324 –333.

Received December 22, 2004 Revision received January 4, 2006 Accepted January 24, 2006 䡲

Journal of Personality and Social Psychology 2006, Vol. 91, No. 1, 33– 48

Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.33

Thinking Within the Box: The Relational Processing Style Elicited by Counterfactual Mind-Sets Laura J. Kray

Adam D. Galinsky

University of California, Berkeley

Northwestern University

Elaine M. Wong University of California, Berkeley By comparing reality to what might have been, counterfactuals promote a relational processing style characterized by a tendency to consider relationships and associations among a set of stimuli. As such, counterfactual mind-sets were expected to improve performance on tasks involving the consideration of relationships and associations but to impair performance on tasks requiring novel ideas that are uninfluenced by salient associations. The authors conducted several experiments to test this hypothesis. In Experiments 1a and 1b, the authors determined that counterfactual mind-sets increase mental states and preferences for thinking styles consistent with relational thought. Experiment 2 demonstrated a facilitative effect of counterfactual mind-sets on an analytic task involving logical relationships; Experiments 3 and 4 demonstrated that counterfactual mind-sets structure thought and imagination around salient associations and therefore impaired performance on creative generation tasks. In Experiment 5, the authors demonstrated that the detrimental effect of counterfactual mind-sets is limited to creative tasks involving novel idea generation; in a creative association task involving the consideration of relationships between task stimuli, counterfactual mind-sets improved performance. Keywords: counterfactuals, mind-sets, relational processing, analytical, creativity

In the present article, we examined the impact of counterfactual thinking on subsequent thinking styles and problem solving. Broadly speaking, we explored how reflecting back on events in which an outcome almost turned out differently impacts future problem solving. More precisely, we explored the nature of the mind-set that results from constructing counterfactual thoughts. Because counterfactuals involve a consideration of both reality and what might have been, they are inherently relational in nature. Our central thesis is that constructing counterfactual thoughts in one context produces a counterfactual mind-set characterized by a tendency to process information relationally in subsequent contexts. Thus, the mental structure of logical relationships created through counterfactual thinking increases the ability to understand and perceive relationships in subsequent contexts, structuring thought around salient associations and the pursuit of connections. Because the counterfactual mind-set occurs regardless of the content or valence of the counterfactual thoughts (Galinsky & Kray, 2004; Galinsky & Moskowitz, 2000; Kray & Galinsky, 2003), what lingers is the form and structure of the counterfactual (i.e., the tendency to consider relationships and connections between a set of stimuli). Although previous research has explored the consequences of the counterfactual mind-set on subsequent performance across a variety of tasks, the present research is the first to identify the mental states and thinking styles that counterfactuals activate and the mechanisms by which counterfactual mind-sets have their effects. We characterize the counterfactual mind-set as a structured form of thought involving a consideration of relationships and

Whenever individuals consider how the past might have turned out differently, they are engaging in counterfactual thinking. For example, people who ponder, “What would life be like if I had married that other person?” or “Would I be better off if I had selected that other job?” are implicitly comparing reality with what might have been. Thoughts of “if only” and “what if” are signposts for counterfactual musings, and their presence in mental life is both pervasive and predictable. A growing body of literature suggests that counterfactuals are not merely fodder for daydreamers stuck in the past but rather serve important functions for directing future behavior (Roese & Olson, 1995).

Laura J. Kray and Elaine M. Wong, Haas School of Business, University of California, Berkeley; Adam D. Galinsky, Kellogg Graduate School of Management, Northwestern University. Elaine M. Wong is now at the School of Communication, Northwestern University. This research was supported in part by National Science Foundation Grants SES-0233294 and SES-0136931. We gratefully acknowledge insightful comments provided by Phil Tetlock on an earlier version of this article and the valuable feedback provided by participants in colloquia at the Haas School of Business, Carnegie Mellon University, and the University of Chicago. We are indebted to the many research assistants who helped out in every stage of the research process. In particular, we are grateful to Joyce Chen, Joe Gacula, Linda Pham, and Alissa Roberts. Correspondence concerning this article should be addressed to Laura J. Kray, Haas School of Business, 545 Student Services Building, #1900, Berkeley, CA 94720. E-mail: [email protected]

33

34

KRAY, GALINSKY, AND WONG

associations between a set of stimuli. We expect this processing style to have three consequences. First, it should create phenomenological experiences and preferences for structured thought (Sternberg, 1988). Second, it should promote lay conceptions of analytic thought, defined as “an examination of a complex, its elements and their relations” (Merriam-Webster, 2005). Third, it should promote structured imagination, or the tendency to build on existing knowledge structures (Ward, 1994). We have chosen the term relational processing style to capture the essence of the counterfactual mind-set because it both describes the nature of counterfactual thoughts and because it is broad enough to encompass the full range of effects produced by the activation of a counterfactual mind-set.

The Conceptual Link Between Counterfactuals and Relational Processing Although the ability to undo events and construct possible worlds is theoretically unlimited, in reality, when and how counterfactuals are constructed is fairly predictable. The commencement of counterfactual thinking is often initiated when an event nearly occurred (Kahneman & Miller, 1986; Roese & Olson, 1997). For example, missing a plane by 5 min tends to evoke more counterfactuals than missing a plane by 1 hr, presumably because the former case more readily conjures up elements of the day leading up to the flight that could have been undone. In addition to near misses, “abnormal” events tend to produce counterfactual thoughts. For example, it is easier to undo missing one’s flight when a new, atypical route to the airport was taken than it is after taking one’s usual route. It is often the mere presence of an obvious mutable component to an event that leads to the spontaneous generation of counterfactuals. Just like the rules that govern when counterfactuals are generated, how counterfactual thoughts are constructed is predictable. Counterfactuals involve a comparison of the relationship between reality and what might have been. Constructing a counterfactual thought implicitly involves laying out a causal chain of events in an action sequence and mutating one step in the process to construct an alternate reality. As such, running a counterfactual simulation in one’s head is the mental equivalent of conducting an experiment. Like the experimental process, counterfactual thinking involves a logical consideration of relationships and causal associations between events (Einhorn & Hogarth, 1986; Mandel, 2003; Mandel & Lehman, 1996; Wells & Gavanski, 1989). For example, Wells and Gavanski showed that an initial event was judged to be causally connected to a subsequent event to the extent that a mutation to the initial event would have undone the occurrence of the subsequent event. As described above, a logical structure underlies when and how counterfactuals are created. Counterfactuals are most likely when potential mutations to a sequence of events are salient or when potential alternative worlds are close in time and space. Implicit comparisons assessing similarities and differences between the alternate world and reality are then made that facilitate the identification of causal connections (Markman & McMullen, 2003). Broadly speaking, relational processing is used in counterfactual thinking in that it involves the consideration of relationships and connections between events. We expect that once the counterfac-

tual mind-set is activated, it persists because the relational processing style that accompanies it is a well-learned and useful strategy for assessing causality, a critical tool for comprehending the world (cf. Kelley, 1973). Past research has demonstrated that situations with an obvious mutable component tend to elicit counterfactual thoughts, which then orient cognition in subsequent contexts. This cognitive orientation has been called a “counterfactual mind-set” (Galinsky & Kray, 2004; Kray & Galinsky, 2003). The idea that the activation of a counterfactual mind-set impacts subsequent cognition and performance has received strong support across a variety of individual and group tasks. More specifically, counterfactual mindsets appear to be an asset when the consideration of alternatives facilitates performance (Galinsky & Kray, 2004; Galinsky & Moskowitz, 2000; Kray & Galinsky, 2003). Counterfactual mind-sets also appear to encourage skepticism about the dominant hypothesis. For example, having just considered an alternate reality can reduce the confirmation bias or the tendency to seek information that is consistent with an existing hypothesis to the relative detriment of information that could potentially disconfirm it (Galinsky & Moskowitz, 2000; Kray & Galinsky, 2003). In fact, the number of counterfactual thoughts generated in response to a mutable scenario has been shown to be an important predictor of subsequent disconfirmatory information search (Kray & Galinsky, 2003). Counterfactual mind-sets also improve decision accuracy by increasing the discussion of unique information critical for group decision making and promoting synergistic coordination, or the tendency of group members to build on and develop relationships between each other’s ideas (Galinsky & Kray, 2004; Liljenquist, Galinsky, & Kray, 2004). Finally, counterfactual thinking has been shown to increase the scrutiny of persuasive message content (Krishnamurthy & Sivaraman, 2002), enabling decision makers to distinguish between strong and weak arguments. Despite the diversity of tasks explored in this growing body of literature, a common facilitative effect has emerged after the activation of a counterfactual mind-set. Overall, it appears that by considering alternative realities in one context, greater clarity regarding task associations is gained in later contexts. Although the counterfactual mind-set has been described generally as promoting “a consideration of alternatives” (Galinsky & Moskowitz, 2000), in the present investigation, we seek to better understand the content of and processing style associated with this counterfactual mind-set and the mechanisms by which it impacts performance.

Research Overview Despite numerous demonstrations that counterfactual mind-sets affect subsequent problem solving, to date, our understanding of the exact nature of a counterfactual mind-set has been limited because little effort has been made to explore the underlying process by which counterfactual mind-sets influence performance. Although previous research has examined the content and structure of counterfactual thoughts (Roese, 1994; Roese & Olson, 1995), no research to date has explored the phenomenological experience produced by counterfactual thinking. We argue that the mind-set promotes a relational processing style, characterized by a tendency to ponder associations and make connections between a set of stimuli. As such, we suggest that counterfactual mind-sets will

COUNTERFACTUAL MIND-SETS

improve performance on analytic tasks, which typically require that one identify and understand the relationships among a set of stimuli. In seeking to understand the unique characteristics of the counterfactual mind-set, we also explored its impact on two types of creative tasks. Our hypothesis that counterfactual mind-sets promote a relational processing style suggests that they might lead to imaginative processes that build from existing knowledge structures. Ward (1994) termed the tendency to rely on existing knowledge in the creative process as “structured imagination.” Although Ward demonstrated that structured imagination is fairly characteristic of how creative tasks are approached in general, we argue that this tendency is intensified by the activation of a counterfactual mind-set. Just as the generation of counterfactual thoughts are structured and conform to certain rules and logic, we contend that imagination following the activation of a counterfactual mind-set is structured around salient knowledge structures. Therefore, counterfactual mind-sets should have a positive effect on creative tasks that require the identification of associations within and between a set of stimuli. However, if a counterfactual mind-set consists of a relational processing style that structures imagination, then it might hinder the generation of novel ideas. This increased attention to associations among task stimuli may actually decrease one’s ability to “think outside the box.” Each of the findings described above is consistent with the basic hypothesis that counterfactual mind-sets foster not just a consideration of alternatives, as previous research has suggested, but more precisely a consideration of relationships and connections between or among a set of stimuli, structuring thought and imagination around these associations. The present set of experiments was conducted to test this hypothesis. In Experiment 1a and 1b, we examined whether counterfactual mind-sets promote a phenomenological experience consistent with a relational processing style. In Experiment 2, we examined whether counterfactual mind-sets improve performance on an analytic task requiring the consideration of relationships and associations. Specifically, we examined performance on a standardized test designed to assess the analytic reasoning skills of potential graduate school applicants. In Experiments 3 and 4, we explored the impact of counterfactual mind-sets on structured imagination. In Experiment 5, we examined the relative impact of counterfactual mind-sets on creative generation versus creative association tasks, which generally differ in the optimal degree of conceptual attention devoted to relationships and associations between task stimuli. Overall, the studies presented here clarify the nature and phenomenology of the counterfactual mind-set by demonstrating that activating a counterfactual mindset promotes a relational processing style and therefore has predictable effects, both beneficial and detrimental, on a range of problem-solving tasks.

Experiment 1a: Counterfactual Mind-Sets and Mental States The purpose of Experiment 1a was twofold. First, we sought to better understand how a counterfactual mind-set is experienced at a phenomenological level. As described above, we expected that the mental structure of logical relationships created through counterfactual thought would promote a relational processing style. In

35

considering how this processing style might translate into a mental state, we reasoned that an individual who is primed to consider logical relationships should feel poised for analytic and critical thinking. A second goal was to rule out the possibility that a third variable is responsible for the effect of counterfactual mind-sets on performance. Because analytic thinking is affected by mood (Schwarz & Bless, 1991), it is possible that counterfactual mind-sets simply depress moods. Previous research has addressed this possibility by manipulating whether participants generate downward versus upward counterfactuals (Roese, 1994). Downward counterfactuals, or thoughts about how events could have been worse, tend to evoke positive feelings such as joy, relief, and surprise. Upward counterfactuals, or thoughts about how events could have been better, tend to evoke negative feelings such as regret, remorse, and disappointment. Because upward and downward counterfactuals tend to evoke different emotional experiences, the fact that the valence of the counterfactual thoughts has not moderated any of the findings to date on counterfactual mind-sets and problem solving (Galinsky & Moskowitz, 2000; Kray & Galinsky, 2003) suggests that affect and counterfactual mind-sets have independent effects on problem solving. In the present experiment, in addition to manipulating the valence of the scenario, we also measured mood to determine whether it operates as a mediator. Recognizing that common psychological measures of affect rely on selfassessments (cf. Watson, Clark, & Tellegen, 1988), we compared self-reported cognitive and affective states after engaging in counterfactual thinking.

Method Design. The experiment had a 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence of prime: positive vs. negative) betweensubjects factorial design. Participants. Participants were 65 students on a large western university campus. Participants were approached in several cafes on campus and asked to complete a short questionnaire. In return for their compliance, they were given a pencil bearing the university’s logo. Procedure. Participants read a scenario that described the actions of Jane, a woman who was attending a rock concert (see Galinsky & Moskowitz, 2000). In each scenario, an individual at the rock concert wins a valuable prize, a trip to Hawaii. In half of the scenarios, Jane is the winner of the prize (positive valence); in the other half of the scenarios, Jane is not the winner of the prize (negative valence). In addition, half of the scenarios describe a sequence of events designed to elicit counterfactual thoughts, and half of the scenarios describe a sequence of events that is not expected to elicit counterfactual thoughts. In the downward counterfactual scenario, Jane wins the trip to Hawaii when the new seat she had just switched to (in order to get a better view of the stage) was chosen. In the upward counterfactual scenario, Jane loses the trip to Hawaii when the seat that she had just switched from wins the trip. In the noncounterfactual conditions, Jane does not switch seats. After reading one of the four scenarios, participants were asked to consider some thoughts likely to be running through Jane’s mind. After considering the scenario, participants were asked to “Please indicate the extent to which your current mental state is characterized by the following attributes.” This measure allowed us to evaluate the degree to which participants’ current cognitive state was consistent with a sense of being analytic. Specifically, participants’ cognitive states were assessed

KRAY, GALINSKY, AND WONG

36

along the following dimensions (␣ ⫽ .85): analytic, critical, focused, and smart.1 We also sought to distinguish the impact of counterfactual thinking on cognitive states from its impact on affective states. As affective experience can be distinguished along valence-based and arousal-based dimensions (Feldman, 1995; Russell, 1980), we measured both of these affective responses separately. To measure valence-based affect, we included a two-item mood measure (␣ ⫽ .80): mood and happy. To measure arousalbased affect, we simply asked participants to assess their perceived arousal. Mood was measured on a 9-point scale ranging from 1 (very negative) to 9 (very positive). All other assessments were done on 9-point scales ranging from 1 (not at all characteristic) to 9 (very characteristic).

Results and Discussion To test our hypotheses regarding the cognitive and affective states elicited by a counterfactual mind-set, we conducted 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence of prime: positive vs. negative) between-subjects analyses of variance (ANOVAs). Table 1 provides a correlation matrix of all experimental variables. Cognitive state. We hypothesized that the process of thinking counterfactually would elicit a sense of being analytic. Consistent with our hypothesis, participants in a counterfactual mind-set (M ⫽ 5.8, SD ⫽ 1.6) rated their cognitive state as being more consistent with a sense of being analytic than did participants exposed to a noncounterfactual scenario (M ⫽ 4.3, SD ⫽ 1.9), F(1, 61) ⫽ 14.12, p ⬍ .001.2 No other effects emerged as statistically significant. Valence-based affective experience. Although we did not expect our counterfactual manipulation to affect participants’ affective experience, we explored whether the protagonist winning or losing the valuable prize would affect participants’ emotional reaction. Participants exposed to a positive outcome scenario indeed reported a more positive emotional experience (M ⫽ 6.9, SD ⫽ 1.5) than did participants exposed to a negative outcome scenario (M ⫽ 5.39, SD ⫽ 2.1), F(1, 61) ⫽ 8.43, p ⫽ .005. Neither the effect for type of prime, F(1, 61) ⫽ 1.2, p ⫽ .28, nor the interaction between type and valence of prime, F(1, 61) ⫽ 0.08, p ⫽ .78, were statistically significant. Arousal-based affective experience. Like affect, we expected that only the valence of the outcome in the rock concert scenario would impact participants’ perceived level of arousal. Consistent with this hypothesis, a main effect for valence of prime emerged as statistically significant, F(1, 61) ⫽ 5.57, p ⫽ .02. Participants who read the scenario in which a positive outcome occurred (M ⫽ 5.37, SD ⫽ 2.97) reported feeling more aroused than participants who

Table 1 Experiment 1a: Correlations Between Variables Variable 1. 2. 3. 4. 5.

Type of prime Valence of prime Cognitive state Valence-based affective state Arousal-based affective state

1

2

3

4

5

— .04 ⴚ.41 .14 ⫺.10

— .08 ⴚ.34 ⴚ.28

— ⫺.20 ⫺.20

— .12

—

Note. Significant correlations ( p ⬍ .05) are in boldface.

read a scenario in which a negative outcome occurred (M ⫽ 3.63, SD ⫽ 2.7). An unanticipated two-way interaction between valence of prime and type of prime also emerged, F(1, 61) ⫽ 4.17, p ⫽ .05. In the noncounterfactual condition, participants reported more arousal after the positive outcome (M ⫽ 6.22, SD ⫽ 2.86) relative to the negative outcome (M ⫽ 2.96, SD ⫽ 2.54), t(31) ⫽ 3.17, p ⫽ .01; however, in the counterfactual condition, the difference between the two outcomes (M ⫽ 4.6, SD ⫽ 2.99 vs. M ⫽ 4.4, SD ⫽ 2.72) was not statistically significant, t(30) ⫽ 0.22, ns. No other effects emerged as statistically significant. The main purpose of this study was to provide support for our assertion that counterfactual mind-sets create a phenomenological experience consistent with a relational processing style. Consistent with this hypothesis, participants in the counterfactual condition reported feeling more poised for critical thinking than did baseline participants. After having engaged in relational processing through the construction of counterfactual thoughts, participants in the resulting mind-set reported more affinity for analytic thinking. Another goal of this study was to bolster support for the assertion that the impact of counterfactual mind-sets on cognitive processes operates independently of emotional experiences. The fact that the counterfactual manipulation had no effect on participants’ reported affective state casts serious doubt on the possibility that mood accounts for the relationship between counterfactual mind-sets and relational processing. Whereas the counterfactual mind-set had no effect on the mood of participants, it had a clear effect on their cognitions. Given the reasonable assumption that participants are similarly adept at assessing their cognitive and affective experiences, this finding suggests that the effects of counterfactual thinking on cognitive processing occur independently of moods.

Experiment 1b: Counterfactual Mind-Sets and Thinking Styles In Experiment 1a, we observed that counterfactual mind-sets led to self-assessed mental states consistent with a relational processing style. Because the measures we used to assess internal states had not been independently validated, we thought it important to replicate the effect with established measures of cognitive processing styles. To this end, we examined the impact of our counterfactual manipulation on Sternberg’s (1988) thinking style preferences. Sternberg proposes that abilities may be used for three distinct functions in mental self-government, including legislative, executive, and judicial thinking styles (O’Hara & Sternberg, 2001). Individuals with executive styles prefer tasks that have clearly defined structure and guidelines from which to solve problems and build. Individuals with legislative thinking styles prefer tasks that 1 We included several other exploratory, distractor variables that were not theoretically relevant. As no statistically significant effects emerged for these additional variables, they will not be discussed further. 2 In a separate sample of 51 participants, we replicated this effect. That is, participants exposed to a counterfactual prime rated themselves higher on terms characteristic of an analytic style of thinking (M ⫽ 6.74, SD ⫽ 1.06) than those exposed to the noncounterfactual prime (M ⫽ 5.88, SD ⫽ 1.06), F(1, 47) ⫽ 7.26, p ⬍ .01.

COUNTERFACTUAL MIND-SETS

have little assigned structure and that allow them to invent new ideas and tend to enjoy creating original works. Individuals with judicial styles like to evaluate and critique others’ ideas and enjoy giving feedback and advice. We hypothesized that the executive thinking style preference would be stronger for individuals in a counterfactual mind-set. We expected that the mental structure of logical relationships created through counterfactual thinking would increase preferences for the structured, rule-based, logical nature of the executive thinking style. Another goal of the present experiment was to explore the connection between the construction of counterfactual thoughts and the strength of the resulting mind-set. Past research has demonstrated that the counterfactual mind-set’s impact occurs regardless of the content or valence of the counterfactual thoughts (Galinsky & Kray, 2004; Kray & Galinsky, 2003), which suggests that what lingers is the implicit mental structure of the counterfactual. An established methodology for assessing the strength of the counterfactual mind-set involves counting the number of counterfactual thoughts generated in the pretask scenario. Because no counterfactual is explicitly stated in the scenario, the tendency to infer counterfactual thoughts following the mutable versus nonmutable scenario can be assessed through this process. Kray and Galinsky (2003) observed that the strength of counterfactual activation mediated the relationship between the experimental manipulation and disconfirmatory information search. In the present experiment, we examined whether counterfactual activation (the number of counterfactual thoughts) mediated the relationship between type of prime and thinking styles. Finally, because we were interested in determining whether writing out the counterfactual thoughts (vs. simply pondering them) intensifies the impact of the prime on thinking styles, we included this as a factor in our design.

Method Design. The experiment had a 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence: positive vs. negative) ⫻ 2 (thought listing: yes vs. no) between-subjects factorial design. We also included an additional no-valence control condition to establish that our valence-based control conditions serve as appropriate baseline comparisons. Participants. Participants were 139 students from a large midwestern university campus. Participants were paid $10 an hour for their participation. Procedure. Participants were greeted in the laboratory by an experimenter who explained that they would complete several questionnaires

37

related to decision making. The experimental manipulations and the dependent variables were embedded in a single packet of questionnaires. Participants read the same rock concert scenario used in Experiment 1a, considered thoughts going through the main character’s mind (for approximately half the participants), and then completed both thinking style and mood measures. Experimental manipulation. The counterfactual prime manipulations were identical to those used in Experiment 1a. We also include an additional control condition, with no counterfactual prime and no valence. In this scenario, participants read the following: “Three weeks ago, Jane bought a general admission ticket to a rock concert of her favorite band. Jane is now at the concert, which is about to begin.” Approximately half the participants were asked simply to read the scenario as in Experiment 1a. The other participants were asked to “List some thoughts going through Jane’s mind.” The sheet on which they listed their thoughts was numbered from 1 to 10, but participants were told to list only as many thoughts as came to their mind. Because none of the scenarios contain actual counterfactual statements, but rather only possess the potential to generate counterfactual thoughts, the number of counterfactual thoughts that participants subsequently listed was our measure of the strength of the counterfactual mind-set. Thinking style. Participants’ preferred thinking style was assessed using a 24-item subset of Sternberg and Wagner’s (1991) Thinking Styles Inventory. The Thinking Styles Inventory comprises three subscales, including Executive Style, Legislative Style, and Judicial Style. Participants were asked to indicate how well each statement characterized their current preferred approach to solving problems and making decisions using a 7-point Likert scale ranging from 1 (extremely uncharacteristic) to 7 (extremely characteristic). The subscales had a high degree of reliability ranging from .72 to .76. Mood. Mood was assessed using a one-item 9-point scale ranging from 1 (very negative) to 9 (very positive).

Results and Discussion Table 2 provides a correlation matrix for all experimental variables. Counterfactual activation. Two independent coders identified the number of counterfactual thoughts listed by participants. The reliability for counterfactual thoughts was high (␣ ⫽ .95), and, therefore, the ratings of the two coders were averaged. We submitted the number of counterfactual thoughts to a 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence: positive vs. negative) between-subjects factorial design. As expected, counterfactual prime participants (M ⫽ 1.78, SD ⫽ 1.1) listed significantly more counterfactual thoughts than noncounterfactual prime participants (M ⫽ 0.19, SD ⫽ 0.42), F(1, 67) ⫽ 69.48, p ⬍ .001.

Table 2 Experiment 1b: Correlations Between Variables Variable 1. 2. 3. 4. 5. 6. 7.

Type of prime Valence of prime Number of counterfactual thoughts Executive style Judicial style Legislative style Mood

Note.

1

2

3

4

5

6

7

— .03 .69 .21 ⫺.01 ⫺.03 .08

— ⴚ.24 .10 .04 ⴚ.23 ⫺.09

— .28 .03 .00 .08

— .10 ⫺.29 .10

— .34 .24

— .25

—

Significant correlations ( p ⬍ .05) are in boldface.

38

KRAY, GALINSKY, AND WONG

Consistent with previous research (Galinsky & Moskowitz, 2000; Roese & Hur, 1997), participants exposed to a negatively valenced event (M ⫽ 1.25, SD ⫽ 1.2) generated more counterfactual thoughts than participants exposed to a positively valenced event (M ⫽ 0.70, SD ⫽ 1.0), F(1, 67) ⫽ 7.85, p ⫽ .007. More important for establishing the independence of counterfactual thinking and mood, however, the two-way interaction was not significant, F(1, 67) ⫽ 1.79, ns. Thinking styles. To test our hypotheses, we conducted separate 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence of prime: positive vs. negative) ⫻ 2 (thought listing: yes vs. no) between-subjects ANOVAs for each thinking style. We hypothesized that exposure to a counterfactual prime would elicit an executive thinking style. In support of our hypothesis, participants in a counterfactual mind-set (M ⫽ 5.28, SD ⫽ 0.68) rated themselves higher on executive thinking style than participants exposed to a noncounterfactual scenario (M ⫽ 4.97, SD ⫽ 0.82), F(1, 114) ⫽ 4.85, p ⫽ .03. The only other effect to emerge was a tendency for participants exposed to a scenario in which the protagonist did not win the trip to Hawaii to indicate a greater preference for the legislative thinking style (M ⫽ 5.38, SD ⫽ 0.81) than participants exposed to the positive emotional experience (M ⫽ 5.04, SD ⫽ 0.66), F(1, 114) ⫽ 5.74, p ⫽ .02. No other effects emerged as statistically significant. Relationship between counterfactual activation and executive thinking style. To better understand the relationship between counterfactual mind-sets and the executive thinking style, we examined whether the amount of counterfactual activation mediated the relationship between the counterfactual primes and an increased executive thinking style. When we regressed executive thinking style on both type of counterfactual prime and number of counterfactual thoughts listed, the effect of number of counterfactual thoughts on the outcome approached significance (␤ ⫽ .15), t(68) ⫽ 1.77, p ⫽ .08, but the partial effect of the type of prime on the outcome was significantly reduced in magnitude once the number of thoughts was controlled (␤ ⫽ ⫺.00), t(68) ⫽ ⫺0.01, ns. A Sobel test determined the reduction in the significance level was statistically significant (Z ⫽ 2.35, p ⬍ .05). This finding suggests that the generation of counterfactual thoughts accounted for the relationship between the mutable scenario and thinking style preferences. Supplementary control condition analyses. We also conducted analyses that included the additional no-valence control condition (M ⫽ 5.00, SD ⫽ 0.83) to build confidence that our counterfactual manipulation was responsible for the increased preference for an executive thinking style. To do so, we conducted planned contrasts comparing the two counterfactual conditions with the three control conditions (no-valence, negative outcome, positive outcome noncounterfactuals), which revealed that participants in the counterfactual conditions scored higher on executive thinking style than those in the control conditions, t(134) ⫽ 2.24, p ⬍ .05. Additionally, we tested whether the additional control differed from the two original control conditions and found no significant differences for their effects on executive thinking style, t(134) ⫽ 0.16, ns. Supplementary mood analyses. To determine whether mood was affected by our experimental manipulation, we first conducted an ANOVA, with mood as the dependent variable and type of prime, valence of prime, and thought listing as the independent

variables. No effects were statistically significant. We also conducted an ANOVA, with executive thinking style as the dependent variable, including our independent variables and the covariate mood. The main effect of type of counterfactual prime remained reliable, F(1, 113) ⫽ 4.47, p ⫽ .04. These findings suggest that mood did not mediate the effect of counterfactual prime on executive thinking style. Three noteworthy findings emerged from this set of experiments. First, the heightened sense that one was poised for analytic thinking (see Experiment 1a) and the greater preference for an executive thinking style (see Experiment 1b) following the activation of a counterfactual mind-set is consistent with the idea that the mind-set promotes a relational processing style. Second, the number of counterfactual thoughts was implicated as the mediating mechanism through which our experimental manipulation elicited a preference for an executive style of thinking (see Experiment 1b). And finally, across both experiments, we failed to observe any indication that mood is a driving force behind counterfactual mind-sets. This conclusion is backed by a lack of a significant effect of the counterfactual manipulation on mood and by the fact that mood did not reduce the effect of counterfactual primes on executive thinking style. In addition, the observation that the no-valence control was identical to the valence-control conditions gives us confidence that the remaining experiments in which valence-based controls were used establish valid baseline conditions.

Experiment 2: Counterfactual Mind-Sets and Analytical Reasoning The demonstration that counterfactual mind-sets affect mental states in a manner consistent with a relational processing style was our first step in understanding the process through which counterfactual mind-sets have their impact. The second step is to explore how counterfactual mind-sets impact analytical reasoning abilities. Because the analytical process involves the examination and identification of logical relationships between task variables, we expected the relational processing style activated by counterfactual thoughts to facilitate analytical reasoning abilities. To test this hypothesis, we had participants complete a version of the Law School Admission Test (LSAT) analytical reasoning section. The LSAT analytical reasoning section assesses one’s ability to understand and apply rules, determine relationships between concepts, analyze situations and draw conclusions, and apply logic to ambiguous or complex situations (Princeton Review, 2005). Specifically, this LSAT section consists of problems designed to “simulate the kinds of detailed analyses of relationships that a law student must perform in solving legal problems” (Law School Admission Council, 2005, p. 10). For example, one might be told to arrange guests at a dinner party and be given rules, including who may sit next to whom. These rules are followed by several questions that examine one’s understanding of the relationships between dinner guests. If counterfactual mind-sets promote a relational processing style, then invoking a counterfactual mind-set should be positively associated with LSAT analytical reasoning performance.

COUNTERFACTUAL MIND-SETS

Method Design. The experiment had a 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence: positive vs. negative) between-subjects factorial design. We also included an additional control condition in which participants simply took the LSAT exam without first reading a scenario. Participants. Participants were 135 students from a large midwestern university campus. Participants were paid $10 an hour for their participation. Procedure. Participants were greeted in the laboratory by an experimenter who explained that they would complete several questionnaires related to decision making. The experimental manipulations and the dependent variables were embedded in one large packet of questionnaires. Participants were randomly assigned to one of the five experimental conditions described in Experiment 1b. After reading one of the scenarios, all participants were asked to consider some thoughts likely to be running through Jane’s mind and then to indicate their current mood on a 9-point scale ranging from 1 (very negative) to 9 (very positive). Finally, participants completed one section of the LSAT analytical reasoning test. LSAT analytical reasoning problems. Each analytical reasoning section of the LSAT has four games with 24 questions that must be accurately diagrammed in order to be answered correctly. Our main dependent variable was task performance. We followed LSAT scoring guidelines within which the overall score consists of the number of questions answered correctly adjusted for the number of guesses. We also measured task effort, as gauged by the number of attempted answers.

Results and Discussion Counterfactual activation. Two independent coders identified the number of counterfactual thoughts in relation to the rock concert scenario. The reliability for counterfactual thoughts was high (␣ ⫽ .95), and, therefore, the ratings of the two coders were averaged. We submitted the number of counterfactual thoughts to a 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence: positive vs. negative) between-subjects factorial design. As expected, counterfactual prime participants (M ⫽ 1.32, SD ⫽ 0.66) listed significantly more counterfactual thoughts than noncounterfactual prime participants (M ⫽ 0.29, SD ⫽ 0.54), F(1, 108) ⫽ 88.08, p ⬍ .001. We also observed a main effect for valence of prime, with the negative outcome (M ⫽ 0.99, SD ⫽ 0.79) generating more counterfactuals than the positive outcome (M ⫽ 0.62, SD ⫽ 0.75), F(1, 108) ⫽ 10.61, p ⫽ .002. The interaction between type and valence of prime was not significant, F(1, 108) ⫽ 0.71, p ⫽ .40. Task performance. To measure analytical performance, we assessed the LSAT score using an ANOVA, which included the type of prime and valence as factors. We hypothesized that counterfactual mind-sets would improve analytical reasoning performance. Consistent with this hypothesis, counterfactual mind-set participants (M ⫽ 7.92, SD ⫽ 2.74) outperformed noncounterfactual mind-set participants (M ⫽ 6.09, SD ⫽ 3.45), F(1, 108) ⫽ 9.51, p ⫽ .003.3 No other effects emerged as statistically significant for this analysis. Task effort. To determine whether our manipulations affected task effort, we conducted an ANOVA in which the dependent variable was the number of LSAT items attempted, and the independent variables were the type of prime and valence. Results indicated an effect for the valence of the prime such that participants exposed to the positive emotional experience attempted more LSAT questions (M ⫽ 15.35, SD ⫽ 4.50) than those exposed to the

39

negative emotional experience (M ⫽ 13.30, SD ⫽ 4.65), F(1, 108) ⫽ 5.53, p ⫽ .02. No other results emerged as statistically significant. Relationship between counterfactual activation and LSAT performance. We next examined whether the amount of counterfactual activation mediated the relationship between the mutable primes and LSAT performance. When we regressed LSAT score on both type of counterfactual prime and number of counterfactual thoughts listed, number of counterfactual thoughts was statistically significant (␤ ⫽ .33), t(109) ⫽ 2.82, p ⫽ .006, but not type of prime (␤ ⫽ .07), t(109) ⫽ 0.58, ns. A Sobel test determined that the reduction in the significance level was statistically significant (Z ⫽ 3.85, p ⬍ .001). This finding suggests that the generation of counterfactual thoughts accounted for the relationship between the mutable scenario and performance. Supplementary control condition analyses. We also conducted analyses that included the additional no-valence control condition (M ⫽ 5.95, SD ⫽ 3.84) to build confidence that our counterfactual manipulation was responsible for the better LSAT performance. To do so, we conducted planned contrasts comparing the two counterfactual conditions with the three control conditions, which revealed that participants in the counterfactual conditions performed better than those in the control conditions, t(130) ⫽ 3.28, p ⫽ .001. Additionally, we tested whether the additional control differed from the two original control conditions and found no significant differences for their effects on LSAT performance, t(130) ⫽ ⫺0.18, ns. Supplementary mood analyses. We sought to determine whether mood played a role in the relationship between counterfactual mind-sets and LSAT task effort and performance. To do so, we first conducted an ANOVA, with mood as the dependent variable and type of prime and valence of prime as betweensubjects factors. No effects were statistically significant. We also conducted analyses of covariance using LSAT task effort (LSAT attempted) and task performance (overall LSAT score) as the dependent variables, type of prime and valence as our independent variables, and mood as a covariate. For the LSAT number attempted, the main effect for valence remained reliable, F(1, 107) ⫽ 5.49 p ⬍ .05. For the overall LSAT score, the main effect of type of prime remained reliable, F(1, 107) ⫽ 7.16, p ⫽ .009. These findings suggest that mood did not mediate the effect of counterfactual prime or valence on LSAT task effort or task performance. One goal of the present study was to provide support for the assertion that counterfactual mind-sets promote a relational processing style, thereby facilitating the recognition and better understanding of ambiguous and complex relationships. In support of this hypothesis, we found that participants in a counterfactual mind-set outperformed participants who were not in a counterfactual mind-set on the LSAT analytic reasoning test. In combination with the results of Experiment 1, which demonstrated that the 3 We also tested our hypothesis by examining only the number of correct LSAT items (unadjusted for guesses). We found the same pattern of results, with participants in a counterfactual mind-set (M ⫽ 9.21, SD ⫽ 2.61) outperforming noncounterfactual mind-set participants (M ⫽ 7.71, SD ⫽ 3.20), F(1, 108) ⫽ 7.42, p ⫽ .008.

40

KRAY, GALINSKY, AND WONG

activation of a counterfactual mind-set was predictive of phenomenologies and preferences consistent with a relational processing style, Experiment 2 demonstrated that performance on a task measuring one’s ability to understand logical relationships and make connections between task stimuli is improved by counterfactual mind-sets.

Experiment 3: Counterfactual Mind-Sets and Structured Imagination Thus far, we have provided evidence consistent with the idea that counterfactual mind-sets promote a relational processing style in terms of both mental states and analytical reasoning performance. Because we expect the effects of a relational processing style to be quite broad, in Experiment 3, we explore whether and how counterfactual mind-sets influence the tendency to build on existing knowledge structures during the creative generation process, or structured imagination. Intuitively it might seem that thoughts of “if only” associated with counterfactual thinking would facilitate creative generation. The construction of alternative possible worlds would seem to be the epitome of creative generation, and it is easy to speculate that counterfactual thinking would encourage imagination to roam unfettered and unencumbered by mental constraints. To the contrary, we contend that the counterfactual mind-sets may not be the springboard to freewheeling generation. A key insight underlying this counterintuitive hypothesis concerns the fact that logical rules and clear structure govern when and how counterfactuals are constructed. As we have repeatedly demonstrated, counterfactual mind-sets have their impact regardless of the content or direction of the preceding counterfactual thoughts. This observation suggests that what lingers following the construction of counterfactual thoughts is their underlying logical form. We argue that the mental structure of logical relationships created through counterfactual thought increases the tendency to structure thought around salient associations and the pursuit of connections. In addition to the conceptualization described above, the counterfactual mind-sets increase a preference for structured thinking, as was empirically demonstrated in Experiment 1b. Structured thinking differs from the mental states that encourage creative generation, which generally requires an expansion of conceptual attention that goes beyond the bounds of what is presently known or salient (Guilford, 1950). Some have even characterized the mind-set that encourages creative generation as a “risky” processing style (Friedman & Forster, 2001; quotations in original). Noting the connections and relationships between stimuli and making structured associations, as counterfactual participants did with relative ease in Experiment 2, may inhibit the tendency to go beyond what is already known or salient, and thereby impair creative generation. Creative generation tasks typically elicit responses that are loosely defined, which creates the potential for an infinite number of unique responses (Guilford, 1950). However, performance on creative generation tasks tends to be impaired because participants borrow from and rely too heavily on existing knowledge structures. For example, participants instructed to create creatures “beyond their wildest imagination” tended to produce results that conformed to the attributes of realistic earth creatures, including

bilateral symmetry, and ordinary sensory receptors and appendages (Ward, 1994). On the basis of Ward’s concept of structured imagination, we expected participants in a counterfactual mind-set to adopt a more structured approach to generating creative output than under baseline conditions. Given the relational processing style characteristic of counterfactual thinking, the resulting mindset should increase the tendency to structure one’s imagination around existing knowledge structures. That is, in contrast to the “thinking outside the box” characteristic of creative generation processes, we expected counterfactual mind-sets to promote “thinking within the box.”

Method Design. The experiment had a 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence: positive vs. negative) ⫻ 2 (type of scenario: rock concert vs. spelling bee) between-subjects factorial design. Participants. Participants were 93 students from a large western university campus enrolled in an introductory organizational behavior course. By participating in the experiment, participants received partial credit toward a class requirement. Procedure. Participants were greeted in the laboratory by an experimenter who told them that they would complete several tasks predictive of future job performance. Participants first read one version of the two prime scenarios. As in previous experiments, approximately half the participants read a version of the rock concert scenario. To ensure that our results generalize to different instantiations of the counterfactual mind-set, the other half of participants read a version of a spelling bee scenario, in which a young boy named Paul competes to advance in the National Junior Spelling Bee. In each scenario, an individual at the spelling bee advances to the next round. To manipulate valence, in half of the spelling bee scenarios, Paul correctly spells the assigned word and advances to the next round, whereas in the other half, he misspells the word and is eliminated from the competition. Additionally, half of the scenarios describe a sequence of events designed to elicit counterfactual thoughts, whereas the other half of the scenarios describe a sequence of events not expected to elicit counterfactual thoughts. In the downward counterfactual scenario, Paul advances to the next round of the competition after his place in line is altered (because he had to use the restroom), and he is asked to spell a word he knows (but had he stayed in his original place in line, he would have been given a word he did not know how to spell). In the upward counterfactual scenario, Paul is eliminated from the competition after his place in line is altered, and he is asked to spell a word he does not know (but had he stayed in his original place in line, he would have been given a word he knew how to spell). In the noncounterfactual conditions, he either spells or misspells a word but does not alter his place in line. After reading the scenario and listing thoughts in the protagonists’ mind, participants began the creative task. Creative task. Following Ward (1994), we asked participants to “imagine going to another galaxy in the universe and visiting a planet very different from earth” and to spend 7 min drawing a picture of an animal that is local to this planet. Immediately upon completion of the drawing task, participants completed a questionnaire in which they provided a short written description of how they went about approaching this creative task and then evaluated their approach along several domains. Participants’ evaluations of their approach to the task were used to determine the extent to which they engaged in structured imagination. Specifically, they indicated the extent to which they considered the following five items of knowledge: known science fiction creatures, general attributes of science fiction creatures, known earth animals, general attributes of earth animals, and consideration for the local environment. Each item was rated on a 7-point scale ranging from 1 (not at all) to 7 (completely).

COUNTERFACTUAL MIND-SETS Creative coding. Because Ward’s (1994) study suggests potential ceiling effects for some characteristics of the imagined creatures (e.g., bilateral symmetry) and demonstrated greater variability for atypical sensory organs, our dependent variable for the drawings was the atypicality of the sensory organs.4 Three trained coders who were blind to the study hypotheses coded the drawings and descriptions for atypical sensory organs. Following Ward’s coding, sensory organs were considered atypical if they (a) lacked a major sensory organ (i.e., eyes, ears, nose), (b) had atypical numbers of a sensory organ (e.g., three eyes), (c) demonstrated an unusual configuration of the senses (e.g., eyes located below the nose), (d) had an exaggerated or unusual ability (e.g., eyes that had laser beams), or (e) served an atypical function (e.g., ears for protection). The total number of atypicalities were tallied for each participant. The codings for the drawing were highly reliable (␣ ⫽ .93), as were the codings for the descriptions (␣ ⫽ .89), and discrepancies were resolved through discussion. The coders’ data were averaged to create one measure for sensory atypicality in the drawing and one measure for sensory atypicality in the description.

Results and Discussion Counterfactual activation. Three independent coders identified the number of counterfactual thoughts after the rock concert and spelling bee scenarios. The reliability for both was high (␣s ⫽ .88 and .87, respectively), so the ratings were averaged. We submitted the number of counterfactual thoughts to a 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence: positive vs. negative) ⫻ 2 (type of scenario: rock concert vs. spelling bee) between-subjects ANOVA. As expected, counterfactual prime participants (M ⫽ 1.32, SD ⫽ 0.96) listed significantly more counterfactual thoughts than noncounterfactual prime participants (M ⫽ 0.94, SD ⫽ 0.71), F(1, 85) ⫽ 4.00, p ⬍ .05. Participants exposed to a negatively valenced event (M ⫽ 1.33, SD ⫽ 0.77) generated more counterfactual thoughts than participants exposed to a positively valenced event (M ⫽ 0.93, SD ⫽ 0.90), F(1, 85) ⫽ 6.30, p ⫽ .01. No other effects were statistically significant. Self-reported structured imagination. We expected participants in the counterfactual mind-set condition to have a more structured imagination than noncounterfactual mind-set participants. To test our hypothesis, we conducted a multivariate ANOVA, including each of the five statements described above, as well as type of prime, valence of prime, and type of scenario as between-subjects factors. Consistent with our hypothesis, the only statistically significant effect to emerge was a tendency for counterfactual mind-set participants (M ⫽ 4.49, SD ⫽ 0.77) to report more structured imagination than noncounterfactual mind-set participants (M ⫽ 4.04, SD ⫽ 0.85), F(5, 80) ⫽ 3.18, p ⫽ .01. An examination of the univariate effects revealed that counterfactual mind-sets promoted a reliance on specific science fiction creatures, F(1, 84) ⫽ 4.34, p ⬍ .04; general attributes of science fiction creatures, F(1, 84) ⫽ 4.28, p ⬍ .05; and local environment considerations, F(1, 88) ⫽ 8.18, p ⬍ .01. No other effects were statistically significant. Creative coding. Although the self-report results support our hypothesis, we further tested it by examining the actual drawings and drawing descriptions. To do so, we used a mixed-model ANOVA, with type of prime, valence of prime, and type of scenario as between-subjects factors and type of coding (drawings vs. descriptions) as a repeated measure. In support of our hypothesis, counterfactual mind-set participants (M ⫽ 1.11, SD ⫽ 1.00) were rated as incorporating fewer sensory atypicalities into their

41

drawings and descriptions than noncounterfactual mind-set participants (M ⫽ 1.69, SD ⫽ 1.61), F(1, 85) ⫽ 4.92, p ⬍ .03. No other effects were statistically significant.5 The primary goal of this study was to demonstrate that the relational processing style associated with counterfactual mindsets can lead to “thinking within the box.” The results of this experiment are consistent with our hypothesis. Participants in a counterfactual mind-set reported structuring their imaginative process around existing knowledge structures to a larger degree than control participants. In addition, they showed evidence of considering the local environment in constructing their drawings. Finally, participants in a counterfactual mind-set incorporated fewer atypicalities into their drawings and descriptions than participants in the noncounterfactual mind-set. Together, these findings reinforce the notion that counterfactual mind-sets promote a form of structured imagination. A secondary goal of the present study was to demonstrate that the results found in the previous studies were not driven by the nature of the rock concert scenario. To do so, we included a different scenario that involved a mutable event at a spelling bee. Regardless of which mutable scenario participants read, counterfactual mind-sets promoted a form of structured imagination. This observation gives us confidence that our results thus far generalize to different instantiations of the counterfactual mind-set.

Experiment 4: Counterfactual Mind-Sets and Creative Generation In Experiment 4, we further explore whether counterfactual mind-sets impact the creative generation process. Because the drawings created in the previous experiment were highly variable, and it was difficult to determine what aspects of science fiction creatures counterfactual mind-sets might have been borrowing from, in the present experiment, we used a creative generation task that can be more easily coded for structured imagination. In the previous experiments, participants reported relying on existing knowledge and examples in constructing their alien creatures. The presence of salient examples in a creative generation task leads to less novel solutions than when examples are not provided a priori (Smith, Ward, & Schumacher, 1993). If counterfactual mind-sets facilitate a relational processing style that seeks connections with readily available cognitive representations, then participants in this mind-set should be more attentive to and influenced by a set of provided examples. As a result, the output of participants in a counterfactual mind-set should be less novel than baseline conditions. To test this hypothesis, we examined performance on a creative generation task involving the creation of novel product labels. Prior to starting the task, participants were provided with a set of examples. In the present experiment, we measured the degree to which individuals were able to ignore these examples in 4 Our data replicate this ceiling effect. For instance, 92% of the drawings demonstrated bilateral symmetry, 1% did not, and the remaining 7% were excluded from the data analysis because the drawings were profiles, and bilateral symmetry could not be determined. 5 We also examined whether participants differed in terms of appendage atypicalities in their drawings and descriptions, but no effects emerged as statistically significant.

42

KRAY, GALINSKY, AND WONG

their creation of labels for the new products. That is, we measured the extent to which participants’ own product labels resembled the examples we provided them. In addition to coding for resemblance to examples, we also coded the product labels on two dimensions: overall creativity and descriptiveness of the labels in representing the product category. In the previous experiment, participants in a counterfactual mindset reported that they were more likely to consider the local environment in drawing their space aliens. This suggests that counterfactual mind-sets gear participants toward structuring imagination to reveal something about the essence of the object. If this is true, then counterfactual mind-sets should produce product labels that are representative of the product. We predicted the counterfactual mind-set participants would both draw on the given examples and be more likely to produce labels that were more descriptive of the product. However, to the extent that the labels were not judged to be particularly novel, they should suffer in terms of their perceived creativity.

Method Overview and design. The experiment had a 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence of prime: positive vs. negative) between-subjects factorial design. Participants. Participants were 29 undergraduate business students from a large western university campus enrolled in an introductory organizational behavior course. The experiment was conducted outside of the classroom setting. By participating in the experiment, partial fulfillment of a course requirement was granted. Procedure. Participants were greeted in the laboratory by an experimenter who explained that they would complete several questionnaires related to decision making. The experimental manipulations and our dependent variables were embedded in the packet of materials that participants received. Participants were given up to 30 min to complete the packet of materials. Experimental manipulations. The manipulations were identical to those used in Experiment 1a. Because Experiment 1b revealed that the strength of the counterfactual mind-set is not dependent on whether thoughts are listed, we chose not to have participants list their thoughts for this experiment. Instead, participants were simply asked to ponder thoughts running through the protagonist’s mind. Creative generation task. We used a modified version of the creative generation task, described in Rubin, Stoltzfus, and Wall (1991). Participants were asked to imagine that they were interviewing with a top marketing firm, and part of the interview involved testing their aptitude for business. To do so, participants were tasked with creating new labels for new products. Specifically, they were instructed to create at least one (and up to three) new label for each of three categories of products (pasta, nuclear element, pain reliever). Six examples were provided for each category, as described in Appendix A. To encourage creative output, participants were advised not to use or copy aspects of the examples provided. For each of the categories, the examples provided had two common endings, which are defined as a letter or cluster of letters that ended at least one multisyllabic word. For example, all of the examples provided of nuclear elements ended in _on or _ium (e.g., radon, plutonium, argon, carbon, radium, uranium). Creative output was operationalized in terms of the number of product names created for each category that did not share the word endings of the examples. We also examined the sheer number of product names created for each category to determine whether motivational differences were evident.

Creative generation coding. Three independent coders evaluated each idea generated by participants on 9-point scales ranging from 1 (not at all) to 9 (extremely) for creativeness and descriptiveness. Specifically, coders were asked to consider “How creative is this response?” and “How descriptive is this response in revealing the type of product?”

Results and Discussion Creative generation. To examine the novelty of name generation, we submitted the total number of names with endings deviating from the supplied examples to a 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence of prime: positive vs. negative) ⫻ 3 (product category: pasta vs. nuclear element vs. pain reliever) mixed model ANOVA, with repeated measures on the third factor. The only significant effect to emerge from this analysis was a main effect for counterfactual prime, F(1, 25) ⫽ 4.92, p ⬍ .05. Across the three categories, counterfactual mind-set participants (M ⫽ 1.26, SD ⫽ 1.44) generated significantly fewer novel names than noncounterfactual mind-set participants (M ⫽ 2.70, SD ⫽ 2.20). We also conducted a similar analysis examining the raw number of ideas generated and observed no statistically significant effects, suggesting our experimental manipulation did not impact effort. Creative versus descriptive ratings. Three independent coders who were blind to condition and our hypotheses evaluated each idea generated by participants for creativeness and descriptiveness. Reliabilities were good (␣ ⫽ .72 and .73, respectively), so we combined the ratings of each coder. We submitted the ratings to a 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence of prime: positive vs. negative) ⫻ 2 (rating: creativity vs. descriptiveness) mixed model ANOVA, with repeated measures on the third factor. A main effect for type of coding emerged, indicating that labels were judged to be more descriptive (M ⫽ 5.75, SD ⫽ 0.69) than they were creative (M ⫽ 5.09, SD ⫽ 0.76), F(1, 25) ⫽ 8.61, p ⬍ .01. More important for testing our hypothesis was the statistically significant two-way interaction between type of prime and type of coding, F(1, 25) ⫽ 4.66, p ⬍ .05. Counterfactual prime labels (M ⫽ 4.86, SD ⫽ 0.54) were judged to be less creative than noncounterfactual prime labels (M ⫽ 5.34, SD ⫽ 0.90), t(27) ⫽ 1.76, p ⫽ .09; yet, counterfactual prime labels (M ⫽ 5.97, SD ⫽ 0.59) were judged to be more descriptive than noncounterfactual prime labels (M ⫽ 5.51, SD ⫽ 0.73), t(27) ⫽ 1.86, p ⫽ .07. Counterfactual mind-sets led to more descriptive labels that were nonetheless lacking in creativity. No other effects emerged as statistically significant for this analysis. The results of this experiment support our hypothesis that counterfactual mind-sets can impair creative generation. Individuals who had previously pondered counterfactual thoughts generated new product labels that were less novel than individuals who had not previously pondered counterfactual thoughts. This pattern emerged regardless of whether novelty was judged in terms of the similarity to the provided examples or a global evaluation of creativity by independent coders. The fact that counterfactual thinking did not impact the number of names generated suggests sheer effort was not responsible for this effect. Although participants were instructed to be as creative as possible and to refrain from borrowing from the examples provided, counterfactual mind-set participants were less effective at breaking out of the mold set by the examples. In essence, the relational

COUNTERFACTUAL MIND-SETS

processing style characteristic of the counterfactual mind-set promoted structured imagination, or thinking within the box. However, within that box, counterfactual mind-sets led to more descriptive and potentially useful labels. Like participants in the previous experiment who reported considering the local environment when drawing space aliens, counterfactual mind-sets led participants to consider the essence of what the product was about. The finding that labels generated by counterfactual mind-set participants were actually judged to be more descriptive of the product category than labels generated by control participants is consistent with the hypothesis that counterfactual mind-sets promote a relational processing style. That is, a heightened attention to the attributes generally associated with the categories for which labels were created (i.e., pasta) led to the generation of labels that were deemed to be representative of the category by lay judges.

Experiment 5: Creative Generation Versus Creative Association We demonstrated in the previous experiment that creative generation is impaired following the invocation of a counterfactual mind-set. Measuring creativity in terms of idea generation is but one approach, as creativity is generally regarded to be a multidimensional construct (Amabile, 1983). The purpose of Experiment 5 was to determine whether counterfactual mind-sets might improve performance on creative association tasks that involve the consideration of relationships and making connections between disparate knowledge structures. If counterfactual mind-sets promote a relational processing style involving the consideration of relationships between task stimuli, then the mind-set should improve performance on creative tasks involving the identification of unusual associations between stimuli, associations that are adaptive and responsive to the present context. Galinsky and Moskowitz (2000) provided some evidence consistent with this hypothesis with respect to the Duncker candle problem. This task gauges the ability of individuals to overcome functional fixedness, characterized by a failure to recognize a use for a particular object in a given context as a result of a fixation on its typical use (Duncker, 1945). An individual is given a candle, a box of tacks, and a book of matches and challenged with affixing the candle to the wall so that it can be lit without dripping wax onto the floor or wall. Because the box initially functions as a container for tacks, the problem solver often fails to recognize its potential as a solution to the problem: The tacks are dumped out of the box, which is affixed to the wall with a couple of tacks, and the candle is placed on top of or in the box and lit. The solution to the task requires the problem solver to see potential relationships other than the obvious ones—the box is not just a repository for tacks but can also be used as stand (Glucksberg & Weisberg, 1966). In addition, the solution involves recognizing a relationship between the candle and the box. Galinsky and Moskowitz (2000) showed that, by invoking a counterfactual mind-set prior to engaging in the task, problem solvers were more likely to recognize the potential use of the box of tacks as a platform for the candle relative to control participants. We argue that performance on the Duncker candle problem was facilitated by the counterfactual mind-set because it promoted a tendency to explore the possible relationships between the critical

43

objects. In the present experiment, we sought further evidence in favor of this interpretation by examining performance on another creative association task, the remote associates task (RAT; M. T. Mednick, Mednick, & Mednick, 1964). The RAT requires an individual to form “mutually distant associative elements into new combinations which are useful and meet specified as well as unforeseen requirements” (S. A. Mednick, 1962). Specifically, the test requires identifying a unique association among three distinct words. For example, the common link for the words sore—shoulder—sweat is cold. Similar to the Duncker candle problem, by considering the relationships between task stimuli, performance on the RAT improves. In the present experiment, we used a within-subject design so that we could explore the relative effect of counterfactual mindsets on a creative generation task versus a creative association task. We aimed to replicate the facilitative effect of the counterfactual mind-set observed previously with the Duncker candle for the RAT and also demonstrate a replication of the impairing effect for the creative generation task used in Experiment 4. By showing both facilitation and debilitation across tasks but within individuals, we aimed to provide strong evidence that the counterfactual mind-set promotes a relational processing style, which has wideranging effects.

Method Overview and design. The experiment had a 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence of prime: positive vs. negative) ⫻ 2 (type of creative task: association vs. generation) mixed design, with repeated measures on the third factor. Participants. Participants were 50 undergraduate students at a midwestern university who were recruited through e-mail solicitations; they participated in the experiment, along with several other unrelated tasks. They were compensated $15 for their time. Procedure. Participants were greeted in the laboratory by an experimenter who explained that they would complete several tasks assessing ability in business contexts. Participants were given a packet of materials that contained the experimental manipulations, a modified version of the RAT (M. T. Mednick et al., 1964), and the identical creative generation task used in Experiment 4. All participants read the prime scenarios and then completed the RAT first before completing the creative generative task.6 Participants were instructed to complete each task before proceeding to the next one. Participants were allowed to work on each task until they had finished or could not answer any more questions. Finally, participants were debriefed and then proceeded to work on several unrelated tasks. Experimental primes. The primes were the four rock concert scenarios used in the previous experiments. As in Experiment 4, participants simply pondered thoughts running through the protagonist’s mind without listing any thoughts. The RAT (M. T. Mednick et al., 1964). The RAT is designed to measure the creative ingenuity of individuals and requires identifying a unique common denominator among three distinct words. We shortened the original task designed by Mednick et al. to include only 10 items (see Appendix B for the items). Consistent with our analysis of the generative

6

Although we did not counterbalance the order of tasks in the present experiment, the findings of Experiment 3 established that performance on the creative generation task was negatively impacted when the task immediately followed the counterfactual manipulation.

44

KRAY, GALINSKY, AND WONG

cognitive task, the dependent variables for the RAT included the number of attempted items and the number of correct items.

Results First, because the effect of our experimental manipulations did not vary depending on product type in the generative cognitive task (replicating the results of Experiment 4), we collapsed across product type in the present set of analyses. To facilitate a comparison of performance across the two creative tasks, we first computed z scores separately for each task and then submitted each z score to a 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence of prime: positive vs. negative) ⫻ 2 (type of task: creative association vs. creative generation task) mixed model ANOVA, with repeated measures on the third factor. Number of correct items and novel names. To measure creative performance on the two tasks, we assessed the number of correct associations on the RAT and the number of novel names generated on the new products task. Consistent with our hypothesis, the only effect to emerge as statistically significant was the expected two-way interaction between type of task and counterfactual prime, F(1, 45) ⫽ 22.37, p ⬍ .001 (see Figure 1). Counterfactual mind-set participants (M ⫽ 0.43, SD ⫽ 0.78) outperformed noncounterfactual mind-set participants (M ⫽ ⫺0.48, SD ⫽ 1.02) on the RAT, F(1, 49) ⫽ 12.51, p ⬍ .01, yet counterfactual mind-set participants (M ⫽ ⫺0.34, SD ⫽ 0.68) performed worse than noncounterfactual mind-set participants (M ⫽ 0.40, SD ⫽ 1.18) on the creative generative task, F(1, 50) ⫽ 7.97, p ⬍ .01.7 Counterfactual mind-sets improved performance on a task requiring the identification of a common association for a set of words but impaired performance on a task requiring the generation of novel ideas. No other effects emerged as statistically significant for this analysis. Number of items attempted and labels generated. We also looked at task effort in terms of items attempted on the RAT and labels generated on the new products task. The only effect to emerge was a two-way interaction between task and counterfactual prime that approached significance, F(1, 45) ⫽ 3.72, p ⫽ .06 (see Figure 1). Whereas the number of items attempted on the RAT was greater in the counterfactual condition (M ⫽ 0.36) than in the noncounterfactual condition (M ⫽ ⫺0.40), F(1, 49) ⫽ 8.16, p ⬍ .01, the difference between the two counterfactual conditions was not statistically significant in the creative generation task (Ms ⫽ ⫺0.05 and 0.02, F ⬍ 1, ns). No other effects emerged as statistically significant for this analysis.

Discussion The results of this study provide strong support for the assertion that counterfactual mind-sets promote a relational processing style, which has differential effects on creative tasks measuring the generation of novel ideas versus the identification of associations. Rather than uniformly impairing creative performance, the effect of counterfactual mind-sets appears to depend on the underlying creative process being assessed. Specifically, counterfactual mindsets improved the ability of participants to identify unusual and useful associations between sets of words in the RAT, yet caused participants to borrow too heavily from the provided examples in

Figure 1. Mean performance as a function of type of task and type of prime in Experiment 5. CF ⫽ counterfactual condition; NCF ⫽ noncounterfactual condition; RAT ⫽ remote associates task.

the new products labels task. The fact that both tasks measure aspects of creativity suggests that the debilitating effect of counterfactual mind-sets is limited to thinking “outside the box.”

General Discussion The question of whether thoughts about alternate worlds borne out of mutated pasts impact the future has important theoretical and applied implications. Although the adage “What’s done is done” may suggest that pondering the past is an unproductive use of time, the present research provides strong evidence to suggest that imagining alternatives to past realities by considering a dif7 The relationship between performance on the creative generation task and the creative association task was not statistically significant, r(49) ⫽ ⫺.12, ns; this relationship did not depend on experimental condition.

COUNTERFACTUAL MIND-SETS

ferent path, choice, or action has a powerful impact on how future analytic and creative problems are solved. In particular, generating counterfactuals in one context appears to alter thought processes to be more relational in subsequent contexts, despite the new context’s irrelevance to the imagined world. Independent of the content or valence of the imagined world, the act of generating counterfactuals produces a lingering tendency to consider relationships and associations and to problem solve from within existing frameworks. The present article provides a range of evidence to support the hypothesis that counterfactual mind-sets promote a relational processing style. We demonstrated in Experiment 1 that counterfactual mind-sets increase a sense of being poised for analytic and critical thinking and preferences for a structured style of thinking. Evidence that the counterfactual mind-set improves performance on an analytic task involving the assessment of relationships between task variables was provided in Experiment 2. We demonstrated in Experiments 3 and 4 that counterfactual mind-sets increase the tendency to structure imagination around existing knowledge structures, leading to more descriptiveness but less novelty. Finally, using a within-subject design, we demonstrated in Experiment 5 that the heightened tendency to build on existing knowledge structures following the activation of the counterfactual mind-set leads to better performance on a creative association task involving the consideration of associations between task stimuli but to worse performance on a creative generation task. The counterfactual thought-listing methodology used in Experiments 1b and 2 also shed light on the process through which counterfactual mind-sets promote a relational processing style. In Experiment 1b, we demonstrated that the effect of counterfactual mind-sets on executive thinking style preferences was mediated by counterfactual activation. In Experiment 2, we demonstrated that the listing of counterfactual thoughts mediated the relationship between the mutable prime and LSAT performance. In combination with previous research in which a similar methodology was used (Galinsky & Kray, 2004; Kray & Galinsky, 2003), the amount of counterfactual activation has proved to be a robust mediating mechanism in the relationship between mutable primes and performance. The greater the number of counterfactual thoughts generated following exposure to a scenario in which an event almost turned out differently, the more ingrained the mental structure of logical relationships becomes, thus increasing preferences for structure and performance on analytical tasks. Counterfactual primes influence not just what we think but how we think. The present experiments consistently demonstrated that what we think, as determined by the direction of the counterfactual elicited (upward or downward), did not moderate any of the effects of how we think, as determined by the mutable nature of the prime. Across two experiments (1a and 2), we also demonstrated that the valence-based conditions produced results identical to both a baseline condition without a pretask scenario and a pretask scenario in which no valence-based event occurs. This finding suggests that the valence of the outcome did not impact the results. Finally, the fact that measurements of mood did not differ across our counterfactual prime manipulation across multiple experiments gives us more confidence that it is the process of thinking counterfactually, and not the content of the counterfactuals, that is responsible for the observed effects. In total, counterfactual mind-sets and the

45

differential emotional experiences associated with upward versus downward counterfactuals appear to operate independently on subsequent thinking styles and cognitive processing. The present article is important because it is the first to our knowledge that delves directly into the phenomenological experience of a counterfactual mind-set. The findings of Experiment 1a and 1b suggest that counterfactuals exert a powerful impact on how individuals perceive their own cognitive state but do not appear to affect perceptions of affective states. Previously, researchers have argued that counterfactual mind-sets involve a state of heightened awareness of multiple possible worlds, thereby promoting mental simulations. But mental simulations brought about by a salient counterfactual tend not to be free-form. Instead, they follow systematic laws of mutability that involve tweaking particular aspects of the counterfactual context to undo a known outcome, thereby promoting a consideration of cause– effect relationships. The present research suggests that counterfactual thinking primes a relational processing style in subsequent contexts that facilitate the examination of the relationship between clues, cues, examples, and props embedded within a problem-solving task. The observation that the counterfactual mind-set promotes a relational processing style sheds light on one detrimental, yet seemingly anomalous, effect resulting from its activation. Specifically, counterfactual mind-sets have been shown to impair performance on the Wason (1966) card selection task involving four cards, each bearing a symbol: “E, K, 4, 7” (Byrne & Tasso, 1994; Galinsky & Moskowitz, 2000). The challenge is to determine what cards must be turned over to determine whether the following conditional statement is true: “If a card has a vowel on one side, then it has an even number on the other side.” Both sets of researchers independently found that counterfactual thinking impaired performance by leading participants to incorrectly select the “4” card. Because the conditional statement to be tested is not bidirectional, the selection of this card is an error of commission. The conditional statement central to the task can be misinterpreted to imply a bidirectional hypothesis (Byrne & Tasso, 1994). We contend that it is the consideration of the relationship and connections between the antecedent and consequent characteristic of a relational processing style that promoted this misguided tendency to entertain two hypotheses at once. The fact that counterfactual mind-set participants did not incorrectly solve the problem because of a failure to select the potentially falsifying “E” and “7” cards or incorrectly selecting the irrelevant “K” card further supports this explanation. We have claimed that counterfactual thinking elicits a particular cognitive style characterized by a consideration of relationships and associations. The idea that cognitive styles can differentially affect performance on tasks that require focus from those that require flexibility is supported by the theorizing and data of Peterson and Nemeth (1996). These researchers found that minority influence encourages flexible thinking and improves performance on tasks in which flexibility is an asset but that majority influence can aid performance on tasks that require focusing on one dimension of a two-dimension task (e.g., the Stroop task). We are not claiming that counterfactual thinking is akin to majority influence but rather suggesting that the work by Peterson and Nemeth is a useful demonstration that cognitive processing styles can have a differential influence, depending on whether a task

46

KRAY, GALINSKY, AND WONG

requires flexibility, focus, or consideration of task-embedded alternatives.

Limitations and Future Directions The research presented here shows a uniformly negative impact of counterfactual mind-sets on creative generation tasks. However, one limitation of the studies presented here is that they only explore one aspect of the creative generation process. Creative generation can be broken down into fluency, or the number of ideas generated, flexibility, or the number of different categories of ideas represented, and novelty, or the uniqueness of the idea generated (Amabile, 1983). Perhaps counterfactual mind-sets differentially affect these components of creativity. The experiments presented here provide some evidence that counterfactual mindsets decrease novelty, as name generation was more constrained by the suffixes of the examples after counterfactual primes. Although no differences were observed in fluency, or the quantity of names generated overall in Experiments 4 and 5, this null finding may be a reflection of the fact that the number of names generated had a limited range (from 1 to 3). The new product labels task also did not allow for a clear gauge of flexibility, or the number of definable categories of names generated. Given that counterfactual mind-sets promote a relational processing style that involves the consideration of relationships and associations, fluency and flexibility may actually be facilitated following the mind-set’s activation. The observation that new product names generated by counterfactual mind-set individuals were judged to be more descriptive of the product category than the names generated by noncounterfactual mind-set individuals suggests a practical approach to idea generation when in a counterfactual mind-set. By broadening the examination to consider the impact of counterfactual thinking on innovation, which involves both the generation of novel ideas and their successful implementation (Amabile, Conti, Coon, Lazenby, & Herron, 1996), researchers may find that the counterfactual mind-set facilitates the pragmatic process of turning a novel idea into reality by modeling the execution process after previously successful ventures. Future research that explores different instantiations of counterfactual mind-sets is needed. In all of the experiments in this article, our manipulation involved the presentation of a scenario that resulted in the spontaneous generation of counterfactual thoughts resulting from the presence of a salient mutable component. Individuals constructed counterfactual thoughts without any direction or guidance from the experimenter. Whereas the results of this set of experiments show consistent effects, it is important to consider their boundaries. For example, other research has explored the effect of counterfactual thinking after encouraging individuals to imagine the implications of various “what-if” scenarios of how the past could have played out differently (cf. Tetlock & Lebow, 2001) or prefactual considerations of what may be (Gleicher, Boninger, Strathman, Armor, Hetts, & Ahn, 1995). Mind-sets resulting from these approaches, which more directly focus thought in an exploratory, imaginative direction, may have a beneficial effect on the subsequent generation of novel ideas. Instead of promoting thinking within the box, they may actually facilitate out-of-the-box thinking. Whether the counterfactual is spontaneously generated or

brought about by “what-if” scenarios may moderate the relationship between counterfactual mind-sets and a relational processing style. Another possible direction for future research is to explore whether different types of counterfactual thoughts have differential effects on relational processing. A particularly important distinction may be the additive versus subtractive nature of the counterfactual. Additive counterfactuals refer to an action that may have been taken to create an alternate world, whereas subtractive counterfactuals refer to an action that may not have been taken to create an alternate world (Roese, Hur, & Pennington, 1999). Although counterfactuals generally aid in making causal judgments, Roese et al. demonstrated that additive counterfactuals more often express causal sufficiency, whereas subtractive counterfactuals more often express causal necessity (McGill, 1998; McGill & Klein, 1993). In the present research, the most common reaction to the rock concert scenario was “If only she had not switched seats. . .,” which expresses a subtractive counterfactual. If the causal necessity associated with subtractive counterfactuals creates a closer association between the antecedent and consequent events than the causal sufficiency associated with additive counterfactuals, then the resulting relational processing style may be more strongly activated for the former type of counterfactual than the latter. One consequence of this possibility is that additive counterfactuals may be more beneficial for creative generation than subtractive counterfactuals have proved to be. Given the powerful effect counterfactual thinking has been shown to have on a wide variety of problem-solving tasks, it is important to consider the implications of these findings. For example, an important practical consideration is how the use of a counterfactual mind-set as a debiasing technique stacks up against other procedures for promoting analytic thinking. Procedures that encourage decision makers to “consider the opposite” (Lord, Lepper, & Preston, 1984; Mussweiler, Strack, & Pfeiffer, 2000) or assign group members the role of a “devil’s advocate” (Cosier, 1978; Janis & Mann, 1977) work by explicitly directing decision makers to become more critical. However, no explicit training or assignment of roles is required when a counterfactual mind-set is subtly activated. Perhaps the activation of a counterfactual mindset to promote analytic thinking may be advantageous in delicate situations in which it is particularly important to avoid the appearance of a heavy hand guiding the decision-making process. Likewise, because blatant attempts to restrict an individual’s freedom often provoke reactance (Brehm, 1966), the counterfactual mindset approach to guiding decision-making processes may be more readily embraced by decision makers than the more directive approaches described above.

Conclusion The present set of experiments provides evidence across multiple domains that counterfactual mind-sets promote a relational processing style, which is characterized by a tendency to consider relationships and associations among a class of stimuli and to structure thought and imagination around those associations. As a result, performance on analytic and creative tasks requiring the identification of logical relationships and associations is aided, but performance on creative tasks requiring the generation of novel

COUNTERFACTUAL MIND-SETS

ideas matters is hindered. On a general level, the present research suggests that reflecting back on events in which an outcome almost turned out differently and mentally constructing an alternate world can impact how future problems are approached. More specifically, thinking about what may have been can prevent one from creating novel ideas but can lead that same person to notice hidden connections. Simply put, counterfactual mind-sets promote thinking within the box.

References Amabile, T. M. (1983). The social psychology of creativity: A componential conceptualization. Journal of Personality and Social Psychology, 45, 357–376. Amabile, T. M., Conti, R., Coon, H., Lazenby, J., & Herron, M. (1996). Assessing the work environment for creativity. Academy of Management Journal, 39, 1154 –1184. Brehm, J. W. (1966). A theory of psychological reactance. New York: Academic Press. Byrne, R. M. J., & Tasso, A. (1994, August). Counterfactual reasoning: Inferences from hypothetical conditional. In A. Ram & K. Eiselt (Eds.), Proceedings of the 16th Annual Conference of the Cognitive Science Society. Hillsdale, NJ: Erlbaum. Cosier, R. A. (1978). The effects of three potential aids for making strategic decisions on prediction accuracy. Organizational Behavior and Human Decision Processes, 22, 295–306. Duncker, K. (1945). On problem solving. Psychological Monographs, 58, 270. Einhorn, H. J., & Hogarth, R. M. (1986). Judging probable cause. Psychological Bulletin, 99, 3–19. Feldman, L. A. (1995). Valence focus and arousal focus: Individual differences in the structure of affective experience. Journal of Personality and Social Psychology, 69, 153–166. Friedman, R. S., & Forster, J. (2001). The effects of promotion and prevention cues on creativity. Journal of Personality and Social Psychology, 81, 1001–1013. Galinsky, A. D., & Kray, L. J. (2004). From thinking about what might have been to sharing what we know: The effects of counterfactual mind-sets on information sharing in groups. Journal of Experimental Social Psychology, 40, 606 – 618. Galinsky, A. D., & Moskowitz, G. B. (2000). Counterfactuals as behavioral primes: Priming the simulation heuristic and consideration of alternatives. Journal of Experimental Social Psychology, 36, 257–383. Gleicher, F., Boninger, D. S., Strathman, A., Armor, D., Hetts, J., & Ahn, M. (1995). With an eye toward the future: The impact of counterfactual thinking on affect, attitudes, and behavior. In N. J. Roese & J. M. Olson (Eds.), What might have been: The social psychology of counterfactual thinking (pp. 283–304). Hillsdale, NJ: Erlbaum. Glucksberg, S., & Weisberg, W. R. (1966). Verbal behavior and problem solving: Effects of labeling in a functional fixedness problem. The Journal of Experimental Psychology, 71, 659 – 664. Guilford, J. P. (1950). Creativity. American Psychologist, 5, 444 – 454. Janis, I. L., & Mann, L. (1977). Emergency decision making: A theoretical analysis of responses to disaster warnings. Journal of Human Stress, 3, 35– 48. Kahneman, D., & Miller, D. T. (1986). Norm theory: Comparing reality to its alternatives. Psychological Review, 93, 136 –153. Kelley, H. H. (1973). The processes of causal attribution. American Psychologist, 28, 107–128. Kray, L. J., & Galinsky, A. D. (2003). The debiasing effect of counterfactual mind-sets: Increasing the search for disconfirmatory information in group decisions. Organizational Behavior and Human Decision Processes, 91, 69 – 81.

47

Krishnamurthy, P., & Sivaraman, A. (2002). Counterfactual thinking and advertising responses. Journal of Consumer Research, 28, 650 – 658. Law School Admission Council. (2005).Preparing for the LSAT. Retrieved April 11, 2006, from http://www.lsac.org/pdfs/2006 –2007/TestPrep06.pdf Liljenquist, K. A., Galinsky, A. D., & Kray, L. J. (2004). Exploring the rabbit hole or possibilities by myself or with my group: The benefits and liabilities of activating counterfactual mind-sets for information sharing and group coordination. Journal of Behavioral Decision Making, 17, 263–279. Lord, C. G., Lepper, M. R., & Preston, E. (1984). Considering the opposite: A corrective strategy for social judgment. Journal of Personality and Social Psychology, 47, 1231–1243. Mandel, D. R. (2003). Judgment dissociation theory: An analysis of differences in causal, counterfactual, and covariational reasoning. Journal of Experimental Psychology: General, 132, 419 – 434. Mandel, D. R., & Lehman, D. R. (1996). Counterfactual thinking and ascriptions of cause and preventability. Journal of Personality and Social Psychology, 71, 450 – 463. Markman, K. D., & McMullen, M. N. (2003). A reflection and evaluation model of comparative thinking. Personality and Social Psychology Review, 7, 244 –267. McGill, A. L. (1998). Relative use of necessity and sufficiency information in causal judgments about natural categories. Journal of Personality and Social Psychology, 75, 70 – 81. McGill, A. L., & Klein, J. G. (1993). Contrastive and counterfactual reasoning in causal judgment. Journal of Personality and Social Psychology, 64, 897–905. Mednick, M. T., Mednick, S. A., & Mednick, E. V. (1964). Incubation of creative performance and specific associative priming. Journal of Abnormal and Social Psychology, 69, 84 – 88. Mednick, S. A. (1962). The associative basis of the creative process. Psychological Review, 69, 220 –232. Merriam-Webster. (2006). Merriam-Webster online dictionary. Retrieved April 11, 2006, from http://www.m-w.com/dictionary/analysis. Mussweiler, T., Strack, F., & Pfeiffer, T. (2000). Overcoming the inevitable anchoring effect: Considering the opposite compensates for selective accessibility. Personality and Social Psychology Bulletin, 26, 1142– 1150. O’Hara, L. A., & Sternberg, R. J. (2001). It doesn’t hurt to ask: Effects of instructions to be creative, practical, or analytical on essay-writing performance and their interaction with students’ thinking styles. Creativity Research Journal, 13, 197–210. Peterson, R. S., & Nemeth, C. J. (1996). Focus versus flexibility: Majority and minority influence can both improve performance. Personality and Social Psychology Bulletin, 22, 14 –23. Princeton Review. (2005). The LSAT in detail. Retrieved May 3, 2006, from http://www.princetonreview.com/law/testprep/testprep.asp? TPRPAGE⫽87&TYPE⫽LSAT-SECTIONS Roese, N. J. (1994). The functional basis of counterfactual thinking. Journal of Personality and Social Psychology, 66, 805– 818. Roese, N. J., & Hur, T. (1997). Affective determinants of counterfactual thinking. Social Cognition, 15, 274 –290. Roese, N. J., Hur, T., & Pennington, G. L. (1999). Counterfactual thinking and regulatory focus: Implications for action versus inaction and sufficiency versus necessity. Journal of Personality and Social Psychology, 77, 1109 –1120. Roese, N. J., & Olson, J. M. (1995). Functions of counterfactual thinking. In N. J. Roese & J. M. Olson (Eds.), The social psychology of counterfactual thinking (pp. 169 –197). Hillsdale, NJ: Erlbaum. Roese, N. J., & Olson, J. M. (1997). Counterfactual thinking: The intersection of affect and function. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 29, pp. 1–59). New York: Academic Press.

KRAY, GALINSKY, AND WONG

48

Rubin, D. C., Stoltzfus, E. R., & Wall, K. L. (1991). The abstraction of form in semantic categories. Memory & Cognition, 19, 1–7. Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and Social Psychology, 39, 1161–1178. Schwarz, N., & Bless, H. (1991). Happy and mindless, but sad and smart? The impact of affective states on analytic reasoning. In J. P. Forgas (Ed.), Emotion and social judgments: International series in experimental social psychology (pp. 55–71). Elmsford, NY: Pergamon Press. Smith, S. M., Ward, T. B., & Schumacher, J. S. (1993). Constraining effects of examples in a creative generation task. Memory & Cognition, 21, 837– 845. Sternberg, R. J. (1988). Mental self-government: A theory of intellectual styles and their development. Human Development, 31, 197–224.

Sternberg, R. J., & Wagner, R. K. (1991). MSG thinking styles inventory manual. Unpublished manuscript. Tetlock, P. E., & Lebow, R. N. (2001). Poking counterfactual holes in covering laws: Cognitive styles and historical reasoning. American Political Science Review, 95, 829 – 843. Ward, T. B. (1994). Structured imagination: The role of category structure in exemplar generation. Cognitive Psychology, 27, 1– 40. Wason, P. C. (1966). Reasoning. In B. Foss (Ed.), New horizons in psychology (pp. 135–151). Middlesex, England: Penguin. 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. Wells, G. L., & Gavanski, I. (1989). Mental simulation of causality. Journal of Personality and Social Psychology, 56, 161–169.

Appendix A Examples Provided in Creative Generation Task Used in Experiments 4 and 5 (1)

Please generate a name for: a new pasta

Examples: radon, plutonium, argon, carbon, radium, uranium

Examples: spaghetti, lasagna, fettuccini, rotini, pastina, rigatoni (2)

(3)

Please generate a name for: a nuclear element

Please generate a new name for: an analgesic (pain reliever)

Examples: Tylenol, Anacin, aspirin, bufferin, panadol, Midol

Appendix B Modified Version of the Remote Associates Task Please identify a common word that links each set of 3 words together. You should answer as many questions as you can. Example: sore—shoulder—sweat Answer: cold

(6)

chocolate—fortune—tin (cookie)

(7)

barrel—root— belly (beer)

(1)

blank—white—lines ( page)

(8)

broken— clear— eye (glass)

(2)

magic—plush—floor (carpet)

(9)

pure— blue—fall (water)

(3)

thread—pine—pain (needle)

(10)

(4)

stop—petty—sneak (thief)

(5)

envy— golf— beans (green)

widow— bite—monkey (spider)

(Note: Answers appear in parentheses.)

Received June 17, 2005 Revision received September 28, 2005 Accepted September 29, 2005 䡲

Journal of Personality and Social Psychology 2006, Vol. 91, No. 1, 49 – 62

Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.49

Self-Regulatory Processes Defend Against the Threat of Death: Effects of Self-Control Depletion and Trait Self-Control on Thoughts and Fears of Dying Matthew T. Gailliot

Brandon J. Schmeichel

Florida State University

Texas A&M University

Roy F. Baumeister Florida State University Nine studies (N ⫽ 979) demonstrated that managing the threat of death requires self-regulation. Both trait and state self-control ability moderated the degree to which people experienced death-related thought and anxiety. Participants high (vs. low) in self-control generated fewer death-related thoughts after being primed with death, reported less death anxiety, were less likely to perceive death-related themes in ambiguous scenes, and reacted with less worldview defense when mortality was made salient. Further, coping with thoughts of death led to self-regulatory fatigue. After writing about death versus a control topic, participants performed worse on several measures of self-regulation that were irrelevant to death. These results suggest that self-regulation is a key intrapsychic mechanism for alleviating troublesome thoughts and feelings about mortality. Keywords: self-regulation, mortality salience, death anxiety, thought suppression, terror management

demand and consume some of an individual’s limited resources for self-regulation (Muraven & Baumeister, 2000).

The thought of death can be frightening. As a consequence, people prefer to avoid thinking about death (e.g., Aries, 1981; Becker, 1973). By what means are people capable of avoiding thoughts of death? The current work assesses the role of selfregulation in minimizing thoughts and anxiety surrounding death. Self-regulation (or self-control) is the capacity to override one’s thoughts, feelings, and habitual patterns of behavior. Selfregulation is a highly adaptive capacity that facilitates success in myriad domains of life, including interpersonal relationships, academic achievement, and coping with and adjusting to stress (e.g., Baumeister, Heatherton, & Tice, 1994; Mischel, Shoda, & Peake, 1988; Shoda, Mischel, & Peake, 1990; Tangney, Baumeister, & Boone, 2004). Awareness of mortality creates the potential for stress (e.g., see Greenberg et al., 2003; Pollak, 1979), and so it seems plausible that one benefit of self-regulation may be to minimize death-related thoughts and anxiety. Hence our central hypotheses in this investigation were (a) that low capacity for self-regulation, either as state or trait, would increase vulnerability to disturbing thoughts and feelings about death and (b) that coping with thoughts of death, like other acts of self-control, should

Self-Regulation: Trait and State Differences Some individuals are highly adept at self-regulation whereas others are not, and these individual differences in trait self-control have been associated with a diverse range of behavior. For example, compared with those lower in trait self-control, people higher in trait self-control are better at coping with anxiety and other negative moods, avoiding addictive behaviors, and responding to other people in prosocial, constructive ways (e.g., Finkel & Campbell, 2001; Mischel et al., 1988; Tangney et al., 2004). The diverse and substantial benefits of high trait self-control underscore the idea that self-regulation is a valuable tool in many aspects of life. Self-regulation can vary not just as trait but also as state. For any individual, self-regulatory success is more likely at some times than at others. Recent work has suggested that self-regulation relies upon a limited resource or strength (for reviews, see Gailliot & Baumeister, in press; Muraven & Baumeister, 2000). Effortful acts of self-regulation appear to consume or deplete this limited resource, thereby impairing later attempts at self-regulation. For instance, participants in one study who performed a task that required self-control (i.e., suppressing or exaggerating responses to an emotional film) were less able to exert self-control on a subsequent task (i.e., squeezing a handgrip) than were participants who first completed a task that did not require self-control (i.e., watching the film without regulating emotional responses; Muraven, Tice, & Baumeister, 1998). Presumably, the initial act of self-control temporarily depleted self-regulatory strength, thereby impairing subsequent self-regulated performance.

Matthew T. Gailliot and Roy F. Baumeister, Department of Psychology, Florida State University; Brandon J. Schmeichel, Department of Psychology, Texas A&M University. This research was supported by National Institutes of Health Grant MH 65559. Thanks to Lauren Brewer, Shenika Thomas, Savana Carroll, and Daniel Donaldson for helping to conduct the studies reported here. Correspondence concerning this article should be addressed to Matthew T. Gailliot or Roy F. Baumeister, Department of Psychology, Florida State University, One University Way, Tallahassee, FL 32306-1270. E-mail: [email protected] or [email protected] 49

50

GAILLIOT, SCHMEICHEL, AND BAUMEISTER

Thus, the ability to self-regulate may differ both across individuals (trait self-control) and within individuals across time (selfcontrol depletion). The current work examined whether poor trait and state self-regulation, respectively, would hamper the management of concerns about death and whether the management of concerns about death would hamper self-regulation.

Self-Regulation and the Management of Mortality Concerns The present work is based on the assumption that the idea of death is threatening and can evoke aversive thoughts and feelings. There are at least two ways in which self-regulation might facilitate the defense against these unwanted states. The first is thought control. Research indicates that people cope with thoughts of death and minimize preoccupation with death in part by suppressing such thoughts or redirecting their thoughts away from death (e.g., Greenberg, Pyszczynski, Solomon, Simon, & Breus, 1994; Harmon-Jones et al., 1997; Pollak, 1979). For instance, on explicit measures of mortality concerns (e.g., measures of death anxiety), most people score relatively low. On more implicit measures (e.g., word-association tests), however, people score significantly higher (e.g., Feifel & Branscomb, 1973; Pollak, 1979). This suggests that people possess an implicit awareness of death and that such thoughts are suppressed from explicit awareness. Moreover, suppressing thoughts about death requires effortful, controlled processing (Greenberg et al., 1994; Harmon-Jones et al., 1997; Pyszczynski, Greenberg, & Solomon, 1999; see also Wegner, 1994). For example, immediately after thinking about mortality, one’s death thoughts are suppressed and are less accessible to awareness (e.g., Arndt, Greenberg, Solomon, Pyszczynski, & Simon, 1997; Greenberg et al., 1994; Harmon-Jones et al., 1997). If cognitive resources are diverted by a concurrent task, however, death thoughts remain highly accessible to awareness (Arndt et al., 1997; Greenberg, Arndt, Schimel, Pyszczynski, & Solomon, 2001). Apparently the cognitive load diverts the resources needed to suppress death thoughts, so death thoughts remain highly accessible (see Smart & Wegner, 1999; Wegner & Zanakos, 1994; Wenzlaff & Wegner, 2000). Low trait self-control and depleted self-control strength might also undermine the suppression of death thoughts. Thought suppression requires self-control (e.g., Baumeister et al., 1994; Wegner, 1994), and so dispositionally low or temporarily depleted self-control should be linked to poorer suppression, leading to increased accessibility of death-related thoughts and anxiety. Effortfully suppressing a thought reduces both its explicit and implicit accessibility (Anderson & Green, 2001; MacLeod, 1989; McBride & Dosher, 1997), and disrupting controlled efforts toward suppressing death thoughts increases the implicit accessibility of those thoughts (e.g., Arndt et al., 1997; Pyszczynski et al., 1999). Low self-control should therefore be associated with increased accessibility of both implicit and explicit death thoughts. Likewise, thought suppression depletes self-control strength (e.g., Gordijn, Hindriks, Koomen, Dijksterhuis, & Van Knippenberg, 2004; Muraven et al., 1998). After thinking about death, people suppress thoughts about death. Self-regulatory resources should be depleted as a result. Self-regulation should therefore be impaired shortly after people think about and suppress thoughts of death.

A second way that self-regulation might facilitate the management of concerns about death is the regulation of emotion. An individual who capably controls his or her emotions is likely to experience less of the fear and anxiety associated with death. Emotion regulation requires self-control, and therefore both low trait self-control and self-control depletion undermine emotion regulation (e.g., Baumeister, Bratslavsky, Muraven, & Tice, 1998; Finkel & Campbell, 2001; Muraven et al., 1998; Tangney et al., 2004). Consequently, low trait self-control and self-control depletion should be associated with increased emotional preoccupation with death. In sum, we posited that self-regulation should facilitate suppressing thoughts about death and, to some extent, regulating emotional responses to concerns about death. Another potential basis for predicting a link between selfregulation and death-thought accessibility, however, involves selfesteem. Terror management theory has suggested that self-esteem is primarily a defense against death, so chronic low self-esteem could leave people more vulnerable to thoughts and anxieties about death (e.g., see Pyszczynski, Greenberg, Solomon, Arndt, & Schimel, 2004). The present work is not concerned with selfesteem per se but with the possibility that self-esteem could account for the effects of self-regulation. Trait self-esteem and trait self-control are positively correlated (Tangney et al., 2004), hence it is theoretically possible that any effects of self-control are due to self-esteem. It is also plausible that high self-esteem results from having more effective self-control, so people with high self-esteem might be less vulnerable to self-control depletion. To test these possibilities, we assessed the role of self-esteem.

Research Overview In nine studies, using both correlational and experimental methods, we tested whether self-regulation facilitates the management of mortality concerns. We first examined whether poor selfregulation would hamper the management of mortality concerns. Our first studies tested the prediction that people with low (vs. high) trait self-control would be more prone to come up with death-related thoughts in response to ambiguously evocative stimuli (Studies 1A and 1B) and would have higher death anxiety (Study 1C). Next, we manipulated temporary self-regulatory capacity by having participants complete either an initial task that required self-regulation or one that did not require self-regulation. To the extent that exercising self-control depletes the capacity for self-regulation (e.g., see Muraven & Baumeister, 2000), we expected to find greater death-thought accessibility (Study 2) and a greater proclivity to perceive death-related themes in an ambiguous stimulus (Study 3) among participants who had (vs. had not) previously expended their self-regulatory resources. Subsequently, we assessed (rather than manipulated) self-control depletion and death-thought accessibility and death anxiety (Study 4). In the final study in this group (Study 5), we sought to show that low trait self-control would leave people more vulnerable to the threat of mortality salience and would result in a heightened tendency to respond to that threat by bolstering support for a political leader (Landau et al., 2004). Such an externalizing response would presumably indicate that the inner defense mechanisms against death were inadequate. Having thus shown the effects of poor self-regulation on mortality defenses, we turned this around to investigate the effects of

THOUGHTS AND FEARS OF DEATH AND SELF-REGULATION

51

mortality defense on self-regulation. If (as argued above) selfregulation consumes a limited resource and defending against the threat of death consumes some of that resource then people’s capacity for self-regulation should be diminished after they defend against such a threat. Participants in the last four studies wrote about either death or a control topic. We predicted that writing about death would activate defensive processes that would deplete self-regulatory resources, as indicated by poorer subsequent performance on the Stroop task (Study 6), impaired logical reasoning abilities (Study 7), reduced success at solving anagrams (Study 8), and less effortful persistence on word puzzles (Study 9).

Studies 1A–1C In Studies 1A–1C we relied on measurement of trait self-control to test the hypothesis that weaker self-control would be associated with greater susceptibility to thoughts about death. We measured death-thought accessibility in different ways in each of these studies. In Study 1A we used ambiguous word puzzles that could be solved with either death-related or death-irrelevant words (e.g., grave, grape). We believed that greater accessibility of death thoughts among persons low in self-control would be reflected in higher rates of death-related word solutions. In a similar vein, in Study 1B we showed participants an ambiguous visual stimulus that could be construed as either related or unrelated to death. We then assessed spontaneous thoughts to determine the frequency of death-related thoughts the image evoked. In Study 1C we relied on a straightforward measure of death anxiety (Templer, 1970) to ascertain whether low self-control would be associated with greater fear related to death.

Method Participants. Participants in these and all subsequent studies were undergraduate students enrolled in introductory psychology courses who received either extra credit or credit toward fulfilling a course requirement. The size and gender makeup of the samples in Studies 1A–1C were as follows: Study 1A (N ⫽ 279; 166 women, 113 men), Study 1B (N ⫽ 13; 12 women, 1 man), and Study 1C (N ⫽ 163; 118 women, 45 men). Materials and procedures. At the start of the academic semester, participants completed measures of trait self-control (the Self-Control Scale; Tangney et al., 2004) and self-esteem (the Rosenberg Self-Esteem Scale; Rosenberg, 1965) during a mass-testing session. We used the brief version of the Self-Control Scale, which contains 13 items (e.g., “I have a hard time breaking bad habits” [reverse scored]; “I am good at resisting temptation”) answered on a scale from 1 (not at all like me) to 5 (very much like me). The Rosenberg Self-Esteem Scale contains 10 items (e.g., “On the whole, I am satisfied with myself.”) answered on a scale from 1 (strongly disagree) to 5 (strongly agree). Higher scores on these measures indicate higher self-control and self-esteem, respectively. Participants in Study 1C also completed the Social Interaction Anxiety Scale (Mattick & Clarke, 1998), as a measure of social anxiety, and the State–Trait Anxiety Index (Spielberger, Gorsuch, & Lushene, 1970), as a measure of general trait anxiety. These measures were used to control for general negative affect. Studies 1A–1C were conducted 4 –12 weeks after the initial mass-testing session. In each study, participants completed questionnaire packets that contained the focal materials along with other, non-death-related measures. In Study 1A, the packet included a list of 20 word fragments, some of which could be completed with death-related thoughts (e.g., sk_ll and gra_e could be solved with the words skull and grave or skill and grape, respectively). We counted and then standardized the number of death-

Figure 1. “All Is Vanity” by Charles Allan Gilbert. This picture, shown to participants in Study 1B, could be interpreted as either a woman or a skull.

related words used to complete the relevant word fragments to serve as a measure of implicit death-thought accessibility (see Greenberg et al., 1994).1 The questionnaire packet completed by participants in Study 1B included a measure of explicit death-thought accessibility. Specifically, these participants were shown a picture that could be interpreted as depicting either (a) a woman sitting in front of a mirror while putting on makeup or (b) a skull (see Figure 1). Participants were instructed to list the first 10 words that came to mind as they viewed the image. A judge who was blind to trait self-control scores determined which thoughts on the thought-listing task were related to death (e.g., skull and skeleton). The number of thoughts related to death constituted the measure of explicit death-thought accessibility. The questionnaire packet completed by participants in Study 1C contained the Death Anxiety Scale (Templer, 1970), which includes 15 items (e.g., “I am very much afraid to die” and “I fear dying a painful death”)

1

Participants completed a list that contained either 5 or 7 word fragments that could be solved with death-related thoughts. Accordingly, the number of death-related thoughts was standardized among participants who completed each list.

GAILLIOT, SCHMEICHEL, AND BAUMEISTER

52

answered as either “true” or “false.” Higher scores indicate greater death anxiety. In these and all subsequent studies, participants were last thanked and fully debriefed. Because of the threatening nature of death, we were careful to explain the importance of and need for the research and to assuage any fears or concerns participants may have had. In this fashion, we tried to ensure that no participant left feeling distressed.

Results In each study, higher trait self-control was associated with less preoccupation with death. Trait self-control correlated negatively with the number of word fragments completed with death-related words, r(279) ⫽ ⫺.13, p ⬍ .05 (Study 1A), the number of death thoughts listed to the ambiguous image, r(13) ⫽ ⫺.59, p ⬍ .05 (Study 1B), and death anxiety, r(163) ⫽ ⫺.36, p ⬍ .01 (Study 1C). Further analyses indicated that each relationship remained significant when controlling for self-esteem (all ps ⱕ .05) and that controlling for self-esteem did not change the strength of any relationship (all ps ⬎ .11). Self-esteem thus did not appear to account for the relationship between self-control and mortal concern. Moreover, the relationship between self-control and death anxiety in Study 1C remained significant when controlling for social anxiety and trait anxiety (scores on the Social Interaction Anxiety Scale and State–Trait Anxiety Index, respectively; ps ⬍ .05). This suggests that the association between high trait self-control and low death anxiety was not simply the result of trait self-control buffering against negative affect in general.

Discussion Studies 1A–1C provided initial evidence for the view that selfregulation protects people from the aversive awareness of death and mortality. High self-control predicted less explicit (death anxiety and responses to an image of a skull) and implicit (responses to word fragments) preoccupation with death. We assume that people periodically confront cues or thoughts that remind them of death and that they therefore exercise selfregulation to prevent these thoughts from lingering in conscious awareness and escalating into greater anxiety. These results suggest that low self-control leaves people less effectively defended against the threatening idea of death and hence more prone to suffer both disturbing thoughts and anxious emotions in connection with death.

Study 2 In Studies 2 and 3 we used laboratory manipulations of selfregulation (rather than correlational evidence) to test the causal hypothesis that low self-control weakens defenses against death. In Study 2 we examined whether temporarily depleted self-regulatory strength would increase the cognitive accessibility of death-related thought. People typically try to minimize or suppress thoughts associated with death. We hypothesized that if self-regulation is needed to suppress death thoughts, then self-regulatory depletion should undermine this suppression and lead to increased accessibility of death-related thoughts. Participants in Study 2 first completed either a task that required self-regulation (thought suppression) or a control task. Thought

suppression requires self-control and has been shown to deplete self-regulatory strength (Muraven et al., 1998). Following this initial task, participants completed the same measure of implicit death thoughts used in Study 1A (word fragment completion). We predicted that participants in the depletion condition would complete the word fragments with death-related words more often than would participants in the no-depletion condition, indicating greater accessibility of death-related thoughts when self-regulatory capacity had been reduced.

Method Participants. Participants were 19 undergraduates (13 women, 6 men) who completed the Rosenberg Self-Esteem Scale during a mass-testing session at the start of the semester. The present study occurred approximately 3 months later. Participants were run individually and told the study was to investigate impression formation and thought patterns. Procedure. The first task served as the manipulation of self-regulatory resources. Specifically, participants completed a thought-recording exercise (borrowed from Wegner, Schneider, Carter, & White, 1987). Participants were given a sheet of paper and asked to write down all of their thoughts for 5 min. Participants randomly assigned to the no-depletion condition received no additional instructions. Participants assigned to the depletion condition were instructed further not to think about a white bear. Each time they did happen to think about a white bear, they were to place a mark on the page and attempt to stop thinking about a white bear. After the thought-listing task, participants rated the difficulty of the task (as a manipulation check), and they completed the Brief Mood Introspection Scale (BMIS; Mayer & Gaschke, 1988). The BMIS contains 20 items indicative of mood (e.g., happy, sad) and arousal (e.g., peppy, drowsy). Participants rated each item to indicate how they were feeling at the present moment, using a scale from 1 (definitely do not feel) to 7 (definitely feel). Participants also indicated to what extent they tried to follow the task instructions (effort) and how well they felt that they followed those instructions (perceptions of task performance). Next, participants completed an unrelated filler task (i.e., making personality judgments of children in a video) that bolstered the cover story. The personality-judgment task took approximately 4 min. Afterward, participants completed the same series of word fragments used in Study 1A.

Results and Discussion Manipulation check. Participants in the depletion condition rated their suppression task as being somewhat more difficult (M ⫽ 3.78, SD ⫽ 1.92) than did the participants in the nodepletion condition (M ⫽ 2.40, SD ⫽ 1.26), t(17) ⫽ 2.53, p ⫽ .08. This pattern suggests that the depleting task required more effortful self-regulation than did the nondepleting task. Implicit death thoughts. We predicted and confirmed that death-related thoughts would be more accessible to the minds of depleted than to those of nondepleted participants. Depleted participants (M ⫽ 2.22, SD ⫽ 1.48) solved more word fragments with death-related words than did nondepleted participants (M ⫽ 0.90, SD ⫽ 1.20), t(17) ⫽ ⫺2.15, p ⬍ .05. This pattern is consistent with the view that self-regulatory resources help reduce deaththought accessibility. Participants who possessed greater selfregulatory strength (i.e., nondepleted participants) exhibited less death-thought accessibility than did participants who had previously expended their self-regulatory strength. Presumably, depleted participants were less able to suppress thoughts of death, and thus death thoughts increased in accessibility. The wordfragment completion task required participants to execute a mental

THOUGHTS AND FEARS OF DEATH AND SELF-REGULATION

search for possible solutions, and solutions related to death appeared to be more accessible when self-regulatory resources were depleted. Self-esteem, mood, arousal, task performance, and effort. Additional analyses suggested that the difference in accessibility of implicit death thoughts between depleted and nondepleted participants was not attributable to differences in self-esteem, mood, arousal, effort, or perceptions of task performance. Regression analysis indicated that self-esteem did not moderate the effect of depletion on the accessibility of death thoughts (t ⬍ 1, ns). Further, depleted and nondepleted participants did not differ in arousal, task performance, or effort (all ps ⬎ .24), nor did any of these factors significantly predict the number of implicit death thoughts (all ps ⬎ .24). Depleted and nondepleted participants did differ in self-reported mood valence ( p ⬍ .05) such that depleted participants were in a more negative mood, but mood did not significantly predict the number of implicit death thoughts ( p ⬎ .32), and the difference in death thoughts between the two conditions was significant even when controlling for mood, F(1, 16) ⫽ 3.19, p ⬍ .05 (one-tailed).

Study 3 The purpose of Study 3 was to provide converging evidence that self-regulatory depletion increases death-thought accessibility. The design of the study mirrored that of Study 2. Participants first completed either a task (controlling attention while watching a video) that has been shown to deplete self-regulatory strength (e.g., Schmeichel, Vohs, & Baumeister, 2003) or a control task. Participants then completed a measure of death-thought accessibility for which they listed their thoughts about an ambiguous image. We predicted that participants who first performed the self-regulatory (vs. the control) task would respond to the ambiguous image with more death-related thoughts because they had depleted selfregulatory strength.

53

BMIS) 4 –5 min after the initial video-watching task. Participants subsequently completed a measure of death-thought accessibility (listing thoughts to an ambiguous drawing) that was identical to the one used in Study 1B except that a different drawing was used (see Figure 2).

Results and Discussion Explicit death thoughts. Depleted participants (M ⫽ 1.34, SD ⫽ 1.00) listed more death-related thoughts while viewing the drawing than did nondepleted participants (M ⫽ .97, SD ⫽ .45), t(65) ⫽ 1.99, p ⫽ .05. Presumably, participants who possessed full self-regulatory strength were able to suppress thoughts of death while viewing the drawing. Depleted participants, lacking selfregulatory strength, were less able to suppress such thoughts. The results of Study 3 thus replicated those of Study 2, using different methods. In both studies, weakening self-control by depleting its resources led participants to have more thoughts about death. Moreover, the finding that depletion increased death thoughts and anxiety after a short delay indicates that the effects of depletion are somewhat enduring. One may very well be susceptible to greater mortal concern long after having exerted self-control. Mood and arousal. Analyses indicated that the effect of depletion on the accessibility of death-related thoughts was not mediated by mood or arousal. Depleted and nondepleted participants did not differ in arousal ( p ⬎ .19), but depleted participants

Method Participants. Sixty-seven undergraduate students (49 women, 18 men) participated in a classroom setting. They were told that the study was to investigate people’s attitudes and opinions. Procedure. The first task served as the manipulation of self-regulatory resources. Specifically, participants watched a 6-min video (without sound) of a woman talking (modified from Gilbert, Krull, & Pelham’s, 1988, study). In the bottom corner of the screen, common one-syllable words (e.g., hair, hat) appeared individually for 10 s. Participants randomly assigned to the depletion condition were instructed to focus their attention on the woman’s face and to refrain from looking at the words. If they happened to look at the words, they were to refocus their attention on the woman as quickly as possible. Attention automatically orients toward novel stimuli appearing in the environment (e.g., see Shiffrin & Schneider, 1977), and so the task required these participants to exert self-control by overriding prepotent orienting of attention to the words and maintain attention instead on the woman’s face only. Participants assigned to the no-depletion condition were instructed to watch the video as they would normally (i.e., as if they were sitting at home watching TV) and hence were not required to exert self-control. After the video-watching task, participants completed the target materials (described below) that were embedded among other measures (not related to death) that helped bolster the cover story about attitudes and opinions. Participants completed measures of mood and arousal (the

Figure 2. Artwork (Gillam) from the May 1894 cover of Judge magazine. This picture, shown in Study 3, could be interpreted as either two men or a skull.

GAILLIOT, SCHMEICHEL, AND BAUMEISTER

54

were in a somewhat more negative mood ( p ⫽ .06). However, neither mood nor arousal significantly predicted the number of death thoughts ( ps ⫽ .22 and .38, respectively), and the difference in death thoughts between the two conditions was significant, even when controlling for mood, F(1, 63) ⫽ 2.93, p ⬍ .05 (one-tailed).

Study 4 The purpose of Study 4 was to provide additional evidence that self-regulatory depletion increases death thoughts and anxiety. Specifically, we assessed (rather than manipulated) temporary self-regulation abilities using a recently developed scale that has been shown to measure self-regulatory depletion (Twenge, Muraven, & Tice, 2004; see also Finkel & Campbell, 2001, for a similar measure). Participants then completed measures of implicit death-thought accessibility (word fragments) and death anxiety. We hypothesized the following: If self-regulation helps to minimize death-thought accessibility and death anxiety, then participants who report more self-regulatory depletion should exhibit more death-thought accessibility and death anxiety than should participants who report relatively less self-regulatory depletion.

Method Participants. One hundred fifty-nine undergraduate students (110 women, 49 men) participated. The study was conducted in a classroom setting, and participants were told that it was to investigate attitudes and vocabulary. They completed questionnaire packets that contained all instructions and materials (i.e., the focal materials along with other, nondeath-related personality measures.) Procedure. Specifically, participants completed a measure that assessed perceptions of the momentary availability of self-regulatory resources (the State Depletion Scale; Twenge et al., 2004). The State Depletion Scale contains 25 items (e.g., “I feel like my willpower is gone.” “My mental energy is running low.”) answered on a scale from 1 (not true) to 7 (very true; ␣ ⫽ .91). Higher scores indicate greater self-regulatory depletion. Subsequent measures in the packet assessed implicit death-thought accessibility and death anxiety. Specifically, participants completed the Death Anxiety Scale (Templer, 1970) and the same word fragments as used in Study 1A and Study 2.

Study 5 The purpose of Study 5 was to test the hypothesis that good self-regulation reduces concerns about death by examining the consequences of increased death concerns rather than the extent of them. Specifically, Study 5 examined the role of trait self-control in one well-documented consequence of mortality salience: increased worldview defense. When mortality is made salient, people bolster faith in their own culture and react more positively toward others who support their cultural norms and values, and they react more negatively toward others who disagree with their cultural norms and values (e.g., Florian & Mikulincer, 1997; Greenberg et al., 1990; Heine, Harihara, & Niiya, 2002; Ochsmann & Mathey, 1994). Likewise, mortality salience has been shown to increase support for political leaders. For instance, after writing about death or the September 11, 2001, terrorist attacks on the United States, participants (from the United States) showed increased support for current President George W. Bush compared with participants who wrote about merely aversive topics such as dental pain (Landau et al., 2004). Study 5 of the current investigation was a partial replication of the Landau et al. (2004) study that assessed the relationships among mortality salience, trait self-control, and support for the President. We made a few hypotheses about the results of Study 5: If self-control is used to reduce death-related thought and anxiety, then participants higher in trait self-control should be more capable of managing concerns about death following a mortality-salience induction than should participants lower in trait self-control. The more effective suppression abilities of individuals high (vs. low) in trait self-control should reduce both implicit and explicit deaththought accessibility (Anderson & Green, 2001; MacLeod, 1989; McBride & Dosher, 1997). To the extent that support for President Bush is exacerbated by death concerns, participants high in trait self-control should not show increased support for the President after thinking about death. Participants first did or did not write about their own death. Then they read a short passage in support of President Bush and indicated how much they agreed with the passage. Our prediction was that mortality salience would boost support for the President but only among participants with low trait self-control.

Method Results and Discussion State self-control depletion correlated positively with implicit death thoughts, r(159) ⫽ .20, and death anxiety, r(159) ⫽ .43 (both ps ⬍ .05). Participants who indicated that they were relatively more depleted solved more word fragments with deathrelated words and reported more death anxiety than did participants whose self-regulatory resources were less depleted. These findings converge with those of the previous studies. Measured self-control depletion revealed the same association between depletion and death-thought accessibility as did manipulated self-control depletion in Studies 2 and 3. Presumably, participants who were relatively less depleted were more capable of managing concerns about death and thus experienced relatively less preoccupation with death. The results thus far are highly consistent with the hypothesis that self-regulation minimizes death-thought accessibility and death anxiety.

Participants. Participants were 88 undergraduates (41 women, 47 men) who completed a brief measure of trait self-control (Tangney et al., 2004) and self-esteem (Rosenberg, 1965) during a mass-testing session earlier in the semester. The present study was conducted in classrooms in groups of 10 –25 participants approximately 2 weeks prior to the 2004 U.S. presidential election (some 7 weeks after the mass-testing session). Participants were told that the study was to investigate personality characteristics and political attitudes. They received a packet that contained all materials for the experiment. Procedure. Participants first completed a demographic questionnaire and two filler personality-related questionnaires. Participants randomly assigned to the mortality-salience condition then completed a questionnaire that asked them to describe the emotions that the thought of their own death aroused in them and to write about what will happen to their bodies as they physically die (see Rosenblatt, Greenberg, Solomon, Pyszczynski, & Lyon, 1989). Participants in the control condition did not complete this questionnaire but instead proceeded immediately to the next items in the packet, two personality-relevant measures intended to bolster the cover story.

THOUGHTS AND FEARS OF DEATH AND SELF-REGULATION These measures were included immediately after the mortality-salience manipulation but before the target dependent variable (described below), because the effects of the mortality-salience manipulation used typically emerge only after a short delay or distraction (see Pyszczynski et al., 1999). To test the effect of mortality salience on worldview defense, we asked participants next to read and evaluate a statement (used in Landau et al.’s, 2004, study) regarding President Bush and his antiterrorism policies. The passage praised the United States and President Bush and his actions against terrorism. After reading the passage, participants responded to the following items on a scale from 1 (strongly disagree) to 5 (strongly agree): “To what extent do you endorse this statement?”; “I share many of the attitudes expressed in the above statement”; and “Personally I feel secure knowing that the President is doing everything possible to guard against any further attacks against the United States.” Participants also indicated their political orientation by placing a mark on a continuous spectrum that ranged from Liberal to Conservative. The responses to the Likert items and spectrum (measured in centimeters across the spectrum) were standardized (z scored) and averaged (␣ ⫽ .91), and the single score served as our dependent measure of support for President Bush (higher scores indicated greater support).

Results A regression analysis was conducted to predict support for the President from trait self-control scores, mortality-salience condition, and their interaction, while controlling for self-esteem.2 The analysis indicated a significant effect of mortality-salience condition (␤ ⫽ 1.35, p ⬍ .05). Participants in the mortality-salience condition indicated supporting President Bush to a greater extent than did participants in the control condition. This replicates the findings of Landau et al. (2004), in which mortality salience increased support for President Bush. The effect of self-control approached significance (␤ ⫽ .29, p ⫽ .06) as did its interaction with mortality-salience condition (␤ ⫽ ⫺1.89, p ⫽ .06). To interpret the interaction between mortality salience and selfcontrol, and also to test directly our specific hypothesis, we assessed the simple effect of mortality-salience condition among participants who were relatively high versus relatively low in trait self-control (one standard deviation above and below the mean on the self-control scale, respectively; Aiken & West, 1991), controlling for trait self-esteem. Results indicate that the effect of mortality salience on support for the President was significant and robust for participants low in self-control ( p ⬍ .05) but was nonsignificant and negligible for those high in self-control ( p ⬎ .67; see Figure 3). Thus, although mortality salience increased support for the President, this effect occurred primarily among participants lower in trait self-control. Participants higher in selfcontrol did not show increased support for the President after thinking about death.

Discussion These findings converge upon the hypothesis that selfregulation facilitates the management of concerns about death. Whereas the results of the previous studies indicate that good trait and state self-control reduces death-thought accessibility and death anxiety, the results of Study 5 suggest that high trait self-control helps to reduce defensive reactions to mortality salience. Mortality salience increases liking for charismatic and decisive leaders (Cohen, Solomon, Maxfield, Pyszczynski, & Greenberg, 2004; Landau et al., 2004). People with high trait self-control were

55

apparently less cognizant of death after mortality salience, however, and so after thinking about death they did not increase support for President Bush.

Study 6 The central hypothesis of this investigation is that defending against the threatening idea of death requires self-regulation and consumes self-regulatory resources. Thus far we have presented evidence that when the capacity for self-regulation is low, thoughts and fears of death are more common. We turn now to a complementary way of testing the hypothesis, which is to show that when people defend themselves against thoughts and feelings about death, they deplete their self-regulatory resources. Past work has shown that following mortality salience, people actively and effortfully suppress thoughts of death (Arndt et al., 1997; Greenberg et al., 1994; Harmon-Jones et al., 1997). The result should be impaired performance on subsequent tasks requiring self-regulation. In Study 6, participants wrote about death or a control topic. We assumed that writing about death but not the control topic would activate disturbing thoughts and feelings that would require selfregulatory exertion to defend the self against them. Participants were then provided a delay that allowed participants sufficient time to suppress thoughts of death (if they had them). After the delay, participants completed the Stroop color–word interference task as a measure of self-regulation. For the Stroop task, participants saw words (i.e., red, blue, green) on a computer screen, and they were to respond by indicating the font color of the word. On some trials (incongruent trials), the meaning of the word differed from its font color (e.g., red appeared in a blue font), so on these trials participants had to exert self-control by overriding the tendency to read the word and respond instead according to the font color. Performance on the Stroop task was operationalized as the number of errors made during incongruent trials. We predicted that participants who wrote about death would make more Stroop errors than would those who wrote about a neutral topic, thereby suggesting that their self-regulatory resources had been depleted by the mortality-salience induction. We also assessed performance on Stroop trials that did not require self-control (i.e., congruent trials or trials for which the font color and meaning of the word were the same). This allowed us to determine whether mortality salience would impair automatic cognitive processing, as might be expected if the self-regulatory impairments following mortality salience are caused by another factor aside from self-control depletion (e.g., mood or arousal). We predicted that mortality salience would not influence performance on congruent trials.

2 Including self-esteem and its higher order interactions with self-control and mortality-salience condition yielded no significant effects as a function of self-esteem (all ps ⬎ .21). In addition, our primary results remained relatively unchanged when excluding self-esteem from the model. We included self-esteem in the model to demonstrate that the relationship between trait self-control and worldview defense (support for President Bush) was not attributable to self-esteem.

56

GAILLIOT, SCHMEICHEL, AND BAUMEISTER

Figure 3. Support for President George W. Bush as a function of mortality-salience condition and trait self-control (Study 5).

Method Participants. Fifty-seven undergraduate students (45 women, 12 men) participated. Five participants were excluded from all analyses because of technical difficulties (e.g., being interrupted while completing the Stroop task) or for not following instructions, leaving a final sample of 52 (42 women, 10 men). Participants were randomly assigned to condition. Procedure. Participants were run individually and were told that the study was to investigate personality and perception. Participants first completed a computer task that was similar to the Stroop task so they could become familiarized with how to respond on the keyboard. Specifically, participants completed 60 trials in which a string of Xs (XXXXX) appeared on the computer screen in either a red, green, or blue font. Participants were to indicate the color of the Xs by pressing one of three computer keys (the R, G, or B key) as quickly as possible. Following each response, the next string of Xs appeared immediately. After practicing the computer task, participants wrote about either death or dental pain. As in Study 5, mortality-salience participants described the emotions that the thought of their own death aroused in them and what will happen to them as they physically die, whereas the dental-pain participants described the emotions that the thought of their own dental pain aroused in them and what would happen to them if they were to experience dental pain (see Rosenblatt et al., 1989). Participants then completed the BMIS, as a measure of mood and arousal, and a filler task (a crossword puzzle) for 4 min to provide sufficient time for them to suppress thoughts of death (Pyszczynski et al., 1999). Participants then completed the final Stroop task. For this task, participants completed 3 blocks of trials in which the word red, blue, or green appeared on the computer screen in either a red, blue, or green font. Participants were to indicate the font color by pressing one of three computer keys (the R, G, or B key) and were asked to respond as quickly and accurately as possible. The first and third blocks consisted of 30 congruent trials in which the meaning and font color of the word were the same. The second block consisted of 60 incongruent trials in

which the meaning and font color of the word were different. At the end of the task, participants completed demographic information, indicated whether they had completed the Stroop task before (“yes” or “no”), and were probed for suspicion. No participant indicated that he or she thought that writing about death or dental pain might influence performance on the Stroop task.

Results Stroop task performance. In all of our analyses involving Stroop task performance, we controlled for whether participants had ever performed the Stroop task before, because there is a large practice effect on Stroop task performance (MacLeod, 1991). An analysis of covariance indicated that mortality-salience participants made more errors (M ⫽ 2.50, SD ⫽ 1.86) on the incongruent trials on the Stroop task than did dental-pain participants (M ⫽ 1.40, SD ⫽ 1.85), F(1, 49) ⫽ 4.26, p ⬍ .05. This result suggests that mortality salience impaired participants’ ability to exert self-control. Additional analyses indicated that the two groups did not differ on the number of errors on either block of congruent trials (both Fs ⬍ 1, ns). This pattern of results suggests that mortality salience did not impair automatic processing (congruent trials) but did impair controlled processing (incongruent trials). Further, participants did make some errors on the congruent trials on the first and third blocks (in the mortality-salience condition, M ⫽ .32 and .63 [excluding one outlier on this measure], respectively; in the dentalpain condition, M ⫽ .25 and .63, respectively). This suggests that the congruent trials were somewhat cognitively demanding, yet mortality salience did not impair performance on these trials. In addition, mortality-salience and dental-pain participants did not differ in their reaction times on either the incongruent or

THOUGHTS AND FEARS OF DEATH AND SELF-REGULATION

congruent trials (Fs ⬍ 1, ns). This suggests that the increased number of errors among mortality-salience participants was not the result of a speed–accuracy trade-off, such as if they had made more errors because they responded faster. Mood and arousal. Mortality-salience and dental-pain participants did not differ in mood valence or arousal (as assessed by the BMIS; ts ⬍ 1, ns). This suggests that mortality-salience participants’ performing worse on the Stroop task than dental-pain participants was probably not attributable to mood or arousal.

Discussion Participants who wrote about death performed worse (i.e., made more errors) on the Stroop task than did control participants but only on trials that required self-control (i.e., incongruent trials). This suggests that mortality salience impaired the ability to exert self-control. Presumably, after thinking about death, participants suppressed thoughts of death, and the act of thought suppression depleted their self-regulatory strength. Participants who wrote about death rather than dental pain did not perform worse on trials that did not require self-control (i.e., congruent trials), which suggests that mortality salience did not have an adverse effect on automatic cognitive processes.

Study 7 The results of Study 6 suggest that mortality salience subsequently impairs self-control but not cognitive processing in general. It is plausible, however, that the congruent Stroop trials were relatively easy and therefore a poor measure of general cognitive abilities. In Study 7 we therefore provided another test of the hypothesis that mortality salience impairs self-control but not other cognitive processes by using a more difficult task to assess general cognitive abilities. Participants first wrote about either death or uncertainty. Uncertainty salience has been shown to be a poignant self-threat (e.g., van den Bos, Poortvliet, Maas, Miedema, & van den Ham, 2005) and therefore seemed an adequate control topic. Participants then completed problems requiring either analytical reasoning or rote memory. Research has indicated that analytical reasoning suffers more from self-regulatory depletion than does rote memory (Schmeichel et al., 2003). Consistent with the idea that coping with thoughts of death depletes self-regulatory resources, we therefore predicted that mortality salience would cause participants to perform worse on problems requiring analytical reasoning but not rote memory.

Method Participants. Participants were 38 undergraduates (24 women, 14 men). Data from 1 participant who failed to follow instructions were excluded from all analyses. Participants were run together in a large group and were told that the study was to investigate personality and cognitive abilities. Procedure. Participants randomly assigned to the mortality-salience condition first completed the essay about death used in the previous studies. Participants assigned to the uncertainty-salience condition described the emotions uncertainty arouses in them and what would happen to them if they were to experience uncertainty. Afterward, participants

57

completed a filler questionnaire for approximately 5 min to allow them to suppress any thoughts about death they may have had. Participants then completed either 6 analytical reasoning problems or 20 verbal definition problems requiring rote memory (e.g., defining moratorium, allude, augment). These problems were borrowed from a Graduate Record Examinations preparation book and were classified as easy. Each problem had 4 or 5 multiple-choice answers. Participants were instructed to choose the most appropriate answer.

Results and Discussion A 2 (mortality vs. uncertainty salience) ⫻ 2 (analytical vs. rote memory problems) between-subjects analysis of variance indicated a significant interaction between mortality-salience condition and problem type, F(1, 33) ⫽ 4.76, p ⬍ .05. Simple contrasts indicated that, among participants who completed the analytical reasoning problems, mortality-salience participants (M ⫽ 1.38, SD ⫽ 1.19) solved fewer problems correctly than did uncertaintysalience participants (M ⫽ 2.56, SD ⫽ 1.13), F(1, 33) ⫽ 6.34, p ⬍ .05. Among participants who completed the rote memory problems (verbal definitions), mortality-salience (M ⫽ 16.20, SD ⫽ 1.55) and uncertainty-salience (M ⫽ 15.50, SD ⫽ 3.27) participants did not differ in the number of problems solved correctly (F ⬍ 1, ns). Mortality salience thus impaired performance on problems requiring self-control (analytical reasoning skills) but not problems requiring only more basic cognitive processing (rote memory for verbal definitions). This underscores the notion that mortality salience depletes self-regulatory strength and that the deficits in self-regulation following mortality salience are particular to selfregulation and not cognitive processing in general. An alternative interpretation of why mortality salience might impair self-regulation, however, might be that thoughts of death rebound after mortality salience (e.g., Arndt et al., 1997; Wegner, 1994) and thereby distract participants from self-regulatory tasks. Similarly, it is plausible that other aftereffects of mortality salience (e.g., worldview activation; Arndt, Greenberg, & Cook, 2002) could somehow interfere with self-regulation. The results of Studies 6 and 7 directly contradict these possibilities, however. Mortality salience did not impair performance on congruent Stroop trials (Study 6) or on rote memory (Study 7), as one might have expected if the hyperaccessibility of death thoughts or worldview constructs impaired cognitive performance. If participants’ minds were full of rebounding thoughts of death, for instance, they probably would not have been able to perform very well at rote memory. To provide another test of the alternative explanation that mortality salience impairs self-regulation via distraction, in Study 8 we tested for moderation by trait self-esteem. There is some evidence that mortality salience might not increase the accessibility of death thoughts or worldview defense among individuals with high selfesteem (see Pyszczynski et al., 2004). We made the following hypothesis: If the hyperaccessibility of death thoughts and not the depletion of self-regulatory resources impairs self-regulation following mortality salience, then self-esteem might moderate the effect of mortality salience on self-regulation.

Study 8 Participants in Study 8 first wrote about either death or dental pain and then solved anagrams. Solving anagrams has been used

GAILLIOT, SCHMEICHEL, AND BAUMEISTER

58

frequently in prior research on self-regulation (e.g., Baumeister et al., 1998; Gordijn et al., 2004). The task requires self-control insofar as one must combine letters into different groupings and then break them apart (overriding) to try a different combination. Persistence and attention control are also required, insofar as one must work diligently on the task and avoid being distracted, and so self-regulation is required to remain focused and succeed. We predicted that participants would solve fewer anagrams after writing about death than after writing about dental pain, consistent with the idea that suppressing thoughts of death requires and therefore depletes self-regulatory strength.

Method Participants. Participants were 46 undergraduates (32 women, 14 men) who completed the Rosenberg Self-Esteem Scale during a masstesting session at the start of the semester. The present study occurred approximately 3 months later. The study was conducted in a classroom setting, and participants were told that the study was to investigate personality and cognition. Procedure. At the beginning of the session, participants were given a list of 80 five-letter anagrams to solve to serve as a baseline measure of anagram-solving ability. They were given 5 min to complete as many anagrams as they could. Next, participants were randomly assigned to respond to the questions about death or dental pain that were used in Study 6. To provide a delay that would allow time for participants to suppress thoughts of death, we then asked participants to complete two filler questionnaires for approximately 5 min. For the final task, participants were given another list of anagrams to solve for 5 min. Last, participants were probed for suspicion. No participant indicated that he or she thought that writing about death or dental pain might influence performance at solving anagrams.

Results and Discussion Anagram performance. An analysis of covariance that used the number of anagrams solved at the start of the session as a covariate indicated that mortality-salience participants solved fewer anagrams at the end of the session (M ⫽ 18.27, SD ⫽ 5.68) than did dental-pain participants (M ⫽ 21.80, SD ⫽ 5.68), F(1, 43) ⫽ 4.41, p ⬍ .05. Mortality-salience and dental-pain participants did not differ in the number of anagrams solved at the start of the session (t ⬍ 1, ns). These results are consistent with the idea that coping with thoughts of death requires and therefore depletes self-control strength. Self-esteem. A regression analysis indicated that the effect of mortality salience on anagram performance was not moderated by self-esteem (t ⬍ 1, ns). This suggests that the effect of mortality salience was not related to self-esteem, as might have been expected if the relationship between mortality salience and anagram performance was attributable to hyperaccessibility of death thoughts or worldview constructs.

Study 9 The purpose of Study 9 was to provide additional evidence that mortality salience depletes self-regulatory strength and that the self-regulatory impairments following mortality salience are not caused by worldview activation or a rebound in thoughts of death. Specifically, after writing about death or dental pain, participants completed a sheet filled with word fragments (e.g., so_a). We told

participants to solve all of the word fragments but that they could stop when they wanted. The word fragments were relatively easy and with enough persistence (self-control), one could eventually solve each one by substituting into the word fragment different letters in the alphabet. Still, the task is lengthy and discouraging, and so persisting requires one to override the impulse (strengthened by the experimenter’s permission) to quit. In this sense, quitting early and leaving more word fragments unsolved was an indicator of impaired self-regulation. Further, as in the previous studies, some of the word fragments could be solved with words related to death. Responses to these items served as a measure of death-thought accessibility that allowed us to test whether impaired self-regulation following mortality salience might be attributable to the hyperaccessibility of death thoughts. As another approach to examine whether the hyperaccessibility of death thoughts (rather than self-control depletion) after mortality salience impairs subsequent self-regulatory performance, we provided an opportunity for participants to engage in worldview defense before solving the word fragments. There is some evidence that engaging in worldview defense after thinking about death reduces the accessibility of death thoughts (Arndt et al., 1997; Greenberg et al., 2001). Therefore, if mortality salience leads to worldview defense, then any subsequent impairments in self-regulation would likely not be due to heightened accessibility of death thoughts.

Method Participants. Participants were 55 undergraduates (24 women, 18 men, 13 unreported). The study was conducted in a classroom setting, and participants were told that it was to investigate the relationship between people’s attitudes and their lexicon. Participants received a packet that contained all materials for the experiment and were allowed to work through the packet at their own pace. Procedure. Participants first completed a filler questionnaire and then the same mortality-salience manipulation used in Study 8. Specifically, participants were randomly assigned to write about either death or dental pain. After this questionnaire, participants completed two more filler questionnaires for approximately 5 min to provide time for participants to suppress any thoughts about death. The next questionnaire constituted the measure of worldview defense. Specifically, participants read two handwritten essays about the United States that were ostensibly written by two foreigners (borrowed from Greenberg, Simon, Pyszczynski, Solomon, & Chatel’s, 1992, study). The order of the two essays was counterbalanced across participants. One essay was in support of the United States and praised Americans, whereas the other essay was not in support of the United States and criticized Americans. Participants evaluated the truth and validity of the essay and the likability, intelligence, and knowledgeability of each essay’s author on 9-point scales. The summed evaluations of each essay served as the measures of favorability toward worldview-consistent and worldviewinconsistent opinions, respectively. In accord with past research (e.g., Greenberg et al., 1994), worldview defense was defined as the difference between these two measures. Larger differences indicate more pronounced worldview defense. For the final task, participants completed the same list of 20 word fragments used in Study 1A. Specifically, 15 word fragments could be solved with only neutral words not related to death (e.g., soda), and 5 word fragments could be solved with either neutral or death-related words. Participants were asked to do their best to solve all of the word fragments

THOUGHTS AND FEARS OF DEATH AND SELF-REGULATION and were given as much time as they needed to complete the task. It was left to their discretion as to when they decided to stop.

59

to account for their reduced persistence. The most parsimonious explanation for these results is that participants’ self-regulatory resources were depleted following mortality salience.

Results and Discussion Worldview defense. Mortality-salience participants engaged in worldview defense (M ⫽ 13.71, SD ⫽ 12.93) to a greater extent than did dental-pain participants (M ⫽ 7.69, SD ⫽ 13.06), t(51) ⫽ ⫺1.65, p ⫽ .05 (one-tailed). This finding is consistent with previous research showing that mortality salience increases worldview defense. Given past findings that worldview defense reduces the accessibility of death thoughts (Arndt et al., 1997; Greenberg et al., 2001), this suggests that the accessibility of death thoughts may have been attenuated among mortality-salience participants. Unsolved word fragments. Mortality-salience participants left more neutral word fragments unsolved (M ⫽ 3.52, SD ⫽ 3.86) than did dental-pain participants (M ⫽ 1.43, SD ⫽ 1.58), t(27.37) ⫽ ⫺2.45, p ⬍ .05 (degrees of freedom have been adjusted to correct for a violation of the homogeneity of variance assumption). This is consistent with the idea that mortality salience, or at least the typical response to it, is depleting. Mortality-salience participants persisted less on the subsequent task than did dentalpain participants. Further, the difference between the two conditions in the number of word fragments left unsolved remained significant even when controlling for the number of target items completed with words related to death (implicit death thoughts) and worldview defense scores (both Fs ⱖ 6.87, ps ⬍ .05). This suggests that any difference between mortality-salience and dental-pain participants in the extent to which death thoughts were cognitively accessible or the extent to which they engaged in worldview defense did not account for the finding that mortality-salience participants left more word fragments unsolved. The impairments in self-regulation among mortality-salience participants therefore appeared independent of death-thought accessibility and worldview defense. In addition, mortality-salience and dental-pain participants did not differ in the number of target items related to death unsolved (t ⬍ 1.05, ns). This suggests that mortality-salience participants did not leave more neutral word fragments unsolved simply because they solved instead more death-related word fragments. Further analysis indicated that greater worldview defense was associated with leaving more neutral word fragments unsolved, r(53) ⫽ .27, p ⫽ .05. This suggests that being more threatened by the thought of death (and hence being more defensive) was associated with greater depletion. The fact that worldview defense did not account for the effect of mortality salience on depletion, however, indicates that engaging in worldview defense was not the cause of depletion. Rather, it was likely the self-regulatory effort of having to cope with the threat of death that depleted selfregulatory resources. In sum, participants who wrote about death solved fewer neutral word fragments than did participants who wrote about dental pain. Results also indicated that the accessibility of death thoughts probably did not account for the differences in self-control. Mortality-salience participants persisted less even though they were given the opportunity to engage in worldview defense, which should have attenuated the accessibility of death thoughts (Arndt et al., 1997; Greenberg et al., 2001). Likewise, the number of implicit death thoughts and the extent of worldview defense did not seem

General Discussion The current work shows that self-regulation plays a key role in managing thoughts of death and death-related anxiety. Using a variety of measures and manipulations, in nine studies we showed that poor self-regulation undermines the management of mortality concerns and that the management of mortality concerns undermines subsequent self-regulation. Specifically, people with low trait self-control had more thoughts about death and higher death anxiety than did people with high trait self-control (Studies 1A– 1C). Low state self-control, created by prior and seemingly irrelevant exercises that depleted self-regulatory strength, likewise made people more susceptible to intrusive thoughts about death (Studies 2–3). Perceiving oneself to be low in self-regulatory strength was associated with a similar upsurge in thoughts of death and death anxiety (Study 4). In addition, one defensive reaction to the threat of death is to increase support for an ostensibly strong and patriotic authority figure (Cohen et al., 2004; Landau et al., 2004), and such defensive shifts in political preferences in response to thoughts of death were found mainly among people low in trait self-control (Study 5). Further, mortality salience led to impairments in self-regulation. Participants who had to think and write about death later performed worse on the Stroop task (Study 6) and analytical reasoning problems (Study 7), solved fewer anagrams (Study 8), and quit sooner on a task of persistence (Study 9), as compared with participants who wrote about a control topic. People effortfully suppress thoughts of death following their activation (Arndt et al., 1997; Greenberg et al., 1994; Harmon-Jones et al., 1997; Pyszczynski et al., 1999), and so it seems likely that the act of suppressing thoughts of death after writing about death depleted participants’ self-regulatory strength. These findings are consistent with the view that thoughts about death and mortality are disturbing and are widely treated as threats that must be kept at bay. Specifically, we propose that the world contains many cues that could evoke thoughts and feelings about death, and that people use self-regulation to prevent these cues from flooding the conscious mind with troublesome thoughts and aversive emotions such as anxiety (Greenberg, Pyszczynski, & Solomon, 1986). But because the capacity for self-regulation is limited, and because some people have more of it than others, the regulatory process is not consistently effective. Hence some stimuli will evoke conscious thoughts about death, at least for some of the people some of the time. Trait and state capacity for selfregulation appears to offer an effective basis for predicting and understanding who will suffer most from such thoughts and fears.

Implications, Limitations, and Alternative Explanations These results are consistent with previous work showing that people actively and effortfully suppress thoughts of death (e.g., Greenberg et al., 1994; Harmon-Jones et al., 1997). Whereas previous work showed that a concurrent cognitive load undermines the suppression of death thoughts (Arndt et al., 1997), the current work indicates that depleted self-regulatory strength and low trait

60

GAILLIOT, SCHMEICHEL, AND BAUMEISTER

self-control also undermine the suppression of death thoughts. And conversely, the suppression of death thoughts depletes self-control strength. Taken together, the picture that emerges is that people actively seek to suppress or minimize awareness of death and that concurrent cognitive activity, previous self-regulatory exertion, or habitually poor self-regulatory abilities undermine such efforts. Though a few alternative explanations may be proposed for the link between self-control and mortal concern, they seem largely unable to account for the bulk of the current findings. Specifically, the results appeared independent from any effects of self-esteem, arousal, mood, or general negative affect (i.e., general anxiety and social anxiety in Study 1C). Some effects may be explainable in terms of social desirability, such as if people giving socially desirable responses reported having both high self-control and low death anxiety, yet this view cannot adequately account for all of the findings. It is unclear, for example, how social desirability would have been related to support for President Bush (Study 5). The social desirability explanation also does not easily account for the fact that mortality salience impaired performance on tasks that required self-control but not other cognitive tasks (Studies 6 and 7). We also controlled for self-esteem in most of the studies. Insofar as reporting having high self-esteem indicates giving socially desirable responses, then the results do not appear caused by differences in social desirability. Further, the effect of depletion following mortality salience did not appear attributable to distraction by thoughts about death (Study 9), active attempts toward defending one’s worldviews (Study 9), or deficits to general cognitive abilities (Studies 6 and 7). The most parsimonious conclusion is that mortality salience impaired self-control by depleting self-regulatory resources. Indeed, mortality salience increased worldview defense (Study 9), a defensive reaction aimed at coping with the threat of death, and greater worldview defense predicted (but did not account for) greater depletion. Participants who seemingly attempted to cope the most with death and therefore found it the most frightening were the most depleted. Studies 1A–1C were correlational, which makes it difficult to draw causal conclusions. Still, they do indicate a personality pattern that seems to make people more vulnerable to unwelcome thoughts and emotions. Preliminary evidence has linked low selfcontrol to a variety of maladaptive behavior patterns (e.g., Gottfredson & Hirschi, 1990; Tangney et al., 2004). It is plausible that one reason people with low versus high self-control engage in such maladaptive behaviors is that they have substantial death anxiety. For example, both low self-control and death anxiety are associated with increased aggression (McGregor et al., 1998; Stucke & Baumeister, in press; Tangney et al., 2004) and risky sexual behavior (Taubman Ben-Ari, 2004; Wills, Gibbons, Gerrard, Murry, & Brody, 2003). Perhaps people with low self-control engage in such behaviors in part as a response to preoccupation with death (see Tice, Bratslavsky, & Baumeister, 2001). If we could generalize the present results to other sources of stress, threat, and trauma, this work would be even more important, but we hesitate to generalize without further evidence. It is plausible that self-regulation might be more closely tied to coping with the threat of death than with other threats or aversive states. The thought of death can be especially threatening and trigger active efforts toward suppression (e.g., Arndt et al., 1997), perhaps more so than most other thoughts. Our use of dental pain and

uncertainty as control conditions was intended to differentiate death from other bad thoughts, and clearly in the present studies death had effects that were not duplicated by dental pain or uncertainty. Death could represent a particularly important problem because it threatens to nullify one’s life and its meaning, and so people may seek to defend against the thought of death more extensively than they would defend against the thought of other misfortunes. Future research may profitably examine whether selfregulation and self-control depletion are relevant to psychological defenses against more than death.

Concluding Remarks Coping with unwanted thoughts and feelings can be an effortful struggle. Death is inevitable and inescapable, and so people must protect themselves from thoughts of death if they are to reduce the anxiety that the awareness of death might evoke. Rational attempts to deny death ultimately fail, however, and so people must engage in other defenses to ward off the threat of death (Pyszczynski et al., 1999). The current work indicates that self-regulation is a useful tool in the psychological defense against death. Through effective self-regulation, people can minimize their awareness of death and reduce anxiety associated with death. But this peace of mind comes at a cost. When thoughts of death arise, it takes effortful self-control to suppress them, and these efforts consume resources that leave people with less self-control afterward— causing people to perform less effectively on many tasks. Conversely, expending resources on any sort of self-control task reduces one’s defenses, thereby opening the door for disturbing thoughts of death.

References Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting interaction. Newbury Park, CA: Sage. Anderson, M. C., & Green, C. (2001, March 15). Suppressing unwanted memories by executive control. Nature, 6826, 366 –369. Aries, P. (1981). The hour of our death. New York: Oxford University Press. Arndt, J., Greenberg, J., & Cook, A. (2002). Mortality salience and the spreading activation of worldview-relevant constructs: Exploring the cognitive architecture of terror management. Journal of Experimental Psychology: General, 131, 307–324. 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. Baumeister, R. F., Bratslavsky, E., Muraven, M., & Tice, D. M. (1998). Ego depletion: Is the active self a limited resource? Journal of Personality and Social Psychology, 74, 1252–1265. Baumeister, R. F., Heatherton, T. F., & Tice, D. M. (1994). Losing control: How and why people fail at self-regulation. San Diego, CA: Academic Press. Becker, E. (1973). The denial of death. New York: Academic Press. Cohen, F., Solomon, S., Maxfield, M., Pyszczynski, T., & Greenberg, J. (2004). Fatal attraction: The effects of mortality salience on evaluations of charismatic, task-oriented, and relationship-oriented leaders. Psychological Science, 15, 846 – 851. Feifel, H., & Branscomb, A. B. (1973). Who’s afraid of death? Journal of Abnormal Psychology, 81, 282–288. Finkel, E. J., & Campbell, W. K. (2001). Self-control and accommodation in close relationships: An interdependence analysis. Journal of Personality and Social Psychology, 81, 263–277.

THOUGHTS AND FEARS OF DEATH AND SELF-REGULATION 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. Gailliot, M. T., & Baumeister, R. F. (in press). Self-control and business ethics: How strengthening the self benefits the corporation and the individual. In R. A. Giacalone, C. L. Jurkiewicz, & C. Dunn (Eds.), Positive psychology, ethics, and social responsibility in organizational life. Greenwich, CT: Information Age. Gilbert, D., T., Krull, D. S., & Pelham, B. W. (1988). Of thoughts unspoken: Social inference and the self-regulation of behavior. Journal of Personality and Social Psychology, 55, 685– 694. Gordijn, E. H., Hindriks, I., Koomen, W., Dijksterhuis, A., & Van Knippenberg, A. (2004). Consequences of stereotype suppression and internal suppression motivation: A self-regulation approach. Personality and Social Psychology Bulletin, 30, 212–224. Gottfredson, M. R., & Hirschi, T. (1990). A general theory of crime. Stanford, CA: Stanford University Press. Greenberg, J., Arndt, J., Schimel, J., Pyszczynski, T., & Solomon, S. (2001). Clarifying the function of mortality salience-induced worldview defense: Renewed suppression or reduced accessibility of death-related thoughts? Journal of Experimental Social Psychology, 37, 70 –76. Greenberg, J., Martens, A., Jonas, E., Eisenstadt, D., Pyszczynski, T., & Solomon, S. (2003). Psychological defense in anticipation of anxiety: Eliminating the potential for anxiety eliminates the effect of mortality salience on worldview defense. Psychological Science, 14, 516 –519. Greenberg, J., Pyszczynski, T., & Solomon, S. (1986). The causes and consequences of a need for self-esteem: A terror management theory. In R. F. Baumeister (Ed.), Public self and private self (pp. 189 –212). New York: Springer-Verlag. 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). Role of consciousness and accessibility of death-related thoughts in mortality salience effects. Journal of Personality and Social Psychology, 67, 627– 637. Greenberg, J., Simon, L., Pyszczynski, T., Solomon, S., & Chatel, D. (1992). Terror management and tolerance: Does mortality salience always intensify negative reactions to others who threaten one’s worldview? Journal of Personality and Social Psychology, 63, 212–220. Harmon-Jones, E., Simon, L., Greenberg, J., Pyszczynski, T., Solomon, S., & McGregor, H. (1997). Terror management theory and self-esteem: Evidence that increased self-esteem reduces mortality salience effects. Journal of Personality and Social Psychology, 72, 24 –36. Heine, S. J., Harihara, M., & Niiya, Y. (2002). Terror management in Japan. Asian Journal of Social Psychology, 5, 187–196. Landau, M. J., Solomon, S., Greenberg, J., Cohen, F., Pyszczynski, T., Arndt, J., et al. (2004). Deliver us from evil: The effects of mortality salience and reminders of 9/11 on support for President George W. Bush. Personality and Social Psychology Bulletin, 30, 1136 –1150. MacLeod, C. M. (1989). Directed forgetting affects both direct and indirect tests of memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 13–21. MacLeod, C. M. (1991). Half a century of research on the Stroop effect: An integrative review. Psychological Bulletin, 109, 163–203. Mattick, R. P., & Clarke, J. C. (1998). Development and validation of measures of social phobia scrutiny fear and social interaction anxiety. Behaviour Research and Therapy, 36, 455– 470. Mayer, J. D., & Gaschke, Y. N. (1988). The experience and metaexperience of mood. Journal of Personality and Social Psychology, 55, 102–111. McBride, D. M., & Dosher, B. A. (1997). A comparison of forgetting in an

61

implicit and explicit memory task. Journal of Experimental Psychology: General, 126, 371–392. 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 motivates aggression against worldview-threatening others. Journal of Personality and Social Psychology, 74, 590 – 605. Mischel, W., Shoda, Y., & Peake, P. K. (1988). The nature of adolescent competencies predicted by preschool delay of gratification. Journal of Personality and Social Psychology, 54, 687– 696. Muraven, M., & Baumeister, R. F. (2000). Self-regulation and depletion of limited resources: Does self-control resemble a muscle? Psychological Bulletin, 126, 247–259. Muraven, M., Tice, D. M., & Baumeister, R. F. (1998). Self-control as a limited resource: Regulatory depletion patterns. Journal of Personality and Social Psychology, 74, 774 –789. Ochsmann, R., & Mathey, M. (1994). Depreciating of and distancing from foreigners: Effects of mortality salience. Unpublished manuscript, Universitat Mainz, Mainz, Germany. Pollak, J. M. (1979). Correlates of death anxiety: A review of empirical studies. Omega: Journal of Death & Dying, 10, 97–121. 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., Arndt, J., & Schimel, J. (2004). Why do people need self-esteem? A theoretical and empirical review. Psychological Bulletin, 130, 435– 468. Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University 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. Schmeichel, B. J., Vohs, K. D., & Baumeister, R. F. (2003). Intellectual performance and ego depletion: Role of the self in logical reasoning and other information processing. Journal of Personality and Social Psychology, 85, 33– 46. Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human information processing: I. Detection, search, and attention. Psychological Review, 84, 1– 66. Shoda, Y., Mischel, W., & Peake, P. K. (1990). Predicting adolescent cognitive and self-regulatory competencies from preschool delay of gratification: Identifying diagnostic conditions. Developmental Psychology, 26, 978 –986. Smart, L., & Wegner, D. M. (1999). Covering up what can’t be seen: Concealable stigma and mental control. Journal of Personality and Social Psychology, 77, 474 – 486. Spielberger, C. D., Gorsuch, R. L., & Lushene, R. E. (1970). Manual for the State–Trait Anxiety Inventory. Palo Alto, CA: Consulting Psychologists. Stucke, T. S., & Baumeister, R. F. (in press). Ego depletion and aggressive behavior: Is the inhibition of aggression a limited resource? European Journal of Social Psychology. Tangney, J. P., Baumeister, R. F., & Boone, A. L. (2004). High self-control predicts good adjustment, less pathology, better grades, and interpersonal success. Journal of Personality, 72, 271–322. Taubman Ben-Ari, O. (2004). Intimacy and risky sexual behavior: What does it have to do with death? Death Studies, 28, 865– 887. Templer, D. I. (1970). The construction and validation of a death anxiety scale. Journal of General Psychology, 82, 165–177. Tice, D. M., Bratslavsky, E., & Baumeister, R. F. (2001). Emotional distress regulation takes precedence over impulse control: If you feel bad, do it! Journal of Personality and Social Psychology, 80, 53– 67.

62

GAILLIOT, SCHMEICHEL, AND BAUMEISTER

Twenge, J. M., Muraven, M., & Tice, D. M. (2004). Measuring state self-control: Reliability, validity, and correlations with physical and psychological stress. Unpublished manuscript, San Diego State University. van den Bos, K., Poortvliet, P. M., Maas, M., Miedema, J., & van den Ham, E. J. (2005). An enquiry concerning the principles of cultural norms and values: The impact of uncertainty and mortality salience on reactions to violations of bolstering of cultural worldviews. Journal of Experimental Social Psychology, 41, 91–113. Wegner, D. M. (1994). Ironic processes of mental control. Psychological Review, 101, 34 –52. Wegner, D. M., Schneider, D., Carter, S. R., & White, T. L. (1987). Paradoxical effects of thought suppression. Journal of Personality and Social Psychology, 53, 5–13.

Wegner, D. M., & Zanakos, S. (1994). Chronic thought suppression. Journal of Personality, 62, 615– 640. Wenzlaff, R. M., & Wegner, D. M. (2000). Thought suppression. Annual Review of Psychology, 51, 59 –91. Wills, T. A., Gibbons, F. X., Gerrard, M., Murry, V. M., & Brody, G. H. (2003). Family communication and religiosity related to substance use and sexual behavior in early adolescence: A test for pathways through self-control and prototype perceptions. Psychology of Addictive Behaviors, 17, 312–323.

Received May 1, 2005 Revision received November 1, 2005 Accepted November 3, 2005 䡲

62

GAILLIOT, SCHMEICHEL, AND BAUMEISTER

Twenge, J. M., Muraven, M., & Tice, D. M. (2004). Measuring state self-control: Reliability, validity, and correlations with physical and psychological stress. Unpublished manuscript, San Diego State University. van den Bos, K., Poortvliet, P. M., Maas, M., Miedema, J., & van den Ham, E. J. (2005). An enquiry concerning the principles of cultural norms and values: The impact of uncertainty and mortality salience on reactions to violations of bolstering of cultural worldviews. Journal of Experimental Social Psychology, 41, 91–113. Wegner, D. M. (1994). Ironic processes of mental control. Psychological Review, 101, 34 –52. Wegner, D. M., Schneider, D., Carter, S. R., & White, T. L. (1987). Paradoxical effects of thought suppression. Journal of Personality and Social Psychology, 53, 5–13.

Wegner, D. M., & Zanakos, S. (1994). Chronic thought suppression. Journal of Personality, 62, 615– 640. Wenzlaff, R. M., & Wegner, D. M. (2000). Thought suppression. Annual Review of Psychology, 51, 59 –91. Wills, T. A., Gibbons, F. X., Gerrard, M., Murry, V. M., & Brody, G. H. (2003). Family communication and religiosity related to substance use and sexual behavior in early adolescence: A test for pathways through self-control and prototype perceptions. Psychology of Addictive Behaviors, 17, 312–323.

Received May 1, 2005 Revision received November 1, 2005 Accepted November 3, 2005 䡲

INTERPERSONAL RELATIONS AND GROUP PROCESSES

Peacocks, Picasso, and Parental Investment: The Effects of Romantic Motives on Creativity Vladas Griskevicius, Robert B. Cialdini, and Douglas T. Kenrick Arizona State University Four experiments explored the effects of mating motivation on creativity. Even without other incentives to be creative, romantic motives enhanced creativity on subjective and objective measures. For men, any cue designed to activate a short-term or a long-term mating goal increased creative displays; however, women displayed more creativity only when primed to attract a high-quality long-term mate. These creative boosts were unrelated to increased effort on creative tasks or to changes in mood or arousal. Furthermore, results were unaffected by the application of monetary incentives for creativity. These findings align with the view that creative displays in both sexes may be linked to sexual selection, qualified by unique exigencies of human parental investment. Keywords: creativity, sexual selection, parental investment, self-presentation, mating goals

mythology—are universally female. Yet if “there is no biological reason why a man can’t provide the elements of inspiration” (p. 9, Prose, 2002), how could it be that the elixir of inspiration seems to be primarily concocted by women and predominantly imbibed by men? The current research presents an evolutionary cognitive framework designed to shed light on the mystery of the muses and also to explain and predict a wider range of behavioral phenomena. This framework is grounded in two underlying premises: First, that a number of human mental traits—including the capacity for and display of creativity—may in part be linked to evolutionary processes of sexual selection and differential parental investment (Miller, 1999, 2000; Trivers, 1972); and second, that evolutionarily relevant contextual cues can serve to activate goals (e.g., the goal to attract a mate) that facilitate behaviors historically associated with success for the attainment of such goals (e.g., Maner et al., 2005; Roney, 2003; Wilson & Daly, 2004). From this foundation, four experiments examine whether cues designed to temporarily activate a goal to attract a mate can increase people’s creativity. Drawing on the theory of differential parental investment, the research also explores which specific cues might stimulate a creative boost for men versus women. Finally, the studies gather evidence regarding the nature of the psychological mechanism that may underlie how mating motives serve to foster creativity.

“In order to create there must be a dynamic force—and what force is more potent than love?” —Igor Stravinsky

The Guinness Book of World Records lists Pablo Picasso as the most prolific artist in history with an astounding 147,800 works of art. Picasso’s career is often depicted as a tortuous series of profoundly inspired artistic periods— blue, rose, cubist, surrealist—in which his subjects underwent extravagant visual transformations at the hands of a creative genius performing at the apogee of his ability. But a closer look at Picasso’s generative periods reveals an intriguing constant: Each new epoch blossoms with paintings of a new woman—not a sitter or a model, but a mistress— each of whom is touted to have served Picasso as an incandescent, albeit temporary, muse (Crespelle, 1969; MacGregor-Hastie, 1988). Picasso’s artistic history, however, is not unique: Creative juggernauts such as Salvador Dalı´, Friedrich Nietzsche, and Dante were also acutely inspired by their own muses (Prose, 2002). The enigmatic notion of a muse is rooted in Greek mythology, in which nine godly muses traversed the land, stirring the creative spirits of mortal artists and scientists. And according to historian Francine Prose (2002), all muses share one striking and inextricable feature: Muses— both in history and in

Vladas Griskevicius, Robert B. Cialdini, and Douglas T. Kenrick, Department of Psychology, Arizona State University. This research was facilitated by a National Science Foundation Graduate Research Fellowship awarded to Vladas Griskevicius and by National Institutes of Health Grant 5R01MH64734 to Douglas T. Kenrick. We thank Steve Neuberg, Cathy Cottrell, Noah Goldstein, and Josh Tybur for their helpful comments on a draft of this article. Correspondence concerning this article should be addressed to Vladas Griskevicius, Department of Psychology, Arizona State University, Tempe, AZ 85287-1104. E-mail: [email protected]

Natural and Sexual Selection of Creative Displays The construct of creativity is composed of multiple dimensions that share a common thread: Creativity is the ability to produce work that is both novel (i.e., original, unexpected) and appropriate (i.e., useful, valuable; Sternberg & Lubart, 1999). Although classical creativity research has not been concerned with the ultimate origins of human creativity (e.g., see Simonton, 2000), those who

Journal of Personality and Social Psychology, 2006, Vol. 91, No. 1, 63–76 Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.63

63

64

GRISKEVICIUS, CIALDINI, AND KENRICK

have offered evolutionary explanations have generally presumed that creative abilities evolved because they somehow enhanced the likelihood of our ancestors’ survival without adding unnecessary metabolic costs (e.g., Byrne, 1995; Gibson & Ingold, 1993; Kingdon, 1993). For example, a creative way to spear a fish or build a hut could have enhanced survival for their creator and his or her kin. Other evolutionary theorists have speculated that creative capacities may be a byproduct of various perceptual and cognitive mechanisms (e.g., Pinker, 1997). However, Miller (2000; Haselton & Miller, in press) has recently argued that such explanations are inadequate for several reasons. For instance, not only have other large-brained animals not evolved similar capacities, but many human displays of creativity are highly valued socially yet are difficult to explain in terms of survival value. For example, a farmer produces more tangible survival benefits in a week than a team of musicians, poets, and sculptors will produce in a lifetime. Yet a provocative melody, poem, or sculpture is likely to elicit greater appreciation than an absolutely perfect melon, potato, or zucchini. Instead of producing direct survival benefits, Miller (2000) and others have proposed that several human mental traits, including creativity, are likely to have at least in part evolved via sexual selection (e.g., Eysenck, 1995; Kanazawa, 2000). Unlike natural selection, whereby traits evolve solely because they enhance the probability of an individual’s survival, Darwin (1871) suggested that some traits, such as the elaborate plumage of peacocks, evolve via sexual selection—they evolve because they enhance an individual’s ability to attract a mate, which may or may not be independent of whether the trait enhances survival (see Andersson, 1994). Supporting this viewpoint, human creativity has several features in common with sexually selected traits across different species. Just as members of various species prefer partners with prominent sexually selected traits, such as brilliant tails, humans show a desire for creativity in a romantic partner (Buss & Barnes, 1986; Li, Bailey, Kenrick, & Linsenmeier, 2002). Sexually selected traits across species also tend to function as markers of “good genes” (Møller & Petrie, 2002; Zahavi & Zahavi, 1997). Correspondingly, Haselton and Miller (in press) have found suggestive empirical evidence indicating that creativity may partly serve a similar good genes function in humans.1 Moreover, because one of the distinct markers of sexually selected traits across species is the conspicuous display of such traits in courtship (Andersson, 1994), the current research explores the extent to which people may be more likely to display creativity when they are primed with cues related to courtship.

Cueing Creativity Through Mating Motivation Mental mechanisms that evolved to solve specific adaptive problems are often highly sensitive to ecological cues indicating a particular adaptive problem or opportunity, such as a potential ¨ hman & threat or mating opportunity (Cosmides & Tooby, 1992; O Mineka, 2001; Schaller, Faulkner, Park, Neuberg, & Kenrick, 2004; Todd & Gigerenzer, 2000). Moreover, much research has shown that various cues can automatically activate certain goal and need states (Chartrand & Bargh, 1996; Schaller, 2003), and that such states can influence perception and behavior without explicit conscious awareness (Bargh, 1990; Bargh & Chartrand, 1999).

Given the central role of reproduction in evolutionary processes, a functional perspective suggests that mating goals are likely to be closely linked to adaptive outcomes (Bugental, 2000; Kenrick, Li, & Butner, 2003). Cues related to mating can serve to both activate a mating goal and its affective responses (Scott, 1980) and trigger specific mating-related cognitive mechanisms (Gutierres, Kenrick, & Partch, 1999; Haselton & Buss, 2000; Kenrick, Sadalla, & Keefe, 1998). Furthermore, mating-related motives appear to facilitate particular perceptions, cognitions, and behaviors associated with reproductive success (Griskevicius, Goldstein, Mortensen, Cialdini, & Kenrick, in press; Maner et al., 2005; Roney, 2003). If displays of creativity have evolved in part because of their benefit in courtship, cues designed to activate mating motives may also trigger displays of creativity.

Courtship Displays and Parental Investment But should mating motives produce creative displays in both sexes? Across most species—and in 95%–97% of all mammals—it is exclusively males who display sexually selected traits during courtship (Cronin, 1993). This sex-differentiated pattern is consistent with the notion of predominantly female muses eliciting creative courtship displays in male artists (Prose, 2002). Given this evidence, it is reasonable to hypothesize that mating cues should elicit creativity only for males—the male-only creative display hypothesis. However, it is also plausible that mating cues could stimulate creative displays for both men and women. Sex differences in mammalian sexually selected traits are primarily linked to a species’ levels of maternal and paternal parental investment—the time and energy devoted to producing viable offspring (Trivers, 1972). In those species in which courtship displays are an exclusively male sport, such as peacocks, males tend to invest the absolute minimum in offspring—sperm. However, in some species these sex roles are reversed such that the males contribute the majority of parental investment (e.g., Mormon cricket, pipefish seahorse, Panamanian poison arrow frog). Consequently, in such species it is the females and not the males who display their elaborate sexually selected traits in courtship (Andersson, 1994; Buss, 2005). Humans are somewhat different from most mammals in that both men and women tend to contribute significant parental investment to their offspring (Geary, 2000). Thus, it is reasonable to predict that courtship might stimulate creative displays for both men and women. Consistent with this reasoning are findings that men and women express relatively equal preferences for creativity in a mate (Li et al., 2002). Thus, a consideration of the particulars of human parental investment suggests the alternative hypothesis that mating cues will elicit creativity for both men and women— the unisex creative display hypothesis.

Study 1 The initial experiment tested whether priming people with cues related to mating would cause an increase in their creative displays relative to people primed with neutral cues unrelated to mating. To 1 More specifically, Haselton & Miller (in press) found that ovulating women tend to unconsciously shift their preferences toward preferring creativity-related traits compared with other desirable “good dad” traits in potential sex partners.

ROMANTIC MOTIVES AND CREATIVITY

test this idea, we used photographs of desirable and available targets to prime mating. Qualitative differences in people’s displays of creativity were assessed via ratings of short stories written by participants in each condition. Comparative cross-species findings, data on human mate preferences, and parental investment theory lent themselves to two competing hypotheses: the maleonly creative display hypothesis (that a mating prime will elicit creativity only for males) and the unisex creative display hypothesis (that a mating prime will increase creative displays by both men and women).

Method Participants Ninety-one participants (Ps; 35 men and 56 women) were recruited from introductory psychology classes as partial fulfillment of their class requirement. All Ps came to the lab in groups of 2 to 6 and were each seated between partitions at a computer. The mean age for women was 19.2 (SD ⫽ 1.7), and the mean age for men was 19.7 (SD ⫽ 2.1).

Design and Procedure The overall design of the experiment was a between-participants 2 (P sex) ⫻ 2 (Prime: mating vs. control) design. At the beginning of the study, all Ps wrote two short stories that were used to assess their dispositional creative ability and their general motivation to display it. After the first two writing tasks, half of the participants were primed with cues related to mating. Following the prime, all Ps wrote an additional two stories that were later rated for their levels of creativity. These ratings served as the main dependent measure in the study. Creativity tasks. Although there are multiple psychological measures designed to evaluate various forms of creativity (see Kerr & Gagliardi, 2004), these measures are generally not designed to allow participants to display creativity in ways akin to a courtship situation. To allow Ps the freedom to display their creativity in multiple ways—to allow them to produce something that could be perceived as both novel and appropriate—we developed a methodology in which Ps were asked to write a short story about an ambiguous image. Ps viewed and wrote stories about two types of images: cartoon drawings and abstract paintings. Each of the cartoon drawings showed a pair of individuals in an ambiguous situation (in a prison cell and in a cafe´), and both abstract paintings were composed of multiple abstract colorful shapes. Each image was sequentially presented on a computer screen, and Ps were asked “What do you think is happening in this image?” All Ps had up to 5 min to write a story for each image, although if they finished earlier they could advance to the next part of the study. Because there is considerable variation in people’s creative abilities, a baseline level was established for each person. Thus, each P saw and wrote about two of the four ambiguous images (one cartoon and one painting) before the prime manipulation, and they wrote about the two other images (the counterpart cartoon and the counterpart painting) after the prime manipulation. The premanipulation stories served as a baseline measure of each P’s creative ability; the postmanipulation stories served as the main dependent measures. The order of the two types of images that Ps saw was always the same: The cartoon drawing was first and the abstract painting was second both before and after the manipulation. However, the order of which version of the cartoon and painting that Ps saw was counterbalanced, and a P never saw the same image twice. Thus, the sequence of the study for all Ps was: Cartoon A (or B), Painting A (or B), prime manipulation, Cartoon B (or A), and Painting B (or A). Mating prime. After establishing a baseline level of creativity, half of the participants were primed with mating cues similar to those that have been successfully used to activate mating goals in similar procedures (Roney, 2003; Wilson & Daly, 2004). To induce a romantic mindset, Ps

65

viewed an array of six total photos of three attractive opposite-sex individuals—two photos of each person. (The photos were collected from Match.com—a dating website—and were prerated by students as being highly attractive). All Ps were then asked to select one person from the array whom they thought was the most desirable romantic partner. After making their selection, the photo of the selected person remained on the screen, and Ps were asked to imagine that they were preparing to go on a first date with this individual. In an effort to make the mating prime more powerful, Ps were asked to write about their idea of the perfect first date with this person. All Ps had up to 3 min to write their descriptions, although if they finished before time was up they had the option of advancing to the next part of the study. Control participants underwent a similar procedure devoid of any romantic connotations. They saw a photo of a street with several buildings and were asked to imagine being on that street. They were then given up to 3 min to write about their idea of the most pleasant weather conditions in which to walk around and look at the buildings. Priming booster shots. After the prime manipulation, all Ps wrote a story about one of the cartoon images. Before going on to write the fourth and last story (about the abstract painting), all Ps underwent a prime “booster shot” to ensure that they were still in a romantic or a control frame of mind. The booster shots were procedurally identical to the original prime manipulations except that they consisted of a different array of photos of attractive individuals and a photo of another building. Creativity measures. All four stories written by each P were rated for creativity by four student judges (two men and two women) who were blind to experimental conditions. The judges were not experts on creativity; instead, they were fellow students intended to resemble the type of person who might find him or herself on a date with one of the participants. The judges rated each story on eight attributes: the extent to which they thought the story was creative, original, clever, imaginative, captivating, funny, entertaining, and charming. Each attribute was rated on a 1 (not at all) to 9 (very) scale. Before beginning the rating process, judges read a selection of the stories for each image to familiarize themselves with what to expect. Thus, the creativity ratings of the stories were relative to the other stories written in the same situation. Measures of effort. The study also included two measures designed to ascertain how much effort Ps put into writing each story. First, the amount of time Ps took to work on each story (maximum 5 min) was recorded. Second, the number of words Ps used to compose each story was also counted.

Results Overview of Stories The stories written by participants showed quite a bit of variability in creativity. Because Ps were not specifically instructed to be creative, a sizable portion of the stories were somewhat bland. For instance, when writing about the cartoons, some Ps merely wrote a sentence or two describing the situation with little additional insight (e.g., for the prison cartoon: “These are two men in prison. They are there because they were suspected of terrorism.” For the cafe´ cartoon: “These two people work together and are on a break from work at a coffee shop.”). However, many Ps were also inspired to spontaneously include more information, which usually resulted in more creative answers, as in the following description of what could be going on in the cafe´: Nigel is trying to decide whether or not to a get a nose job. He just can’t decide. However, his friend Reginald had one and his nose was simply stunning. Reginald is a very particular sort of fellow you know. That latte he’s drinking had to be just so. Soy milk with a dollop of foam and merely a whisper of cinnamon. Too much of one ingredient might completely throw off the balance of his day. When

GRISKEVICIUS, CIALDINI, AND KENRICK

66

one is so particular about cinnamon, you could only imagine how he’d prefer his nose. All of these things Nigel noted to himself as Reginald went on and on.

Despite the fact that Ps wrote about what could be interpreted as a comedic cartoon, the majority of the answers lacked any element of humor. Only a few participants responded to the cartoon question by writing a witty one-sentence tagline common in comic strips (e.g., for the two imprisoned men in shackles: “I am badly in need of a pedicure.”). The two abstract images had fewer standard responses. Although something about a psychedelic experience was periodically mentioned, a good portion of Ps made up a more interesting short description (e.g., “I think in this painting there’s a butterfly breaking out of its cocoon. However, I think this is a metaphor for someone breaking free of their past” or “I see a basketball game in a poor neighborhood with a lot of graffiti. The people are playing because they hope that basketball will help them get out of this neighborhood.”). Multiple Ps wrote responses that were more interesting, such as the following: The setting is a seedy, underground jazz club, where bands have to compete with drug dealers for the patrons’ attention. A good quintet is performing, with a tenor saxophone, two trumpets, a trombone and a drummer. The instruments are old and worn, but the music that they make is enough to turn the attention of the crack dealers and the junkies. The music is haphazard and at times seems arrhythmic and amelodic, but it fits the scene like a velvet glove.

Story Creativity The eight measures of creativity showed high cohesiveness for each of the four judges (alphas for the eight ratings for each judge ranged from .88 –.94).2 Thus, the eight ratings were combined into a creativity index for each of the four coders. The ratings of all four coders also showed high interrater reliability (␣ ⫽ .86 for the cartoon; ␣ ⫽ .88 for the painting), and the ratings for the four judges were combined into a creativity index for the cartoon and a creativity index for the painting. We next tested whether the postmanipulation creativity ratings of the stories about the cartoon and the painting were differentially influenced by P sex and the prime manipulation. This was done via a repeated-measures mixed-model 3-factor analysis of covariance (ANCOVA) with the premanipulation creativity ratings as a covariate; P sex and prime were entered as between-participants factors and type of image was entered as a within-participants factor. As expected, type of image did not interact with P sex or prime (all ps ⬎ .35), meaning that creativity for both types of images was similarly affected by the prime and by P sex. Thus, the creativity indices for the two types of stories were combined in the remainder of the analysis. The mean rating of creativity for the premanipulation (covariate) stories was 3.83 (SD ⫽ 1.41). Men tended to be more creative than women (M ⫽ 4.30, SD ⫽ 1.65 for men vs. M ⫽ 3.54, SD ⫽ 1.15 for women; p ⫽ .011). To test the specific hypotheses of the study, we performed two planned comparisons— one for men and one for women— using the premanipulation creativity scores as a covariate, which took into account the differences at baseline. For men, the mating prime significantly elevated creativity compared with men in the control, t(86) ⫽ 2.81, p ⫽ .006, ␩2 ⫽ .088 (see Figure 1). However, for women the same romantic prime had no

Figure 1. Men’s and women’s story creativity depending on type of prime in Study 1 with standard error bars (adjusted means).

effect on their creativity compared with the women in the control, t(86) ⫽ .17, p ⫽ .88. Thus, results supported the male-only creative display hypothesis, indicating that mating cues increased creativity for men but not for women.

Expended Effort The amount of effort Ps expended on each story was assessed by the amount of time they took to write the stories and the number of words used. We first assessed whether the prime manipulation had a similar impact on the expended effort measures for the two types of postmanipulation images. One ANCOVA was performed on time taken to write the story using the premanipulation time as a covariate, and another ANCOVA was performed on number of words in a story using the premanipulation number of words as a covariate. Both analyses indicated that type of image did not interact with P sex or with prime (all ps ⬎ .3), meaning that the primes had a similar effect on both men’s and women’s effort on stories about the cartoon and the painting. Thus, the effort measures for the two types of stories were collapsed. A two-factor ANCOVA on time spent writing the stories with the premanipulation time as a covariate revealed no significant interaction or main effects of P sex and prime (all ps ⬎ .35). A two-factor ANCOVA for the number of words used in the stories with the premanipulation words as a covariate also showed no significant main effects of P sex and prime (all ps ⬎ .13) or interaction ( p ⬎ .70). Overall, despite a difference for men in the creative quality of the stories in the mating prime condition, there was no indication that this creative jump was related to exerting more effort—spending more time or words— on the stories. 2

Of the eight measures, humor had the lowest correlation with the other seven measures. This was primarily because very few Ps (less than 20%) received a rating above 1 on humor. However, when Ps did display humor it was generally considered very creative, and humor still had a correlation above .50 with each of the other seven measures.

ROMANTIC MOTIVES AND CREATIVITY

Discussion The results of Study 1 supported the male-only creative display hypothesis, whereby a mating prime stimulated only men to write stories that were judged as more creative. Notably, this creative boost occurred despite there being no actual incentive for men to be more creative; that is, the stimuli were photographs and the romantic situations were imaginary. This creative increase was also not accompanied by a parallel increase in either the time spent or the number of words used to write the stories. Thus, it does not seem that a mating prime generated greater creativity because it induced participants to exert more effort into writing the story. Instead, mating cues appear to inspire an increase in the quality of men’s displays.

Study 2 Although the results of Study 1 were supportive of the maleonly creative display hypothesis, the findings were limited. For instance, research indicates that men are more sexually aroused by visual images than women (Hamann, Herman, Nolan, & Wallen, 2004). Because the key part of the mating prime procedure consisted of viewing photographs of attractive individuals, it is possible that this procedure may have been stronger at inducing a romantic mindset for males than for females. However, if the priming procedure was successful for both men and women, the lack of a creative spike for females is both slightly puzzling and highly instructive. It is puzzling because prior research indicates that both men and women indicate a relatively equal preference for creativity in a mate (Li et al., 2002). Moreover, because women’s creative displays are likely to be attractive to potential mates, it seems odd that courtship would not spur creativity. However, the lack of a creative increase for women in Study 1 is also instructive: Perhaps it reflects a tendency for women to not display their creativity when encountering a particular type of potential mate but not when encountering other types of potential mates.

Short-Term and Long-Term Mating Strategies A closer inspection of the research on human mating indicates abundant evidence that women and men behave differently in their pursuit of long-term and short-term mates (Buss & Schmitt, 1993; Gangestad & Simpson, 2000; Kenrick, Sadalla, Groth, & Trost, 1990). Whether a relationship is likely to be short-term (e.g., a one-night stand) or long-term (e.g., a marriage) has vastly different implications for the expected parental investment of men and women. If offspring result from a short-term mateship, the female is likely to contribute the majority of parental investment, whereas the male contributes practically nothing. Within a long-term relationship, however, both parents expect to contribute significantly to offspring. A short-term relationship is more akin to courtship in the vast majority of other mammalian species, in which fathers contribute little or nothing to the offspring. Under such circumstances, females tend to be highly selective about their mating partners, choosing only those mates who manifest characteristics associated with good genes (Gangestad & Simpson, 2000). Consequently, in most mammalian species and in many other vertebrate species such as the peacock, it is the male who does the displaying, and the

67

female who does the evaluating. When people are pursuing a short-term strategy, therefore, it would be expected that only the males would increase their displays of creativity. It seems likely that just this kind of a romantic relationship was made salient in Study 1, in which participants saw photos of multiple attractive opposite sex strangers and had no information indicating whether these strangers were potential long-term mates. Moreover, attractive males may be more likely to be perceived to be inclined toward unrestricted mating strategies (Gangestad, Haselton, & Buss, in press). Unlike most other mammals, however, human mating often involves substantial investment by fathers— up to and including a lifetime monogamous commitment of the male’s effort and resources. High male parental investment, though rare in mammals, tends to coevolve with slow-developing high-cost offspring that are helpless at birth (Geary, 2000). In species with high male parental investment, fathers and mothers invest a great deal in a long-term mate, and both sexes tend to be choosy about the characteristics of a long-term partner. It would follow that when people are pursuing a long-term mate, both men and women would be under selective pressure to display desirable sexually selected characteristics, including the ability to be creative. Study 2 tested whether priming participants with cues for an explicitly short-term versus a long-term mating situation would produce a different pattern in creative displays for males versus females. Given cross-species findings and differences in expected parental investment for the two types of mating strategies, two predictions were generated. For men, it was predicted that both short-term and long-term mating primes would elicit creativity. However, for women it was predicted that only the long-term prime would produce a creative boost. To prime the two different mating situations, all participants read a short scenario in which they imagined themselves desiring a short-term or a long-term mate. To avoid the potentially problematic sex difference in arousal to visual cues, we did not use photographs.

Method Participants Two hundred participants (64 men and 136 women) were recruited from introductory psychology classes as partial fulfillment of their class requirement.

Design and Procedure The study design was a between-participants 2 (P sex) ⫻ 3 (Prime: short-term mating vs. long-term mating vs. control) design. The general procedure of the study was very similar to that of Study 1, including using the identical method to measure creativity. However, the method used to prime mating was different. Mating primes. All Ps read and imagined themselves in one of three scenarios (short-term mating, long-term mating, and control) that were of similar length (about 850 words). In the short-term scenario, Ps imagined themselves during the last day of their vacation on an exotic island. On this last day they met a desirable person and spent a romantic afternoon and had dinner with the new romantic interest. The scenario ends as the two lovers are passionately kissing on a moonlit beach. The short-term scenario repeatedly emphasized that the two people would likely never see each other again. In the long-term scenario, Ps imagined meeting someone desirable on the university campus. Ps imagined spending a wonderful afternoon and a

GRISKEVICIUS, CIALDINI, AND KENRICK

68

romantic evening with this person, including a candlelight dinner and a sweet kiss goodnight. Throughout the scenario, the reader ponders that this person may be a good long-term partner, and the scenario ends as the reader is anticipating going out on an “official” first date with this person. In the control scenario, Ps imagined getting ready to go to a muchanticipated concert with a same-sex friend. During the night of the show, Ps imagined that they could not find the tickets. However, the scenario has a happy ending as the friend shows up with the tickets, and they both head off in a great mood anticipating a delightful musical experience. To test whether the scenarios were successful at eliciting romantic feelings, self-report measures of romantic and sexual affect were collected from a total of 58 male and female Ps. After reading one of the three scenarios, Ps indicated to what extent they were experiencing sexual and romantic arousal on a 7-point scale with the endpoints at 1 (not at all) and 7 (very much). Results indicated that there were no differences in the measures across the two romantic scenarios for men or women (all ps ⬎ .35). However, compared with Ps in the control scenario, Ps who read the mating scenarios reported significantly more romantic arousal (mating M ⫽ 5.88, SD ⫽ .61; control M ⫽ 1.85, SD ⫽ .48; p ⬍ .01) and significantly more sexual arousal (mating M ⫽ 4.93, SD ⫽ .78; control M ⫽ 1.18, SD ⫽ .29; p ⬍ .01). Priming booster shots. As in Study 1, the current experiment had a booster of the prime manipulation after Ps wrote the first postprime creative story. In the booster for both romantic prime conditions, Ps were asked to imagine themselves back in the scenario that they read earlier. Then, Ps were given up to 3 min to write in detail about what physical characteristics they would desire in this person. In the control condition, Ps were also asked to go back to the scenario they read earlier. However, they wrote about the physical characteristics of the anticipated concert venue. Dependent measures. As in Study 1, four student judges (two men and two women) who were blind to experimental condition rated the stories on the same creativity dimensions. The amount of time Ps spent writing each story and the number of words used were also recorded.

Results Creativity The eight creativity measures again showed high cohesiveness for each of the four judges (alphas for the eight ratings for each judge ranged from .90 –.94), so the eight measures were combined into a creativity index for each judge. The ratings of all four judges also showed high interrater reliability (␣ ⫽ .87 for the cartoon; ␣ ⫽ .88 for the painting). Thus, the ratings for the four judges were combined into a cartoon creativity index and a painting creativity index. To test whether the prime had a similar effect on men’s and women’s creativity on each type of postmanipulation image, we used the same analysis strategy from Study 1. A repeated-measures mixed-model 3-factor ANCOVA with type of image as a withinparticipants factor and the premanipulation ratings of creativity as a covariate again indicated that type of image did not interact with the other two factors (all ps ⬎ .60). Thus, the creativity indices for the two types of stories were combined into one measure.3 The mean rating of creativity for the premanipulation (covariate) stories was 4.18 (SD ⫽ 1.44), and males again tended to be more creative than females at baseline (M ⫽ 4.52, SD ⫽ 1.48 for males vs. M ⫽ 4.02, SD ⫽ 1.40 for females; p ⫽ .021). To test the specific hypotheses of the study, we performed a series of planned contrasts using the premanipulation stories as covariate, which took into account the differences at baseline. It was predicted that both of the romantic scenarios would increase men’s creativity. As seen in Figure 2, men in the long-

Figure 2. Men’s and women’s story creativity depending on type of prime in Study 2 with standard error bars (adjusted means).

term condition did indeed write significantly more creative stories than men in the control condition, t(193) ⫽ 3.77, p ⫽ .003, ␩2 ⫽ .047. Although men in the short-term mating condition also displayed more creativity than men in the control condition, this difference was only marginally significant, t(193) ⫽ 1.69, p ⫽ .096, ␩2 ⫽ .016.4 Nevertheless, men in the long-term and the short-term conditions did not differ from one another, t(193) ⫽ 1.41, p ⫽ .16, and the combination of the two mating conditions was significantly different from the male control condition, t(193) ⫽ 2.71, p ⫽ .009, ␩2 ⫽ .037. Thus, both a short-term and a long-term mating prime boosted men’s creative displays to some extent. As expected, women in the short-term mating condition showed no significant difference in creativity from women in the control, t(193) ⫽ .34, p ⫽ .73. This predicted lack of difference conceptually replicated the findings for women from Study 1. However, a planned comparison of the long-term mating prime and the control conditions did not indicate the predicted increase in creativity for women, t(193) ⫽ .98, p ⫽ .32. Thus, neither the 3 The lack of interactions also indicates that imagining the physical characteristics of the potential mate in the booster shot did not have a different impact on creativity than just reading the original scenario. 4 The slightly lower creativity for men in the short-term versus the long-term condition can be primarily attributed to 2 Ps, who reported currently being in a committed romantic relationship and showed a sizable creative drop after imagining pursuing a short-term fling. However, the identical short-term scenario did produce significant increases in male creativity in Studies 3 and 4, and Ps’ current relationship status did not significantly affect the results in any of the studies. In addition, analyses of Ps’ sociosexual orientation (Simpson & Gangestad, 1991, 1992) and creativity in the different conditions also did not indicate any significant effects, perhaps because the scenarios primed specific mating strategies irrespective of a person’s sociosexual inclination.

ROMANTIC MOTIVES AND CREATIVITY

short-term nor the long-term mating primes boosted women’s creativity.5

Expended Effort Two three-factor ANCOVAs (one for time and one for number of words) again indicated that type of image (cartoon or painting) did not interact with either P sex or prime (all ps ⬎ .60). Thus, the effort measures for the two types of stories were collapsed. An ANCOVA for time spent writing the stories again indicated no significant interaction or main effects of P sex or prime (all ps ⬎ .40). An ANCOVA for the number of words used in the stories also indicated no significant interaction or main effects (all ps ⬎ .30).

Discussion The results of Study 2 provided further qualified evidence that cues related to mating can lead to an increase in displayed creativity, at least in males. As predicted, men showed an increase in creativity when primed with thoughts of pursuing either a shortterm or a long-term mate. Also as predicted, women did not show an increase in creativity in the short-term prime condition. However, contrary to prediction, women failed to show an increase in creativity in the long-term condition. The overall pattern for men and women was consistent with the results of Study 1. Also following the results of Study 1, none of the increases in creativity produced by the mating primes was accompanied by any indication of increased effort to produce the display.

Study 3 Although the results of Study 2 supported a mating-inspired increase in creativity for men, the supposed long-term mating prime did not lead to the predicted increase in women’s creative displays. Why not? First, it is possible that as in many species, women may simply not display this sexually selected characteristic in courtship. Instead, women may have evolved creative abilities primarily to judge the quality of men’s creative displays (Miller, 2000). However, it is possible that our long-term mating prime manipulation, which involved preparing for a first date with a man who had long-term potential, may have been insufficient to reach the threshold for women’s mating-linked displays of creativity. Because women incur significantly higher reproductive costs if they are abandoned by their mates, women selecting a long-term mate should be especially focused on and sensitive to the trustworthiness and commitment levels of the man (Haselton & Buss, 2000; Hrdy, 1999; Hurtado, Hill, Kaplan, & Hurtado, 1992). Following this logic, women would need to be assured of the trustworthiness and the commitment level of a potential mate before committing themselves strongly. This reasoning is consistent with findings that women are, compared with men, slower to report feelings of love during courtship and have a higher desire for commitment from their romantic partner before consenting to sex (Peplau, 2003). Women are also, compared with men, relatively more conservative about believing professions of love by members of the opposite sex (Haselton & Buss, 2000). Although the long-term scenario used in Study 2 specified that the imagined mate was a potential long-term partner, the overall perceived mate quality of the person— especially as it relates to trustworthiness

69

and commitment—may have been too ambiguous to ensure that women would perceive this person to be a truly high-quality mating prospect. Study 3 was undertaken to examine this possibility. Study 3 was conceptually similar to Study 2, except that a new key condition was added in which men and women read about a long-term mate who was a clearly committed long-term partner. The results of Study 2, along with male and female differences in parental investment, led to two separate predictions— one for men and one for women. For men, it was predicted that creativity would increase after all of the three mating primes. However, the pattern for women was predicted to be distinctly different: The committed long-term mating prime should be the only one to produce a higher level of creativity when compared with the other conditions, which should again not differ from one another. In an attempt to explore whether specific mating cues can produce creative boosts in other domains of creativity besides story-writing, Study 3 also used a different measure of creativity— the Remote Associates Test (RAT; Mednick, 1962). The RAT was originally developed as an objective test of creativity, whereby creativity is defined in a much narrower scope: Creativity is the ability to make rapid appropriate associations between various concepts. Each RAT question consists of providing people three words (e.g., “dress, dial, flower”) and giving them a limited amount of time (15 seconds in the current study) to come up with the one correct word linked to all three of the original words (“sun”). Success on the RAT has been shown to correlate reliably with success on classic insight problems, and the RAT is often used in the study of creative problem solving (Bowden & Beeman, 1998; Schooler & Melcher, 1995). Finally, to assure that the results of Studies 1 and 2 were not caused by some peculiar effect produced by the content of the control conditions, Study 3 used a no-prime control condition. That is, participants in the control did not read a scenario.

Method Participants One hundred and fifty-seven participants (85 men and 72 women) were recruited from introductory psychology classes to participate as partial fulfillment of their class requirement.

Design and Procedure The design of the study was a between-participants 2 (P sex) ⫻ 4 (Prime: short-term vs. potential long-term vs. committed long-term vs. control) design. The conceptual procedure was similar to that of Study 2, except for the addition of a new romantic scenario for the committed long-term condition and the use of the RAT to assess creativity. RAT. The RAT can be administered using any number of questions while giving Ps any amount of time to answer them. In the current study, Ps saw 40 RAT-like questions and had 15 s to answer each question. The

5

In addition to judging creativity, the stories were also judged on the perceived intelligence of the participant. The measures of intelligence showed high concordance with the measures of creativity, whereby the mating primes led males to be perceived as more intelligent. However, it may have been difficult to meaningfully separate creativity and intelligence within the judging context, making it uncertain whether mating primes indeed boosted a purer form of intelligence display.

70

GRISKEVICIUS, CIALDINI, AND KENRICK

first 20 questions were used to establish a baseline score for each P regarding their general RAT performance. Then, after the priming manipulation, Ps worked on a different set of 20 RAT questions; performance on this last set of 20 constituted the dependent measure in the study. All of the specific RAT questions were adapted from Bowden and Beeman (2003), who provide normative statistics regarding the percentage of people who tend to solve specific RAT questions within various time limits. The 40 questions used in the current study were shown to have been solved 40%– 60% of the time within a 15-s time limit by university students. Ps in all conditions responded to the same first set of 20 questions (baseline) and to a different set of 20 subsequent questions (dependent measure). The two sets were matched to be of equal difficulty, and the order of the questions within each set of 20 was randomized. Mating primes. The same procedure from Study 2 was used to prime mating. Two of the romantic scenarios (short-term and potential long-term) were identical to those used in Study 2, except that the original long-term scenario has now been labeled potential long-term. To prime cues of a committed high-quality long-term mate, a brief paragraph was added to the original long-term scenario. The paragraph contained three new pieces of information: (a) the couple had already been dating for a while, which was intended to signify commitment; (b) the P had met and approved of the target’s friends, which was intended to signify trustworthiness; and (c) the target had met the P’s friends who had given their approval, which was intended to signify that the target was good relationship material. Unlike in Studies 1 or 2, Ps in the control condition did not read a story or see any photos. Instead, they were given a short break and saw a blank screen before continuing to the next set of RAT questions. Priming booster shot. As in Studies 1 and 2, all Ps received a booster shot to refresh the mating prime after completing the first 10 of the second set of the dependent-measure RAT questions. The specific nature of the booster shot was identical to that in Study 2, in which people were asked to write about the physical characteristics of the imagined person from the romantic scenario. In the control condition, participants were merely given another short break and saw a blank screen before continuing with the remaining 10 RAT questions.

Results Creativity For the 20 baseline RAT questions, participants solved 8.72 (SD ⫽ 3.44) questions, and there were no sex differences in performance. In addition to testing the predictions of the study via planned contrasts, we also tested whether the predicted contrasts explained most of the between-cell variance by testing the residual contrast, which was predicted to be nonsignificant (see Levin & Neumann, 1999). Male creativity. For men, it was predicted that the three mating primes would produce a higher level of creativity than in the control condition. As seen in Figure 3, a planned contrast using the baseline RAT measures as covariates indicated that this was indeed the case, F(1, 80) ⫽ 2.58, p ⫽ .012, ␩2 ⫽ .077. Creativity in the three mating conditions did not differ from one another ( p ⫽ .99). The test of the residual contrast was also not significant, Fresidual (2, 80) ⫽ 0.55, ns, meaning that the predicted contrast accounted for most of the between-cell variance. Female creativity. For women, it was predicted that only the prime for the committed long-term mate would produce a creative boost above that of the other three conditions. As seen in Figure 3, a planned contrast with baseline RAT scores as the covariate indicated that this was indeed the case, F(1, 72) ⫽ 2.03, p ⫽ .048, ␩2 ⫽ .056. As would be predicted from Studies 1 and 2, the remaining three conditions—short-term, potential long-term, and control— did not differ from one another ( p ⫽ .70). The test of the residual contrast was also not significant, Fresidual (2, 72) ⫽ 2.96, ns, meaning that the predicted contrast explained most of the between-cell variance in creativity. Thus, women showed a creative increase only when they were primed with thoughts of a committed long-term mate who was clearly high-quality relationship material.

Figure 3. Men’s and women’s performance on the Remote Associates Test (RAT) depending on type of prime in Study 3 with standard error bars (adjusted means).

ROMANTIC MOTIVES AND CREATIVITY

71

Differences in Imagined Mate Across Scenarios

Mood and Arousal

To examine whether participants indeed perceived the key differences across the romantic scenarios, we collected additional data to ascertain what Ps actually perceived to be the differences and similarities of the imagined mate in each of the three scenarios. A total of 92 separate male and female Ps underwent the same priming procedure used in the current study, whereby each person only read one of the three scenarios. Afterward, Ps indicated the quality of various characteristics regarding the person whom they imagined desiring in the scenario. More specifically, Ps were asked 13 questions in random order that were generally phrased in the following way: “To what extent is this person __________ ,” and Ps provided their responses on a scale of 1 (not at all) to 9 (very). The questions asked about the imagined mate’s perceived trustworthiness (trustworthy, truthful, honest; ␣ ⫽ .93), level of commitment (committed, faithful, likely to cheat [reverse scored]; ␣ ⫽ .80), and desirability as a long-term partner (good relationship material, the right person for me, confidence that they’re a good boyfriend or girlfriend; ␣ ⫽ .91). In addition, Ps were also asked to report the extent to which the potential mate was seen as being creative, intelligent, funny, and physically attractive. To test whether men and women differed in their perceptions of the mate on the characteristics across the scenarios, 13 two-factor analyses of variance (P sex ⫻ Prime) were performed. Results indicated no significant interactions (all ps ⬎ .30). Thus, male and female ratings were combined for the remainder of the analysis. As seen in Table 1, Ps did indeed perceive different levels of trustworthiness, commitment, and long-term mate value across the three conditions: The short-term mate was always perceived as having the least of the three qualities, and the committed long-term mate was always perceived as having the most of the three qualities. When compared with the potential long-term scenario, the person imagined in the committed long-term scenario was perceived as significantly more trustworthy, t(89) ⫽ 2.48, p ⫽ .015, ␩2 ⫽ .064, more committed, t(89) ⫽ 2.10, p ⫽ .038, ␩2 ⫽ .047, and of higher long-term mate quality, t(89) ⫽ 2.08, p ⫽ .040, ␩2 ⫽ .047. Thus, men and women did perceive the key differences between the two long-term primes. Analyses of Ps’ perceptions of physical attractiveness, creativity, intelligence, and humor indicated no significant differences among the three conditions (all ps ⬎ .30), with all of the ratings being relatively high (see Table 1 for ratings of attractiveness and creativity as an example).

One possible mechanism for how the romantic primes may have produced the specific patterns of creativity was by influencing men’s and women’s mood or arousal. To investigate this possibility, a separate group of a total of 63 male and female Ps underwent the priming procedure from Study 3. Afterward, Ps rated to what extent they felt positive arousal (energetic, excited, passionate; ␣⫽ .75), negative arousal (upset, tense, nervous; ␣ ⫽ .83), and positive mood (happy, upbeat, joyful; ␣ ⫽ .83). Ratings were provided on a 1 (not at all) to 7 (very much) scale. Results indicated that none of the romantic primes produced much negative arousal for men or women (Ms between 1.98 and 2.40). However, the patterns for positive mood and positive arousal were instructively different across the different conditions (see Table 2). For men, the two long-term primes produced similar levels of mood and arousal, and ratings for both primes were slightly (nonsignificantly) higher than those for the short-term prime. For women, mood and arousal for the two long-term mating primes also did not differ from each other, and both were significantly higher than the mood and arousal produced by the shortterm prime ( p ⬍ .05). This overall pattern for women is quite different from the pattern of women’s performance on the RAT. Specifically, although the two long-term conditions produced highly similar patterns of positive mood and positive arousal, women’s RAT performance in these two conditions differed. Moreover, despite men’s and women’s different performance on the RAT in the short-term mating condition, the short-term scenario produced no sex differences in positive arousal or mood (all ps ⬎ .60). Taken as a whole, the RAT performance patterns produced by the romantic scenarios appear to be different from the mood and arousal patterns produced by these primes.

Table 1 Perceived Attributes in Imagined Mate Depending on Scenario Type of romantic scenario Perceived attribute

Short term

Potential long term

Committed long term

Trustworthy Committed Relationship material Intelligent Creative

6.00 (1.58) 5.32 (1.48) 5.83 (1.82) 8.84 (0.76) 6.85 (1.46)

6.69 (1.18) 6.43 (1.16) 7.01 (1.46) 8.00 (1.02) 6.75 (1.27)

7.53 (1.25) 7.15 (1.55) 7.77 (1.14) 8.16 (1.32) 6.97 (1.14)

Discussion Results from Study 3 provided a clearer picture of how mating cues influence creativity. Study 3 replicated both the short-term and the (potential) long-term findings from Study 2. It is important to note that these results were replicated with a different measure of creativity—the RAT—and a no-prime control condition. As predicted, Study 3 also showed that when participants were primed with the desire for a committed long-term mate—a mate who was perceived to be more trustworthy, committed, and generally better relationship material— both men and women showed higher levels of creativity. This finding supports the notion that when males or females are pursuing a mating strategy in which they each expect to contribute significant parental investment, both sexes are likely to display a sexually selected trait such as creativity. However, men and women appear to have different mate-quality thresholds for displaying creativity: For men, the requirements are relatively low—the desire to attract any desirable woman will do; for women, the requirements are high— only a desire for a long-term mate who is clearly a high-quality mate produces the display. Additional data also indicated that the precise pattern of creative boosts for men and women could not be explained solely by the arousal or positive mood that were produced by each of the romantic scenarios.

Study 4 Note. Means are on a 1–9 scale in which higher numbers indicate more of the perceived characteristic. Numbers in parentheses denote standard deviations.

Thus far, findings from all three studies show that specific cues related to mating can produce an increase in creativity for men and

GRISKEVICIUS, CIALDINI, AND KENRICK

72

Table 2 Men’s and Women’s Mood and Arousal Depending on Scenario Type of romantic scenario

Positive mood Males Females Positive arousal Males Females

Short term

Potential long term

Committed long term

4.42 (1.65) 4.27 (1.65)

5.00 (1.27) 5.70 (0.85)

5.07 (1.16) 5.73 (0.77)

4.18 (1.94) 3.67 (1.90)

4.54 (1.61) 5.03 (0.96)

4.57 (1.41) 5.20 (0.92)

Note. Means are on a 1–7 scale in which higher numbers indicate a more intense state. Numbers in parentheses denote standard deviations.

women. However, would mating cues continue to produce a creative boost when compared with people who are motivated to do well on the creative task? To test this question, the current study used a methodology almost identical to that of some conditions of Study 3, except that it added a key new condition in which participants were provided with a monetary incentive to do well on the RAT but in which no romantic prime was used—the monetary incentive condition. The study had two competing hypotheses with different implications for the process by which mating cues may stimulate creativity. First, it was plausible that Ps in the monetary incentive and the men in the (short-term) mating condition would perform significantly better on the RAT than the Ps in the control condition. Such a result would indicate that mating cues might facilitate creativity by motivating people to somehow work harder at the task. However, a different outcome was also deemed plausible. The second possibility was that men in the mating condition would perform significantly better than men in any of the other conditions. Such a result would indicate that mating cues do not increase creativity simply by leading people to try harder at the task.

RAT problems. This question served as a manipulation check of the incentive instructions. Analyses indicated that motivation for men and women did not interact with condition ( p ⬎ .70), and that men’s and women’s motivation did not differ from each other across the three conditions (for men, M ⫽ 4.79, SE ⫽ .16; for women, M ⫽ 4.80, SE ⫽ .15), F(2, 161) ⫽ .085, p ⫽ .92. Thus, the motivation scores for men and women were combined. Results also indicated that Ps in the mating and control conditions also did not differ from each other in motivation (for mating, M ⫽ 4.59, SE ⫽ .18; for control, M ⫽ 4.71, SE ⫽ .19; p ⫽ .63). Thus, these two groups were combined into a no-incentive condition. A comparison of the monetary incentive condition and the combined no-incentive group revealed that participants in the monetary incentive condition reported being significantly more motivated to do well on the RAT items (for monetary incentive, M ⫽ 5.09, SE ⫽ .19; for no incentive, M ⫽ 4.65, SE ⫽ .13), F(1, 165) ⫽ 3.90, p ⫽ .051, ␩2 ⫽ .023. Thus, the manipulation to induce motivation to do well on the task had the desired effect.

Creativity For the 20 baseline RAT questions, participants solved 8.01 (SD ⫽ 3.12) questions and there were no sex differences in performance. To test the specific hypotheses of the study, we performed a series of planned comparisons using the baseline RAT scores as a covariate. First, examining the men, a planned contrast revealed the predicted increase in RAT performance in the mating condition when compared with the control condition, t(160) ⫽ 2.83, p ⫽ .005, ␩2 ⫽ .048. However, as seen in Figure 4, men in the mating condition also performed significantly better than men in the monetary incentive condition, t(160) ⫽ 2.21, p ⫽ .029, ␩2 ⫽ .030. Although men in the monetary incentive condition performed slightly better than men in the control condition, this difference

Method Participants One hundred and sixty-seven participants (78 men and 89 women) were recruited from introductory psychology classes to participate as partial fulfillment of their class requirement.

Design and Procedure The design of the study was a between-participants 2 (P sex) ⫻ 3 (Condition: mating vs. control vs. monetary incentive) design. The procedures in the (short-term) mating and the control prime conditions were identical to those of Study 3. In the monetary incentive condition, the procedure was similar to that of the control, except that before Ps started the last 20 RAT problems, they received the following instructions: “On the remaining problems, we would like you to try your hardest to do the best you can. If you are in the top 30% of everyone’s scores, you will be entered in a raffle to win $60!”

Results Motivation Check After Ps finished the RAT questions, they were asked to what extent they were motivated to do well on the 20 postmanipulation

Figure 4. Men’s and women’s performance on the Remote Associates Test (RAT) depending on condition in Study 4 with standard error bars (adjusted means).

ROMANTIC MOTIVES AND CREATIVITY

was not significant ( p ⫽ .31). For women, there were no significant differences across conditions, F(2, 85) ⫽ 2.24, p ⫽ .44.

Discussion The results of Study 4 indicated that men primed with mating cues performed better on the creativity task even when compared with people who were trying harder at the task. Previous research (Amabile, 1996) has indicated that extrinsic motivation, such as a monetary incentive, does not always result in higher creative quality, so perhaps the finding that an incentive to do well on the RAT did not significantly improve performance is not surprising. However, the fact that a mating prime led to a creative boost above that produced by merely trying hard at the task suggests that the process by which mating cues stimulate creativity is not related to external motivation. Consistent with the findings from Studies 1 and 2, the mating-inspired creative boost appears to be unrelated to expending more effort on the creative task.

General Discussion The current research set out to explore whether priming men and women with a variety of theoretically relevant mating cues would lead to an increase in creativity. This question was tested in four studies, in which men and women first either looked at photographs of attractive opposite sex individuals (Study 1) or imagined being in a particular romantic scenario (Studies 2– 4), and then performed tasks assessing creativity on subjective (Studies 1 & 2) or objective (Studies 3 & 4) measures. For men, cues designed to stimulate a motive to attract either a desirable short-term or a long-term mate produced an increase in creativity. That is, after just thinking of attracting a desirable woman as any kind of a romantic partner, men showed an increase in creativity— even if they themselves could not actually benefit romantically from this creative burst in the current setting. In contrast, women only increased their creative output after imagining wanting to attract a clearly high-quality (i.e., trustworthy and committed) long-term mate. Women did not show a creative increase when primed to think about attracting a short-term mate or a potential long-term mate who had yet to prove his worth as good relationship material. These findings are consistent with the theories of sexual selection and differential parental investment. When pursuing a shortterm mating strategy, women are expected to provide the vast majority of parental investment. In other species, selectively high female investment is also associated with courtship displays primarily by males. For example, male bowerbirds, who do not contribute direct resources to offspring, have evolved to construct elaborate bowers during courtship; they then artfully decorate the bowers with a colorful assortment of flower petals, berries, and snail shells, or even “paint” it with regurgitated bluish residue (Borgia, 1986). Females inspect the bowers and preferentially mate with males who have the largest, most symmetric, and best decorated displays because, like the peacock’s tail, the creative bower displays seem to serve as a reliable indicator of fitness (Borgia, 1995). When pursuing a long-term mate, however, both men and women expect to invest significantly in offspring. In other species, high male parental investment is associated with both sexes being choosier when selecting a mate and with both sexes tending to display desirable qualities in courtship. Such a pattern can be seen

73

in gibbons, in which each sex invests substantially in offspring. During courtship, both male and female gibbons have evolved to sing complex and elaborate musical patterns to each other—a behavior called dueting (Geissmann, 2000; Raemaekers & Raemaekers, 1984). Despite the fact that human males often invest substantially in shared offspring, women always stand a reasonable danger of substantially higher costs if they choose a long-term mate who does not intend to commit (Kenrick et al., 1990). In this light, it makes some sense that women require assurance that a prospective mate is really going to invest in offspring before investing the energy in creative displays.

The Psychological Mechanism The current research examined several plausible psychological processes by which mating cues could stimulate creativity. Results from three of the studies indicated that mating cues are unlikely to elicit creativity by stimulating one to try harder to be creative. In Studies 1 and 2, participants neither spent more time nor used more words to write creative stories. In addition, participants given a monetary incentive in Study 4 still performed worse than people primed with mating cues. Although it is possible that a chance to win $60 may not have been a strong incentive, participants in this condition nevertheless reported being more motivated to try harder. Study 3 indicated that the mood and arousal states produced by each of the specific romantic scenarios showed a distinctly different pattern from that of creativity. For instance, although women were equally positively aroused by the potential and the committed long-term mate scenarios, only the committed scenario produced a creative boost for women. Moreover, men and women reported relatively equal mood and arousal after reading the short-term scenario, although only men showed a creative increase in that situation. Although the mood and arousal ratings were selfreported, these findings suggest that the particular pattern of increases in creativity cannot be explained solely by changes in mood and arousal— or at least the kind of general arousal typically measured in psychological research. A third process, which we believe is most likely to underlie these effects, can be inferred from Studies 3 and 4, in which the RAT was used to assess creativity. Because the purpose of each RAT question is to ascertain whether a person can rapidly locate a particular concept in their mind, the RAT is an appropriate and rather specific test of the accessibility of various concepts. The results of Study 4 indicated that merely trying harder did not improve performance on the RAT; however, a mating prime did improve performance, most likely because it enabled superior accessibility to relevant but remote informational links. The results of Studies 1 and 2 are also consistent with the possibility that mating cues enabled better accessibility to various concepts: By having access to a richer set of creative avenues brought about by the romantic primes, people could write stories that were more creative.

Alternative Explanations The current research has adopted a framework inspired by existing theory regarding sexual and natural selection accounts of creativity (Miller, 2000). It would surely be possible to derive predictions regarding creativity and mating from several other theoretical perspectives, though none seem to offer as straightforward an account of the pattern of results obtained in this series of studies. For example, it

74

GRISKEVICIUS, CIALDINI, AND KENRICK

is possible that any vivid fantasy could prime a creative process and lead to an increase in creativity. Thus, for women, perhaps only imagining a high-quality long-term mate may have led to a fantasy that was vivid. However, because both the potential and committed long-term mating primes were very similar to each other but only one of them produced an increase in creativity, this explanation seems unlikely. Not only did both primes produce highly similar mood and arousal for women, but both primes are likely to have produced relatively equally vivid fantasies. Perhaps an association between creativity and mating arises because of simple mechanisms of associative priming (Srull & Wyer, 1979; see Higgins, 1996 for a review). For example, Chartrand, Fitzsimons, and Fitzsimons (2004) have shown that participants primed with thoughts about Apple computers—a brand marketed as the computer for creative individuals— became more creative compared with people primed with other computer brands. Although priming participants with photos of attractive individuals or romantic scenarios may very well activate concepts of creativity, it is difficult to see how an associative model framework could account for the very specific pattern of sex differences and sex similarities found across the different conditions. The functional framework used in this research is by no means an alternative to the associative network model of cognition. Both models imply that there are certain links between motivation, cognition, and behavior. However, the functional model does more than just assert that priming specific ideas will lead to the activation of associatively linked semantic and affective categories. Rather, the functional model leads to more finely articulated predictions about how activating specific functional goals should lead to specific goal-consistent—and sex-consistent— cognitive and behavioral responses (Maner et al., 2005). A social learning model might also suggest that men and women have been differentially rewarded for producing creativity, although it is again difficult to predict the precise prime-specific pattern of sex differences and similarities that we found from this perspective. Further, the final study suggested that tangible rewards actually failed to increase creativity. A social role theory might also posit that it is part of the male role to be creative. However, without a consideration of sex differences in parental investment, it would be difficult for that theory to have predicted a priori why creativity is part of the male role to be creative in all mating contexts but part of the female role to be creative only in committed long-term mating contexts. Neither social role theories nor social learning theories are mutually exclusive with evolutionary accounts, as evolutionary theorists presume that social roles across societies are to some extent a function of evolved adaptations in men and women and that many behaviors involve an adaptive interplay of learning and ¨ hman evolved predispositions (Kenrick, Trost, & Sundie, 2004; O & Mineka, 2001). We are not aware, however, of social role or social learning theorists who have offered predictions for the pattern of results obtained here—a pattern which follows directly from considerations of sexual selection and parental investment.

Limitations and Future Directions One of the limitations of the current research is that it did not test whether people become more creative during actual courtship (as suggested by anecdotal evidence). Although future research needs to explore this question, the current framework would pre-

dict that the presence of a desirable mate is likely to produce the strongest creative boost, as long as other forces, such as social anxiety, are not working against the display. The current studies also explored only two specific forms of creativity in one culture. However, the proposed evolutionary framework would predict that mating-related cues should spur creativity across cultures, although the specific forms of creativity will surely depend on cultural or local norms (Norenzayan & Heine, 2005; Norenzayan, Schaller, & Heine, in press). Moreover, there may also be multiple person variables that could be used to better predict the display. For instance, mating cues might lead to a boost in people’s most favored or best-practiced form of creativity: a musician might become more musically creative, a comedian might be seen as funnier, a poet might be inspired to display verbal fluidity, and a dancer may become more expressive through movement. Future research might also examine why creativity isn’t always “turned on” in men as it appears to confer significant benefits. One possibility is that there are costs associated with allocating one’s energies to permanent creative displays, such as decreased capacity to attend to other matters or decreased functioning of short-term memory. Another possibility is that there are limitations in most people’s abilities to be creative, and that creativity is in a sense a form of “truth in advertising” for a highly functioning brain (see Miller, 2000). Although the current findings are consistent with a sexual selection account of creativity, these studies were not designed to ascertain whether creativity is an exclusively sexually selected versus naturally selected trait. Indeed, it is often difficult to draw a line between sexual selection and natural selection, partly because mating choices are often based on traits that are themselves naturally selected adaptations (such as symmetrical and healthy feather displays or the ability to fly well in birds). Thus, it is plausible that some more rudimentary form of creativity in humans may have originally evolved because of its association with survival ability (e.g., enhancing ability to create new tools or verbally negotiate one’s way out of a conflict). Given that such abilities were adaptive for whatever reason, the members of one sex would have been well served to use them as cues to infer mate value in the other sex. At that point, it would have become useful to display creativity as a courtship tactic. Thus, creativity would be a function of both natural and sexual selection. That is, creativity might provide a survival advantage and function as a heritable fitness indicator (Haselton & Miller, in press). The fact that our research indicates that creativity is likely to be displayed by males across mating contexts, but only selectively by females in contexts associated with quality long-term mates, does suggest, however, that sexual selection plays some role in human creative displays (whatever their ultimate origin). A larger question pertains to how creativity fits in with other possibly related constructs that may also be in part sexually selected. Although the kinds of creativity measured in the first two studies (story-writing) may be similar to artistic displays, the creativity displayed via the RAT in the latter two studies appears to be more closely related to problem solving rather than artistic ability. From the perspective adopted here, creativity is likely to be related to intelligence (Miller, 2000). In this light, future research might address whether mating motives provide a boost in many types of creative intelligence or perhaps whether only specific forms of creative intelligence are associated with courtship displays. If such displays evolved as fitness indicators, one interesting

ROMANTIC MOTIVES AND CREATIVITY

possibility is that those individuals who show greater creative boosts in response to mating primes might also be more likely to possess characteristics associated with heritable fitness.

Conclusion This set of studies began with a puzzling question: Why are muses predominantly women who inspire men? Four experiments indicated that when men merely thought of pursuing a desirable romantic partner, either for a long-term relationship or a short-term fling, such thoughts consistently produced a “muse effect,” whereby they boosted male creativity. These findings are consistent with the experiences of Picasso and other male artists who were inspired by female muses. The findings also generally align with the view that creative displays may be linked to sexual selection, which tends to select for more intensive male displays, as witnessed in the courtship tail displays of peacocks. Unlike in peacocks, however, because men often invest heavily in their offspring, women are also subject to sexual selection pressures when it comes to long-term mating. Consistent with parental investment theory, when women were primed with cues designed to instill a desire to attract a high-quality long-term mate, these cues also served to produce a muse effect for women by spurring their creativity.6

6 The renowned Victorian poet Elizabeth Barrett Browning may provide one such example (Winwar, 1950). Mired in artistic mediocrity in her 20s, Elizabeth received a letter from an adoring fan, Robert Browning, who professed his love for her. Not surprisingly, Elizabeth was unconvinced of his intentions until a year—and many of Robert’s letters—later, at which point the two lovers agreed to elope in the near-distant future. It was precisely during this phase of her relationship when Elizabeth was inspired to pen Sonnets from the Portuguese, which would become her greatest and most critically acclaimed creative work.

References Amabile, T. M. (1996). Creativity in context: Update to the social psychology of creativity. Boulder, CO: Westview. Andersson, M. (1994). Sexual selection. Princeton, NJ: Princeton University Press. Bargh, J. A. (1990). Auto-motives: Preconscious determinant of social interaction. In E. T. Higgins & R. M. Sorrentino (Eds.), Handbook of motivation and social cognition: Foundations of social behavior (Vol. 2, pp. 93–130). New York: Guilford Press. Bargh, J. A., & Chartrand, T. L. (1999). The unbearable automaticity of being. American Psychologist, 54, 464 – 479. Borgia, G. (1986). Sexual selection in bowerbirds. Scientific American, 254, 92–100. Borgia, G. (1995). Complex male display and females choice in the spotted bowerbird: Specialized functions for different bower decorations. Animal Behavior, 49, 1291–1301. Bowden, E. M., & Beeman, M. J. (1998). Getting the right idea: Semantic activation in the right hemisphere may help solve insight problems. Psychological Science, 9, 435– 440. Bowden, E. M., & Beeman, M. J. (2003). Normative data for 144 compound remote associate problems. Behavior Research Methods, Instruments, and Computers, 35, 634 – 639. Bugental, D. B. (2000). Acquisition of the algorithms of social life: A domain-based approach. Psychological Bulletin, 126, 187–219. Buss, D. M. (2005). Evolutionary psychology: The new science of the mind (2nd ed.). Boston: Allyn & Bacon.

75

Buss, D. M., & Barnes, M. (1986). Preferences in human mate selection. Journal of Personality and Social Psychology, 50, 559 –570. Buss, D. M., & Schmitt, D. P. (1993). Sexual strategies theory: An evolutionary perspective on human mating. Psychological Review, 100, 204 –232. Byrne, R. (1995). The thinking ape: Evolutionary origins of intelligence. Oxford, United Kingdom: Oxford University Press. Chartrand, T. L., & Bargh, J. A. (1996). Automatic activation of impression formation and memorization goals: Nonconscious goal priming reproduces effects of explicit task instructions. Journal of Personality and Social Psychology, 71, 464 – 478. Chartrand, T. L., Fitzsimons, G. M., & Fitzsimons, G. J. (2004, February). The effects of priming anthropomorphized objects on behavior. Paper presented at the meeting for the Society for Personality and Social Psychology, Austin, TX. Cosmides, L., & Tooby, J. (1992). Cognitive adaptations for social exchange. In J. H. Barkow, L. Cosmides, & J. Tooby (Eds.), The adapted mind (pp. 163–228). Oxford, United Kingdom: Oxford University Press. Crespelle, J. P. (1969). Picasso and his women. New York: Hodder & Stoughton. Cronin, H. (1993). The ant and the peacock: Altruism and sexual selection from Darwin to today. New York: Cambridge University Press. Darwin, C. (1981). The descent of man, and sexual selection in relation to sex (Vols. 1 & 2). Princeton, NJ: Princeton University Press (Original work published 1871) Eysenck, H. J. (1995). Genius: The natural history of creativity. Cambridge, United Kingdom: Cambridge University Press. Gangestad, S. W., Haselton, M. G., & Buss, D. M. (in press). Evolutionary foundations of cultural variation: Evoked culture and mate preferences. Psychological Inquiry. Gangestad, S. W., & Simpson, J. A. (2000). On the evolutionary psychology of human mating: Trade-offs and strategic pluralism. Behavioral and Brain Sciences, 23, 573–587. Geary, D. C. (2000). Evolution and proximate expression of human paternal investment. Psychological Review, 126, 55–77. Geissmann, T. (2000). Duet songs of the siamang, Hylobates syndactylus: I. Structure and organisation. Primate Report, 56, 33– 60. Gibson, K. R., & Ingold, T. (Eds.). (1993). Tools, language, and cognition in human evolution. Cambridge, United Kingdom: Cambridge University Press. Griskevicius, V., Goldstein, N. J., Mortensen, C. R., Cialdini, R. B., & Kenrick, D. T. (in press). Going along versus going alone: When fundamental motives facilitate strategic (non)conformity. Journal of Personality and Social Psychology. Gutierres, S. E., Kenrick, D. T., & Partch, J. J. (1999). Beauty, dominance, and the mating game: Contrast effects in self-assessment reflect gender differences in mate selection. Personality & Social Psychology Bulletin, 25, 1126 –1134. Hamann, S., Herman, R. A., Nolan, C. L., & Wallen, K. (2004). Men and women differ in amygdala response to visual sexual stimuli. Nature Neuroscience, 7, 411– 416. Haselton, M. G., & Buss, D. M. (2000). Error management theory: A new perspective on biases in cross-sex mind reading. Journal of Personality and Social Psychology, 78, 81–91. Haselton, M. G., & Miller, G. F. (in press). Women’s fertility across the cycle increases the short-term attractiveness of creative intelligence compared to wealth. Human Nature. Higgins, E. T. (1996). Knowledge activation: Accessibility, applicability, and salience. In E. T. Higgins & A. W. Kruglanski (Eds.), Social psychology: Handbook of basic principles (pp. 133–168). New York: Guilford Press. Hrdy, S. B. (1999). Mother nature: Maternal instincts and how they shape the human species. New York: Ballantine Books. Hurtado, A. M., Hill, K., Kaplan, H., & Hurtado, I. (1992). Trade-offs

76

GRISKEVICIUS, CIALDINI, AND KENRICK

between female food acquisition and child care among Hiwi and Ache foragers. Human Nature, 3, 185–216. Kanazawa, S. (2000). Scientific discoveries as cultural displays: A further test of Miller’s courtship model. Evolution and Human Behavior, 21, 317–321. Kenrick, D. T., Li, N. P., & Butner, J. (2003). Dynamical evolutionary psychology: Individual decision-rules and emergent social norms. Psychological Review, 110, 3–28. Kenrick, D. T., Sadalla, E. K., Groth, G., & Trost, M. R. (1990). Evolution, traits, and the stages of human courtship: Qualifying the parental investment model. Journal of Personality, 58, 97–117. Kenrick, D. T., Sadalla, E. K., & Keefe, R. C. (1998). Evolutionary cognitive psychology: The missing heart of modern cognitive science. In C. Crawford & D. L. Krebs (Eds.), Handbook of evolutionary psychology (pp. 485–514). Hillsdale, NJ: Erlbaum. Kenrick, D. T., Trost, M. R., & Sundie, J. M. (2004). Sex-roles as adaptations: An evolutionary perspective on gender differences and similarities. In A. H. Eagly, A. Beall, & R. Sternberg (Eds.), Psychology of gender (pp. 65–91). New York: Guilford Press. Kerr, G., & Gagliardi, C. (2004). Measuring creativity in research and practice. Unpublished manuscript, Arizona State University, Tempe. Kingdon, J. (1993). Self-made man and his undoing. New York: Simon & Schuster. Levin, J. R., & Neumann, E. (1999). Testing for predicted patterns: When interest in the whole is greater than the sum of its parts. Psychological Methods, 4, 44 –57. Li, N. P., Bailey, J. M., Kenrick, D. T., & Linsenmeier, J. A. (2002). The necessities and luxuries of mate preferences: Testing the trade-offs. Journal of Personality and Social Psychology, 82, 947–955. MacGregor-Hastie, R. (1988). Picasso’s women. London: Lennard Publishing. Maner, J., Kenrick, D., Becker, D., Robertson, T., Hofer, B., Neuberg, S., et al. (2005). Functional projection: How fundamental social motives can bias interpersonal perception. Journal of Personality and Social Psychology, 88, 63–78. Mednick, S. A. (1962). The associative basis of the creative process. Psychological Review, 69, 220 –232. Miller, G. F. (1999). Sexual selection for cultural displays. In R. Dunbar, C. Knight, & C. Power (Eds.), The evolution of culture (pp. 71–91). Edinburgh, United Kingdom: Edinburgh University Press. Miller, G. F. (2000). The mating mind: How sexual choice shaped the evolution of human nature. New York: Doubleday. Møller, A. P., & Petrie, M. (2002). Condition dependence, multiple sexual signals, and immunocompetence in peacocks. Behavioral Ecology, 13, 248 –253. Norenzayan, A., & Heine, S. J. (2005). Psychological universals across cultures: What are they and how do we know? Psychological Bulletin, 131, 763–784. Norenzayan, A., Schaller, M., & Heine, S. J. (in press). Evolution and culture. In M. Schaller, J. Simpson, & D. Kenrick (Eds.), Evolution and social psychology. New York: Psychology Press. ¨ hman, A., & Mineka, S. (2001). Fears, phobias, and preparedness: O Toward an evolved module of fear and fear learning. Psychological Review, 108, 483–522.

Peplau, L. A. (2003). Human sexuality: How do men and women differ? Current Directions in Psychological Science, 12, 37– 40. Pinker, S. (1997). How the mind works. New York: Norton. Prose, F. (2002). The lives of the muses: Nine women and the artists they inspired. New York: Harper Collins Publishers. Raemaekers, J. J., & Raemaekers, P. M. (1984). Vocal interactions between two male gibbons, Hylobates lar. Natural History Bulletin of the Siam Society, 32, 95–106. Roney, J. R. (2003). Effects of visual exposure to the opposite sex: Cognitive aspects of mate attraction in human males. Personality and Social Psychology Bulletin, 3, 393– 404. Schaller, M. (2003). Ancestral environments and motivated social perception: Goal-like blasts from the evolutionary past. In S. J. Spencer, S. Fein., M. P. Zanna, & J. M. Olson (Eds.), Motivated social perception (pp. 215–231). Mahwah, NJ: Erlbaum. Schaller, M., Faulkner, J., Park, J. H., Neuberg, S. L., & Kenrick, D. T. (2004). Impressions of danger influence impressions of people: An evolutionary perspective on individual and collective cognition. Journal of Cultural and Evolutionary Psychology, 2, 231–247. Schooler, J. W., & Melcher, J. (1995). The ineffability of insight. In S. M. Smith, T. B. Ward, & R. A. Finke (Eds.), The creative cognition approach (pp. 249 –268). Cambridge, MA: MIT Press. Scott, J. P. (1980). The function of emotions in behavioral systems: A systems theory analysis. In R. Plutchik & H. Kellerman (Eds.), Emotion: Theory, research, and experience (Vol. 1). New York: Academic Press. Simonton, D. K. (2000). Creativity: Cognitive, personal, developmental, and social aspects. American Psychologist, 55, 1. Simpson, J. A., & Gangestad, S. W. (1991). Individual differences in sociosexuality: Evidence for converging and discriminant validity. Journal of Personality and Social Psychology, 67, 870 – 883. Simpson, J. A., & Gangestad, S. W. (1992). Sociosexuality and romantic partner choice. Journal of Personality, 60, 31–51. Srull, T. K., & Wyer, R. S. (1979). The role of category accessibility in the interpretation of information about persons: Some determinants and implications. Journal of Personality and Social Psychology, 37, 1660 – 1672. Sternberg, R. J., & Lubart, T. I. (1999). The concept of creativity: Prospects and paradigms. In R. J. Sternberg (Ed.), Handbook of creativity (pp. 3–15). Cambridge, United Kingdom: Cambridge University Press. Todd, P. M., & Gigerenzer, G. (2000). Precis of “simple heuristics that make us smart.” Behavioral & Brain Sciences, 1, 727–780. Trivers, R. L. (1972). Parental investment and sexual selection. In B. Campbell (Ed.), Sexual selection and the descent of man: 1871–1971 (pp. 136 –179). Chicago: Aldine. Wilson, M., & Daly, M. (2004). Do pretty women inspire men to discount the future? Proceedings of the Royal Society, Series B, 271(Suppl. 4), 177–179. Winwar, F. (1950). The immortal lovers: Elizabeth Barrett & Robert Browning; a Biography. New York: Harper. Zahavi, A., & Zahavi, A. (1997). The handicap principle: A missing piece of Darwin’s puzzle. New York: Oxford University Press.

Received June 30, 2005 Revision received October 24, 2005 Accepted October 31, 2005 䡲

Journal of Personality and Social Psychology 2006, Vol. 91, No. 1, 77–96

Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.77

Navigating the Interdependence Dilemma: Attachment Goals and the Use of Communal Norms With Potential Close Others Jennifer A. Bartz and John E. Lydon McGill University Four studies investigated attachment in the context of new relationship development. Anxiously attached individuals overwhelmingly used communal norms and avoided using exchange norms when interacting with a potential close other; however, when a potential close other used communal norms, anxious individuals experienced increased interpersonal anxiety. Anxious individuals also used discrete communal behaviors to diagnose relationship potential. By contrast, secure individuals were more comfortable in potential communal situations. Moreover, implicit thoughts about closeness were associated with improved performance on a mental concentration task for secure individuals, whereas implicit closeness thoughts were associated with poorer performance for anxious individuals. Finally, avoidant individuals disliked the potential close other when the other used communal norms and downplayed relational motives for the other’s communal behavior. Keywords: attachment, communal, exchange, relationship development, interdependence dilemma

the human experience, understanding the factors that are involved in the development of these bonds is an important endeavor. At the outset of a relationship, there is a great deal of uncertainty: “Is the other person equally interested in developing a relationship?” “How do I communicate interest?” “Will I be rejected?” People want to demonstrate interest and commitment but are reluctant because trust is not yet established. Ironically, as Holmes (1991) noted, although feelings of trust influence one’s level of involvement, trust cannot be assessed unless one is at least somewhat involved. Thus, one is presented with an interdependence dilemma in which the desire to express interest and commitment must be weighed against the risk of rejection. Importantly, though the ability to navigate the interdependence dilemma has consequences for relationship progress: If one is not willing to take a leap of faith, the relationship is unlikely to get off the ground. Although some find it relatively easy to deal with this uncertainty, others find it more difficult. What factors are associated with the ability to navigate the interdependence dilemma? Attachment theory has made substantial contributions to the understanding of close relationships (Bowlby, 1969, 1973; Fraley & Shaver, 2000; Hazan & Shaver, 1987, 1994; Mikulincer & Shaver, 2003; Simpson & Rholes, 1998). Although considerable research has been conducted to investigate attachment in the context of established adult relationships, less research has been conducted to probe the role of attachment in the context of new relationship development. Yet the notion that attachment models play a part in new relationships, guiding interpersonal perceptions, expectations, and behaviors with new partners, is a basic assumption of adult attachment theory (Collins & Read, 1994; Fraley & Shaver, 2000; Hazan & Shaver, 1987; Mikulincer & Shaver, 2003; Pietromonaco & Feldman Barrett, 2000). The goal of this research was to explore attachment differences in the development of closeness, specifically with regard to the interdependence dilemma people face at the outset of a close relationship.

The need to establish close bonds is a basic human characteristic (Baumeister & Leary, 1995; Bowlby, 1969, 1973; Maslow, 1962; Sullivan, 1953). The achievement of close bonds has been found to be associated with improved mental and physical well-being (House, Landis, & Umberson, 1988; Kiecolt-Glaser & Newton, 2001), whereas the failure to achieve closeness has been linked to mental (Bowlby, 1969; Davilia, Burge, & Hammen, 1997; Leary, 1990) and physical illness (Lynch, 1979) and suicide (Trout, 1980). Because of the central role that close relationships play in

Jennifer A. Bartz and John E. Lydon, Department of Psychology, McGill University, Montreal, Quebec, Canada. Portions of this research were presented in symposia presentations at the Fourth Annual Meeting of the Society for Personality and Social Psychology, Los Angeles, California, February 2003, and at the Fifth Annual Meeting of the Society for Personality and Social Psychology, Austin, Texas, January 2004. In addition, Study 3 was presented at the Seventh Annual Meeting of the Society for Personality and Social Psychology, Palm Springs, California, January 2006. This research is based on Jennifer A. Bartz’s doctoral dissertation. This research was supported in part by a Social Sciences and Humanities Research Council of Canada doctoral fellowship and the Mary Louise Taylor McGill Major fellowship to Jennifer A. Bartz and by grants from the Social Sciences and Humanities Research Council of Canada and the Fonds Pour la Formation de Chercheurs et l’Aide a la Recherche (Quebec, Canada) to John E. Lydon. We thank Mark Baldwin and Jeff Simpson for their helpful comments and suggestions, as well as Niall Bolger and Ringo Moon-Ho Ho for advising about statistical analyses. We also gratefully acknowledge Hugo Gagnon, Sarah Kerner, Dominique Pipher, Kristin Tallman, Laura Monner, Rachel Firmer, Clara Wagner, and Miriam Rozenek for their help in conducting this research. Correspondence concerning this article should be addressed to Jennifer A. Bartz, who is now at the Department of Psychiatry, Mount Sinai School of Medicine, One Gustave L. Levy Place, Box 1230, New York, NY 10029. E-mail: [email protected]

77

78

BARTZ AND LYDON

Close Relationships and the Use of Communal and Exchange Norms

ably to signal interest in a communal relationship, but people use exchange norms when the potential for friendship does not exist.

What is a close relationship? One feature distinguishing close relationships from more casual relationships is the set of norms that govern the giving and receiving of benefits. According to Clark and colleagues (Clark, 1984a, 1984b; Clark & Mills, 1979; Clark, Mills, & Powell, 1986), close relationships, such as those between family members, friends, and romantic partners, are associated with communal norms, whereas more casual relationships, such as those between business partners, acquaintances, and strangers, are associated with exchange norms. Communal norms reflect a genuine concern for the welfare of the other. Benefits are given on the basis of need or to please, and people do not keep track of individual contributions. Giving help to or doing a favor for the other person does not necessitate that the other person reciprocate. Likewise, receiving help or a favor does not require one to respond in kind. By comparison, exchange norms reflect the idea that no obligation is felt toward the needs of the other person. Benefits are given in return for benefits received or with the expectation of repayment, and people tend to keep track of individual contributions. In essence, the use of communal norms means that benefits and aid are given freely and that if there is reciprocation, it is performed to meet the needs of the other person, whereas the use of exchange norms means that benefits and aid are not given freely and that receiving a benefit or aid calls for prompt reciprocation, preferably in kind, to eradicate the outstanding debt. Clark and colleagues have found considerable support for this theory. In one study (Clark, 1984b), participants worked with a confederate on a task for a shared reward. On completion of the task, participants were responsible for dividing the reward between the two group members. The dependent variable was whether participants chose to work with the same color pen as or a different color pen from their partner. Choosing a different color pen more often than chance was thought to indicate exchange norms, as individual contributions would be clear, whereas choosing the same color pen more often than chance was thought to indicate the avoidance of exchange norms and the use of communal norms,1 as individual contributions to the task would be obscured. Supporting Clark’s theory, participants avoided using exchange norms when they believed friendship with their partner was possible—presumably to signal their interest in a relationship with the other person— but participants used exchange norms when they thought friendship was unlikely. In a second study investigating the behavior of existing friends (communal condition) and strangers (exchange condition), Clark (1984b) found that whereas strangers generally chose to work with a different color pen from their partner (i.e., used exchange norms), pen choice for existing friends was random. It was theorized that the existing friends did not go out of their way to actively avoid using exchange norms because they had an established relationship and, therefore, should not need to send a message signaling interest in friendship (Clark, 1984b). It is important to note, however, that random pen choice is consistent with the use of communal norms because it suggests participants were not trying to keep track of individual contributions. In sum, these studies suggest that people avoid using exchange norms when the potential (and desire) exists for friendship with another person, presum-

New Relationship Development and the Interdependence Dilemma Lydon, Jamieson, and Holmes (1997) conducted a series of studies focusing on the awkward position people face when they hope to establish a communal relationship with an acquaintance. These researchers found that when people desire a communal relationship with someone, consistent with Clark’s (1984b) theory, they appear to know the communal script (i.e., give freely without concern for reciprocation), and they try to behave communally, for example, by going out of their way to do a favor for or help the potential friend. However, putting themselves on the line in this way creates anxiety, and to reduce their anxiety, people look for signs of interest and commitment (“Does he or she care about me?”), especially in patterns of social exchange. As Holmes (1991) noted, patterns of social exchange are often used to diagnose signs of caring because they yield information about the other’s willingness to make personal sacrifices and to respond to one’s needs. However, Holmes (1981, 1991) warned, this microlevel perspective on social exchange patterns and on the other’s behavior can have the ironic effect of increasing feelings of vulnerability. Specifically, he argued, when people look to discrete behaviors to diagnose relationship potential, they tend to attach more meaning to those behaviors than is appropriate and, as a result, the importance of the behavior for the relationship becomes exaggerated; moreover, he stated, when discrete behaviors are analyzed in isolation, it tends to lead to perceived imbalances in what was given relative to what has been received (Holmes, 1981, 1991). Thus, at the outset of a relationship, there is a strong desire for closeness but uncertainty about the other’s motives, and this creates a great deal of anxiety. The hope is that the relationship will move forward, but for this to happen, feelings of anxiety and uncertainty need to be regulated. This leads to a vigilant monitoring of the other’s behavior in an effort to confirm the other’s interest and commitment. Ironically, though, this vigilant monitoring and microlevel perspective can perpetuate the very feelings of anxiety and uncertainty they are performed to control.

Adult Attachment Theory According to adult attachment theory, over the course of repeated interactions with significant others, individuals develop 1 The distinction between the avoidance of the use of exchange norms and the use of communal norms is important. Simply not keeping track of individual contributions (indicated by random pen choice) is thought to reflect the use of communal norms, as it suggests participants were not anticipating using task contribution information when it came time to distribute the reward. By comparison, the active avoidance of the use of exchange norms (indicated by participants’ choosing the same color pen significantly more often than chance) is thought to reflect an effort to avoid looking as though one would prefer an exchange relationship (Clark, 1984b). As Clark (1984b) stated, “when people are trying to form a communal relationship, they are not only concerned with following communal norms but also with avoiding any perception on others’ parts . . . that they might prefer an exchange relationship” (p. 553).

NAVIGATING THE INTERDEPENDENCE DILEMMA

mental models for close relationships that contain beliefs about whether the self is worthy of love and affection and whether others are trustworthy and reliable (Hazan & Shaver, 1987, 1994; Mikulincer & Shaver, 2003). Moreover, in addition to the importance of attachment models in guiding cognition, theorists have begun to stress the emotion and behavioral regulation properties of the attachment system when describing differences in attachment (Fraley & Shaver, 2000; Mikulincer & Shaver, 2003). Whereas adult attachment models have been conceptualized in different ways, current theory emphasizes a dimensional approach in which differences in attachment lie along the two dimensions of anxiety and avoidance (Brennan, Clark, & Shaver, 1998; Fraley & Waller, 1998). Attachment anxiety is associated with a heightened desire for closeness and intimacy combined with concerns about attachmentfigure (un)availability. Furthermore, attachment anxiety is theorized to be associated with the use of hyperactivation strategies that aim to secure attention from an unresponsive or inconsistently responsive attachment figure (Mikulincer & Shaver, 2003). Specifically, as Mikulincer and Shaver (2003) described, attachment anxiety is associated with hypervigilance and increased sensitivity to cues of acceptance and rejection, attachment figures are intensely monitored, and efforts are made to maintain contact with the attachment figure. Furthermore, as these researchers noted, attachment anxiety is associated with a preoccupation with selfworth, sensitivity to one’s internal distress, and emotion-focused coping (e.g., ruminating over worries and concerns). Finally, because the attachment system is chronically engaged in the pursuit of attachment-related goals, individuals high in attachment anxiety tend to have few resources left for the pursuit of such nonattachment endeavors as exploration and affiliation (Mikulincer & Shaver, 2003). Attachment avoidance, on the other hand, is associated with a strong need for independence and the tendency to minimize the importance of closeness. As Mikulincer and Shaver (2003) explained, attachment figures are seen as unreliable and unable to provide protection, and so, to deal with ensuing feelings of vulnerability, attachment needs are denied, and self-reliance is pursued. Thus, in contrast to attachment anxiety, attachment avoidance is associated with the use of deactivation strategies that function to prevent distress from the failure to attain closeness by shutting down the attachment system (Mikulincer & Shaver, 2003). In this case, monitoring of the attachment figure is avoided, and efforts are made to prevent confrontations with threatening information that could activate the attachment system and cause distress (Mikulincer & Shaver, 2003). In contrast to attachment anxiety, attachment avoidance is associated with distance coping in which efforts are made to suppress worries and concerns (Mikulincer & Shaver, 2003). Finally, attachment security (reflecting low anxiety and low avoidance) is associated with comfort with closeness and autonomy and the use of primary attachment strategies (e.g., seeking closeness and support) in times of need, rather than the secondary strategies of hyperactivation or deactivation. Basically, a history of attachment-figure availability reinforces the beliefs that people are generally well intentioned and that the world is a safe place, thus allowing more securely attached individuals to develop new rela-

79

tionships and to engage in such non-attachment-related activities as affiliation and exploration (Mikulincer & Shaver, 2003).

The Present Investigations We believe the attachment system should be critical in guiding perceptions, expectations, and behaviors with potential relationship partners, especially with respect to navigating the interdependence dilemma. Although attachment models arise over the course of repeated interactions with significant others and primarily reflect expectations about close others, we believe they nevertheless should come into play with potential partners—that is, individuals with whom one would like to be close. Collins and Read (1994) argued that chronic attachment models should be especially influential in interpersonal situations in which little is known about the other person (also see Pietromonaco & Feldman Barrett, 2000). Interpersonal situations occurring at the outset of a relationship are ambiguous, and little is known about the other person; thus, people should be especially likely to draw on their working models of attachment to guide expectations and to gauge behavior. Collins and Read (1994) also noted that attachment models should come into play when the situation is relevant to chronic attachment goals (e.g., seeking acceptance or establishing independence). To the extent that people see these situations as opportunities to satisfy chronic attachment goals, attachment models and associated beliefs, plans, and strategies should come into play (Mikulincer & Shaver, 2003).

Hypotheses Situations involving the desire and opportunity for closeness should activate the chronic goals (for closeness and acceptance) and the hyperactivation strategies associated with attachment anxiety and should magnify the approach–avoidance conflict associated with this attachment orientation. Thus, we predicted that in potential communal situations (i.e., when interacting with an attractive, friendly, available partner), individuals high in anxiety and low in avoidance ( preoccupied) would be especially likely to signal their interest in a communal relationship by avoiding the use of exchange norms. At the same time, given their uncertainty about the reliability of others and their chronic concerns about selfworth, anxious individuals should be most susceptible to interdependence dilemma concerns inherent in these situations. Interdependence dilemma concerns should also be particularly salient because, in contrast to their more secure counterparts, anxiously attached individuals’ strong desire for closeness means they have more to lose if the relationship does not materialize. Thus, we predicted that, ironically, when the potential close other used communal norms, anxious individuals would feel especially anxious and uncertain (“Is this person really communicating interest in me?”) and concerned about self-worth (“How am I coming across? Is this person attracted to me?”). Furthermore, given their uncertainty about others’ interpersonal motives, anxious individuals should require greater reassurance and, thus, should be especially likely to focus on discrete events in an effort to diagnose relationship potential. Finally, anxious individuals’ interpersonal uncertainty and anxiety should have consequences: In situations involving the possibility of closeness, anxiously attached individuals’

BARTZ AND LYDON

80

interpersonal concerns should interfere with their ability to engage in non-attachment-related activities. Potential communal situations should also arouse the chronic goals and deactivation strategies associated with attachment avoidance. Highly avoidant individuals, especially those who are low in anxiety (dismissive), are distrustful of close others, desire independence and self-reliance, and prefer to maintain a distance between themselves and close others. Thus, it was predicted that avoidant individuals would use exchange norms with a potential close other to signal their aversion to intimacy and to establish boundaries. Similarly, when a potential close other communicated interest in closeness, it was theorized that although avoidant individuals would likely be concerned about the other encroaching on their independence, they might nevertheless regulate their distress by avoiding confrontations with these unpleasant thoughts. Thus, we predicted that when a potential close other signaled interest, avoidant individuals would disparage the other so as to avoid dealing with the possibility of closeness (if the other is seen as unattractive or unlikable, one does not have to think about what might happen in a relationship with that person). Finally, given their desire to downplay the importance of closeness, avoidant individuals should be less likely to make relational attributions for a potential close other’s communal behavior. In contrast to their more insecure counterparts, secure individuals should demonstrate greater interpersonal confidence in potential communal situations because of their positive beliefs about themselves and others. In contrast to their more anxious counterparts, secure individuals should be less desperate to communicate interest in closeness. That is, they might not feel compelled to go out of their way to signal interest in a close relationship because of their trust that the relationship would develop. Moreover, their interpersonal confidence should translate into greater tolerance for uncertainty, reducing the need to microscopically analyze the other’s behavior for signs of interest. In contrast to their more avoidant counterparts, secure individuals should not feel threatened by another’s interest in closeness because they are less concerned with preserving their independence. Thus, we predicted that in contrast to their more insecure counterparts, secure individuals would feel relatively comfortable when a potential close other expressed interest in closeness.

Study 1 This study was designed to investigate the association between attachment and the use of communal and exchange norms with a potential close other. The procedure for this study was based on Clark’s (1984b) paradigm. The participant and the confederate worked on a group task for a shared reward; the dependent variable was whether the participant chose to work with the same color pen as the confederate. Choosing a different color pen was considered indicative of exchange norms because individual contributions would be clear, whereas choosing the same color pen was considered indicative of the avoidance of exchange norms because individual contributions would be obscured. It was predicted that anxious individuals would go out of their way to avoid appearing exchange oriented (i.e., they would use the same color pen as their partner) to signal interest in closeness, whereas avoidant individuals would adopt an exchange orientation (i.e., they would use a

different color pen from their partner) to signal their aversion to closeness. Finally, it was predicted that although secure individuals might not actively avoid using exchange norms, nevertheless, they would not adopt an exchange orientation (i.e., their pen choice would be random).

Method Participants Seventy university students volunteered to participate. Single participants were targeted to ensure that they would be available and interested in a communal relationship with the confederate (Clark, 1986). Three participants were excluded from the analyses because they used their own pen to work on the group task. Thus, there were 67 participants in the final sample (32 men and 35 women, M age ⫽ 19.5 years). Fifty-six participants described themselves as single, and 11 participants described themselves as dating.2 Participants received either extra credit or $10 (Canadian) for their participation.

Procedure University students were recruited to participate in a study supposedly investigating group performance and monetary incentives. In trying to create the potential for a communal relationship, we followed Clark’s (1984b, 1986) induction and had the participant interact with an attractive, opposite-sex partner (a confederate) who, the participant was subtly informed (see discussion of the group information sheet, below), was single and a recent transfer student. In addition, the participant and the confederate had a 2-min interaction in the hallway prior to entering the testing room, purportedly because the experimenter was running late. This interaction was designed to encourage a sense of connection between the participant and the confederate. During the interaction, the confederate attempted to engage the participant in a conversation first by asking whether the participant had ever participated in a psychology experiment and then continuing with casual conversation (e.g., asking, “Where are you from?”, “What is your major?”, etc.). After the hallway interaction, the participant and the confederate were brought into the lab. In accordance with Clark (1984b), the participant and the confederate were told that the study was designed to investigate the effects of monetary incentives on group performance and attitudes and that they would be working on two group tasks for which they would be rewarded. They were further told that one participant would be in charge of dividing the reward from the first task and that the other participant would be in charge of dividing the reward from the second task. It was emphasized that they were free to divide the reward however they wanted (in actuality, they never divided the reward). The first group task was then explained. This task consisted of finding a series of number sequences imbedded in a matrix. Fifty cents would be given for each number sequence found. The participant was always in charge of dividing the reward on the first group task, and the confederate was always in charge of dividing the reward on the second group task (which never occurred). The participant and the confederate were then informed that, to save time, they would be working separately so that one person could complete the personality questionnaires while the other began the first group task.

2

The 11 dating participants were included in the analyses because participants were sometimes confused about what dating meant (i.e., the question did not specify dating one person exclusively). Including the dating participants did not alter the results. Although we targeted single participants in Studies 2, 3, and 4, those who described themselves as dating were included in the analyses.

NAVIGATING THE INTERDEPENDENCE DILEMMA

81

The experimenter then escorted the confederate to a separate testing room and returned with an informed-consent form, the personality measures, and a group information sheet for the participant to complete. The group information sheet, which requested demographic information including relationship status and length of attendance at the university, was used to make salient the potential for a communal relationship with the confederate (see Clark, 1984b, 1986). Specifically, the group information sheet was given to the participant after the confederate had completed the top portion, indicating that she or he was single and a recently arrived transfer student. In this way, the participant was made aware that the confederate was available for a communal relationship. After completing the personality measures, which included the Relationship Questionnaire (RQ; Bartholomew & Horowitz, 1991), a measure of chronic attachment, and the group information sheet, the participant was given the first task, already partially completed by the confederate, to finish. Three red pens and three black pens were placed in a pen-holder on the participant’s desk. The participant thus had the choice to work with either the same color pen as or a different color pen from the confederate (the confederate alternated working with either a red or black pen). The participant was allotted 4 min to work on the task, after which he or she was informed that the study was over, probed for suspicions, debriefed, and compensated.

avoidant-dismissive, preoccupied, and avoidant-fearful, respectively (one participant did not answer the attachment categorization item). The Pearson chi-square test revealed that pen choice was associated with attachment, ␹2(3, N ⫽ 66) ⫽ 13.67, p ⬍ .005 (see Table 1). Moreover, as predicted, the proportion of preoccupied participants who chose the same color pen (93%) was significantly greater than the proportion expected by chance (50%; Z ⫽ 3.22, p ⬍ .001). With respect to the other three groups, pen choice did not differ significantly from chance. The proportion of secure participants who chose the same color pen (39%) was not significantly less than the proportion expected by chance (Z ⫽ 1.17, ns), the proportion of dismissive participants who chose the same color pen (31%) was not significantly less than the proportion expected by chance (Z ⫽ 1.37, ns), and the proportion of fearful participants (46%) who chose the same color pen was not significantly less than the proportion expected by chance (Z ⬍ 1, ns). Finally, the Pearson chi-square test unexpectedly revealed a marginal effect for sex, ␹2(1, N ⫽ 67) ⫽ 3.39, p ⫽ .066. Overall, men were slightly more likely to use communal norms (63%) than were women (40%).

Measures

Continuous Analyses

The RQ (Bartholomew & Horowitz, 1991) consists of four short paragraphs describing the secure, preoccupied, avoidant-dismissive, and avoidant-fearful attachment styles. Participants rated the extent to which they resembled each of the four styles in their close relationships (i.e., relationships with parents, siblings, close friends, relatives, or romantic partners) based on a 5-point scale ranging from 0 (not at all) to 4 (completely). Participants also selected the one attachment style that best described how they felt in their close relationships. Following Fraley and Shaver (1997), we created composites of anxiety and avoidance by subtracting participants’ dismissive ratings from their preoccupied ratings and by subtracting their secure ratings from their fearful ratings. We then standardized these composite scores.

Inspection of the anxious and avoidant composites showed that those who chose the same color pen were more anxious, M(34) ⫽ .31, than those who chose a different color pen, M(33) ⫽ ⫺.32, t(65) ⫽ 2.67, p ⫽ .01, but there was no difference between the same and different pen choice groups on the avoidant dimension (t ⬍ 1.5). Analyses of the individual attachment items similarly found that those who chose the same color pen were more preoccupied (M ⫽ .41) than those who chose a different color pen (M ⫽ ⫺.42), t(60.79) ⫽ 3.73, p ⬍ .001, whereas there were no differences between the same and different pen choice groups on attachment security, dismissive, or fearful ratings (all ts ⬍ 1.7). It is interesting to note, however, that when we investigated whether sex qualified any of these findings, results revealed a significant interaction between sex and pen choice on dismissive ratings, F(1, 63) ⫽ 7.44, p ⬍ .01. Specifically, men who chose a different color pen were significantly more dismissive, M(12) ⫽ .84, than men who chose the same color pen, M(20) ⫽ ⫺.10, t(63) ⫽ ⫺2.76, p ⬍ .05, whereas dismissive scores did not differ for women as a function of pen choice (t ⬍ 1.2).

Results To investigate whether attachment was associated with participants’ pen choice, we adopted two data-analysis strategies. First, participants were grouped according to their self-categorized attachment style, and the Pearson chi-square test was conducted to investigate whether pen choice was associated with chronic attachment. Following Clark (1984b), we also conducted difference of proportion tests to determine whether group differences in pen choice differed significantly from chance. Second, participants were grouped according to whether they chose the same color pen as or a different color pen from the confederate, and independentsamples t tests were conducted to investigate group differences on the continuous attachment scores (i.e., the anxious and avoidant composites and the individual attachment items). Finally, although we did not have specific predictions related to sex, we investigated whether sex was associated with the use of communal and exchange norms and whether sex qualified any of the findings, recognizing that sex was confounded with confederate.

Categorical Analyses In this study, 39% of participants categorized themselves as secure, and 20%, 21%, and 20% categorized themselves as

Discussion In sum, our predictions for the anxiously attached were supported: Preoccupied individuals were more likely to choose the Table 1 Pen Color Choice as a Function of Attachment (Study 1) Pen color choice Attachment style

Same

Different

Secure Preoccupied Avoidant-dismissive Avoidant-fearful Total

10 13 4 6 33

16 1 9 7 33

BARTZ AND LYDON

82

same color pen as their partner than one would expect by chance, and those who chose the same color pen as their partner were more anxious and more preoccupied. Thus, in accordance with Clark’s (1984b) theorizing, these findings suggest that when interacting with an attractive, available, opposite-sex partner, anxiously attached individuals go out of their way to avoid appearing exchange oriented, presumably to signal interest in their partner. The findings for the other three groups were less clear-cut. As predicted, pen choice for secure individuals was random; however, pen choice for fearful individuals was also random. Finally, although dismissive individuals were more likely to choose a different color pen than preoccupied individuals—suggesting their preference for an exchange orientation— contrary to our predictions, their pen choice was not significantly different from chance. Moreover, those who chose a different color pen were not more dismissive. That said, additional analyses revealed that men who chose a different color pen were more dismissive. However, very few women (n ⫽ 2) categorized themselves as dismissive, making it difficult to investigate this effect for women. On a related note, unexpectedly, women overall were slightly less likely than their male counterparts to use the same color pen as their partner; this may also have made it difficult to detect an association between women’s avoidant attachment and the use of exchange norms (i.e., if women were less likely to use communal norms, dismissive women would have had to have been that much less communal). Why were women slightly less likely to use communal norms? If anything, one would expect women to be more communal. One possible factor may have been the confederate. Male and female participants interacted with different confederates; although we tried to select comparable confederates, slight differences (e.g., one confederate may have been liked more than the other) could have influenced participants’ desire for closeness and their use of communal norms.

Study 2 In Study 1, when interacting with an available, desirable partner, anxious individuals, compared with their more secure and avoidant counterparts, went out of their way to express interest in closeness, but how do anxious individuals respond when an available, desirable partner expresses interest in them? The goal of this study was to investigate the relationship between attachment and participants’ response to the use of communal or exchange norms by a potential close other. In this study, we created a new psychological situation by modifying the procedures from Study 1. Specifically, the confederate worked on the group task after the participant, and the completed group task was returned to the participant, who was in charge of tabulating the results. In this way, the participant was able to see whether the confederate used the same or different color pen (experimental manipulation of communal vs. exchange norms). State feelings of interpersonal anxiety, partner liking, and partner perceptions were assessed. It was theorized that whereas secure individuals would feel comfortable with a potential close other’s use of communal norms, insecure (anxious and avoidant) individuals would feel more distressed. Anxious individuals have a strong desire for closeness but are uncertain about others’ interest and tend to doubt their selfworth. As we have argued, anxious individuals should be espe-

cially susceptible to interdependence dilemma concerns in situations involving possible closeness because of their chronic interpersonal insecurities. Thus, it was predicted that, compared with the exchange condition, anxious individuals in the communal condition would experience more interpersonal anxiety because of the potential for closeness in this condition. Avoidant individuals, on the other hand, strive for independence and do not like it when others try to get too close. Thus, it was predicted that, compared with the exchange condition, avoidant individuals in the communal condition would respond to their partner’s overtures with decreased liking and would try to cast their partner’s communal behavior in a negative light. Finally, although situations involving possible closeness should be distressing to avoidant individuals given their preference for independence, avoidant individuals tend to suppress worries and concerns. Thus, we did not have specific predictions regarding avoidant individuals’ affective response in this study.

Method Participants Sixty-three university students volunteered to participate. Single participants were targeted; 2 participants were excluded because they were in a serious relationship, and 2 participants were excluded because they suspected that their partner was a confederate. There were 59 participants in the final sample (26 men and 33 women, 50 single and 8 dating [1 participant did not answer this question], M age ⫽ 21.05 years). Participants were randomly assigned to the communal (n ⫽ 28) or exchange conditions (n ⫽ 31). Participants were given either extra credit or $10 (Canadian) for their participation.

Procedure University students were recruited to participate in a study supposedly investigating personality, group performance, and incentives. As in Study 1, the participant had a 2-min interaction with an attractive, opposite-sex confederate on arrival at the testing session. After the interaction, the participant and the confederate were brought into the lab and were informed that they would be working on two group tasks for which they would be rewarded; when the first group task was completed, one person would be given the reward to divide between the members of the group, and when the second group task was completed, the other person would be given the opportunity to divide the reward. Different from Study 1, it was explained that even though participants would be working together on both tasks, because the study was interested in the effects of two different working conditions, they would be working in two separate rooms during the first task, so they would be unable to communicate, and they would be working in the same room during the second task. The group task, which was similar to the number matrix task used in Study 1, was then explained. Participants were told that they would be given 25¢ for each number sequence found. Again, the participant was responsible for dividing the reward on the first group task (the second group task never took place). The participant was then escorted to a room next door, was given an informed-consent form to complete, and then began the first group task. The experimenter left and returned after 6 minutes with the group information sheet (similar to that used in Study 1), partially completed by the confederate, and the personality measures, which again included a measure of chronic attachment. The experimenter then brought the group task to the confederate, who used either the same color pen as (communal condition) or a different color pen from (exchange condition) the participant. (Although participants were not specifically told that their partner had an

NAVIGATING THE INTERDEPENDENCE DILEMMA option with regard to pen choice—primarily because it was difficult to convey this information without drawing attention to the importance of pens in the study— care was taken to display the pens on the participant’s desk in an artificial manner so that the participant would assume that the other’s desk space was similarly laid out.) The completed group task was then returned to the participant, who was instructed to tabulate how many number sequences the team had found. In this way, the participant was able to see whether the confederate used a pen of the same color or one of a different color. At this point, the participant was told that before beginning the second group task in which he or she would be working with the other participant in the same room, it was necessary to complete a social interaction prestudy questionnaire to control for individual differences in expectations about working together (Clark & Mills, 1979). The questionnaire assessed situational feelings of interpersonal anxiety (i.e., state affect and self-esteem), partner perceptions, and partner liking. After completing these measures, the participant was probed for suspicions, debriefed, and compensated.

Measures Experience in Close Relationships scale (ECR; Brennan et al., 1998). This 36-item questionnaire is designed to assess the two dimensions of attachment avoidance and anxiety. Avoidant items reflect comfort with closeness and dependency, and anxious items reflect anxiety about being abandoned. Participants indicate on a 7-point scale how much they agree or disagree with each item, in terms of how they experience romantic relationships. We added the ECR to this study because of the limitations associated with the RQ (i.e., assessment of discrete dimensions and categories). After reverse scoring items for which lower numbers reflected greater attachment avoidance or anxiety, we computed attachment avoidance and anxiety scores for each participant by taking the mean response on the 18 avoidance (␣ ⫽ .90) and 18 anxiety (␣ ⫽ .90) items. Avoidance and anxiety scores were not correlated, r(57) ⫽ .19, ns. Situational feelings of interpersonal anxiety. To assess situational feelings of interpersonal anxiety, we created a composite that included state affect and aspects of state self-esteem. Specifically, the composite included the Anxious and Uncertain subscales from the Profile of Mood States (POMS; McNair, Lorr, & Droppleman, 1971). (These items were interspersed among items from the Depressed, Hostile, Happy, and Confident subscales of the POMS.) The composite also included the Appearance subscale from the State Self-Esteem Scale (SSES; Heatherton & Polivy, 1991). Examples from this subscale include “I feel that others respect and admire me,” “I feel good about myself,” “I am pleased with my appearance right now,” and “I feel unattractive” (this last items was reverse scored).3 The Cronbach’s alpha coefficient for the Anxious, Uncertain, and Appearance Self-Esteem subscales (reverse scored) was .73. To calculate the index, we summed participants’ ratings from the Anxious and Uncertain subscales and subtracted their Appearance Self-Esteem rating (standardized scores were used in the calculation to adjust for scale differences in these measures). Partner perceptions and partner liking. Partner perceptions and partner liking were assessed in the Social Interaction Prestudy Questionnaire. In this questionnaire, participants rated how well certain traits applied to their partner (the confederate). Because we were particularly interested in how avoidant participants would construe their partner’s communal behavior, we selected traits from the Extended Personality Attributes Questionnaire (Spence, Helmreich, & Holahan, 1979), which taps positive communion (warm, helpful, and understanding), negative communion (spineless, fussy, and servile), positive agency (independent, competitive, and self-confident), and negative agency (arrogant, greedy, and hostile). Participants indicated their response by placing a slash through a line, anchored from 0 (not at all) to 50 (completely). Cronbach’s alpha coefficients for communal-positive, communal-negative, agency-positive, and agency-negative composites were .78, .58, .36, and .66, respectively.

83

Participants also indicated the extent to which (a) they liked the person with whom they were working, (b) they would like to continue a conversation with that person on another occasion, (c) their partner was the kind of person they would want to have as a friend, and (d) their partner was the kind of person they would want to work with on another project by placing a slash through a line, anchored from 0 (not at all) to 50 (completely). We created a partner-liking score by taking participants’ mean response to the four partner-liking items (␣ ⫽ .87).

Results This study was designed to investigate attachment differences in participants’ response to the use of communal or exchange norms by a potential close other. The dependent variables were situational interpersonal anxiety, partner liking, and partner perceptions. It was predicted that attachment anxiety would be more strongly associated with interpersonal anxiety in the communal condition compared with the exchange condition, whereas attachment avoidance would be more strongly associated with partner liking (negative correlation) and negative partner perceptions in the communal condition compared with the exchange condition. To investigate these predictions, we used the Fisher z-transformation test (Cohen, Cohen, West, & Aiken, 2003) to compare the withincell correlations between attachment and interpersonal anxiety, partner perceptions, and partner liking between experimental conditions.

Attachment Anxiety and Interpersonal Anxiety Overall, attachment anxiety was associated with situational feelings of interpersonal anxiety across experimental conditions, r(57) ⫽ .58, p ⬍ .001. However, as predicted, the association between attachment anxiety and interpersonal anxiety was significantly greater in the communal condition, r(25) ⫽ .80, than in the exchange condition, r(28) ⫽ .37, Z ⫽ 2.59, p ⫽ .01.4 Moreover, although avoidant attachment was also associated with interpersonal anxiety, r(57) ⫽ .37, p ⬍ .005, this association did not differ between experimental conditions, Z ⫽ 1.12, ns. It is important to note that the difference in the association between attachment anxiety and interpersonal anxiety in the communal and exchange 3 Although the SSES taps performance, social competence, and appearance, it is thought that different components of state self-esteem should be affected by different experimental situations (Heatherton & Polivy, 1991). We focused on the appearance component because it was theorized that appearance concerns should be especially salient in the communal condition. That is, in both the communal and exchange conditions, participants expected to work on a task with their partner, and thus, performance and social interaction concerns should be salient. However, appearance concerns (e.g., “I feel unattractive”) should be more salient in the communal condition than in the exchange condition given the potential for closeness and increased likelihood of being evaluated by another person on the basis of one’s appearance in the communal condition. 4 This effect was moderated somewhat by sex ( p ⬍ .10). Anxiously attached women experienced more interpersonal anxiety than their secure counterparts across conditions, r(31) ⫽ .53, p ⫽ .001, whereas for anxiously attached men, this relationship was weak in the exchange condition, r(11) ⫽ .35, ns, but very strong in the communal condition, r(11) ⫽ .92, p ⬍ .001.

BARTZ AND LYDON

84

conditions remained when controlling for attachment avoidance. That is, a comparison of the partial correlations between attachment anxiety and interpersonal anxiety (controlling for attachment avoidance) in the communal, r(24) ⫽ .77, and exchange conditions, r(27) ⫽ .39, remained significant (Z ⫽ 2.17, p ⬍ .05).

Attachment Avoidance and Partner Liking and Partner Perceptions Partner liking. Overall, avoidant attachment was negatively associated with partner liking, r(57) ⫽ ⫺.28, p ⬍ .05; however, as predicted, this association was stronger in the communal condition, r(25) ⫽ ⫺.54, than in the exchange condition, r(28) ⫽ ⫺.02, Z ⫽ 2.12, p ⬍ .05. By contrast, attachment anxiety was not associated with partner liking, r(57) ⫽ ⫺.03, ns. Moreover, attachment anxiety did not qualify the differential association between avoidant attachment and partner liking in the communal and exchange conditions. That is, a comparison of the partial correlations between attachment avoidance and partner liking (controlling for attachment anxiety) in the communal, r(24) ⫽ ⫺.56, and exchange conditions, r(27) ⫽ ⫺.02, remained significant (Z ⫽ 2.18, p ⬍ .05). Finally, although this effect was not qualified by sex, men in general liked the female confederate more than women liked the male confederate, t(57) ⫽ 2.28, p ⬍ .05. (This finding supports our speculation in Study 1 that any sex differences may have been due to differences in male and female participants’ liking of the confederates rather then sex per se.) Partner perceptions. Overall, avoidant attachment was associated with ascribing more negative communal traits (e.g., servile) to the partner, r(56) ⫽ .31, p ⬍ .05 (one participant did not respond to this question); however, as predicted, the association between avoidant attachment and perceptions of the partner’s negative communal traits was stronger in the communal condition, r(25) ⫽ .56, than in the exchange condition, r(29) ⫽ .09, Z ⫽ 1.95, p ⫽ .051. Unexpectedly, attachment anxiety was also associated with negative communal partner perceptions, r(56) ⫽ .28, p ⬍ .05. Although this association did not differ between experimental conditions (Z ⫽ .70, ns), controlling for attachment anxiety qualified the aforementioned association between attachment avoidance and perceptions of the partner’s negative communal traits (Z ⫽ 1.58, ns). No sex differences were found in the association between attachment avoidance and perceptions of the partner’s negative communal traits. Finally, with respect to agency traits, avoidant attachment was associated with seeing the partner as more negatively agentic (e.g., greedy), r(57) ⫽ .28, p ⬍ .05, and less positively agentic (e.g., less independent), r(57) ⫽ ⫺.31, p ⬍ .05, but these associations did not differ between experimental conditions.

Discussion As predicted, attachment anxiety was strongly associated with interpersonal anxiety when a potential close other used communal norms. Why were anxiously attached individuals not more distressed when their partner used exchange norms, as this might suggest their partner’s disinterest in a communal relationship? We believe it is unlikely that the anxious participants interpreted their partner’s exchange behavior as a rejection. The norm when inter-

acting with a stranger is exchange, so anxious individuals may have thought their partner’s behavior in the exchange conditions (i.e., choosing to work with a different color pen) was consistent with the situation and thus did not take it as a personal slight. It was when the confederate deviated from the norm, expressing interest in closeness, that they felt especially anxious, uncertain, and concerned about their appearance. Thus, it appears that when opportunities arise for anxiously attached individuals to pursue their goals for closeness and acceptance, they experience more interpersonal anxiety. Avoidant individuals had a very different response to the potential close other’s use of communal norms. Attachment avoidance was associated with decreased partner liking in the communal condition, whereas there was no association between attachment avoidance and partner liking in the exchange condition. Thus, it was not simply that avoidant individuals disliked their partner; rather, they disliked their partner when their partner attempted closeness. Avoidant individuals value independence and selfsufficiency and prefer to keep a distance between themselves and close others; they also tend to avoid confrontation with worries or concerns. We believe their partner’s communal overtures likely threatened avoidant individuals’ need for self-reliance, and they responded by disparaging their partner. Rather than response to possible closeness, an alternate explanation of the findings from this study is that the confederate’s obscuring of contributions in the communal condition made it difficult for participants to divide the reward because each person’s contribution to the task was ambiguous. We believe it is unlikely that this would explain anxious individuals’ heightened interpersonal anxiety. Referring back to Study 1, the large majority of anxious individuals used the same color pen as their partner, and given that they were also responsible for dividing the reward in Study 1, they would have been presented with the same quandary. Avoidant individuals, on the other hand, may indeed have been perturbed when their partner used the same color pen because it made it difficult for them to divide the reward. However, given that distributing benefits on the basis of task contribution is an exchange norm, this would seem to support the notion that avoidant individuals prefer to operate from an exchange rather than a communal perspective. These findings also shed light on Study 1. In Study 1, anxiously attached individuals chose to work with the same color pen as their partner more often than chance. In line with Clark’s (1984b) theorizing, we interpret this behavior as an effort to signal interest in closeness by avoiding the use of exchange norms. An alternate explanation, however, is that anxiously attached individuals obscured contributions because they were concerned they would be evaluated negatively if they performed poorly. Study 2 suggests that this was not the case: Anxiously attached individuals had less interpersonal anxiety when their performance was distinguishable from their partner’s performance (i.e., in the exchange condition). In sum, Studies 1 and 2 suggest that situations involving the opportunity for closeness elicit an approach–avoidance conflict in anxiously attached individuals. In Study 1, anxiously attached individuals were more likely that their secure and avoidant counterparts to symbolically approach their partner, whereas in Study 2, attachment anxiety was associated with greater feelings of interpersonal anxiety following a potential close other’s communal

NAVIGATING THE INTERDEPENDENCE DILEMMA

overtures. We believe the confederate’s communal behavior likely aroused anxious individuals’ interdependence dilemma concerns (“Is this person interested in me?”) and their deep-seated ambivalence about close others. On the one hand, given their desire for closeness, anxiously attached individuals were probably highly motivated to detect signs of interest in the other’s behavior, but on the other hand, given their chronic uncertainty about others, their hopes may have been tempered by concerns that they would be found unworthy (Holmes, 1991). Avoidant individuals, by contrast, disparaged their partner when their partner used communal norms presumably because their partner’s communal overtures threatened their desire for independence. Studies 3 and 4 were designed to explore these ideas.

Study 3 A core prediction of the interdependence dilemma is that in potential communal situations, there is the desire for closeness but uncertainty about the other’s motives and this uncertainty needs to be resolved for the relationship to move forward. Consequently, as Holmes (1991) noted, people seek confirmation about the other’s motives, often using patterns of social exchange as a testing ground—that is, they look for evidence that the other genuinely cares as indicated by, for example, the other’s use of communal norms (“Is the other responsive to my needs?”). As we have argued, because of their strong desire for closeness and their chronic uncertainty, anxiously attached individuals should be most susceptible to interdependence dilemma concerns. Indeed, Study 2 found attachment anxiety to be more strongly associated with feelings of interpersonal anxiety in situations involving the possibility of closeness. If anxious individuals are more susceptible to interdependence dilemma concerns, then they should be especially likely to engage in this process of uncertainty reduction, evaluating even relatively trivial behaviors for their diagnostic information about relationship potential. Indeed, the notion that anxiously attached individuals are especially alert to environmental cues conveying information about acceptance and rejection is also a central prediction of attachment theory (Fraley & Shaver, 2000). The goal of Study 3 was to investigate the idea that potential communal situations arouse anxiety and uncertainty in anxiously attached individuals, which then sets in motion a need to microscopically analyze and diagnose the other’s behavior— even trivial behaviors—for signs of interest and commitment. Moreover, we wanted to probe whether anxious individuals’ hypervigilance and need to reduce uncertainty were associated with assigning greater relational significance to more trivial situations that are less clearly communal or whether this process would be more apparent in situations clearly suggestive of communal potential. Although everyone should respond with interdependence dilemma concerns when the behavior of the other is clearly suggestive of communal potential, we believe that even trivial behaviors should elicit interdependence dilemma concerns for the anxiously attached. In this study, we departed from the laboratory methodology and instead used a guided-visualization, self-report methodology to assess preexisting potential communal relationships from participants’ own social world. Moreover, although we sought to create the potential for a romantic relationship in Studies 1 and 2, some participants may have thought of the confederate as a potential

85

friend and not as a romantic prospect. Thus, in this study, participants were instructed to nominate someone with whom they hoped to establish a deeper relationship (a potential close other); the nature of that relationship (friendship vs. romantic) was left open, which allowed us to compare romantic versus nonromantic potentially communal relationships. Participants then visualized themselves in one of two scenarios. In one scenario, participants needed to borrow notes for a class they have missed, and the potential close other offered to loan them the notes; in the second scenario, the potential close other suggested being study partners and exchanging phone numbers. The offer to lend someone notes is a relatively trivial, mundane event that can be attributed to a variety of nonrelationship factors, whereas the suggestion to exchange phone numbers is less common and the motives for the behavior, we believe, are more suggestive of relationship potential. Affect, significance of the event for the future of the relationship, and relational attributions for the other’s communal behavior were assessed. It was hypothesized that overall, participants would believe the event had greater significance for the future of the relationship and would make more relational attributions when the potential close other suggested being study partners than when the potential close other offered to lend them notes. Similarly, with respect to affect, it was hypothesized that the study partner scenario would elicit less indifference and greater unease compared with the notes scenario because the prospect of closeness would arouse interdependence dilemma concerns. Attachment, however, was hypothesized to moderate these effects. Secure individuals (low anxious and low avoidant) were predicted to be the baseline group—that is, they should infer greater significance and make more relational attributions in the less ambiguous study partner scenario than in the more trivial notes scenario. They should also experience less indifference and greater unease in the study partner scenario, in line with interdependence dilemma predictions. Anxious individuals, by comparison, should infer greater significance, make more relational attributions for the potential close other’s behavior, and feel more uneasy regardless of scenario; that is, even the discrete communal behavior scenario should set in motion interdependence dilemma concerns. Finally, avoidant individuals should be less likely to infer significance and should make fewer relational attributions for the potential close other’s communal behavior because of their tendency to downplay the importance of closeness; for the same reason, they should also be more indifferent to the potential close other’s communal behavior. Lastly, it was predicted that avoidant individuals would be less happy in the notes scenario, in which they must rely on the other for help, because of their aversion to dependency.

Method Participants Two hundred seventy-nine participants were recruited on a volunteer basis to complete one of two relationship surveys. Single participants were targeted to ensure participants were adequately interested in pursuing a relationship with their chosen person. We excluded 2 participants who were unable to nominate a potential close other, 23 participants who were insufficiently interested in developing a relationship with the potential

BARTZ AND LYDON

86

close other (i.e., their interest ratings were below the midpoint; these participants did not differ from the other participants in their attachment avoidance or anxiety scores), 1 participant who did not complete the attachment questionnaire, and 1 participant who was married, had known the potential close other for 30 years, and was more than 10 standard deviations above the mean participant age. Thus, 252 participants completed one of two relationship surveys (126 men and 125 women, 219 single and 32 dating, M age ⫽ 20.94 years). (There were missing data on some of the questions because of nonresponse; however, there were never more than 5 [2%] missing data points for the relational indices, which were the main dependent variables).

Procedure Prospective participants were asked if they would be interested in completing a survey about new relationship development. Participants were told that it was important to complete the survey in one sitting, alone, and in a quiet place. Agreeing participants completed an informed-consent form, were given the survey and an envelope, and were instructed to return the survey in the sealed envelope to ensure anonymity. Surveyors were unaware of the hypotheses when administering the survey, and participants were debriefed on returning the survey. The survey began with a brief introduction, which stated that the goal of the research was to learn more about the development of new relationships. Participants were informed that the survey involved nominating someone with whom they were not currently friends but with whom they could imagine being close friends, visualizing themselves and this person in a scenario, and answering some questions about the scenario as well as some demographic questions and some personality measures. Participants were encouraged to fully immerse themselves in the scenario (e.g., by visualizing themselves and their chosen person in the situation, imagining the surroundings, etc.). Moreover, to help participants mentally simulate the experience, blank spaces were inserted throughout the scenario, and participants were instructed to write their chosen person’s first name in the spaces provided. After the introduction, participants were asked to think of a same-sex or opposite-sex acquaintance. Specifically, they were instructed to think of Someone with whom you are not currently friends, but someone with whom you could imagine being close friends if you got to know each other better. This person could be a casual friend with whom you would like to establish a deeper friendship, or, possibly, a romantic relationship. . . . Although you are not close friends right now, you think you might really enjoy spending time together in the future. It remains to be seen how your relationship will develop. Because of the nature of the scenarios (borrowing class notes and exchanging phone numbers to be study partners), participants were directed to select a peer, that is, someone relatively close in age, so that the scenario would be appropriate and realistic. After selecting their chosen person but before going on to read the social interaction scenario, participants were instructed to visualize their chosen person. To aid in the visualization process, participants were presented with a series of questions about their chosen person (e.g., “What is it like being with this person?” and “How do you feel when you are with this person?”). The purpose of the guided visualization was to help participants immerse themselves so as to discourage top-of-the-head responses (Lydon et al., 1997). After the visualization, participants read one of the following two scenarios. Notes (relatively trivial communal behavior). “Imagine that you and __________ are taking a class together. You have missed a few classes

since the last midterm and __________ has the notes for the classes you missed.” After rating how comfortable they would feel in that situation, participants were asked to imagine that their chosen person had offered to lend them the notes. Study partner (less ambiguous gesture of friendship). “Imagine that you and __________ are taking a class together. One day after class __________ gives you his or her phone number and suggests being study partners for the upcoming exam.” After reading the scenario, participants answered questions about their affective response, the significance of the event for the future of the relationship, and attributions for the potential close other’s behavior (see Measures, below). At the end of the questionnaire, participants indicated (a) whether they were able to think of someone with whom they desired a deeper friendship, (b) how long they had known the person, (c) how interested they were in developing a relationship with the person, (d) what type (romantic or friendship) of relationship they desired, and (e) how realistic the scenario was to them. We also assessed chronic attachment (the measure was presented as a relationship style questionnaire) and demographic information at this point.

Measures Significance of the event for the future of the relationship. Participants rated, on a scale from 1 (not at all) to 7 (a lot), the extent to which they thought the event would say something about the progress of the relationship and how much closer they thought they would feel to their chosen person after the event compared with prior to the event. Using the same scale, participants also rated the extent to which they agreed with the following statements: “We will probably get closer,” “This event cements our relationship,” and “This event has no particular implications or significance for the relationship” (this item was reverse scored). An index of the significance of the event for the development of the relationship was created by averaging participants’ responses to these five items (␣ ⫽ .74). Relational attributions for the potential close other’s behavior. Participants rated, on a scale from 1 (disagree strongly) to 7 (agree strongly), the extent to which they agreed with seven statements reflecting different attributions for their chosen person’s behavior. A relational attribution composite was created by taking the mean of participants’ relational attributions (i.e., “because he or she likes me [i.e., this event occurred because of the way this person feels about me],” “because he or she cares about me,” “to become closer,” and “as a gesture of friendship”) and subtracting their situational attribution (i.e., “I don’t think this event had much to do with who the other person is, or the way he or she feels about me; outside circumstances were probably the primary cause of this event”).5 Affect. Participants rated how indifferent, uneasy, and happy they would feel in the situation described in the scenario, using a scale from 1 (not at all) to 7 (extremely). ECR (Brennan et al., 1998). The 36-item questionnaire used in Study 2 was used to assess attachment avoidance (␣ ⫽ .91) and anxiety (␣ ⫽ .91).

5

Participants also responded to a dispositional attribution statement (“because he or she is a nice person”) and a self-interested attribution statement (“out of self-interest [i.e., to improve his or her standing in the class]”). However, these items were not included in the relational attribution composite because we believe they likely tapped participants’ chronic beliefs about others and not simply beliefs about whether the behavior had meaning for relationship development. Indeed, there was no correlation between making relational attributions and making dispositional or self-interested attributions, r(245) ⫽ ⫺.01, ns, and r(245) ⫽ .04, ns, respectively.

NAVIGATING THE INTERDEPENDENCE DILEMMA

Results A series of hierarchical multiple regressions were conducted to investigate the influence of attachment anxiety and avoidance on participants’ affective response to the potential close other’s communal behavior, their beliefs about the significance of that behavior for relationship development, and the kind of attributions they made about the potential close other’s communal behavior. Attachment anxiety and avoidance were standardized and entered in the first step of the regression along with scenario (notes vs. study partner; contrast coded as 1 and ⫺1, respectively); the two-way interactions of attachment anxiety and avoidance, scenario and attachment anxiety, and scenario and attachment avoidance were entered in the second step; and the three-way interaction of scenario, attachment anxiety, and attachment avoidance was entered in the third step of the regression. We also investigated whether the type of relationship participants thought about (friendship or romantic) or participants’ sex influenced any of the dependent variables and whether relationship type or participants’ sex qualified any of the effects of attachment and/or scenario. There were no significant main or interaction effects of relationship type on any of the dependent variables. Similarly, with one exception, there were no main or interaction effects of sex on any of the dependent variables. Specifically, attachment avoidance and attachment anxiety were negatively correlated with feeling uneasy for male participants, whereas women’s uneasiness ratings were not correlated with avoidance or anxiety. Basically, highly avoidant or anxious men were less likely to report feeling uneasy compared with their less avoidant and less anxious counterparts.

Relationship Characteristics One hundred two participants thought of a potential friend, and 149 participants thought of a potential romantic partner (if participants selected both, they were categorized as seeking a romantic relationship because it was assumed that if participants indicated any interest in a romantic relationship, then that was their true preference). The mean interest in developing a relationship was 5.27 (range ⫽ 4.00 –7.00). The mean relationship length was slightly less than a year and a half (497 days), and the median was approximately 6 and a half months (196 days; range ⫽ 1 day–11 years). The mean scenario realism score was 4.82 (range ⫽ 1.00 – 7.00). Attachment was not associated with the type of relationship participants were interested in developing (both ts ⬍ 1), nor was attachment associated with interest in developing a relationship, perceived scenario realism, or relationship length (all rs ⬍ .11). Relationship type, however, was associated with interest in developing a relationship, t(230.167) ⫽ 6.37, p ⬍ .001. Those seeking a romantic relationship were significantly more interested in developing a relationship with their chosen person (M ⫽ 5.56) than were those seeking a friendship (M ⫽ 4.83).

87

pants inferred greater significance from the suggestion to be study partners than from the notes offer, providing construct validation for the two scenarios—that is, as intended, participants viewed the study partner scenario, which involved a less ambiguous instance of the potential close other’s communal behavior, as having a greater impact on relationship development than the notes scenario. As predicted, however, analyses also revealed a significant effect for attachment anxiety (B ⫽ .17, sr ⫽ .17, p ⬍ .005). Anxiously attached individuals inferred greater significance from the potential close other’s communal behavior regardless of scenario. In addition, there was a marginal interaction between scenario and attachment anxiety (B ⫽ .10, sr ⫽ .09, p ⫽ .095). Simple effects tests found that attachment anxiety was positively associated with beliefs about the significance of the event in the notes scenario, r(127) ⫽ .27, p ⬍ .005, but was not associated with significance beliefs in the study partner scenario, r(121) ⫽ .11, ns, primarily because everyone had more optimistic beliefs for the development of the relationship in the study partner scenario. Results also revealed an unexpected three-way interaction between scenario, attachment anxiety, and attachment avoidance (B ⫽ .12, sr ⫽ .11, p ⬍ .05). This interaction is explained by the fact that for the notes scenario, high avoidant, low anxious (dismissive) participants were considerably less likely than their peers to believe that the event had significance for relationship development.

Relational Attributions for the Potential Close Other’s Behavior Analyses investigating the relational attribution composite also revealed an effect for scenario (B ⫽ ⫺.29, sr ⫽ ⫺.29, p ⬍ .001), again providing construct validation and suggesting that participants perceived their potential close other’s suggestion to be study partners as more relationally motivated than the offer to lend them notes. There was also an interaction between scenario and attachment anxiety (B ⫽ .16, sr ⫽ .16, p ⬍ .01; see Figure 1). Simple effects tests revealed that attachment anxiety was positively associated with relational attributions in the notes scenario, r(125) ⫽ .24, p ⬍ .01, but was not associated with relational attributions in

Significance of Event for Relationship Development Analyses investigating the composite measure of beliefs about the significance of the event for relationship development revealed an effect for scenario (B ⫽ ⫺.43, sr ⫽ ⫺.42, p ⬍ .001). Partici-

Figure 1. Relationship attributions (for potential close other’s behavior) as a function of scenario and attachment anxiety (Study 3).

BARTZ AND LYDON

88

the study partner scenario, r(118) ⫽ ⫺.10, ns, again, primarily because everyone made more relational attributions in the study partner scenario. Thus, similar to the significance of the event for relationship development analyses, anxiously attached individuals made more relational attributions regardless of the quality of the other’s communal behavior. Finally, as predicted, analyses also revealed a negative association between attachment avoidance and making relational attributions (B ⫽ ⫺.22, sr ⫽ ⫺.22, p ⬍ .001), suggesting avoidant individuals’ reluctance to make relational attributions for the other’s communal behavior.

Affective Response to Potential Close Other’s Communal Behavior Overall, participants felt less indifferent (B ⫽ .20, sr ⫽ .20, p ⬍ .005) and more uneasy (B ⫽ ⫺.15, sr ⫽ ⫺.15, p ⬍ .05) in the study partner scenario than in the notes scenario. Again, this finding suggests that the study partner scenario had more emotional relevance for participants than did the notes scenario. The fact that participants were more uneasy in the study partner scenario also reinforces the notion that interdependence dilemma situations are associated with increased anxiety. As predicted, attachment anxiety was associated with feeling uneasy across scenarios (B ⫽ .13, sr ⫽ .12, p ⬍ .05). Finally, avoidant attachment was associated with feeling more indifferent across scenarios (B ⫽ .13, sr ⫽ .13, p ⬍ .05) and, interestingly, was also associated with feeling more uneasy across scenarios (B ⫽ .22, sr ⫽ .22, p ⫽ .001). Moreover, results revealed a significant Scenario ⫻ Avoidance interaction on happiness ratings (B ⫽ ⫺.14, sr ⫽ ⫺.14, p ⬍ .05). As predicted, avoidant attachment was negatively associated with happiness ratings in the notes scenario, r(125) ⫽ ⫺.21, p ⬍ .05, in which participants received help from the other, but was not associated with happiness ratings in the study partner scenario, r(121) ⫽ .04, ns.

Discussion As predicted, overall, participants felt the suggestion to be study partners and exchange phone numbers was more significant for relationship development than the potential close other’s offer to lend class notes; however, attachment moderated these effects. Anxious individuals, to a greater extent than their more secure counterparts, believed the event had significance for relationship development and were more likely to make relational attributions for the potential close other’s communal behavior regardless of scenario. Even in the notes scenario, which involved a relatively mundane, trivial communal behavior, anxious individuals were more likely to detect evidence of caring and to use that behavior to diagnose relationship potential. Finally, avoidant individuals were less likely to make relationship attributions for the potential close other’s behavior, reflecting their desire to minimize the importance of closeness and their more pessimistic expectations about others. Interestingly enough, when we decomposed the relational attribution composite and investigated relational and situational attributions separately, although the pattern of results was the same, the Scenario ⫻ Attachment Anxiety interaction only attained statistical significance for the situational attribution. This suggests that anxious individuals were somewhat more willing than others

to make positive relational attributions for the other’s offer to lend them notes but that they were especially unwilling to discount the behavior as purely situational. That is, they appear to have been clinging to the notion that there must be some meaning in the gesture but at the same time were hesitant to conclude that the other was definitely interested in a relationship. These findings are consistent with the ambivalence that characterizes attachment anxiety and point to a possible source for the uncertainty anxious individuals evidenced in Study 2 in response to possible closeness. These findings are also consistent with Lydon et al.’s (1997) research, which found that when participants were faced with an interdependence dilemma, they were especially likely to believe that communal gestures had some implication for relationship development. With respect to affect, overall, participants felt less indifferent in the study partner scenario suggesting that they perceived the offer to be study partners as having greater significance for the relationship. It is interesting to note that participants also felt more uneasy when the potential close other suggested being study partners. This finding is in agreement with Lydon et al. (1997), and we believe it is indicative of the feelings of uncertainty and anxiety people experience in association with the interdependence dilemma. Avoidant individuals, characteristically, reported feeling more indifferent to the potential close other’s communal behavior across scenarios; they also felt more uneasy about the potential close other’s communal behavior across scenarios. This indifference is consistent with avoidant individuals’ tendency to downplay the importance of closeness, but their unease suggests that avoidant individuals are nonetheless uncomfortable with situations promoting closeness. Finally, avoidance was negatively associated with happiness in the notes scenario. This decreased happiness in response to a potential close other’s helping behavior is consistent with the avoidant preference for self-reliance. How do these findings illuminate Study 2? The anxiously attached assigned greater meaning and significance to possible communal overtures. They interpreted even a relatively trivial communal behavior as a sign of potential closeness. This is consistent with Holmes’s (1991) predictions about chronic uncertainty in the context of the interdependence dilemma. Although everyone is thought to be susceptible to interdependence dilemma concerns— indeed, secure individuals also read meaning into the potential close other’s behavior in this study—it appears that anxiously attached individuals, who are chronically uncertain, were more susceptible to interdependence concerns as evidenced by their tendency to hopefully latch onto even relatively trivial communal gestures. Unfortunately, as noted earlier, this microanalytic process may actually fuel the interpersonal anxiety anxious individuals are trying to regulate. The findings from Study 3 also shed light on the avoidant individuals in Study 2. The findings from Study 3 suggest that avoidant individuals have a strong desire to downplay the importance of closeness and that this may have contributed to their negative partner evaluations in Study 2. Finally, it is important to note that—at least in Study 3—avoidant individuals were not significantly less interested in developing a relationship with their chosen person than were their more secure or anxious counterparts, thus reducing the likelihood that the avoidants’ response to the potential close other’s communal be-

NAVIGATING THE INTERDEPENDENCE DILEMMA

havior was simply due to insufficient interest in developing a relationship.

Study 4 In Study 1, we found that when interacting with a potential close other, anxious individuals went out of their way to signal interest in closeness. In Study 2, we found that ironically, when a potential close other signaled interest in closeness, attachment anxiety was associated with more interpersonal anxiety than when the potential close other acted more neutrally. The findings from Study 3 suggest that situations involving possible closeness elicit a great deal of uncertainty and that anxiously attached individuals may go into a mode of uncertainty reduction, looking for evidence of the other’s interest and caring, even in more trivial communal behaviors. Taken together, Studies 1, 2, and 3 suggest that in potential relationship development situations, anxiously attached individuals are especially susceptible to interdependence dilemma concerns. What, however, are the implications of these interdependence dilemma concerns? A central prediction of attachment theory is the notion of a secure base. When attachment needs are met, people are free to engage in non-attachment-related activities (e.g., exploration); however, when the attachment system is chronically engaged in the pursuit of attachment goals (as is often the case with those who feel anxiously attached), fewer resources are available for non-attachment-related activities (Mikulincer & Shaver, 2003). The goal of Study 4 was to investigate the consequences of regulating concerns about closeness for anxious individuals’ personal functioning. In addition, although Study 3 investigated the kind of inferences and attributions people make about a potential close other’s communal behavior, it was limited because of the self-report methodology used and because participants did not interact with a real person. Thus, in Study 4, we again explored how participants construed another’s communal behavior; however, rather than asking participants explicitly about their perceptions, we decided to use an implicit measure. Not only do implicit measures avoid drawing attention to specific research questions but also they circumvent participants’ self-presentation motives. Thus, in this study, we used a lexical decision task to assess the activation of attachment themes related to proximity– closeness and distance– rejection. Because the experimental manipulation concerned the possibility of closeness, it was hypothesized that closeness themes would primarily be activated; however, we also wanted to explore the possible activation of rejection concerns. A final goal of this study was to provide further evidence— using a different operationalization of the confederate’s use of communal norms—that situations affording the possibility of closeness, compared with situations in which closeness is less likely, are distressing to anxiously attached individuals. To meet these goals, we returned to the laboratory methodology and had participants again interact with an opposite-sex confederate in a study supposedly investigating cognitive abilities and distributive reasoning. Participants were informed that they would be performing some cognitive tasks (which included a task of mental concentration and a lexical decision task) and then answering some general knowledge (Trivial Pursuit; Hasbro, Inc., Pawtucket, RI) questions. In the communal condition, participants

89

were given the option to work as a team, and the confederate reached out to the participant and suggested working together for the Trivial Pursuit portion of the study. In the baseline performance-anxiety condition, participants were not given the option to work as a team but were simply informed that they would have to answer the Trivial Pursuit questions individually (hence, performance anxiety). We hypothesized that anxiously attached individuals would respond to the prospect of closeness in the communal condition with increased rumination and, consequently, poorer performance on the mental concentration task. By contrast, we predicted that secure individuals would not be distressed about the prospect of closeness in the communal condition—in fact, we thought that they might benefit from contact with a supportive other, resulting in enhanced performance. Finally, although we thought that avoidant individuals would likely be distressed about the prospect of closeness in the communal condition, we did not predict impaired performance on the mental concentration task because their typical coping response is to distance themselves from unpleasant thoughts. In sum, all participants were expected to exhibit performance deficits on the mental concentration task in the performanceanxiety condition, but secure individuals’ performance was expected to be facilitated in the communal condition, whereas anxiously attached individuals’ performance was expected to be impaired in the communal condition. Moreover, it was predicted that secure and anxious individuals’ performance on the mental concentration task in the communal condition would be differentially associated with proximity accessibility, with proximity accessibility being associated with better performance for the secure participants but worse performance for the anxious participants. By comparison, avoidant individuals’ performance was not predicted to be affected by the communal condition because it was thought they would likely keep in check or suppress concerns about closeness.

Method Participants Sixty-nine male university students volunteered to participate in the study (M age ⫽ 19.7 years). Again, single participants were targeted (54 single and 15 dating). Participants were randomly assigned to either the communal (n ⫽ 32) or the performance-anxiety condition (n ⫽ 37) and received $10 (Canadian) for their participation.

Procedure As in the previous studies, the participant had a 2-min interaction with an attractive, opposite-sex confederate, who was also supposedly a participant. After the interaction period, the participant and confederate were brought into the testing room. The experimenter explained that the study was investigating cognitive abilities and distributive reasoning and that first they would be working on some cognitive tasks individually to assess mental concentration and then would be working on a Trivial Pursuit task. After being given a brief description of the Trivial Pursuit task (see Experimental Manipulation, below), the participant was escorted to a separate room and was left alone to complete an informed-consent form and a thought-listing task (unrelated to this study). The experimenter returned after a few minutes with some personality measures (including the ECR) and the group information sheet for the

90

BARTZ AND LYDON

participant to complete. Once the participant had finished these items, the experimenter explained the d2 test (Brickenkamp, 1981), which was presented as a test of mental concentration. This test consists of a matrix of rows containing random sequences of the letters d and p with one, two, or no apostrophes above and/or below each letter; the objective is to go through each row and cross out all the ds with two apostrophes. The experimenter timed the participant as he worked on the d2 test. After completing a short evaluation form (see d2 evaluation form, below), the participant began the lexical decision task, which was also presented as a test of mental concentration. Under the pretext of checking on the other participant, the experimenter left so that the participant could work on the lexical decision task alone (participants completed the d2 test and lexical decision task in the absence of the confederate). The experimenter returned shortly after to probe for suspicions, debrief, and compensate the participant.

Experimental Manipulation Communal condition. In this condition, the participant and the confederate were informed that the second part of the study was investigating distributive reasoning and that they would be working on a task in which they would each be asked Trivial Pursuit questions. Moreover, they were told that they had the option to work with their partner on this task. Specifically, the experimenter said, “For this task, you have the option to work with your partner; so, if you choose to work with your partner, although each question will be directed at one of you, you will be able to help each other.” The experimenter then paused, and the confederate looked at the participant, smiled, and suggested working together as a team, in this way expressing the desire to be communal. The experimenter confirmed that they would be working as a team on the Trivial Pursuit task (all participants agreed to work with the confederate) and asked if there were any questions. Finally, as in Studies 1 and 2, in this condition, the participant interacted with an attractive, opposite-sex partner (a confederate) who was presented as single and a recent transfer student, in this way increasing the desirability and perceived likelihood of a communal relationship with the confederate. Performance-anxiety control condition. Our objective in designing this condition was to create a noncommunal condition—that is, an interaction in which the possibility of closeness was unlikely— however, we did not want participants to feel rejected by their partner. Thus, the participant and the confederate were informed that the second part of the study was investigating general knowledge and would involve a Trivial Pursuit game. In this condition, they were not given the option to work together; rather, they were told that the experimenter would be asking them the Trivial Pursuit questions individually and that each in turn would have the opportunity to answer the questions. Although the primary goal was to create a noncommunal condition in which the confederate did not express the desire to be communal by suggesting to work as a team, this meant having the participant work individually on the Trivial Pursuit task and answer questions in front of the confederate and experimenter. As a result, this condition was conceptualized as a performance-anxiety condition. Finally, in this condition, we followed Clark’s (1984b) manipulation of an exchange relationship and presented the confederate as dating and not a transfer student to decrease the likelihood that the participant would see her as available for a communal relationship. The confederate was also slightly less talkative during the hallway interaction (i.e., she did not initiate a conversation, but if the participant asked her questions, she would respond, so as to not appear rude).

Materials ECR (Brennan et al., 1998). The 36-item questionnaire used in Studies 2 and 3 was used to assess attachment avoidance (␣ ⫽ .91) and anxiety (␣ ⫽ .89).

The d2 mental concentration test (Brickenkamp, 1981). This test consists of a matrix of 14 rows of random sequences of the letters d and p with one, two, or no apostrophes above and/or below each letter. Participants were instructed to cross out all the ds with two apostrophes and were given 15 s to go through each row (the experimenter timed participants with a stopwatch). Participants were instructed to work as quickly and accurately as possible. The d2 test is theorized to be an index of resistance against interference and has been used to assess rumination (Kuhl, 1981). Performance on the d2 test was calculated by summing the total number of hits (i.e., the total number of items crossed out) and subtracting the total number of mistakes (i.e., the number of non-d2s erroneously marked), in this way taking into account both errors of omission (i.e., the number of d2s missed) and errors of commission. The d2 evaluation form. This questionnaire assessed participants’ experience during the d2 test. Of interest, participants rated on a scale from 0 (not at all) to 4 (extremely) how much attention they devoted to the task and how much time they spent thinking about how well they were doing on the task by placing a slash through a line from 0% to 100%. An attention– distraction score was calculated by subtracting how much participants reported they were thinking about how well they were doing from how much attention they said they devoted to the d2 (higher numbers reflecting greater attention). Lexical decision task. This task was used to assess the cognitive accessibility of proximity and distance–rejection themes. Participants were presented with a series of letter strings on a PC and were instructed to judge as quickly and as accurately as possible whether each letter string was a word or nonword by pressing the appropriate key on the keyboard. Participants were given 12 practice trials to familiarize themselves with the task, and then they began the 87 experimental trials. Each trial began with a star presented in the middle of the screen, followed by the target stimulus after a pause of 500 ms. The target stimulus lasted on the screen for 1,000 ms during which participants indicated whether the target was a word or nonword. After they had indicated their response, a blank screen followed for 500 ms, and then, the next trial began. The target letter strings were taken from Mikulincer, Birnbaum, Woddis, and Nachmias (2000) and consisted of 3 proximity words (closeness, love, and affection)6 and 6 distance–rejection words (separation, rejection, abandonment, distance, loneliness, and alone). For exploratory purposes, we also included 5 coping strategy words (intimacy, escape, worry, security, and control), taken from Mikulincer (1998), which were theorized to reflect the coping strategies associated with different attachment orientations.7 In addition, 9 communion and 9 agency words were included for pilot-testing purposes, along with 10 neutral words and 45 nonwords. The 10 neutral words had no positive or negative connotations and no link to proximity or rejection themes (e.g., elephant, book, and picture). The 45 nonwords were created by taking common English words and changing one letter. All words and nonwords were matched for number of letters, and trials were randomly ordered across participants. This task was programmed using E-prime and run on a Dell Pentium 3 XPS T700r PC.

6 Mikulincer et al. (2000) used six proximity words; however, during pilot testing, three of these words (kiss, hug, and caress) were repeatedly mentioned as curious and out of place given the context of the study; thus, these words were dropped from the lexical decision task. 7 It was thought that worry and security might be more accessible for the anxiously attached, whereas escape and control might be more accessible for the avoidantly attached, and that this might interact with experimental condition. However, no mean differences in reaction times (RTs) to these words were found, nor were any correlations between these RTs and mental concentration found as a function of attachment.

NAVIGATING THE INTERDEPENDENCE DILEMMA

91

Manipulation check. After participants were debriefed, they rated the extent to which they (a) liked their partner, (b) were attracted to their partner, and (c) would like to be friends with their partner by placing a slash through a line ranging from 0 (not at all) to 100 (very much). Participants were given privacy to answer the questionnaire and were assured that their responses would remain anonymous and confidential. A composite manipulation check measure was created by averaging these three items (␣ ⫽ .71).

Results Manipulation Check The independent samples t test investigating the manipulation check composite revealed a significant difference between experimental conditions, t(67) ⫽ 3.48, p ⫽ .001. Supporting the experimental manipulation, those in the communal condition liked their partner more, were more attracted to their partner, and were more interested in a relationship with their partner, M(32) ⫽ 69.55, than those in the performance-anxiety condition, M(37) ⫽ 58.62. The interactive effects of experimental condition and attachment on attraction to partner were also investigated in an analysis of variance (ANOVA) after splitting the attachment dimension scores at the median. The ANOVA revealed a three-way interaction between condition, attachment anxiety, and attachment avoidance, F(1, 61) ⫽ 4.63, p ⬍ .05. Focused comparisons showed that low anxious–low avoidant participants and high anxious– high avoidant participants were more attracted to the confederate in the communal condition, M(7) ⫽ 77.19 and M(9) ⫽ 70.30, respectively, than in the performance-anxiety condition, M(12) ⫽ 55.08 and M(11) ⫽ 57.52, respectively, t(61) ⫽ 3.58, p ⬍ .001, and t(61) ⫽ 2.19, p ⬍ .05, whereas there were no between-condition differences in partner attraction for low anxious– high avoidant participants or for high anxious–low avoidant participants. There were no attachment group differences within the communal and performance-anxiety conditions.

Mental Concentration–Rumination To investigate the influence of condition and attachment on mental concentration–rumination a 2 (condition: communal vs. performance anxiety) ⫻ 2 (attachment anxiety: low vs. high) ⫻ 2 (attachment avoidance: low vs. high) ANOVA was conducted on participants’ d2 scores (i.e., the sum of d2 hits minus mistakes).8 The ANOVA investigating the effects of experimental condition and attachment anxiety and avoidance on d2 performance yielded a main effect for condition, F(1, 58) ⫽ 4.70, p ⬍ .05. Overall, performance was worse in the performance-anxiety condition, M(37) ⫽ 183.36, than in the communal condition, M(29) ⫽ 200.66; however, this effect was qualified by an interaction between condition and attachment anxiety, F(1, 58) ⫽ 4.49, p ⬍ .05, but not by an interaction with avoidance (F ⬍ 1.5). As depicted in Figure 2, participants low in attachment anxiety in the communal condition performed significantly better, M(13) ⫽ 212.67, than their low anxious counterparts in the performance-anxiety condition, M(20) ⫽ 178.48, t(58) ⫽ 3.05, p ⬍ .01, and, importantly, also performed significantly better than their high anxious counterparts in the communal condition, M(16) ⫽ 188.64, t(58) ⫽ 2.05, p ⬍ .05. There was no difference between

Figure 2. Mental concentration (d2 test performance) as a function of experimental condition and attachment anxiety (Study 4).

the low and high anxious participants in the performance-anxiety condition, nor was there a difference between conditions for the high anxious participants (both ts ⬍ 1.5). Thus, as predicted, overall, participants were less able to concentrate in the performance-anxiety condition, suggesting their preoccupation with answering the Trivial Pursuit questions in front of their partner. However, whereas the communal condition facilitated performance for the more secure participants, it impaired performance for the more anxiously attached.

Self-Reported Attention–Distraction The ANOVA investigating the effects of condition and attachment on self-reported attention– distraction yielded a marginally significant interaction between condition and attachment anxiety, F(1, 61) ⫽ 3.32, p ⫽ .073. In the communal condition, high anxious participants reported being less attentive and more distracted, M(18) ⫽ 29.39, than their less anxious counterparts, M(14) ⫽ 62.36, t(61) ⫽ 2.04, p ⬍ .05. There were no other group differences (ts ⬍ 1.5).

Lexical Decision Analyses9 Drawing on Greenwald, McGhee, and Schwartz (1998), we recoded reaction times (RTs) on the lexical decision task less than 300 ms as 300 ms and RTs greater than 3,000 ms as 3,000 ms. 8 Three participants with errors-of-commission scores (mistakes) greater than 60 (i.e., more than 4 standard deviations above the group mean of 5.75) were excluded from the d2 analyses because they most likely misunderstood the instructions to the d2 test. The fourth highest participant had an error-of-commission score of 17. 9 Although native English speakers were targeted, 13 participants reported that English was their second language. One participant who rated his English proficiency as moderate (3 on a scale from 1 to 5) was dropped from the lexical decision analyses. The 12 remaining participants who

BARTZ AND LYDON

92

Overall, the error rates were low (the mean was 2.7%).10 RTs for words with incorrect responses were replaced with the mean RT for that word. Error rates were not related to experimental condition or attachment. The mean RT across all conditions was 740.96 ms (SD ⫽ 146.84). Within-cell correlations were calculated to investigate the association between the accessibility of proximity or distance–rejection and d2 performance. Specifically, standardized residuals of the proximity and distance–rejection RTs were created using control word RT as the predictor (negative numbers reflecting quicker RTs and increased accessibility). Participants’ d2 scores were also standardized (negative numbers reflecting poorer performance). The cross-products of these standardized scores were taken (e.g., the product of proximity accessibility and d2 score). We then conducted 2 (condition: communal vs. performance anxiety) ⫻ 2 (attachment anxiety: low vs. high) ⫻ 2 (attachment avoidance: low vs. high) ANOVAs on the within-cell correlations between (a) Proximity RT ⫻ d2 Performance and (b) Distance–Rejection RT ⫻ d2 Performance.11 The ANOVA investigating the correlation between proximity accessibility and d2 performance yielded a significant interaction between condition and attachment anxiety, F(1, 57) ⫽ 4.49, p ⬍ .05. In the communal condition, proximity accessibility was associated with better d2 performance for low anxious individuals, M(13) ⫽ ⫺.69, but proximity accessibility was associated with poorer d2 performance for high anxious individuals, M(15) ⫽ .47, t(57) ⫽ 2.37, p ⬍ .05. Thus, in the communal condition, proximity accessibility facilitated performance for the less anxiously attached but impaired performance for the more anxiously attached. In addition, a second interaction was found between attachment anxiety and avoidance independent of experimental condition, F(1, 57) ⫽ 4.29, p ⬍ .05. Among those high in anxiety, proximity accessibility was associated with increased performance for those low in avoidance, M(15) ⫽ ⫺.46, but was associated with decreased performance for those high in avoidance, M(17) ⫽ .76, t(57) ⫽ 2.68, p ⬍ .01. Finally, there were no significant main or interaction effects of condition and/or attachment on the cross-product of distance– rejection accessibility and d2 performance (all Fs ⬍ 2.1).

Discussion As predicted, performance on the mental concentration task was impaired for everyone in the performance-anxiety condition, but anxiously attached individuals also performed poorly in the communal condition. Secure participants, on the other hand, appeared to benefit from the communal condition—performing better than their anxious counterparts in the communal condition and better than their secure counterparts in the performance-anxiety condition. What was it about the communal condition that helped the secures but hurt the anxiously attached participants? Analyses looking at the association between d2 performance and proximity accessibility suggest that whereas thoughts about closeness facilitated the secure individuals’ performance, these thoughts undermined the anxious individuals’ performance. These findings high-

reported that English was not their native language did not differ from the native English speakers on error rates (t ⬍ 1) or RT (t ⬍ 1).

light the importance of a secure base: When secure participants interacted with someone who expressed liking and acceptance, as opposed to someone who was simply neutral, they went on to excel at the mental concentration task. Unfortunately, the potential for closeness did not have the same effect on their anxious counterparts. Anxious individuals ruminated even in response to a relatively positive interpersonal event—that is, an attractive, available other expressing interest in them. Finally, situations involving possible closeness did not trigger concerns about rejection for the anxiously attached, a point to which we return in the General Discussion. Avoidant individuals did not exhibit performance deficits in the communal condition. Given that a core feature of avoidant attachment is discomfort when others try to get too close and that avoidant individuals liked their partner less when their partner worked as a unit in Study 2, it is likely that the confederate’s suggestion to work as a team would have been somewhat distressing to the avoidant participants. However, avoidant attachment is associated with a deactivation of the attachment system, and avoidant individuals tend to cope with distress by suppressing their concerns, thus, their distress in the communal condition may not have compromised their performance on the mental concentration task.

General Discussion Our goal in conducting this research was to investigate whether attachment is related to people’s ability to tolerate the uncertainty inherent in the early phases of relationship development. People who hope to establish a relationship try to follow the communal script to signal interest, but the desire to express interest is tempered by their uncertainty about the prospect for a relationship and the other’s motives. Because differences in attachment reflect beliefs about whether the self is worthy of affection and whether others are trustworthy and reliable, it was theorized that attachment would be related to the ability to tolerate the interdependence dilemma. Overall, our predictions have been supported.

Summary and Discussion of Findings When anxiously attached individuals interacted with an attractive, desirable, opposite-sex partner, they obscured individual contributions to a group task. Drawing on Clark’s (1984b) theorizing, we believe they did this to signal their interest in closeness (i.e., by obscuring individual contributions, they conveyed their preference for a communal relationship over an exchange relationship). The notion that anxiously attached individuals would go out of their way to signal interest in a communal relationship with a potential close other is consistent with their chronic desire for closeness and acceptance. These results are also in accord with Mikulincer and Nachshon’s (1991) research findings that anxiously attached individuals are especially prone to self-disclosure and tend to self10 Three real words and seven nonwords had exceptionally high error rates (greater than 15%) and consequently were dropped from the analyses. The real words that were dropped were not target words. 11 We also looked at the effects of experimental condition and attachment on the accessibility of the proximity– closeness, distance–rejection, and control strategy words. These analyses yielded one marginal effect. There were no significant main effects or interactions.

NAVIGATING THE INTERDEPENDENCE DILEMMA

disclose indiscriminatingly to strangers. Indeed, these researchers proposed that anxiously attached individuals use self-disclosure opportunities to merge with others. In the same way, anxiously attached individuals may have obscured contributions in the current investigation to symbolically merge with their partner. Yet were anxiously attached individuals obscuring contributions to avoid their partner’s negative evaluation should they perform poorly on the group task? The findings from Study 2 suggest this explanation is unlikely. If anxious individuals were obscuring contributions to avoid negative evaluation, they should have been less anxious in the communal condition in Study 2 because their performance was indistinguishable from their partner’s, but this was not the case. So, when interacting with an attractive, desirable, opposite-sex partner, anxious individuals went out of their way to signal interest. However, when such a partner signaled interest by following the communal script, anxious individuals were not reassured. In Study 2, attachment anxiety was associated with greater interpersonal anxiety (i.e., state feelings of anxiety, uncertainty, and concerns about appearance) when a potential close other used communal norms (behaved as a unit) compared with when the other used exchange norms. In Study 4, anxiously attached individuals performed poorly on an unrelated mental concentration task when a potential close other expressed interest in closeness by suggesting to work as a team; moreover, results from the lexical decision task in Study 4 suggest that it was thoughts about closeness in particular that undermined anxious individuals’ performance in the communal condition. Given that securing closeness is a chronic goal for anxiously attached individuals, why would they experience more interpersonal anxiety when a potential close other expressed interest in a communal relationship? We believe that anxious individuals felt especially vulnerable in these potential communal situations and that this vulnerability led to increased feelings of anxiety, uncertainty, and concerns about how their partner would evaluate them. In Study 1, anxious individuals made greater overtures than their secure and avoidant counterparts; as a result, they may have felt more anxious about whether those overtures would be met with equal interest. That is, by putting themselves on the line this way, they had more to lose. On a related note, Vorauer, Cameron, Holmes, and Pearce (2003) found that anxious individuals tend to imbue their overtures to a potential romantic partner with increased significance and meaning. Even when they do not objectively make greater overtures, anxious individuals think they have communicated more interest because they believe others will take into account their inhibitions when judging their behavioral intentions. Thus, in the present investigation, when anxious individuals did make greater overtures, it is likely they felt especially vulnerable. At the heart of the matter, we believe that situations involving possible closeness amplify anxious individuals’ interpersonal ambivalence and elicit an approach–avoidance conflict leading to heightened feelings of interpersonal anxiety. Study 1 suggests that anxious individuals were looking for closeness in these encounters, and this may have lead to an internal dialectic between their deepest hopes and fears (Holmes, 1991). On the one hand, their desire for acceptance should have led them to look for evidence of caring and to infer meaning from their partner’s behavior—indeed, Study 3 suggests this was the case. In Study 3, anxiously attached

93

individuals inferred more meaning and were less likely to discount even relatively trivial communal behaviors as purely situational. On the other hand, it is likely that these judgments were guarded by feelings of vulnerability and the acute possibility that they may ultimately be found unworthy. As Holmes (1991) stated, “the specter of things going well reminds [uncertain] people of the costs of their being wrong” (p. 84). This idea is echoed in the attachment literature as well. For example, Mikulincer and Shaver (2003) speculated that for anxiously attached individuals, positive states may quickly bring to mind other instances in which things began well but ended badly. Although Study 2 suggests that anxious individuals were concerned about evaluation, Study 4 did not find evidence that anxiously attached individuals were specifically concerned about rejection, so how and when might concerns about rejection come into play? Holmes (1991) described the interdependence dilemma as a dialectic that unfolds over time. People signal their interest and then look for signs of interest and commitment in the other’s behavior; this back and forth continues as the relationship unfolds (or not, as the case may be). This research focused on the very early stages of relationship development, a point at which anxious individuals may have been quite hopeful. Perhaps research that focuses on interactions occurring later in relationship development would yield increased concerns about rejection because there is greater investment and, consequently, greater risk. Anxiously attached individuals’ mounting interpersonal anxiety may eventually bias their perceptions in a negative direction. It may also be that in the same way that anxious individuals magnify relatively trivial positive behaviors in this process of uncertainty reduction, they also magnify relatively trivial negative behaviors. This notion is consistent with Fraley and Shaver’s (2000) conceptualization of the anxiety dimension of attachment as being critically involved in the appraisal and monitoring of attachment related cues and the idea that individuals who are high in attachment anxiety have a low threshold for detecting cues of rejection and acceptance. Indeed, Campbell, Simpson, Boldry, and Kashy (2005) found support for this idea in a recent diary study in which it was found that anxious individuals tended to magnify both positive and negative relationship-relevant behaviors in their established relationships. Our research adds to this literature by showing that anxious individuals’ vigilant monitoring of attachment-related behaviors occurs even in the context of potential new relationships. In contrast to their more anxious counterparts, secure participants did not actively avoid using exchange norms when interacting with an attractive, available, opposite-sex partner, but importantly, they also did not use exchange norms (i.e., their pen choice was random). Secure participants also felt comfortable with the potential close other’s use of communal norms. They were not distressed when their partner expressed the desire to be a unit (Studies 2 and 4), and in fact, they actually benefited from the communal condition in Study 4. Were secure participants simply not interested in a relationship with their partner? Because they were the most likely group to have a strong network of friends, secure participants may not have been especially motivated to form a new relationship. However, we believe this explanation is unlikely for a few reasons. First, in the current studies, we tried to create situations in which the potential for a close relationship was desirable and feasible. Drawing on Clark (1984b, 1986), we tar-

94

BARTZ AND LYDON

geted single participants and paired them with an attractive, opposite-sex confederate who was single and a recent transfer student. So, even if secure participants had several close friends, they were still not involved in a romantic relationship and thus should have been interested in the confederate. Second, secure participants did not differ from their anxious counterparts on partner liking (Studies 2 and 4) or interest in developing a relationship (Study 3). Finally, thoughts about closeness, love, and affection were associated with a dampening of rumination in Study 4. If secure participants were not at least somewhat interested, they should not have benefited from the potential for closeness in this study. Rather than secure individuals being disinterested in their partner, we believe secure individuals’ self-esteem, trust in others, and previous successful close relationship experiences account for their behavior in the present investigations. In contrast to their more anxious counterparts, in Study 1, secure individuals felt comfortable letting the situation unfold without a great deal of effort on their part. Indeed, secure individuals may have been more like the existing friends (whose pen choice was also random) in Clark’s (1984b) original study. Clark (1984b) theorized that pen choice was random for existing friends because existing friends know they are friends, so there is no need to send a message conveying interest in closeness. In the current investigation, secure participants may have felt confident that the relationship would develop and, as a result, did not feel the need to go out of their way to make it happen. The findings from Study 3 also support this idea. Secure participants were less likely than their anxious counterparts to invest their potential close other’s discrete communal behavior with increased significance for relationship development, suggesting that although they may have been interested in developing a relationship, they were not preoccupied with evaluating the meaning of discrete events and using that information to infer the other’s motives and to diagnose relationship potential. Another possible factor distinguishing the secure participants from their more anxious counterparts may have been the goals that were activated in our experimental situations. In a study looking at attachment and the pursuit of attachment and affiliation goals, Mikulincer and Selinger (2001) found that secure adolescents pursued attachment goals in attachment-appropriate contexts and affiliation goals in affiliation-appropriate contexts, whereas anxious adolescents focused exclusively on attachment goals and tended to pursue attachment goals in attachment and affiliation contexts. In the current studies, secure participants may have viewed these situations as opportunities to pursue affiliation goals, whereas anxious participants may have viewed them as opportunities to pursue attachment goals. Finally, contrary to our predictions, when avoidant individuals interacted with a potential close other, overall, they did not use exchange norms. Perhaps avoidant individuals only feel the need to establish boundaries when they perceive that others are trying to get too close. Indeed, the findings from Study 2 suggest this may be the case. Compared with when their partner used exchange norms, when their partner tried to work as a unit (i.e., used communal norms), avoidant individuals liked their partner less. Perhaps avoidant individuals felt exploited when their partner used the same color pen; by obscuring how much each person contributed to the task, it would be easier for the confederate to freeload.

This hypothesis would be consistent with avoidant individuals’ pessimistic beliefs about others; however, if this were the case, one would think avoidant individuals would have ascribed more negative agency traits (e.g., greedy) and not more negative communal traits (e.g., servile) to their partner in the communal condition. Avoidant individuals strive for independence, tend to downplay the importance of closeness, and use avoidance strategies to cope with concerns. It may be that when avoidant individuals enter situations involving possible closeness (situations that, in theory, should be distressing), they attempt to suppress the possibility of closeness. Consistent with this idea, in Study 3, avoidant individuals downplayed relational motives for the potential close other’s communal behavior. Moreover, although they professed to be more indifferent to the potential close other’s communal behavior in Study 3, avoidant attachment was associated with feeling uneasy in Studies 2 and 3, suggesting that their professed indifference may have been a fac¸ade. Indeed, this notion is in line with research suggesting that avoidant individuals’ defensive strategies may actually be quite fragile (Mikulincer, Dolev, & Shaver, 2004).

Research Strengths and Limitations Three of the four studies presented had participants interact with a real person (a confederate), and behavioral responses were assessed; nevertheless, there were limitations to this approach. Studies 1 and 2 did not assess how participants were construing their behavior or the behavior of the confederate, mainly because of concerns about drawing attention to the research questions and possibly undermining participants’ behavioral responses. Were participants intending to convey interest by obscuring contributions? Did they perceive their partner’s communal behavior as an expression of interest? Study 3 shed some light on the kind of attributions participants were making and what meaning they saw in the potential close other’s communal behavior. That said, we believe that although people use communal and exchange behaviors to convey how they feel about closeness, this may often occur at an implicit level; thus, explicitly asking participants about their intentions may provide limited information. The lexical decision task used in Study 4 is one way to examine the implicit cognitions and motivations that guide these interactions. In addition, in our attempt to create a noncommunal condition that paralleled the communal condition in Study 4, concerns about performance were aroused, and anxious participants performed poorly in both conditions. Would they have performed poorly if they had simply been asked to do the d2 test without interacting with another participant? Although anxious participants did not perform more poorly than the secures in the performance-anxiety condition, it is possible that anxious individuals are generally poor at mental concentration tasks, and future research should be conducted to explore this possibility. Nonetheless, we believe the important finding from Study 4 is that unlike their more secure counterparts, anxious participants did not benefit from the communal condition—for them, thoughts about closeness, love, and affection did not dampen performance anxiety and rumination.

Theoretical Contributions and Future Directions A major proposition of attachment theory is that attachment models are carried forward into new relationships, guiding percep-

NAVIGATING THE INTERDEPENDENCE DILEMMA

tions, expectations, and behaviors with new partners (Collins & Read, 1994); however, little is known about how the attachment system operates during the early phases of relationship development. This research establishes the importance of attachment at the outset of a relationship and points to some circumstances under which attachment models should come into play. Attachment models appear to be influential when little is known about what can be expected from the other person (Collins & Read, 1994; Pierce & Lydon, 2001; Pietromonaco & Feldman Barrett, 2000). The match between the situation and chronic attachment goals related to security attainment also appears to be important. Finally, behaviors related to the exchange of social commodities that are used to define the nature of the relationship appear to be associated with attachment. One question arising from the current research is whether these findings apply to all close relationships, friendships, or romantic relationships in particular. To increase the chances that participants would desire a relationship with the confederate, the possibility of a romantic relationship was made salient. That said, as noted earlier, some participants may have thought of the confederate as a potential romantic partner, whereas others may have thought of the confederate as a friend. Study 3 spoke to this issue. Participants were allowed to nominate either a potential friend or potential romantic partner, and relationship type did not qualify the results, suggesting that the present findings would apply to close relationships in general. It is worth noting that participants were more interested in developing a romantic relationship in Study 3. Moreover, adult attachment relationships tend to be romantic (Fraley & Shaver, 2000; Hazan & Zeifman, 1999). Given this, we believe the present findings should be especially relevant to potential romantic relationships, but may also apply to close friendships, depending on the level of interest and the extent to which the other is perceived as a potential attachment figure. Is attachment associated with the use of communal and exchange norms in established relationships? The early stages of a relationship are fraught with uncertainty, but in existing relationships, people presumably have established a certain level of trust and are more knowledgeable about their partner’s intentions and level of commitment. That said, we believe attachment would similarly be associated with the use of communal and exchange norms in established relationships to the extent that satisfying chronic attachment goals continues to be important. Uncertainty may also be a critical factor. Perhaps if anxiously attached individuals felt particularly uncertain—for example, when making a major personal or relational decision—they may be especially likely to monitor their partner’s discrete behaviors for reassurance.

Concluding Comments In conclusion, the early stages of a relationship are marked by uncertainty and an interdependence dilemma in which people must wager on the motives of the other person and decide whether being communal is worth the risk. They must also tailor their own behavior to communicate interest but not appear overly concerned with social transactions. Although people share the same script for close relationships, the belief structures and goals associated with chronic attachment orientations provide direction on how that script will be played out as the relationship unfolds.

95 References

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., & Leary, M. R. (1995). The need to belong: Desire for interpersonal attachments as a fundamental human motivation. Psychological Bulletin, 117, 497–529. Bowlby, J. (1969). Attachment and loss: Vol. 1. Attachment. New York: Basic Books. Bowlby, J. (1973). Attachment and loss: Vol. 2. Separation: Anxiety and anger. New York: Basic Books. 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. Brickenkamp, R. (1981). Test d2 (7th ed.). Go¨ttingen, Germany: Hogrefe. Campbell, L., Simpson, J. A., Boldry, J., & Kashy, D. A. (2005). Perceptions of conflict and support in romantic relationships: The role of attachment anxiety. Journal of Personality and Social Psychology, 88, 510 –531. Clark, M. S. (1984a). A distinction between two types of relationships and its implications for development. In J. C. Masters & K. Yarkin-Levin (Eds.), Boundary areas in social and developmental psychology (pp. 241–270). Orlando, FL: Academic Press. Clark, M. S. (1984b). Record keeping in two types of relationships. Journal of Personality and Social Psychology, 47, 549 –557. Clark, M. S. (1986). Evidence for the effectiveness of manipulations of communal and exchange relationships. Personality and Social Psychology Bulletin, 12, 414 – 425. Clark, M. S., & Mills, J. (1979). Interpersonal attraction in exchange and communal relationships. Journal of Personality and Social Psychology, 37, 12–24. Clark, M. S., Mills, J., & Powell, M. C. (1986). Keeping track of needs in communal and exchange relationships. Journal of Personality and Social Psychology, 51, 333–338. Cohen, J., Cohen, P., West, S., & Aiken, L. (2003). Applied multiple regression/correlation analysis for the behavioural sciences (3rd ed.). Hillsdale, NJ: Erlbaum. Collins, N. L., & Read, S. J. (1994). Cognitive representations of attachment: The structure and functions of working models. In K. Bartholomew & D. Perlman (Eds.), Advances in personal relationships (Vol. 5, pp. 53–90). London: Jessica Kingsley. Davilia, J., Burge, D., & Hammen, C. (1997). Why does attachment style change? Journal of Personality and Social Psychology, 73, 826 – 838. Fraley, R. C., & Shaver, P. R. (1997). Adult attachment and the suppression of unwanted thoughts. Journal of Personality and Social Psychology, 73, 1080 –1091. 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 typographical model. In J. A. Simpson & W. S. Rholes (Eds.), Attachment theory and close relationships (pp. 77–114). New York: Guilford Press. Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring individual differences in implicit cognition: The implicit association test. Journal of Personality and Social Psychology, 74, 1464 –1480. Hazan, C., & Shaver, P. R. (1987). Romantic love conceptualized as an attachment process. Journal of Personality and Social Psychology, 52, 511–524. Hazan, C., & Shaver, P. R. (1994). Attachment as an organizational framework for research on close relationships. Psychological Inquiry, 5, 1–22.

96

BARTZ AND LYDON

Hazan, C., & Zeifman, D. (1999). Pair bonds as attachments: Evaluating the evidence. In J. Cassidy & P. R. Shaver (Eds.), Handbook of attachment: Theory, research, and clinical applications (pp. 336 –354). New York: Guilford Press. Heatherton, T. F., & Polivy, J. (1991). Development and validation of a scale for measuring state self-esteem. Journal of Personality and Social Psychology, 60, 895–910. Holmes, J. G. (1981). The exchange process in close relationships: Microbehavior and macromotives. In M. J. Lerner & S. C. Lerner (Eds.), The justice motive in social behavior (pp. 261–284). New York: Plenum Press. Holmes, J. G. (1991). Trust and the appraisal process in close relationships. In W. H. Jones & D. Perlman (Eds.), Advances in personal relationships (Vol. 2, pp. 57–104). London: Jessica Kingsley. House, J. S., Landis, K. R., & Umberson, D. (1988, July 29). Social relationships and health. Science, 241, 540 –545. Kiecolt-Glaser, J. K., & Newton, T. L. (2001). Marriage and health: His and hers. Psychological Bulletin, 127, 472–503. Kuhl, J. (1981). Motivational and functional helplessness: The moderating effects of state versus action orientation. Journal of Personality and Social Psychology, 40, 155–170. Leary, M. (1990). Responses to social exclusion: Social anxiety, jealousy, loneliness, depression, and low self-esteem. Journal of Social and Clinical Psychology, 9, 221–229. Lydon, J. E., Jamieson, D. W., & Holmes, J. G. (1997). The meaning of social interactions in the transition from acquaintanceship to friendship. Journal of Personality and Social Psychology, 73, 536 –548. Lynch, J. J. (1979). The broken heart: The medical consequences of loneliness. New York: Basic Books. Maslow, A. H. (1962). Toward a psychology of being. New York: Van Nostrand. McNair, D. M., Lorr, M., & Droppleman, L. F. (1971). Profile of Mood States. San Diego, CA: Educational and Industrial Testing Services. Mikulincer, M. (1998). Attachment working models and the sense of trust: An exploration of interaction goals and affect regulation. Journal of Personality and Social Psychology, 74, 1209 –1224. Mikulincer, M., Birnbaum, G., Woddis, D., & Nachmias, O. (2000). Stress and accessibility of proximity-related thoughts: Exploring the normative and intraindividual components of attachment theory. Journal of Personality and Social Psychology, 78, 509 –523.

Mikulincer, M., Dolev, T., & Shaver, P. R. (2004). Attachment-related strategies during thought suppression: Ironic rebounds and vulnerable self-representations. Journal of Personality and Social Psychology, 87, 940 –956. Mikulincer, M., & Nachshon, O. (1991). Attachment styles and patterns of self-disclosure. Journal of Personality and Social Psychology, 61, 321– 331. Mikulincer, M., & Selinger, M. (2001). The interplay between attachment and affiliation systems in adolescents’ same-sex friendships: The role of attachment style. Journal of Social and Personal Relationships, 18, 81–106. 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. Pierce, T., & Lydon, J. E. (2001). Global and specific relational models in the experience of social interactions. Journal of Personality and Social Psychology, 80, 613– 631. Pietromonaco, P. R., & Feldman Barrett, L. (2000). The internal working models concept: What do we really know about the self in relation to others? Review of General Psychology, 4, 155–175. Simpson, J. A., & Rholes, W. S. (1998). Attachment in adulthood. In J. A. Simpson & W. S. Rholes (Eds.), Attachment theory and close relationships (pp. 3–21). New York: Guilford Press. Spence, J. T., Helmreich, R. L., & Holahan, C. K. (1979). Negative and positive components of psychological masculinity and femininity and their relationships to self-reports of neurotic and acting out behaviors. Journal of Personality and Social Psychology, 37, 1673–1682. Sullivan, H. S. (1953). The interpersonal theory of psychiatry. New York: Norton. Trout, D. L. (1980). The role of social isolation in suicide. Suicide and Life-Threatening Behavior, 10, 10 –23. Vorauer, J. D., Cameron, J. J., Holmes, J. G., & Pearce, D. G. (2003). Invisible overtures: Fears of rejection and the signal amplification bias. Journal of Personality and Social Psychology, 84, 793– 812.

Received September 21, 2004 Revision received October 26, 2005 Accepted October 28, 2005 䡲

Instructions to Authors For Instructions to Authors, please visit www.apa.org/journals/psp and click on the “Instructions to Authors” link in the Journal Info box on the right.

96

BARTZ AND LYDON

Hazan, C., & Zeifman, D. (1999). Pair bonds as attachments: Evaluating the evidence. In J. Cassidy & P. R. Shaver (Eds.), Handbook of attachment: Theory, research, and clinical applications (pp. 336 –354). New York: Guilford Press. Heatherton, T. F., & Polivy, J. (1991). Development and validation of a scale for measuring state self-esteem. Journal of Personality and Social Psychology, 60, 895–910. Holmes, J. G. (1981). The exchange process in close relationships: Microbehavior and macromotives. In M. J. Lerner & S. C. Lerner (Eds.), The justice motive in social behavior (pp. 261–284). New York: Plenum Press. Holmes, J. G. (1991). Trust and the appraisal process in close relationships. In W. H. Jones & D. Perlman (Eds.), Advances in personal relationships (Vol. 2, pp. 57–104). London: Jessica Kingsley. House, J. S., Landis, K. R., & Umberson, D. (1988, July 29). Social relationships and health. Science, 241, 540 –545. Kiecolt-Glaser, J. K., & Newton, T. L. (2001). Marriage and health: His and hers. Psychological Bulletin, 127, 472–503. Kuhl, J. (1981). Motivational and functional helplessness: The moderating effects of state versus action orientation. Journal of Personality and Social Psychology, 40, 155–170. Leary, M. (1990). Responses to social exclusion: Social anxiety, jealousy, loneliness, depression, and low self-esteem. Journal of Social and Clinical Psychology, 9, 221–229. Lydon, J. E., Jamieson, D. W., & Holmes, J. G. (1997). The meaning of social interactions in the transition from acquaintanceship to friendship. Journal of Personality and Social Psychology, 73, 536 –548. Lynch, J. J. (1979). The broken heart: The medical consequences of loneliness. New York: Basic Books. Maslow, A. H. (1962). Toward a psychology of being. New York: Van Nostrand. McNair, D. M., Lorr, M., & Droppleman, L. F. (1971). Profile of Mood States. San Diego, CA: Educational and Industrial Testing Services. Mikulincer, M. (1998). Attachment working models and the sense of trust: An exploration of interaction goals and affect regulation. Journal of Personality and Social Psychology, 74, 1209 –1224. Mikulincer, M., Birnbaum, G., Woddis, D., & Nachmias, O. (2000). Stress and accessibility of proximity-related thoughts: Exploring the normative and intraindividual components of attachment theory. Journal of Personality and Social Psychology, 78, 509 –523.

Mikulincer, M., Dolev, T., & Shaver, P. R. (2004). Attachment-related strategies during thought suppression: Ironic rebounds and vulnerable self-representations. Journal of Personality and Social Psychology, 87, 940 –956. Mikulincer, M., & Nachshon, O. (1991). Attachment styles and patterns of self-disclosure. Journal of Personality and Social Psychology, 61, 321– 331. Mikulincer, M., & Selinger, M. (2001). The interplay between attachment and affiliation systems in adolescents’ same-sex friendships: The role of attachment style. Journal of Social and Personal Relationships, 18, 81–106. 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. Pierce, T., & Lydon, J. E. (2001). Global and specific relational models in the experience of social interactions. Journal of Personality and Social Psychology, 80, 613– 631. Pietromonaco, P. R., & Feldman Barrett, L. (2000). The internal working models concept: What do we really know about the self in relation to others? Review of General Psychology, 4, 155–175. Simpson, J. A., & Rholes, W. S. (1998). Attachment in adulthood. In J. A. Simpson & W. S. Rholes (Eds.), Attachment theory and close relationships (pp. 3–21). New York: Guilford Press. Spence, J. T., Helmreich, R. L., & Holahan, C. K. (1979). Negative and positive components of psychological masculinity and femininity and their relationships to self-reports of neurotic and acting out behaviors. Journal of Personality and Social Psychology, 37, 1673–1682. Sullivan, H. S. (1953). The interpersonal theory of psychiatry. New York: Norton. Trout, D. L. (1980). The role of social isolation in suicide. Suicide and Life-Threatening Behavior, 10, 10 –23. Vorauer, J. D., Cameron, J. J., Holmes, J. G., & Pearce, D. G. (2003). Invisible overtures: Fears of rejection and the signal amplification bias. Journal of Personality and Social Psychology, 84, 793– 812.

Received September 21, 2004 Revision received October 26, 2005 Accepted October 28, 2005 䡲

Instructions to Authors For Instructions to Authors, please visit www.apa.org/journals/psp and click on the “Instructions to Authors” link in the Journal Info box on the right.

Journal of Personality and Social Psychology 2006, Vol. 91, No. 1, 97–110

Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.97

Intergroup Helping as Status Relations: Effects of Status Stability, Identification, and Type of Help on Receptivity to High-Status Group’s Help Arie Nadler

Samer Halabi

Tel Aviv University

University of Haifa

Integrating research on social identity processes and helping relations, the authors proposed that low-status group members who are high identifiers will be unwilling to receive help from the high-status group when status relations are perceived as unstable and help is dependency-oriented. The first experiment, a minimal group experiment, found negative reactions to help from a high-status outgroup when status relations were unstable. The 2nd and 3rd experiments, which used real groups of Israeli Arabs and Israeli Jews, replicated this finding and showed that high identifiers were less receptive to help from the high-status outgroup than low identifiers. The 4th experiment, a help-seeking experiment with real groups of competing high schools, found that the least amount of help was sought from a high-status group by high identifiers when status relations were perceived as unstable and help was dependencyoriented. Theoretical and applied implications are discussed. Keywords: reactions to help, help-seeking, perceived status stability, ingroup identification, dependency-/ autonomy-oriented help

In his classic essay The Gift, anthropologist Marcel Mauss (1907/1957) described the tribal custom of potlatch, in which a clan leader engages in lavish displays of gift-giving to other clan leaders. Mauss noted that the “motives for such excessive gifts . . . are in no way disinterested. . . . To give is to show one’s superiority” (p. 72). This is group-based behavior in that tribal leaders establish their hierarchical position “to the ultimate benefit . . . of their own clan” (italics added, p. 4). For recipients of potlatch, “to accept without returning . . . [is] to become client and subservient” (p. 72). Phrased in the language of modern social psychology, this suggests that helping relations can be mechanisms through which groups create, maintain, and change status relations. The present research examined the willingness of low-status group members to seek and receive help from a high-status outgroup as affected by the perceived stability of intergroup status relations, the ingroup identification of the recipient, and type of help (i.e., autonomy- or dependency-oriented help). The research hypotheses follow the model of intergroup helping as status relations (Nadler, 2002), which draws on social identity research (Tajfel & Turner, 1986) and helping relations (Bierhoff, 2002; Penner, Dovidio, Piliavin, & Schroeder, 2005; Schroeder, Penner, Dovidio, & Piliavin, 1995). The present research represents the first empirical effort to assess the major tenets of this model.

SOCIAL IDENTITY PERSPECTIVE Over the last three decades, the social identity perspective on intergroup relations (Turner & Reynolds, 2001) has progressed in two complementary lines of research and theory: social identity theory (Tajfel & Turner, 1979, 1986) and self-categorization theory (Tajfel, 1978; Turner, Hogg, Oakes, Reicher, & Wetherell, 1987). Self-categorization theory holds that people’s identity fluctuates across a continuum ranging from individual (e.g., “I am a generous person”) to social (e.g., “I am a fan of a particular soccer team”) identity, and research has explored variables that affect the movement between these two poles and its consequences (e.g., Onorato & Turner, 2004; Roccas, 2003). Social identity theory originated with the pioneering work of Tajfel and his colleagues, who argued that in their quest for positive identity, group members positively distinguish themselves from outgroups by discriminating against them (i.e., Tajfel & Turner, 1979). In support of this, research has demonstrated that the division of people into groups, even on a relatively insignificant dimension (e.g., as specific vs. global perceivers), is sufficient to produce ingroup favoritism (i.e., discrimination against that outgroup) and outgroup devaluation (e.g., Jetten, Spears, & Manstead, 1999). More recent research indicates that the phenomena of ingroup favoritism and outgroup devaluation are affected by the status of the ingroup and the outgroup (Bourhis & Gagnon, 2003). Whereas members of high-status groups show greater discrimination toward low- than equal-status groups (Sachdev & Bourhis, 1991), lowstatus groups sometimes exhibit outgroup favoritism: When they have internalized their low status, they favor the high-status outgroup on dimensions of comparison that are related to their status inferiority (Sachdev & Bourhis, 1987). The inconsistency that is created between outgroup favoritism and people’s general need for positive identity can be resolved by individual mobility (joining the high-status group), social creativity (e.g., reframing the intergroup

Arie Nadler, Department of Psychology, Tel Aviv University, Tel Aviv, Israel; Samer Halabi, Department of Psychology, University of Haifa, Haifa, Israel. This research was supported by the Argentina Chair for Research on the Social Psychology of Conflict and Cooperation, Tel-Aviv University. We thank Sonia Roccas for helpful comments on earlier versions of this article. Correspondence concerning this article should be addressed to Arie Nadler, Department of Psychology, Tel Aviv University, Ramat Aviv 69978, Israel. E-mail: [email protected] 97

98

NADLER AND HALABI

comparison in a way that favors the ingroup), or social competition (working to elevate the status of the ingroup through social change; Tajfel & Turner, 1986). Social competition is more likely when members of the low-status group perceive the existing status hierarchy as relatively unstable and illegitimate and intergroup boundaries as impermeable (Tajfel & Turner, 1986). When status relations are perceived as illegitimate and unstable, members of low-status groups are likely to view the existing social hierarchy as changeable, and because the impermeability of intergroup boundaries prevents individual mobility, they are expected to channel their motivation for social equality toward elevating ingroup status. Regarding the differential effects of perceived legitimacy and perceived stability, research has indicated that variations in levels of perceived stability have stronger effects on the behavior and perceptions of low-status group members toward the high-status outgroup (e.g., Ellemers, Wilke, & van Knippenberg, 1993; Mummendey, Klink, Mielke, Wenzel, & Blanz, 1999). Because the present focus is on behavior and perceptions directed at the highstatus outgroup helper, we centered our attention on the effects of perceived stability on receptivity to help from the high-status group.

HELPING RELATIONS AS POWER RELATIONS Helping relations are inherently unequal social relations. The helper has sufficient resources to confer on a recipient, who is dependent on the helper’s goodwill. This inequality makes receiving help a potentially self-threatening experience for the beneficiary (Nadler & Fisher, 1986). In this line, empirical research has found that when help is self-threatening, people respond negatively to its receipt (e.g., Nadler, 1987; Nadler & Fisher, 1986) and prefer to endure hardships rather than seek it (Nadler, 1991). Yet, this research has converged almost exclusively on interpersonal helping encounters. In an exception to this interpersonal focus, Schneider, Major, Luhtanen, and Crocker (1996) studied reactions to interracial help and found that African American participants who received assumptive help (i.e., unsolicited help) from a European American experienced lower self-esteem than African Americans who received assumptive help from a fellow African American. Similar findings were reported in a study that examined the reactions of Arab-Israelis to the receipt of help from a Jewish-Israeli as opposed to an Arab-Israeli helper (Halabi, 2003). The study found that Arab-Israeli recipients reported lower self-evaluations when they were helped by Jewish-Israelis (the dominant group in Israeli society) than when they were helped by Arab-Israelis (the less dominant group). Taken together, these studies suggest two implications. First, it seems that when there is a salient distinction between social groups, as is the case for African Americans and European Americans in the United States and Arabs and Jews in Israel, interpersonal helping encounters between members of these groups tend to be experienced as an intergroup interaction. In support of this, Suleiman (2004) noted that in Israeli society, interactions between Israeli Arabs and Israeli Jews are perceived by participants as intergroup encounters even when they take the form of interpersonal dialogue (Suleiman, 2004). Second, these studies may be interpreted to indicate that because dependency on the high-status outgroup reinforces its dominant position, help from a member of the dominant outgroup is threatening to recipients from the low-status group. This conclusion is echoed in discussions that note that being on the receiving end of affirmative

action programs can be a stigmatizing experience for its beneficiaries (Pratkanis & Turner, 1996; Steele, 1992). The autonomy or dependency orientation of help may also influence whether dependence on a more privileged outgroup will threaten the recipient’s social identity (Nadler, 1997, 1998). Dependency-oriented help consists of providing a full solution to the problem at hand and reflects the helper’s view that the needy cannot help themselves. When recipients agree that they cannot make it on their own, dependency-oriented help is consistent with their view of themselves and they may readily seek and accept it. However, when recipients believe they can succeed independently, dependency-oriented help is inconsistent with their view of themselves as capable actors. In this case, potential recipients of help are likely to reject offers of dependency-oriented help and refrain from seeking it. Autonomy-oriented help is partial and temporary (e.g., taking the form of instructions or hints) and reflects the helper’s view that given the appropriate tools, recipients can help themselves (Brickman et al., 1982). Autonomy-oriented help allows recipients to retain their independence despite their reliance on the more resourceful helper (Nadler, 1997, 1998). Therefore this type of help is not likely to clash with recipients’ view of themselves as capable and equal actors. Applied to the present research context, this suggests that dependency-oriented help, but not autonomy-oriented help, will be inconsistent with the lowstatus group’s motivation for equality. Because this motivation grows higher when status relations are perceived as unstable, members of low-status groups are expected to be reluctant to seek or receive dependency-oriented help under these conditions.

Intergroup Helping Relations as Status Relations Figure 1 is a schematic representation of intergroup helping as status relations (Nadler, 2002). It suggests two clusters of intergroup helping relations: (a) when status relations are perceived as stable and legitimate and (b) when they are viewed as unstable and illegitimate. In the first case, the high-status group is expected to try to maintain its social advantage by providing dependencyoriented help to the low-status group. Under these conditions, the low-status group is expected to be receptive to dependencyoriented help. When status relations are perceived as unstable and illegitimate, members of the high-status group view their privileged position as being threatened and are expected to try to reaffirm their social advantage through increased efforts to provide dependency-oriented help to the low-status group. Under these conditions, members of low-status groups—which should be motivated to gain equal status—are expected to react negatively to the receipt of dependency-oriented help from the high-status group and be willing to seek such help only when it is autonomy-oriented.

The Role of Ingroup Identification The model of intergroup helping does not account for the effects of individual variation between group members. Yet, research on social identity shows that not all ingroup members respond to threats to ingroup identity the same way. These responses are affected by individuals’ identification with their group. Members who identify strongly with the ingroup are more defensive when ingroup identity is threatened. They express stronger identification with the “threatened” ingroup, more ingroup fa-

INTERGROUP HELPING RELATIONS

99

Figure 1. Intergroup helping relations as affected by perceived legitimacy and stability of power relations between groups. From “Inter-group helping relations as power relations: Maintaining or challenging social dominance between groups through helping,” by A. Nadler, 2002, Journal of Social Issues, 58, 487–502. Copyright 2002 by Blackwell Publishing. Reprinted with permission.

voritism, and increased stereotyping against the outgroup, whereas low identifiers lower their identification with the ingroup (Ellemers, Spears, & Doosje, 1999). Also, threat to social identity increases the perceptions, particularly among those who identify highly with their group (Doosje, Spears, Ellemers, & Koomen, 1999), of the homogeneity of members of the ingroup and of the outgroup (Rothgerber, 1997). Applied to the present context of intergroup helping, we expect the reluctance of low-status group members to receive or seek dependencyoriented help from the high-status group to be more characteristic of high ingroup identifiers than low identifiers. We do not expect similar differences when help is autonomy-oriented.

PRESENT RESEARCH AND HYPOTHESES The present studies focused on the reactions and behaviors of low-status group members to offers of help from the high-status group as affected by the perceived stability of intergroup status relations, their degree of ingroup identification, and the dependency- or autonomy-oriented nature of help. The first study was a minimal group experiment and assessed the link between the perceived stability of status relations and reactions to receiving help from a high-status group. The second experiment examined the same link with real groups (i.e., Israeli Jews and Israeli Arabs). The third experiment used the same ArabIsraeli and Jewish-Israeli intergroup context to examine the effects of ingroup identification on the reactions of low-status

group members to help from the high-status outgroup. The fourth experiment used a different real group context (two competing high schools) to examine help-seeking behavior as affected by the perceived stability of intergroup status relations, degree of ingroup identification, and the autonomy- versus dependency-oriented nature of help. Our main predictions rest on our theorizing that when status relations are perceived as relatively unstable, dependence on the high-status outgroup is inconsistent with group members’ quest for equality and results in a threat to social identity. This threat should be expressed in relatively low affect, drive group members to positively distinguish the ingroup by discriminating against and devaluing the outgroup, and perceive the ingroup and the outgroup as more homogeneous (Studies 1 and 2). These reactions to help will be more characteristic of high than low ingroup identifiers (Study 3). Finally, high identifiers are expected to be the least willing to seek needed help from the high-status outgroup when status relations are perceived as unstable and dependency-oriented help is offered (Study 4).

STUDY 1 Method Participants and Design The first experiment used the minimal group paradigm and consisted of a 2 (help vs. no help) ⫻ 2 (stable vs. unstable status relations) between-

NADLER AND HALABI

100

participants experimental design. Sixty-seven Israeli undergraduate students ages 19 to 24 years (44 female and 23 male, equally distributed across the four experimental cells) participated in the experiment.

interaction. Following this, participants were fully debriefed and the experiment ended.1

Manipulation Checks Procedure Participants were seated in individual booths and were told that they were participating in a study of the links between perceptual style, integrative thinking abilities, and social decision making. We informed participants that people usually fall into either of two perceptual style categories, global perceivers or specific perceivers, and that global perceivers have higher integrative abilities. We then had participants take the dot estimation task (Jetten, Spears, & Manstead, 1996), which would ostensibly determine which perceptual style they belonged to. On completing the task, all participants learned that they were specific perceivers (i.e., that they belonged to the group with lower integrative abilities).

Independent Variable Manipulation Stability of status relations. In the stable status condition, participants were told that the designation of individuals as specific or global perceivers tends to remain consistent across multiple administrations of the dot estimation task. In the unstable status condition, participants were informed that the distinction between the two groups is somewhat inconsistent and may change from test to test (for a similar manipulation of perceived status stability, see Boen & Vanbeselaere, 2000; Turner & Brown, 1978). Help versus no help. Participants were told that in the next part of the experiment they would work in pairs to complete a task requiring integrative ability and that each pair would include one specific and one global perceiver; the other pair member was said to be working in another room. Each participant would be asked to solve 20 anagrams, and participants were informed that students like themselves could usually solve 12 of these 20 anagrams in the allotted time. However, they had to solve at least 10 anagrams to proceed to the next phase of the experiment. To make the second phase of the experiment seem more appealing, we told participants that those who made it to the next phase would have the chance of winning a large monetary prize. Following these instructions, the 20 anagrams appeared consecutively on the computer screen. Each anagram was presented for 5 s. On the basis of a pilot test, we predetermined that 8 of the anagrams were easily soluble and the other 12 extremely difficult. At the end of the anagram test, participants learned that they had solved 8 anagrams correctly, whereas the outgroup member (i.e., a global perceiver) had solved 14 anagrams correctly. At this point a message appeared on the screen stating that to compensate for decrements in performance due to time pressure, participants would be given 2 more minutes to work on the anagrams they had not been able to solve. We told the participants that because students in past sessions had asked to communicate with their pair member, they could now do so by writing a message that the experimenter would deliver to the other participant. The experimenter then left the room and returned, 2 min later, with an envelope containing a message from the other participant (the help condition). The message stated: “It seems that you’re having some difficulty—let me help.” The note included the solution to 4 of the difficult anagrams. In the no-help condition, participants were exposed to the same information but did not receive a message from the other participant. At this point, before they could write a message of their own, participants were asked to respond to a number of questions. These included the dependent measures (measure of affect, ingroup favoritism, evaluation of the outgroup, and perceived homogeneity of the ingroup and outgroup). Participants were asked “how you experienced participation in the experiment” by choosing one of three possibilities: “The experiment feels like an interaction between (1) groups, (2) representatives of two groups, or (3) two individual students.” This allowed us to ascertain that participants perceived the social interaction within the experiment as an intergroup

Perceived stability of status relations. Participants were asked to name the group they belonged to (i.e., global or specific perceivers) and rate the degree to which they thought that the differences between global and specific perceivers would remain constant throughout the experiment, on a 7-point scale ranging from 1 (not at all constant) to 7 (very constant). Help versus no-help manipulation. At the beginning of the debriefing, participants were asked to report if they had received any information from their partner (i.e., the outgroup member) and to describe the nature of this communication.

Dependent Measures We assessed recipients’ reactions to receiving help in terms of affect, behavior (ingroup favoritism), and attitudes (evaluation of the outgroup and perceived homogeneity). Affect. Participants were asked to rate “How I feel now” on nine bipolar adjectives: good/bad, happy/sad, negative/positive, strong/weak, calm/nervous, angry/not angry, satisfied/dissatisfied, secure/insecure, and successful/unsuccessful. Because of the high internal consistency between items (Cronbach’s ␣ ⫽ .89), ratings were summed to obtain a single measure of affect, with higher scores indicating higher affect. Ingroup favoritism. As noted previously, the experiment was presented as assessing the links among perceptual style, integrative thinking abilities, and social decision making. It was indicated to participants that as part of a social decision-making task, they were now asked to divide a hypothetical sum of 3,000 New Israeli Shekels (NIS; equivalent to roughly $700) between a member of “your group—that is, a specific perceiver—and a person who belongs to the group of global perceivers.” Participants were to choose one of seven alternatives: three representing ingroup favoritism (1,800/1,200 NIS, 1,700/1,300 NIS, and 1,600/1,400 NIS for the ingroup and outgroup, respectively), one representing equal division (1,500 NIS to each group), and three representing outgroup favoritism (1,200/1,800 NIS, 1,300/1,700 NIS, and 1,400/1,600 NIS to the ingroup and outgroup, respectively; see Federico, 1998, for a similar assessment of ingroup favoritism). Evaluation of the outgroup. This was measured on six 7-point bipolar adjectives. Although the internal consistency for these items was relatively high (Cronbach’s ␣ ⫽ .82), it increased when ratings for the aggressive/ nonaggressive item were deleted (to ␣ ⫽ .89). This, together with previous findings that reactions to assumptive help from a high-status outgroup are characterized by feelings of hostility toward the helper (Schneider et al., 1996), suggested that the ratings of the outgroup on aggressiveness represented a relatively different dimension than the overall evaluation. This was borne out in a factor analysis, using varimax rotation, which yielded two distinct factors: The first, accounting for 58% of the variance, was a general evaluation factor (i.e., ratings on honest/dishonest, high abilities/ low abilities, successful/unsuccessful, trustworthy/untrustworthy, and dependable/not dependable), and the second, accounting for 18% of the

1

Of the participants, 21% (13 out of 62) wrote back a note to the other participant (i.e., the outgroup member). Six were in the stable status/help cell, 3 in the unstable status/help cell, and 4 in the stable status/no-help cell. This small number does not allow for statistical analyses, yet the contents of the messages are revealing and are consistent with our theoretical logic. All 6 messages in the stable status/help cell consisted of hopeful expectations for future cooperation, words of thanks, and the like. The 4 messages in the stable status/no-help condition were all informative requests (e.g., “Did you do well?” and “Was it hard?”). All 3 messages in the unstable status/help condition included an element of resentment and were a variation on “Thanks, but I can handle it myself.”

INTERGROUP HELPING RELATIONS variance, consisted of ratings of the outgroup ␮s aggressiveness (i.e., aggressive/nonaggressive). Higher scores on the first factor denote more favorable evaluation of the outgroup, and higher scores on the aggressiveness item denote more perceived aggressiveness. Perceived homogeneity of ingroup and outgroup. Perceived homogeneity was measured by asking participants to indicate the degree to which specific perceivers were similar to each other (ingroup homogeneity) and the degree to which global perceivers were similar to each other (outgroup homogeneity). Both scales ranged from 1 (not at all similar to each other) to 7 (very similar to each other; see Ellemers, Spears, & Doosje, 1997, for a similar assessment).

Results All of the participants correctly identified themselves as members of the specific-perceivers group. Five of the 67 participants perceived the interaction as an interpersonal interaction. They were evenly distributed across the four experimental cells (2 in one cell and 1 in each of the other three cells). The analyses reported here are based on the results of all 67 participants.2 Analyses revealed no significant main effects or interactions involving gender, and it was therefore not included as a factor in subsequent analyses in this research.

Manipulation Checks Status Manipulation Participants in the stable status condition, more than participants in the unstable status condition, thought that the differences between the two groups would remain the same throughout the experiment; means were 4.60 and 3.50, respectively, F(1, 65) ⫽ 8.07, p ⬍ .01.

Help Manipulation All participants in the help condition correctly remembered that they had received a communication from their partner and that this communication contained help.

Dependent Measures Measure of Affect A 2 (help vs. no help) ⫻ 2 (stable vs. unstable status) analysis of variance (ANOVA) yielded a main effect for status stability, F(1, 63) ⫽ 20.08, p ⬍ .001, which is qualified by a significant Stability ⫻ Help interaction, F(1, 63) ⫽ 3.80, p ⬍ .05. This interaction reflected the finding that participants who had received help in the unstable status condition had lower ratings of affect than those who had received help in the stable status condition; means were 3.20 and 4.58, respectively, t(63) ⫽ 4.12, p ⬍ .001. The comparable difference in the no-help condition was not significant; means were 4.10 and 4.80, respectively.

Ingroup Favoritism A 2 (help vs. no help) ⫻ 2 (stable vs. unstable status) ANOVA on the measure of ingroup favoritism revealed no significant effects. Although the predicted Help ⫻ Stability interaction was not significant, F(1, 63) ⬍ 1, we tested the a priori hypotheses by comparing the mean score of ingroup favoritism in the unstable status/help cell to the stable status/help cell. This comparison

101

revealed that in line with our hypothesis, participants in the unstable status/help cell tended to be more discriminatory toward the outgroup than those in the stable status/help cell; means were 3.52 and 2.81, respectively, F(1, 31) ⫽ 3.70, p ⬍ .07. The comparable difference in the no-help condition was not significant; means were 3.10 and 2.70, respectively, F(1, 32) ⬍ 1. Another approach to assessing the experimental hypothesis was to compare the percentages of ingroup favoritism choices. We collapsed the ingroup favoritism index, which assessed the magnitude of discrimination on a 7-point scale, into the three categories of ingroup favoritism, equal division, and outgroup favoritism. Because no participant made an outgroup favoritism choice, participants’ allocation decisions were compared between the two categories of ingroup favoritism and equal division. As predicted, participants in the help/unstable status cell made significantly more ingroup favoritism choices than those in the help/stable status cell; average percentages were 62.5% and 23.5%, respectively, ␹2(1, N ⫽ 33) ⫽ 5.10, p ⬍ .05. The equivalent comparison in the no-help condition was not significant; average percentages were 47% and 30%, respectively, ␹2(1, N ⫽ 34) ⫽ 1.12, ns.

Evaluation of the Outgroup A 2 (help vs. no help) ⫻ 2 (stable vs. unstable status) ANOVA on the general evaluation score indicated that when status relations were perceived as unstable, the outgroup was evaluated less positively than when they were perceived as stable; means were 4.30 and 5.00, respectively, F(1, 63) ⫽ 11.54, p ⬍ .001. Also, participants who had received help rated the outgroup less favorably than those who had not received help; means were 4.40 and 4.80, respectively, F(1, 63) ⫽ 4.70, p ⬍ .05. The Stability ⫻ Help interaction was not significant, F(1, 63) ⬍ 1. A 2 (help vs. no help) ⫻ 2 (stable vs. unstable status) ANOVA on perceived aggressiveness of the outgroup revealed a significant interaction, F(1, 63) ⫽ 3.70, p ⬍ .05. Participants who had received help viewed the outgroup as more aggressive when status relations were perceived as unstable than when they were perceived as stable; means were 3.60 and 4.50, respectively, t(63) ⫽ 2.12, p ⬍ .05. The difference between the stable and unstable status cells in the no-help condition was nonsignificant (means were 4.60 and 4.40, respectively).

Perceived Ingroup and Outgroup Homogeneity A 2 (stable vs. unstable status) ⫻ 2 (help vs. no help) ANOVA on the item assessing perceived ingroup homogeneity revealed no significant effects. A similar ANOVA on the perceived homogeneity of the outgroup revealed a significant Status Stability ⫻ Help interaction, F(1, 63) ⫽ 8.27, p ⬍ .005. This interaction derived from the finding that participants in the unstable status conditions who had received help from the outgroup perceived the outgroup as more homogeneous than those who had received help in the stable status conditions; means were 4.70 and 3.00, respectively, t(63) ⫽ 3.77, p ⬍ .001. The comparable difference was not significant in the no-help condition (means were 3.50 and 3.60, respectively; see Table 1). 2 The main analyses with and without these 5 participants revealed similar patterns of findings.

NADLER AND HALABI

102

Table 1 Means (and Standard Deviations) of Affect Scores, Percentage of Ingroup Favoritism Choices, Perceived Aggressiveness of Outgroup, and Perceived Outgroup Homogeneity as a Function of Perceived Status Stability and Help (N ⫽ 62): Study 1 Experimental condition Help Dependent measure Affect Percentage of discriminatory choices Perceived aggressiveness of outgroup Perceived outgroup homogeneity

No help

Unstable

Stable

Unstable

Stable

3.20 (0.55)

4.58 (1.01)

4.10 (0.81)

4.80 (1.20)

62.5% 3.60 (1.20) 4.70 (1.06)

23.5% 4.50 (0.99) 3.00 (1.50)

47.0% 4.60 (0.99) 3.50 (1.06)

30.0% 4.40 (1.50) 3.60 (1.41)

Note. Lower scores on the measure of perceived aggressiveness denote greater level of perceived aggressiveness.

Discussion The results of the first experiment support the central hypothesis. When the status hierarchy was perceived as relatively stable, the receipt of help from the high-status outgroup did not influence recipients’ affect, ingroup favoritism, and perceptions of the outgroup. Yet when the status hierarchy was perceived as unstable, being helped by a member of the high-status outgroup led recipients to feel worse. This relatively negative affect may reflect the greater threat to social identity under these conditions, leading to efforts for positive ingroup distinctiveness (expressed in more discrimination toward the outgroup in the unstable/help cell than the stable/help cell). The parallel comparison in the no-help condition was not significant. The findings for perceived aggressiveness indicate that the outgroup was perceived as most aggressive by participants who had perceived status relations as unstable and had received help from the high-status outgroup. This finding is conceptually important and suggests that when help thwarted the low-status group members’ motivation for equality (i.e., status relations were perceived as unstable), recipients viewed the helper as “forcing” his or her generosity on them. Finally, consistent with research that threat to social identity leads to viewing the source of threat as more homogeneous (Doosje et al., 1999), participants in the unstable status condition who had received help viewed the outgroup as more homogeneous than those who had not received such help. The conclusions from the first experiment are limited by the nature of the help that was studied and the type of groups that were used. Regarding the nature of help, the participants received help without having asked for it (i.e., assumptive help; Schneider et al., 1996). This raises the possibility that help from a high-status group when status relations are perceived as unstable poses a threat to the recipient’s social identity only when it is imposed by the highstatus helper. We address this issue in the fourth experiment and discuss it more extensively in the General Discussion section. Concerning the type of groups used, this experiment used ad hoc, experimentally created groups. While this reliance on the minimal group paradigm increases confidence in the internal validity of the observed phenomena and places this research within

the empirical and theoretical tradition of the social identity perspective, it limits the generalizability of our findings. Several authors suggest that there are differences between empirical relationships observed with ad hoc groups in the laboratory and similar relationships in studies that examine real groups (Jetten et al., 1996; Mullen, Brown, & Smith, 1992). To address this possibility and replicate the findings, the second experiment assessed the same hypothesis with real groups. It examined the reactions of Arab-Israeli participants to the receipt of help from an Arab-Israeli or a Jewish-Israeli helper (i.e., lower and higher status groups in Israeli society, respectively) under varying degrees of perceived stability of intergroup status relations. To extend generalizability, we used the same theoretical logic but changed the focus of the empirical comparison. Whereas the first experiment focused on a comparison between the help and no-help conditions, the second experiment focuses on comparing reactions to receiving help from a high-status outgroup versus help from an ingroup helper under varying levels of perceived stability of status relations. We predict that consistent with the theoretical model (Nadler, 2002) and the findings of the first experiment, ArabIsraelis’ social identity will be most threatened when they receive help from a Jewish-Israeli helper and status relations are perceived as unstable. We expect this high threat to be reflected in depressed affect and to result in high motivation for positive ingroup distinctiveness, which will be expressed by increased ingroup favoritism and devaluation of the outgroup.

STUDY 2 Method Participants and Design Participants were 71 Arab-Israeli high school students in a school in northern Israel, ages 16 and 17. The experiment was conducted on school premises and consisted of a 2 (Arab-Israeli vs. Jewish-Israeli helper) ⫻ 2 (stable vs. unstable status hierarchy) between-participants design. The proportion of male to female participants in each of the experimental cells was the same.

Procedure Participants took part in the experiment in small groups of 5– 6 individuals who were seated in individual booths. The experiment was portrayed as an assessment of the validity of the Israeli psychometric tests, which are used as entrance examinations for institutions of higher learning in Israel. The test was said to include an assessment of verbal, quantitative, and social skills. The research was said to be sponsored by the National Testing Center in Israel and conducted nationwide. After making these short introductory comments, the experimenter excused himself, saying that he must attend a prescheduled meeting with the school’s principal and that his coworker would administer the study.

Independent Variable Manipulations Stability of status relations. Participants were asked to read an information page on psychometric testing, which was presented as part of the measurement of verbal skills. The information page ended with a statement on the achievements of different subgroups in Israeli society on psychometric tests and noted that the average scores of Arab-Israeli students were lower than the average scores of Jewish-Israelis. In the stable status condition, the text stated that this gap had remained constant over the years, and in the unstable status condition, the text stated that the gap was

INTERGROUP HELPING RELATIONS

103

consistently narrowing. Because it is commonly known that members of disadvantaged groups within Israeli society, including many Israeli Arabs, score lower on these tests than members of advantaged groups, this information did not constitute a deception. Participants were then asked to answer a few questions ostensibly assessing their understanding of the material they had just read. One of these questions comprised the manipulation check on the stability of status relations. Group affiliation of helper and help-giving. The experimenter introduced himself, by using a different name, to half of the participants as an Arab-Israeli and to the other half as a Jewish-Israeli. Participants were asked to work on a sample of 12 psychometric problems, 6 of which were extremely difficult and 6 of which were readily soluble. Participants were given 15 min to solve all 12 problems and had been told that students can usually do so in about 10 min. Six minutes into their work, the experimenter, walking between the booths, discreetly pointed to the correct answer on 4 difficult problems (i.e., provided assumptive help). Subsequently, participants were asked to fill out the dependent measures. The experimenter was blind to the stability condition to which participants belonged.

sociology student who was investigating how different groups in Israel perceived each other. They all agreed and were asked to evaluate Israeli Jews on seven 7-point bipolar adjective scales: wise/stupid, honest/dishonest, good/bad, devious/trustworthy, truthful/deceitful, egoistic/altruist, and materialistic/nonmaterialistic. The average of these items was used as an index of outgroup evaluation (Cronbach’s ␣ ⫽ .91).

Manipulation Checks

Group Affiliation of Helper

Stability of status relations. Following the status stability manipulation, participants were asked to answer a number of questions on the material they had just read. One of the questions asked them to rate their agreement, on a 7-point scale ranging from 1 (strongly disagree) to 7 (strongly agree), with the statement that “I think that the gaps between Arab-Israelis and Jewish-Israelis in Israeli society have remained constant over the years.” Group affiliation of helper. Toward the end of the experiment, participants were asked to report their impressions of the session they had just participated in. These included questions asking them for the group affiliation of the test administrator and whether or not he had helped them.

Dependent Variables As in Study 1, here we also assessed affective, behavioral, and attitudinal reactions (i.e., affect, ingroup favoritism, and outgroup evaluation, respectively) to the receipt of help from the high-status outgroup, as influenced by different levels of perceived stability of status relations. Affect. Participants were asked to rate their affect on the same nine 7-point bipolar adjectives that were used in Study 1, and the average of these items was used as an affect score (Cronbach’s ␣ ⫽ .88). Ingroup favoritism. This measure was presented as assessing social decision making. Participants were asked to imagine that they had just learned that high schools in their area (i.e., the Haifa–Nazareth region in northern Israel) needed funds to operate a special educational program. The schools were arranged in four pairs of schools, each consisting of an Arab and a Jewish school (identified by name). Participants were asked to allocate funds by dividing 1,000 NIS between the two schools in each pair by choosing one out of seven possible allocation possibilities. Three represented ingroup favoritism (50/950 NIS, 200/800 NIS, or 350/650 NIS to the Jewish-Israeli and the Arab-Israeli school, respectively), one represented equal division (500 NIS to each school), and three represented outgroup favoritism (950/50 NIS, 800/200 NIS, or 650/350 NIS to the Jewish-Israeli and Arab-Israeli school, respectively). The correlations between choices made on each of these four pairs of Arab-Israeli/JewishIsraeli schools were significant and high (ranging from .57 to .77). Ingroup favoritism scores were computed by subtracting the average amount allocated to the ingroup (an Arab-Israeli school) from the average amount allocated to the outgroup (a Jewish-Israeli school). Thus, positive scores reflect ingroup favoritism and negative scores reflect outgroup favoritism. The higher the score, the greater the amount of ingroup favoritism. Outgroup evaluation. At the end of the experiment, we asked participants if they would be willing to fill out a scale for a separate study by a

Results Manipulation Checks Perceived Stability of Status Relations Participants in the stable status condition perceived the status gap between Arabs and Jews in Israel as being more constant over the years than did participants in the unstable status condition; means were 6.05 and 3.30, respectively, t(69) ⫽ 2.26, p ⬍ .001.

All of the participants correctly remembered the group affiliation of the experimenter (i.e., the helper), and 97% (69 participants) also indicated that he had helped them.3

Dependent Measures Affect A 2 (Arab-Israeli vs. Jewish-Israeli helper) ⫻ 2 (stable vs. unstable status) ANOVA on affect scores revealed a main effect for helper’s group affiliation, F(1, 67) ⫽ 5.15, p ⬍ .05, which was qualified by a two-way interaction, F(1, 67) ⫽ 3.97, p ⬍ .05. The interaction was due to the finding that affect scores in the JewishIsraeli helper/unstable status condition were significantly lower than affect scores in the Jewish-Israeli helper/stable status condition; means were 4.26 and 5.48, respectively, t(67) ⫽ 3.70, p ⬍ .001. The comparable difference in the Arab-Israeli helper condition was not significant; means were 5.24 and 5.54, respectively, t(67) ⬍ 1.

Ingroup Favoritism A 2 (Arab-Israeli vs. Jewish-Israeli helper) ⫻ 2 (stable vs. unstable status) ANOVA revealed a significant two-way interaction, F(1, 67) ⫽ 7.27, p ⬍ .01. This interaction is due to the finding that participants who had received help from a Jewish-Israeli under conditions of an unstable status hierarchy displayed greater ingroup favoritism than those who had received help from a JewishIsraeli under conditions of a stable status hierarchy; means were 554.2 and 252.8, respectively, t(67) ⫽ 2.92, p ⬍ .01. When the helper was an Arab-Israeli, the ingroup favoritism scores were not significantly different when status relations were perceived as 3

The analyses on the ratings of all 71 participants yielded the same effects as analyses using only the 69 participants who correctly remembered having been helped.

NADLER AND HALABI

104

Table 2 Means (and Standard Deviations) of Affect Scores, Amount of Funds Allocated to the Ingroup, and Evaluation of the High-Status Outgroup as a Function of Helper’s Group Affiliation and Status Stability (N ⫽ 71): Study 2 Experimental condition Jewish-Israeli helper Dependent measure Affect Amount given to the ingroup Evaluation of the outgroup (Jewish-Israelis)

Arab-Israeli helper

Unstable

Stable

Unstable

Stable

4.26 (0.67) 554.20 (334.50) 2.75 (0.98)

5.48 (0.99) 252.80 (281.00) 4.33 (1.38)

5.24 (1.30) 295.60 (368.30) 4.32 (1.51)

5.54 (0.78) 200.00 (234.90) 4.55 (0.93)

unstable versus stable; means were 295.6 and 200.0, respectively, t(67) ⬍ 1.4

Outgroup Evaluation A 2 (Arab-Israeli vs. Jewish-Israeli helper) ⫻ 2 (stable vs. unstable status) ANOVA revealed a main effect for the helper’s group affiliation, F(1, 67) ⫽ 9.47, p ⬍ .01, which is qualified by an interaction between status stability and the helper’s group affiliation, F(1, 67) ⫽ 5.24, p ⬍ .05. The interaction is due to the finding that the evaluation of the outgroup (i.e., Israeli Jews) was lower when status relations were perceived as unstable than when they were perceived as stable; means were 2.75 and 4.33, respectively, t(67) ⫽ 3.80, p ⬍ .001. The parallel difference between the stable and unstable status conditions when the helper was an Arab-Israeli was not statistically significant; means were 4.32 and 4.55, respectively, t(67) ⬍ 1 (see Table 2).

Discussion The main findings of the second study are consistent with our predictions and the findings of the first study. The greater threat to social identity was evident on affective, behavioral, and attitudinal reactions to receiving help from the high-status outgroup. Participants who had been helped by the high-status outgroup helper (i.e., a Jewish-Israeli) felt worse under conditions of unstable status relations than participants who had received help from the high-status outgroup helper under conditions of stable status relations, and worse also than those who had been helped by an ingroup helper (i.e., an Arab-Israeli). The depressed affect in the unstable status/outgroup helper cell is consistent with the suggestion that these conditions present the greatest level of threat to social identity. Congruent with the idea that this threat will result in increased efforts to attain positive ingroup distinctiveness, participants in this cell displayed a higher level of discrimination against the outgroup and evaluated it less positively than participants who had received help from the high-status outgroup under the stable status conditions. Variations in the perceived stability of status relations did not affect the discrimination and evaluation of the outgroup by participants who had been helped by an ingroup member (i.e., an Arab-Israeli). There are two important differences between Studies 1 and 2: (a) The first study was a minimal group experiment and the second

was conducted with real groups, and (b) the first study tested the hypothesis by comparing responses between the stable and unstable status cells in the help and no-help conditions, whereas the second study tested the hypothesis by comparing between the stable and unstable cells in the ingroup and high-status outgroup helper conditions. The support for the major experimental hypothesis, given these differences, raises our confidence in the theoretical logic that underlies the phenomena under study. The findings of the second experiment, by themselves, are open to the alternative interpretation that these discrepancies reflect the difference between receiving help from an ingroup versus an outgroup helper, irrespective of the outgroup’s status. Yet, the agreement with the findings of the first experiment, and the conceptually consistent effects of the perceived stability of status relations, makes this alternative interpretation less plausible. Reactions to threat to social identity are determined by level of ingroup identification. In the face of threat to social identity, high identifiers attempt to positively distinguish their group from the source of threat (i.e., the outgroup) more so than low identifiers (Ellemers et al., 1999). Applied to the context of intergroup helping, this suggests that the threat posed by help from the high-status outgroup will be higher for high than low identifiers. High identifiers are expected to report worse affect when receiving help from a high-status outgroup and to discriminate against and devalue the source of threat (i.e., the outgroup helper) more so than those whose ingroup identification is relatively low. The third study examined this hypothesis.

4

Another index of ingroup favoritism is the value that is derived from the choice that participants make on the 7-point scale, where 1 represents extreme ingroup favoritism (i.e., 950 and 50 NIS to the ingroup and outgroup, respectively), 4 represents equality (i.e., 500 NIS to each), and 7 represents an extreme outgroup allocation choice (i.e., 50 and 950 NIS to the ingroup and outgroup, respectively). Decisions on the four allocation items were internally consistent (Cronbach’s ␣ ⫽ .87) and summed to obtain a single index of ingroup favoritism. A 2 (stable vs. unstable status) ⫻ 2 (Arab vs. Jewish helper) ANOVA revealed a significant interaction, F(1, 67) ⫽ 6.29, p ⬍ .01. The pattern of findings was similar to those obtained on the measure of average funds allocated to the ingroup, as reported in the Results section (i.e., the Jewish helper/unstable status cell had the highest ingroup favoritism score).

INTERGROUP HELPING RELATIONS

STUDY 3 Method Participants and Design Sixty-four high school students from an Arab-Israeli high school in northern Israel, ages 16 and 17, participated in a 2 (high identifiers vs. control) ⫻ 2 (ingroup vs. outgroup helper) between-participants experimental design. The proportion of male to female participants in each of the four experimental cells was the same.

Procedure The procedure was identical to the one used in Study 2. The experiment was presented as an investigation of three dimensions in psychometric tests: verbal, quantitative, and social skills. The first part was described as an assessment of verbal skills and consisted of an ingroup identification manipulation.

Independent Variable Manipulations Ingroup identification. In the high identification condition, participants were asked to read half a page on the contributions of Arab culture to humankind in medicine, mathematics, and the arts. In the control condition, participants read a neutral section. Following the reading, participants were asked to answer a few questions on the section they had just read. One of these questions was a check on the ingroup identification manipulation (see Jetten, Spears, & Manstead, 1997, for a similar manipulation). Group affiliation of helper and help-giving. The experimenter presented himself to half of the participants as an Israeli Jew and to the other half as an Israeli Arab. Subsequently, and in an identical manner to the procedures in the second experiment, participants began working on the quantitative part of the assessment. The experimenter provided assumptive help on 4 out of the 20 sequences of geometric problems that the participants were working on.

Manipulation Checks For the manipulation check on ingroup identification, participants were asked to rate their agreement with the statement “I am proud to be an Arab” on a 7-point scale. The manipulation check for group affiliation of helper was identical to Study 2.

Dependent Measures As in the first two experiments, we again assessed the affective, behavioral, and attitudinal reactions to receiving help from the high-status outgroup. Measures of affect, ingroup favoritism, and outgroup evaluation were identical to those used in Study 2.

Dependent Measures Affect A 2 ⫻ 2 ANOVA on the affect measure revealed a main effect for the group affiliation of the helper, F(1, 60) ⫽ 5.66, p ⬍ .05, due to the lower affect scores of participants who had been helped by a member of the outgroup than by an ingroup helper (means were 4.74 and 5.38, respectively). The two-way interaction failed to exceed an acceptable level of significance, F(1, 60) ⫽ 1.92, p ⬍ .17; however, to assess the a priori hypothesis that only in the high identification condition would recipients of help from the highstatus outgroup experience lower affect than recipients of help from an ingroup helper, we conducted planned comparisons between relevant cell means. Consistent with the hypothesis, participants in the high identification/Jewish-Israeli helper group had lower affect than participants in the high identification/ArabIsraeli helper cell; means were 4.34 and 5.39, respectively, t(60) ⫽ 2.70, p ⬍ .01. There was no significant difference between the affect scores in the Arab-Israeli and Jewish-Israeli helper cells in the control condition; means were 5.36 and 5.14, respectively, t(60) ⬍ 1.

Ingroup Favoritism A 2 (high identification vs. control) ⫻ 2 (Arab-Israeli vs. Jewish-Israeli helper) between-participants ANOVA revealed a significant two-way interaction, F(1, 60) ⫽ 4.09, p ⬍ .05. Planned comparisons between relevant cell means indicate that this interaction is due to the outcome that high identifiers who had received help from the high-status outgroup (i.e., from a Jewish-Israeli helper) discriminated against the outgroup more than high identifiers who had received help from an ingroup helper; means were 436.0 and 291.0, respectively, t(60) ⫽ 2.06, p ⬍ .05. The comparable difference for participants in the control condition was not significant; means were 230.0 and 318.0, respectively, t(60) ⬍ 1.5

Outgroup Evaluation A 2 ⫻ 2 ANOVA revealed a similar two-way interaction, F(1, 60) ⫽ 4.41, p ⬍ .05. Planned comparisons indicate that high identifiers who received help from the Jewish-Israeli helper rated the outgroup lower than did high identifiers who had received help from an Arab-Israeli helper; means were 3.68 and 4.62, respectively, t(60) ⫽ 2.02, p ⬍ .05. The parallel difference in the control condition was not significant; means were 4.55 and 4.20, respectively, t(60) ⬍ 1 (see Table 3).

Discussion

Results Manipulation Checks

105

Study 3 indicates that the threat to social identity inherent in receiving help from the high-status outgroup depends on ingroup

Ingroup Identification 5

The difference between the high identification and control conditions was highly significant, t(62) ⫽ 4.97, p ⬍ .0001 (means were 6.62 and 4.18, respectively).

Helper’s Group Affiliation All of the participants correctly remembered the group affiliation of the helper, and 98% reported that he had helped them.

Here, too, an index of ingroup favoritism was created by summing the choices made on the four allocation items (Cronbach’s ␣ ⫽ .79; i.e., scores ranging from 1 to 7, where 1 represents the highest level of ingroup favoritism). A 2 (high vs. low identification) ⫻ 2 (Jewish vs. Arab helper) ANOVA on these scores revealed a significant interaction, F(1, 59) ⫽ 3.97, p ⬍ .05. This interaction reflects a similar pattern of findings as those obtained on the average allocation of funds, as reported in the Results section (i.e., the highest ingroup favoritism score was found in the high identification/Jewish helper cell).

NADLER AND HALABI

106

Table 3 Means (and Standard Deviations) of Affect Scores, Amount of Funds Allocated to the Ingroup, and Evaluation of the High-Status Outgroup as a Function of Helper’s Group Affiliation and Ingroup Identification (N ⫽ 64): Study 3 Experimental condition Jewish-Israeli helper Dependent measure Affect Amount given to the ingroup Evaluation of the outgroup (Jewish-Israelis)

Arab-Israeli helper

High ingroup identification

Low ingroup identification

High ingroup identification

Low ingroup identification

4.34 (0.85) 435.90 (387.70) 3.68 (1.50)

5.14 (1.32) 229.70 (273.10) 4.55 (1.38)

5.39 (1.04) 290.60 (202.80) 4.20 (1.11)

5.36 (0.98) 318.70 (248.70) 4.62 (1.19)

identification. High Arab-Israeli identifiers had lower affect after receiving help from the dominant outgroup than after receiving help from an ingroup helper, but affect for participants in the control condition did not differ in these two cells. This lower affect suggests that receiving help from a high-status outgroup is more threatening for high identifiers than for participants in the control condition who had not undergone a manipulation to increase ingroup identification. Similar to the findings in the first two experiments, this greater threat to social identity led to greater efforts for positive ingroup distinctiveness, which found expression in more discrimination and devaluation of the source of threat (i.e., the high-status outgroup). It may be noted that the manipulation of ingroup identification (i.e., praising Arab contributions to global art and culture) might be construed as having led to perceptions of higher ingroup status. Yet, we maintain that the effects are attributable instead to increased identification. This interpretation is supported by the consistency of the findings with the results of Study 4, which are based on a different manipulation of identification (i.e., informing high school students that graduates of their school are committed to their school). The theoretical model guiding the present research program also predicts that low-status group members will avoid seeking help from the dominant outgroup when dependency on the high-status group poses a threat to social identity (Nadler, 2002). Taken together with the findings of the previous experiments, this model suggests that high identifiers will be most reluctant to seek help from the high-status outgroup when status relations are perceived as unstable. This reluctance is likely to be evident only when the assistance is dependency-oriented (i.e., a full solution is provided). When assistance is autonomy-oriented (i.e., consisting of a partial solution or hints), seeking help from the high-status outgroup is not likely to pose a threat for social identity, and high-identifying members of low-status groups are likely to seek such help even when perceived status relations are unstable. These predictions are the focus of the fourth study.

STUDY 4 The fourth experiment examined the willingness of low-status group members to seek autonomy- or dependency-oriented help from a high-status outgroup as a function of the interaction between (a) perceived stability of status relations and (b) ingroup

identification. Thus, it represents a full examination of the predictions of the intergroup helping model (Nadler, 2002), together with extensions suggested by the findings of Study 3 (i.e., the role of ingroup identification). This examination was carried out using a behavioral measure of willingness to seek help.

Method Participants and Design Fifty-six students from a high school in a midsize town in northern Israel participated in the study. The design was a 2 (high ingroup identification vs. control) ⫻ 2 (stable vs. unstable status relations) between-participants design. Each of the four experimental cells included a similar number of male and female students. The dependent measures consisted of the frequency that participants chose to (a) not seek help, (b) seek dependencyoriented help, and (c) seek autonomy-oriented help. As in Studies 2 and 3, this experiment also used real groups. However, unlike the previous experiments, which used national group affiliation, the present experiment used high school affiliation as the intergroup context.

Procedure During the first part of the experiment the experimenter, who had been introduced as an employee of the Israeli Ministry of Education, described the study as an assessment of alternative forms of psychometric testing. Participants were told that they would be working on tasks assessing verbal abilities and interactive analytical abilities. Following this general introduction, participants were told that the study was being carried out in different geographical areas in the country and that their school and another school had been chosen to represent northern Israel. The other school was said to be a prestigious and reputable school in the city of Haifa (i.e., a high-status outgroup).

Independent Variables Manipulation Stability of status relations. In the stable status condition, participants learned that a comparative analysis had revealed that over the last 5 years their school’s overall performance on various criteria was consistently lower than that of the higher status school (e.g., as measured by admission to selective university programs). In the unstable status condition, participants were told that over the last 5 years the gap between their school and the high-status school had grown consistently narrower. Ingroup identification. Following the manipulation of status stability, the manipulation of ingroup identification was introduced. In the guise of

INTERGROUP HELPING RELATIONS taking the verbal part of the psychometric test, participants were asked to read a short article that they would later be questioned on as part of the “verbal abilities” section of the psychometric test. In the high ingroup identification condition, the short article recounted the history of their high school, praising the school and noting the commitment that students and graduates felt for it. The article was said to have been taken from the local community newspaper. In the control condition, participants read an article of similar length that focused on environmental issues. After they had finished reading the articles, participants were asked to answer a few questions on what they had just read. This short questionnaire included checks on the status stability and ingroup identification manipulations.

Manipulation Checks Stability of status relations. Participants were asked to respond to two questions on a 7-point scale: the chances that a graduate of their high school and a graduate of the high-status school would be admitted to prestigious university departments, and how many years it would be before the scholastic achievements of the two schools would become relatively equal. Responses to these two items were significantly correlated (r ⫽ ⫺.55, p ⬍ .001), and ratings were summed to obtain a single index of perceived status stability. Ingroup identification. Participants answered two questions on a 7-point scale: the degree to which they had a sense of belonging to their school and whether they would recommend enrolling at their school to others. Answers to these two questions were highly correlated (r ⫽ .80, p ⬍ .001), and ratings were summed to obtain a single index of ingroup identification.

Dependent Measures: Avoidance of Help-Seeking, Dependency-Oriented Help-Seeking, and AutonomyOriented Help-Seeking Participants were told that the next part of the study would assess interactive analytic thinking. Participants were given five index cards, each of which displayed an arithmetic problem that had to be solved within 60 s. Two of the problems were readily soluble and were solved by all participants. The other three cards contained insoluble problems. After it became clear that participants could not solve the three problems, and in preparation for the upcoming “interactive” part of the session, the students were asked to indicate one of three choices for each problem they had been unable to solve: (a) not receiving any assistance from a student from the other high school, who was said to be working on the same task (i.e., avoidance of seeking help); (b) receiving the solution to the unsolved problem from the other high school student (i.e., seeking dependencyoriented help); or (c) receiving a hint from the other high school student that might help them find the solution on their own (i.e., seeking autonomyoriented help). The indices for avoidance of help-seeking, dependent help-seeking, and autonomous help-seeking consisted of the average number of times that participant chose either of these three alternatives. Thus, scores for either of these options could range from 0 (i.e., never choosing this option) to 3 (i.e., choosing that option on all three problems).

Results Manipulation Checks Perceived Stability of Status Relations Participants in the stable status group perceived the gap between the ingroup and the outgroup as more stable than did participants in the unstable status condition, F(1, 52) ⫽ 9.29, p ⬍ .01 (means were 3.11 and 2.04, respectively).

107

Ingroup Identification Participants in the high identification condition identified with their group more than participants in the control condition, F(1, 52) ⫽ 14.90, p ⬍ .001 (means were 6.01 and 4.77, respectively).

Dependent Measures Avoidance of Help-Seeking A 2 (stable vs. unstable status) ⫻ 2 (high identification vs. control) ANOVA on the avoidance of help-seeking scores revealed a significant Stability ⫻ Identification interaction, F(1, 52) ⫽ 6.15, p ⬍ .01. This indicates that for participants in the control condition, scores for avoidance of help-seeking did not differ between the stable and unstable status conditions; means were 0.67 and 0.50, respectively, t(52) ⫽ 0.80, ns, whereas high identifiers exhibited significantly higher avoidance when status relations were perceived as unstable than when they were perceived as stable; means were 1.31 and 0.36, respectively, t(52) ⫽ 2.40, p ⬍ .05.

Dependent Help-Seeking Consistent with our hypothesis, the incidence of help-seeking in the unstable status/high identification condition was zero, rendering the use of ANOVA impossible. To counter this problem, we conducted two t tests for independent samples in which equal variance is not assumed; these compared the amount of dependency-oriented help sought in the stable versus unstable status cells, within the high identification and control conditions separately. In the control condition, the amount of dependencyoriented help did not differ between the stable and unstable status cells; means were 1.20 and 1.43, respectively, t(26.51) ⫽ 0.52, ns. In the high identification condition, participants in the stable status cell sought more dependency-oriented help than those in the unstable status cell; means were 1.21 and 0.00, respectively, t(13.00) ⫽ 3.82, p ⬍ .01. Because students had to choose among three alternatives, there were two degrees of freedom on this measure. We expected significant Stability ⫻ Identification interaction on the avoidance index, and a similar interaction, but in an opposite direction, on the dependent help-seeking index. We expected, and found, that high identifiers will have the highest avoidance scores when status relations were unstable and lowest dependency help-seeking scores when relations were unstable. Under these conditions, the ANOVA on the third measure (i.e., autonomy-oriented helpseeking) could not have produced any significant effects. We therefore did not conduct an ANOVA on this index.

Discussion The findings of Study 4 support the hypotheses. Participants were most reluctant to seek help from a member of the high-status outgroup when they had been induced to identify with the ingroup and viewed intergroup status relations as unstable. It is important to note that this reluctance was evident only in the case of dependency-oriented help. In fact, not a single participant in the unstable status/high identification condition sought dependencyoriented help. We interpret this to reflect that under these conditions, seeking help conflicted with the motivation of high identifiers to elevate the status of their ingroup and attain equality with

NADLER AND HALABI

108

the outgroup. When assistance was framed as autonomy-oriented, high identifiers sought similar amounts of help from the outgroup as participants in the control condition, regardless of the stability of status relations. While the first three studies examined segments of the theoretical links suggested by the present analysis (i.e., the stability of power relations in Studies 1 and 2, and ingroup identification in Study 3), the fourth experiment assessed all three conceptual elements in tandem. It found support for the joint effects of (a) structural characteristics of intergroup power relations (i.e., perceived stability), (b) characteristics of the person in need of help (i.e., ingroup identification), and (c) the nature of the help (i.e., autonomy- or dependency-oriented) on help-seeking. Also, the study extends the generalizability of the earlier findings in two major aspects. First, it allows for the generalization of the theoretical principles from the realm of reactions to help (as in the first three experiments) to actual help-seeking behavior. Second, while Study 1 focused on experimentally created ad hoc groups and Studies 2 and 3 on relations between Israeli Arabs and Jews, Study 4 examined status differences between different high schools. The overall similarity in the pattern of findings across these different kind of groups and operationalizations reinforces our confidence that the phenomenon under study represents a basic phenomenon of intergroup relations.

GENERAL DISCUSSION The above four experiments support the hypotheses of the intergroup helping model (Nadler, 2002). Taken together, these studies indicate that members of low-status groups are least receptive to help from a high-status outgroup when status relations are perceived as unstable and help is dependency-oriented. Further, this unwillingness to seek or receive help from the high-status group is particularly characteristic of high ingroup identifiers. We interpret this to indicate that under the abovementioned conditions, dependency on a high-status helper is inconsistent with the motivation of the low-status group for social equality and that such help poses a threat to social identity. Although the overall empirical picture is consistent with this interpretation, the studies contain some methodological ambiguities that require discussion. In the second and fourth experiments, the perceived instability of status relations was induced by informing participants that the gap between their group and the highstatus group was narrowing over time. It is possible that such a manipulation could have resulted in the perception of higher ingroup status or may have led to greater identification with the ingroup. Our interpretation that these effects are attributable to the different perceptions of the stability of status relations is reinforced on a number of grounds. First, this manipulation is consistent with previous manipulations of perceived status stability and instability in this context (Boen & Vanbeselaere, 2000; Federico, 1998; Turner & Brown, 1978). Second, our manipulation checks showed that participants perceived the manipulation as intended. Third, the effects of status stability in Studies 2 and 4 are consistent with those in the first experiment, which was a minimal group experiment and in which a different manipulation of stability was used (i.e., the likelihood that participants would be assigned to the low-status group in the future). In social systems in which the power hierarchy is perceived as stable and legitimate, it is the privileged group’s duty to cater to

the needs of the low-status group. The low-status group’s receptiveness to such help serves as a behavioral acknowledgment of its inferior position. This is captured in Mauss’s (1907/1957) statement that by accepting gifts without returning them in kind, the recipients “become client and subservient” (p. 72). Such social situations are likely to have characterized relations between racial and gender groups in past centuries (e.g., African Americans and European Americans, males and females) and in societies in which the power structure is relatively rigid (such as tribal societies). It is interesting to note that such a unidirectional flow of assistance from the strong to the weak is also evident in nature, in which a less dominant member in the community of birds that was studied rarely gives food to the more dominant member of its group (Zahavi & Zahavi, 1997). This idea is consistent with the argument made by Worchel (1984) that helping may be a way for dominant groups to maintain their supremacy and with analyses of paternalism in gender and race relations (e.g., Jackman, 1994; Pratto & Walker, 2001; van der Berghe, 1967). In the same line, Leach, Snider, and Iyer (2002) argued that one of the consequences of feeling secure in one’s privileged social position is benevolence toward the disadvantaged. They wrote: “The security of the fortunates’ advantage allows a somewhat benevolent reaction to the disadvantaged in the form of pity” (p. 7). In the same way that receptivity to help from the high-status outgroup is a behavioral affirmation of the disadvantaged position of the low-status group, refusing help from the advantaged group may signify a challenge to the existing status quo. As our data indicate, this is more likely when status relations are perceived as unstable. The empirical support for our hypotheses is underscored by the differences that existed between the four experiments, as reflected in the types of groups studied, the status dimensions used, the comparisons used to assess the main hypotheses, and the dependent measures that were assessed. Regarding the types of groups studied, the first study was a minimal group experiment, whereas the other three experiments studied real groups. Regarding the dimension of status, the second and third experiments studied low-status groups within a given society (i.e., Arab citizens in Israel), whereas the fourth experiment focused on status differences between two high schools. The central comparisons across the four experiments also varied (i.e., receiving vs. not receiving help in Study 1, and reactions to receiving help from a high-status outgroup and ingroup helper in Studies 2 and 3). Furthermore, both help-seeking behavior and reactions to help were predicted by the same theoretical logic. In closing, it may be noted that future research should analyze the mediating role of relevant psychological motivations (e.g., positive ingroup distinctiveness) and emotions (e.g., feeling humiliated by dependency) on the observed relationships.

Direct or Indirect Means to Maintain or Challenge Intergroup Status Relations The logic guiding the present research has implications for both high-status groups and help-giving behavior. It suggests that highstatus groups may try to retain a jeopardized position of dominance by providing dependency-oriented help to the source of threat (i.e., the low-status group). This leads to an important question: When will a high-status group try to maintain dominance by providing help, and when will it do so by the more direct means of asserting its superiority? The same question also applies to the low-status

INTERGROUP HELPING RELATIONS

group, which at times may directly challenge existing inequality by confronting the high-status group with a demand for equal power, and at other times less directly by refusing dependency on the more dominant group. Although the direct and less-direct forms of maintaining intergroup power relations are not mutually exclusive, the question of when groups are more likely to use one or the other method remains open. One possible determinant is the social and normative context within which intergroup relations take place. If, for example, the social context is confrontational (e.g., groups are in open conflict), they may be likely to use more explicit and direct methods to maintain or challenge existing status relations. If, however, the social context discourages open confrontation, groups may be likely to use less explicit and confrontational means to maintain or challenge power relations. Helping relations are such means.

Intergroup and Interpersonal Helping Relations Although the present analysis focused on intergroup helping, we theorize that the same processes apply to the analysis of helping relations between differentially powerful individuals (e.g., managers and employees). Consistent with self-categorization theory, the analysis of the helping interaction will be informed by variables that are relevant to this level of analysis (e.g., ingroup identification) when interactants’ social identity is salient. However, when the interactants’ individual self is salient, the helping interaction may be construed as an interpersonal interaction, requiring consideration of variables that are relevant to this level of analysis (e.g., self-esteem). In both cases, however, the pattern of helping relations depends on the perceived stability and legitimacy of existing power relations between the parties and the dependency or autonomy orientation of the help.

Social Change and Helping Relations Times of social change give rise to motivational conflicts between more and less socially dominant groups, and this may be expressed in the framework of helping relations as well. Whereas a stronger party may seek to defend its social advantage by providing dependency-oriented help, the less dominant party may likely be motivated to refuse such help. This may fuel tension between the two parties. In this scenario, one can imagine the more dominant group feeling baffled and angry at the low-status group’s resistance to its generosity, and the less dominant party viewing the generosity of its counterpart as a manipulative effort to retain social advantage. These processes are described as operating in affirmative action programs (Pratkanis & Turner, 1996), and similar barriers have been underscored on the road to peace-building, wherein the stronger party’s efforts to assist its former adversary may be spurned out of hand as mechanisms to retain dominance (Lederach, 1997; Nadler & Saguy, 2004). The fact that helping is both an expression of caring and a demonstration of superiority makes it an especially effective instrument of dominance in the hands of a more advantaged group. As our data show, members of low-status groups are sensitive to this danger and under certain conditions resent such help and are unwilling to seek it.

References Bierhoff, H. (2002). Prosocial behavior. New York: Psychology Press. Boen, F., & Vanbeselaere, N. (2000). Responding to membership of a

109

low-status group: The effects of stability, permeability and individual ability. Group Processes and Intergroup Relations, 3, 41– 63. Bourhis, R. Y., & Gagnon, A. (2003). Social orientations in the minimal group paradigm. In R. Brown & S. Gaertner (Eds.), Intergroup processes (pp. 89 –112). Oxford, England: Blackwell. Brickman, P., Rabinowitz, V. C., Karuza, J., Coates, D., Cohn, E., & Kidder, L. (1982). Models of helping and coping. American Psychologist, 37, 368 –384. Doosje, B., Spears, R., Ellemers, N., & Koomen, W. (1999). Perceived group variability in intergroup relations: The distinctive role of social identity. European Review of Social Psychology, 10, 41–73. Ellemers, N., Spears, R., & Doosje, B. (1997). Sticking together or falling apart: In-group identification as a psychological determinant of group commitment versus individual mobility. Journal of Personality and Social Psychology, 72, 617– 626. Ellemers, N., Spears, R., & Doosje, B. (Eds.). (1999). Social identity: Context, commitment, content. Oxford, England: Blackwell. Ellemers, N., Wilke, H., & van Knippenberg, A. (1993). Effects of the legitimacy of low group or individual status on individual and collective identity enhancement strategies. Journal of Personality and Social Psychology, 64, 766 –778. Federico, C. (1998). The interactive effects of social dominance orientation, group status, and perceived stability on favoritism for high-status groups. Group Processes and Intergroup Relations, 2, 119 –143. Halabi, S., (2003). Reactions to receiving assumptive help in or out-group member: The moderating effects of task difficulty. Unpublished master’s thesis, Tel Aviv University, Israel. Jackman, M. R. (1994). The velvet glove: Paternalism and conflict in gender, class, and race relations. Berkeley: University of California Press. Jetten, J., Spears, R., & Manstead, A. S. R. (1996). Intergroup norms and intergroup discrimination: Distinctive self-categorization and social identity effects. Journal of Personality and Social Psychology, 71, 1222–1233. Jetten, J., Spears, R., & Manstead, A. S. R. (1997). Strength of identification and intergroup differentiation: The influence of group norms. European Journal of Social Psychology, 27, 603– 609. Jetten, J., Spears, R., & Manstead, A. S. R. (1999). Group distinctiveness and in-group discrimination. In N. Ellemers, R. Spears, & B. Doosje (Eds.), Social identity (pp. 107–126). Oxford, England: Blackwell. Leach, C. W., Snider, S. L., & Iyer, A. (2002). “Poisoning the conscience of the fortunate”: The experience of relative advantage and support for social equality. In I. Walker & H. J. Smith (Eds.), Relative deprivation: Specification, development and integration (pp. 1–14). New York: Cambridge University Press. Lederach, J. P. (1997). Building peace: Sustainable reconciliation in divided societies. Washington, DC: U.S. Institute of Peace. Mauss, M. (1954). The gift: Forms and functions of exchange in archaic societies. Glencoe, IL: Free Press. (Original work published 1907) Mullen, B., Brown, R., & Smith, C. (1992). Ingroup bias as a function of salience, relevance, and status: An integration. European Journal of Social Psychology, 22, 103–122. Mummendey, A., Klink, A., Mielke, R., Wenzel, M., & Blanz, M. (1999). Socio-structural characteristics of intergroup relations and identity management strategies: Results from a field study in East Germany. European Journal of Social Psychology, 29, 259 –287. Nadler, A. (1987). Determinants of help seeking behavior: The effects of helper’s similarity, task centrality and recipient’s self esteem. European Journal of Social Psychology, 17, 57– 67. Nadler, A. (1991). Help-seeking behavior: Psychological costs and instrumental benefits. In M. S. Clark (Ed.), Review of personality and social psychology (Vol. 12, pp. 290 –312). New York: Sage. Nadler, A. (1997). Autonomous and dependent help seeking: Personality characteristics and the seeking of help. In B. Sarason, I. Sarason, & R. G.

110

NADLER AND HALABI

Pierce (Eds.), Handbook of personality and social support (pp. 258 – 302). New York: Plenum Press. Nadler, A. (1998). Relationship, esteem and achievement perspectives on autonomous and dependent help seeking. In S. A. Karabenick (Ed.), Strategic help seeking: Implications for learning and teaching (pp. 61–95). Mahwah, NJ: Erlbaum. Nadler, A. (2002). Inter-group helping relations as power relations: Maintaining or challenging social dominance between groups through helping. Journal of Social Issues, 58, 487–502. Nadler, A., & Fisher, J. D. (1986). The role of threat to self-esteem and perceived control in recipient reactions to aid: Theory development and empirical validation. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 19, pp. 81–123). New York: Academic Press. Nadler, A., & Saguy, T. (2004). Trust building and reconciliation between adversarial groups: A social psychological perspective. In H. Langholtz & C. E. Stout (Eds.), The psychology of diplomacy (pp. 29 – 46). New York: Praeger. Onorato, R. S., & Turner, J. C. (2004). Fluidity in the self-concept: The shift from personal to social identity. European Journal of Social Psychology, 34, 257–278. Penner, L. A., Dovidio, J. F., Piliavin, J. A., & Schroeder, D. A. (2005). Prosocial behavior: Multilevel perspective. Annual Review of Psychology, 56, 365–392. Pratkanis, A. R., & Turner, M. E. (1996). The proactive removal of discriminatory barriers: Affirmative action as effective help. Journal of Social Issues, 52, 111–132. Pratto, F., & Walker, A. (2001). Dominance in disguise: Power, beneficence and exploitation in personal relationships. In A. Y. Lee-Chai & J. A. Bargh (Eds.), The use and abuse of power: Multiple perspectives on the causes of corruption (pp. 93–112). Washington, DC: Psychology Press. Roccas, S. (2003). Identification and status revisited: The moderating role of self-enhancement and self-transcendence values. Personality and Social Psychology Bulletin, 29, 726 –736. Rothgerber, H. (1997). External intergroup threat as an antecedent to perceptions in ingroup and outgroup homogeneity. Journal of Personality and Social Psychology, 73, 1206 –1212. Sachdev, I., & Bourhis, R. (1987). Status differentials and intergroup behavior. European Journal of Social Psychology, 17, 277–293. Sachdev, I., & Bourhis, R. (1991). Power and status differentials in minority and majority group relations. European Journal of Social Psychology, 21, 1–24. Schneider, M. E., Major, B., Luhtanen, R., & Crocker, J. (1996). Social

stigma and the potential costs of assumptive help. Personality and Social Psychology Bulletin, 22, 201–209. Schroeder, D. A., Penner, L. A., Dovidio, J. F., & Piliavin, J. A. (1995). The psychology of helping and altruism: Problems and puzzles. New York: McGraw-Hill. Steele, C. (1992, April). Race and the schooling of Black Americans. Atlantic Monthly, 68 –78. Suleiman, R. (2004). The planned meeting as a microcosmos. In R. Halabi (Ed.), Israeli and Palestinian identities in dialogue (pp. 47– 67). Piscataway, NJ: Rutgers University Press. Tajfel, H. (1978). Interindividual behavior and intergroup behavior. In H. Tajfel (Ed.), Differentiations between social groups: Studies in the social psychology of intergroup relations (pp. 27– 45). New York: Academic Press. Tajfel, H., & Turner, J. C. (1979). An integrative theory of intergroup conflict. In W. G. Austin & S. Worchel (Eds.), The social psychology of intergroup relations (pp. 33– 47). Monterey, CA: Brooks/Cole. Tajfel, H., & Turner, J. C. (1986). The social identity theory of intergroup behavior. In W. G. Austin & S. Worchel (Eds.), The social psychology of intergroup relations (pp. 7–24). Chicago: Nelson-Hall. Turner, J. C., & Brown, R. (1978). Social status, cognitive alternatives and intergroup relations. In H. Tajfel (Ed.), Differentiation between social groups: Studies in the social psychology of intergroup relations (pp. 201–234). London: Academic Press. Turner, J. C., Hogg, M. A., Oakes, P. J., Reicher, S. D., & Wetherell, M. S. (1987). Rediscovering the social group: A self-categorization theory. Oxford, England: Blackwell. Turner, J. C., & Reynolds, K. J. (2001). The social identity perspective in inter-group relations: Theories, themes and controversies. In R. Brown & S. Gaertner (Eds.), Intergroup processes (pp. 133–153). Oxford, England: Blackwell Van der Berghe, P. L. (1967). Race and racism. New York: Wiley. Worchel, S. (1984). The darker side of helping: The social dynamics of helping and cooperation. In D. Bar-Tal, J. Karylowski, & J. Reykowski (Eds.), The development and maintenance of prosocial behavior (pp. 379 –397). New York: Plenum. Zahavi, A., & Zahavi, A. (1997). The handicap principle: A missing piece of Darwin’s puzzle. Oxford, England: Oxford University Press.

Received April 26, 2005 Revision received October 29, 2005 Accepted November 4, 2005 䡲